Academia.eduAcademia.edu

Picosecond Photoresponse in Polycrystalline Silicon

Profile image of Thomas HsiangThomas Hsiang

Abstract

Cover art courtesy of Gerhard David, University of Michigan: Nonlinear transmission line consisting of 10 Schottky diodes in a coplanar waveguide; (top left) metallization structure; results of 2D electro-optic field mappings of (top right) the fundamental input frequency (6 GHz), (bottom left) the generated second harmonic (12 GHz) and (bottom right) the generated third harmonic (18 GHz).

Figures (381)
T. A. Gardiner, Ju. V. Vandyshev, G. W. Wicks, and P. M. Fauchet
T. A. Gardiner, Ju. V. Vandyshev, G. W. Wicks, and P. M. Fauchet
controlled Ti:Sapphire oscillator, which provides nearly transform limited 8-fs pulses. These pulses were amplified at a repetition rate of 1] kHz in a multipass amplifier pumped by the second harmonic of a Q-switched Nd: YLF laser [l]. The transform-limited output pulses have a duration of 20 fs, energy up to 300 uJ and a spectrum centered at 780 nm. The amplified pulses were coupled into a 160-zm-diameter. 60-cm-long fused-silica hollow fiber. The fiber was kept straight in a V-groove made in an aluminium bar which was placed in a pressurized chamber with fused-silica windows (1 mm thick) coated for broadband antireflection. The hollow fiber was filled with krypton at different pressures. By properly matching the input beam to the EH,,; mode of the fiber. an overall fiber transmission of 65% was measured. which is close to the value (~ 73 %) predicted by the theory. Figure 1. Spectral broadening in krypton at 2.1 bar and input peak power Pp= 2 GW. The spectrum of the imput pulses is shown as dashed curve.
controlled Ti:Sapphire oscillator, which provides nearly transform limited 8-fs pulses. These pulses were amplified at a repetition rate of 1] kHz in a multipass amplifier pumped by the second harmonic of a Q-switched Nd: YLF laser [l]. The transform-limited output pulses have a duration of 20 fs, energy up to 300 uJ and a spectrum centered at 780 nm. The amplified pulses were coupled into a 160-zm-diameter. 60-cm-long fused-silica hollow fiber. The fiber was kept straight in a V-groove made in an aluminium bar which was placed in a pressurized chamber with fused-silica windows (1 mm thick) coated for broadband antireflection. The hollow fiber was filled with krypton at different pressures. By properly matching the input beam to the EH,,; mode of the fiber. an overall fiber transmission of 65% was measured. which is close to the value (~ 73 %) predicted by the theory. Figure 1. Spectral broadening in krypton at 2.1 bar and input peak power Pp= 2 GW. The spectrum of the imput pulses is shown as dashed curve.
Figure 2. Group delay T, vs. frequency: (i) (dotted line) desired Ty variation, (ii) (solid line) optimized variation of T, of the entire system from the output of the waveguide to the second harmonic crystal in the autocorrelator, (iii) (dashed line) the same as above except that the chirped mirror was removed. The frequency broadened pulses emerging from the hollow fiber were collimated by a silver-coated spherical mirror and propagated through a dispersive delay line which introduced an appropriate group delay 7,= dd¢/dw (¢(@) is the phase retardation). To compress the frequency broadened pulses to their transform limit, 7, must be precisely controlled over a bandwidth exceeding 130 THz.
Figure 2. Group delay T, vs. frequency: (i) (dotted line) desired Ty variation, (ii) (solid line) optimized variation of T, of the entire system from the output of the waveguide to the second harmonic crystal in the autocorrelator, (iii) (dashed line) the same as above except that the chirped mirror was removed. The frequency broadened pulses emerging from the hollow fiber were collimated by a silver-coated spherical mirror and propagated through a dispersive delay line which introduced an appropriate group delay 7,= dd¢/dw (¢(@) is the phase retardation). To compress the frequency broadened pulses to their transform limit, 7, must be precisely controlled over a bandwidth exceeding 130 THz.
Figure 3. Measured (solid line) and calculated (dots) interferometric autocorrelation trace of the compressed pulses, and evaluation of the pulse duration is also given. deviates from a linear behavior due mainly to cubic dispersion in the nonlinear medium. To realise the ideal 7, a delay line was constructed, consisting of two pairs of fused silica prisms of small apex angle (20°) and a chirped mirror. specifically designed to introduce negative group delay dispersion (GDD) in combination with positive cubic and quartic dispersion. The use of thin prisms instead of Brewster-angle-prisms determines a substantial reduction of the positive material GDD in the system. As a result. less negative GDD is required and high-order phase errors (mainly cubic dispersion) of the prisms can be reduced in proportion to the reduced negative GDD. The variation of 7, with frequency of the entire system between the output of the fiber and the f-BaB.O, crystal in the autocorrelator (prism separation 1.8 m, total propagation in fused silica: 15 mm. total propagation length in air 6 m) is shown as solid line in Fig.2. The curve fits the required delay variation well over a bandwidth of 120 THz. The dispersive delay line has a high throughput (785%) over the wavelength range of 630-1030 nm. The combination of low loss and appropriate dispersion control over a relative bandwidth Aw/w,> 0.3 offers the potential for bandwidth-limited single-cycle optical _ pulse generation.
Figure 3. Measured (solid line) and calculated (dots) interferometric autocorrelation trace of the compressed pulses, and evaluation of the pulse duration is also given. deviates from a linear behavior due mainly to cubic dispersion in the nonlinear medium. To realise the ideal 7, a delay line was constructed, consisting of two pairs of fused silica prisms of small apex angle (20°) and a chirped mirror. specifically designed to introduce negative group delay dispersion (GDD) in combination with positive cubic and quartic dispersion. The use of thin prisms instead of Brewster-angle-prisms determines a substantial reduction of the positive material GDD in the system. As a result. less negative GDD is required and high-order phase errors (mainly cubic dispersion) of the prisms can be reduced in proportion to the reduced negative GDD. The variation of 7, with frequency of the entire system between the output of the fiber and the f-BaB.O, crystal in the autocorrelator (prism separation 1.8 m, total propagation in fused silica: 15 mm. total propagation length in air 6 m) is shown as solid line in Fig.2. The curve fits the required delay variation well over a bandwidth of 120 THz. The dispersive delay line has a high throughput (785%) over the wavelength range of 630-1030 nm. The combination of low loss and appropriate dispersion control over a relative bandwidth Aw/w,> 0.3 offers the potential for bandwidth-limited single-cycle optical _ pulse generation.
Figure 4. Measured full width at half maximum of the output beam (triangle) as a function of the distance from the tip of fiber. calculated values (dots) from free space propagation of a beam with an initial profile equal to the EH), mode of the fiber. Increasing the input peak power to 4 GW and decreasing the pressure to 1.1 bar in order to maintain constant the Kerr non-linearity, slightly longer pulses (5.3 fs) with twice as much energy (40 uJ) were obtained.
Figure 4. Measured full width at half maximum of the output beam (triangle) as a function of the distance from the tip of fiber. calculated values (dots) from free space propagation of a beam with an initial profile equal to the EH), mode of the fiber. Increasing the input peak power to 4 GW and decreasing the pressure to 1.1 bar in order to maintain constant the Kerr non-linearity, slightly longer pulses (5.3 fs) with twice as much energy (40 uJ) were obtained.
The "Soliton" Laser and Polarization Additive Pulse Mode-Locking Figure 1. Schematic of soliton laser using polarization additive pulse mode-locking; insets are the SHG trace and the oscilloscope of pulse train. by spontaneous emission. This is a unique example of a macroscopic laser whose noise is attributable to spontaneous emission (as opposed to the fundamentally noisy microscopic laser diodes). Fiber ring lasers are self-starting, they do not require movable etalons as triggers. The reasons for the self- starting behavior are presented. Finally, we describe an environmentally stable version of the stretched pulse ring laser. Figure 1 shows the schematic of one realization of a fiber ring laser. The erbium fiber is diode pumped at 980 nm wavelength via a dichroic coupler. An isolator serves to suppress the counterclockwise traveling wave and also serves as a_polarizer- analyzer. "Rabbit-ear" polarization controllers are positioned on both sides of the isolator to provide effective saturable absorber action via nonlinear polarization rotation. The effective saturable absorber action is illustrated in Fig. 2 which shows schematically how the linear polarization of a mode is transformed into elliptic polarization by a waveplate, rotated via the Kerr effect in a nonlinear medium, and then passed through an analyzer. The nonlinear induced polarization R of an isotropic Kerr medium when written in the circular polarization basis is
The "Soliton" Laser and Polarization Additive Pulse Mode-Locking Figure 1. Schematic of soliton laser using polarization additive pulse mode-locking; insets are the SHG trace and the oscilloscope of pulse train. by spontaneous emission. This is a unique example of a macroscopic laser whose noise is attributable to spontaneous emission (as opposed to the fundamentally noisy microscopic laser diodes). Fiber ring lasers are self-starting, they do not require movable etalons as triggers. The reasons for the self- starting behavior are presented. Finally, we describe an environmentally stable version of the stretched pulse ring laser. Figure 1 shows the schematic of one realization of a fiber ring laser. The erbium fiber is diode pumped at 980 nm wavelength via a dichroic coupler. An isolator serves to suppress the counterclockwise traveling wave and also serves as a_polarizer- analyzer. "Rabbit-ear" polarization controllers are positioned on both sides of the isolator to provide effective saturable absorber action via nonlinear polarization rotation. The effective saturable absorber action is illustrated in Fig. 2 which shows schematically how the linear polarization of a mode is transformed into elliptic polarization by a waveplate, rotated via the Kerr effect in a nonlinear medium, and then passed through an analyzer. The nonlinear induced polarization R of an isotropic Kerr medium when written in the circular polarization basis is
Figure 2. Effective saturable absorber action via polarization rotation in Kerr medium.
Figure 2. Effective saturable absorber action via polarization rotation in Kerr medium.
Figure 4. The phase matching condition between the soliton and the linear continuum.
Figure 4. The phase matching condition between the soliton and the linear continuum.
Figure 3. The spectrum of the pulse (a) without, and (b) with filtering. the excitation is phase matched. The spectrum of the pulse in the resonator clearly shows such resonant sidebands. As a practical matter, they can be greatly suppressed by introducing amplitude filtering with a
Figure 3. The spectrum of the pulse (a) without, and (b) with filtering. the excitation is phase matched. The spectrum of the pulse in the resonator clearly shows such resonant sidebands. As a practical matter, they can be greatly suppressed by introducing amplitude filtering with a
Figure 5. Schematic of fiber ring with positive and negative GVD. The soliton laser self-limits when the nonlinear phase shift approaches a fraction of 27. The stretched pulse laser greatly extends the energy at which self-limiting occurs. Figure 5 shows a schematic of a fiber ring composed of two pieces of equal and opposite GVD. A short pulse circulating in this ring alternately stretches and compresses as indicated. The degree of stretching and compression increases with decreasing pulse width, increasing pulse bandwidth. Thus, the shorter the pulse circulating in the ring, the stronger the effect. For a given peak intensity at the position of the shortest pulse, the net accumulation of
Figure 5. Schematic of fiber ring with positive and negative GVD. The soliton laser self-limits when the nonlinear phase shift approaches a fraction of 27. The stretched pulse laser greatly extends the energy at which self-limiting occurs. Figure 5 shows a schematic of a fiber ring composed of two pieces of equal and opposite GVD. A short pulse circulating in this ring alternately stretches and compresses as indicated. The degree of stretching and compression increases with decreasing pulse width, increasing pulse bandwidth. Thus, the shorter the pulse circulating in the ring, the stronger the effect. For a given peak intensity at the position of the shortest pulse, the net accumulation of
Figure 6. Schematic of stretched pulse fiber ring laser and the SHG traces obtained for different lengths of output fiber. Figure 6 shows a schematic of the stretched pulse ring laser and the experimentally observed SHG traces for different lengths of the output fiber of negative GVD. Note that the output follows the positively dispersive erbium fiber. The pulse is chirped at the output coupler, and, at first, the output fiber compresses the pulse. After the pulse has been compressed to its transform-limited width, further propagation in the fiber again expands the pulse.
Figure 6. Schematic of stretched pulse fiber ring laser and the SHG traces obtained for different lengths of output fiber. Figure 6 shows a schematic of the stretched pulse ring laser and the experimentally observed SHG traces for different lengths of the output fiber of negative GVD. Note that the output follows the positively dispersive erbium fiber. The pulse is chirped at the output coupler, and, at first, the output fiber compresses the pulse. After the pulse has been compressed to its transform-limited width, further propagation in the fiber again expands the pulse.
Figure 7. Comparison of spectra of (a) soliton laser and (b) stretched pulse laser. The Noise Von der Linde [12] has pioneered a simple measurement of the amplitude fluctuations and timing jitter of a modelocked pulse train. The pulse train is detected and the RF spectrum of the detector current is recorded. The spectral components are
Figure 7. Comparison of spectra of (a) soliton laser and (b) stretched pulse laser. The Noise Von der Linde [12] has pioneered a simple measurement of the amplitude fluctuations and timing jitter of a modelocked pulse train. The pulse train is detected and the RF spectrum of the detector current is recorded. The spectral components are
Figure 9. Spectrum of first and 35" components of RF spectrum of soliton laser. Figure 8. The general characteristics of the contributions of amplitude noise and timing jitter to the RF spectrum of detector current.
Figure 9. Spectrum of first and 35" components of RF spectrum of soliton laser. Figure 8. The general characteristics of the contributions of amplitude noise and timing jitter to the RF spectrum of detector current.
Self-starting spaced by Af =27/T, apart. The noise spectrum around each spectral component consists of a contribution due to the amplitude fluctuations that is independent of the order of the spectral component (the darkened portions in Fig. 8), and of a contribution of the timing jitter, that increases in
Self-starting spaced by Af =27/T, apart. The noise spectrum around each spectral component consists of a contribution due to the amplitude fluctuations that is independent of the order of the spectral component (the darkened portions in Fig. 8), and of a contribution of the timing jitter, that increases in
Figure 10. The stretched pulse sigma laser. version of the stretched pulse laser has been commercialized by Clark-MXR, Inc. Yet, in order to achieve the ultimate in environmental stability, the use of Polarization Maintaining (PM) fiber would be desirable. Unfortunately, the simple P-APM action based on the rotation of the polarization ellipse cannot be employed in such fibers. We have attempted to achieve P-APM action by interfering two Kerr-shifted polarization eigenmodes. These attempts failed on account of environmentally induced changes in the birefringence of the fiber. Eventually, a Sigma Laser similar to the one worked on by Duling et al. [14] proved successful. Our laser [15] differs from Duling’s in using different GVD’s in order to achieve stretched pulse operation (Fig. 10).
Figure 10. The stretched pulse sigma laser. version of the stretched pulse laser has been commercialized by Clark-MXR, Inc. Yet, in order to achieve the ultimate in environmental stability, the use of Polarization Maintaining (PM) fiber would be desirable. Unfortunately, the simple P-APM action based on the rotation of the polarization ellipse cannot be employed in such fibers. We have attempted to achieve P-APM action by interfering two Kerr-shifted polarization eigenmodes. These attempts failed on account of environmentally induced changes in the birefringence of the fiber. Eventually, a Sigma Laser similar to the one worked on by Duling et al. [14] proved successful. Our laser [15] differs from Duling’s in using different GVD’s in order to achieve stretched pulse operation (Fig. 10).
Figure 11. The SHG traces and spectrum of the pulse from the sigma laser.
Figure 11. The SHG traces and spectrum of the pulse from the sigma laser.
radius of curvature high reflecting folding mirror provides astigmatic compensation and a 7.5 cm radius of curvature 0.2% output coupler focuses the cavity mode to a spot size of about 35 pm at the surface of the flat SBR. The structure of the SBR, (Figure 1) consists of a 99.5% reflecting Bragg mirror of alternating quarter-wave layers of GaAs and AlAs andtwo uncoupled Ing. 52 Gag,4g As/InP quantum wells located 15 nm from the top surface of a half-wave strain relief layer grown on the final layer of the Bragg mirror.[4-8] The excitonic absorption of the quantum wells is centered near 1500 nm. The output of a diode pumped Nd: YVQi CW laser (Spectra Physics) at
radius of curvature high reflecting folding mirror provides astigmatic compensation and a 7.5 cm radius of curvature 0.2% output coupler focuses the cavity mode to a spot size of about 35 pm at the surface of the flat SBR. The structure of the SBR, (Figure 1) consists of a 99.5% reflecting Bragg mirror of alternating quarter-wave layers of GaAs and AlAs andtwo uncoupled Ing. 52 Gag,4g As/InP quantum wells located 15 nm from the top surface of a half-wave strain relief layer grown on the final layer of the Bragg mirror.[4-8] The excitonic absorption of the quantum wells is centered near 1500 nm. The output of a diode pumped Nd: YVQi CW laser (Spectra Physics) at
Figure 2. Autocorrelation and optical spectrum of the modelocked output. The 2.7 GHz pulsetrain is shown in the inset.
Figure 2. Autocorrelation and optical spectrum of the modelocked output. The 2.7 GHz pulsetrain is shown in the inset.
Figure 3. Plot of the pulsewidth (open markers, FWHM) and time-bandwidth product (solid markers) of the modelocked output with either one (circle), two (square) or three (triangle) pulses circulating in the cavity versus total cavity GVD. Bistable regions are indicated with dashed vertical lines. Simple CW cavity alignment produces self- starting modelocking of the laser believed to be initiated by the ultrafast saturation dynamics of the SBR.[6-8] With approximately 7 W of incident pump power, 200 fs pulses (FWHM assuming a sech2 pulseshape) are produced (see Figure 2). Due to the low cavity GVD and nonlinearities
Figure 3. Plot of the pulsewidth (open markers, FWHM) and time-bandwidth product (solid markers) of the modelocked output with either one (circle), two (square) or three (triangle) pulses circulating in the cavity versus total cavity GVD. Bistable regions are indicated with dashed vertical lines. Simple CW cavity alignment produces self- starting modelocking of the laser believed to be initiated by the ultrafast saturation dynamics of the SBR.[6-8] With approximately 7 W of incident pump power, 200 fs pulses (FWHM assuming a sech2 pulseshape) are produced (see Figure 2). Due to the low cavity GVD and nonlinearities
Figure 5. RF spectrum of the laser output detected by a fast photodiode illustrating at least 60 dB suppression of cavity harmonics. The unequal magnitudes of the peaks is due the response of an RF amplifier. References Figure 4. a) Interferometric cross-correlation trace of temporally adjacent and equally spaced multiple pulses. b) Autocorrelation trace with laser operating as in (a). c) Interferometric cross-correlation of adjacent multiple pulses with unequal temporal spacings. d) Same as (c) but with different relative pulse spacings. In all cases, there are three pulses in the cavity.
Figure 5. RF spectrum of the laser output detected by a fast photodiode illustrating at least 60 dB suppression of cavity harmonics. The unequal magnitudes of the peaks is due the response of an RF amplifier. References Figure 4. a) Interferometric cross-correlation trace of temporally adjacent and equally spaced multiple pulses. b) Autocorrelation trace with laser operating as in (a). c) Interferometric cross-correlation of adjacent multiple pulses with unequal temporal spacings. d) Same as (c) but with different relative pulse spacings. In all cases, there are three pulses in the cavity.
Figure 6. Relative optical phase relationship between adjacent multiple pulses versus time measured using a fixed delay cross-correlator. The relative phase is constant with time and returns to the previous relative phase relationship after a disruption of the laser's operation.
Figure 6. Relative optical phase relationship between adjacent multiple pulses versus time measured using a fixed delay cross-correlator. The relative phase is constant with time and returns to the previous relative phase relationship after a disruption of the laser's operation.
Figure 1. Experimental setup for intracavity gain dynamic Measurement. MQW: Multiple quantum well saturable absorber, PBS: Polarizing beam splitter, SLA: Semiconductor laser amplifier, OC: Output coupler, G: Diffraction grating, S: Slit, I: Optical isolator, P: Polarizer, WP: Half wave plate. The experimental set up is shown in Fig. 1. dynamics in hybrid modelocked semiconductor diode lasers. To support these measurements, intracavity intensity pulse profiles and their corresponding spectrograms were measured to provide direct experimental observation of the effects suggested by the intracavity dynamics.
Figure 1. Experimental setup for intracavity gain dynamic Measurement. MQW: Multiple quantum well saturable absorber, PBS: Polarizing beam splitter, SLA: Semiconductor laser amplifier, OC: Output coupler, G: Diffraction grating, S: Slit, I: Optical isolator, P: Polarizer, WP: Half wave plate. The experimental set up is shown in Fig. 1. dynamics in hybrid modelocked semiconductor diode lasers. To support these measurements, intracavity intensity pulse profiles and their corresponding spectrograms were measured to provide direct experimental observation of the effects suggested by the intracavity dynamics.
Figure 2. Time resolved intracavity dynamic measurements: (a) SLA gain, (b) transmittance of saturable absorber.
Figure 2. Time resolved intracavity dynamic measurements: (a) SLA gain, (b) transmittance of saturable absorber.
Figure 3. Temporal shape of pulses: (a) before saturable absorber. (b) after saturable absorber.
Figure 3. Temporal shape of pulses: (a) before saturable absorber. (b) after saturable absorber.
Figure 4. Spectrally resolved cross correlation at different locations in cavity: (a) laser output. (b) before saturable absorber and (c) after saturable absorber. wavelength variation of 0.8 nm at the second harmonic wavelength, implying a total wavelength chirp of 1.6 nm. at the fundamental wavelength. It should be noted that the chirp exists over the duration of the pulse, implying a nonlinear dispersion of ~ 5 ps/nm.
Figure 4. Spectrally resolved cross correlation at different locations in cavity: (a) laser output. (b) before saturable absorber and (c) after saturable absorber. wavelength variation of 0.8 nm at the second harmonic wavelength, implying a total wavelength chirp of 1.6 nm. at the fundamental wavelength. It should be noted that the chirp exists over the duration of the pulse, implying a nonlinear dispersion of ~ 5 ps/nm.
Figure 5. Intensity profile and instantaneous frequency of pulses: (a) initial symmetric pulse, (b) after dispersive media and (c} after saturable absorber. broadening that would be expected for an instantaneous nonlinearity. In addition, note that the instantaneous frequency is also modified by the pulse propagation through the dispersive media. In Fig. 5(c), is the resultant optical pulse and instantaneous frequency after the optical pulse has passed through the saturable absorber. Note the faster rising edge of the pulse, as compared to Fig. 5(b), and also note that the chirp is now predominantly linear over the main portion of the optical pulse. Thus, a simple model of SPM, GVD and saturable absorption can provide a clearer understanding of the pulse shaping and chirping dynamics of hybrid modelocked diode lasers.
Figure 5. Intensity profile and instantaneous frequency of pulses: (a) initial symmetric pulse, (b) after dispersive media and (c} after saturable absorber. broadening that would be expected for an instantaneous nonlinearity. In addition, note that the instantaneous frequency is also modified by the pulse propagation through the dispersive media. In Fig. 5(c), is the resultant optical pulse and instantaneous frequency after the optical pulse has passed through the saturable absorber. Note the faster rising edge of the pulse, as compared to Fig. 5(b), and also note that the chirp is now predominantly linear over the main portion of the optical pulse. Thus, a simple model of SPM, GVD and saturable absorption can provide a clearer understanding of the pulse shaping and chirping dynamics of hybrid modelocked diode lasers.
Figure 1. Photocurrent spectrum in the vicinity of the beat frequency that results when the laser output is direct detected with a fast photodiode.
Figure 1. Photocurrent spectrum in the vicinity of the beat frequency that results when the laser output is direct detected with a fast photodiode.
Figure 2. (a) Optical spectrum and (b) autocorrelation trace of the dual-frequency Er/Yb glass laser. where a7 is the effective coefficient of thermal expan- sion for the laser cavity. For example, for an invar cavity ap ~ 107%, which gives AQ/Q ~ 107°, or AQ = 60 Hz for 2 = 60 GHz, if the cavity temperature is stabilized to 10-3 K. Note however that the above estimate ig- nores the temperature fluctuations of the gain medium that are induced by the power fluctuations of the pump laser (diode-pumped Nd:YLF laser in our case), Such fluctuations may set a lower limit on the potentially achievable beat-frequency stability.
Figure 2. (a) Optical spectrum and (b) autocorrelation trace of the dual-frequency Er/Yb glass laser. where a7 is the effective coefficient of thermal expan- sion for the laser cavity. For example, for an invar cavity ap ~ 107%, which gives AQ/Q ~ 107°, or AQ = 60 Hz for 2 = 60 GHz, if the cavity temperature is stabilized to 10-3 K. Note however that the above estimate ig- nores the temperature fluctuations of the gain medium that are induced by the power fluctuations of the pump laser (diode-pumped Nd:YLF laser in our case), Such fluctuations may set a lower limit on the potentially achievable beat-frequency stability.
Figure 3. | Experimental setup. The phase modulator was driven at 100 MHz.
Figure 3. | Experimental setup. The phase modulator was driven at 100 MHz.
Figure 4. | Comb-like DTF design showing the distance de- pendence of the dispersion and nonlinear coefficients. The lengths of the DSF segments were 2, 1.2, and 0.4km, re- spectively; and those of the STF segments were 0.6, 0.1, and 0.1km, respectively. 5.6 ps, which gives a pulsewidth of 3.2ps, assuming a sech? pulse-intensity shape for the evolved pulse. Also, by fitting a sech? envelope to the peaks in Fig. 5(a), the FWHM of the evolved spectrum is determined to be 0.77nm, which gives a time-bandwidth product of 0.3140.02 for the soliton-like pulse train. Furthermore, since the repetition period is 17.1 ps, a mark-to-space ratio of > 5 is obtained. The mark-to-space ratio can be further increased by elongating the DTF with use of a larger number of paired segments (DSF followed by STF of appropriate length) in the DTF, as was the case in Refs. 7 and 8. Similarly, increasing the power launched into the DTF beyond the 120 mW used in our experiment by employing a higher-power optical ampli- fier would result in a larger mark-to-space ratio.
Figure 4. | Comb-like DTF design showing the distance de- pendence of the dispersion and nonlinear coefficients. The lengths of the DSF segments were 2, 1.2, and 0.4km, re- spectively; and those of the STF segments were 0.6, 0.1, and 0.1km, respectively. 5.6 ps, which gives a pulsewidth of 3.2ps, assuming a sech? pulse-intensity shape for the evolved pulse. Also, by fitting a sech? envelope to the peaks in Fig. 5(a), the FWHM of the evolved spectrum is determined to be 0.77nm, which gives a time-bandwidth product of 0.3140.02 for the soliton-like pulse train. Furthermore, since the repetition period is 17.1 ps, a mark-to-space ratio of > 5 is obtained. The mark-to-space ratio can be further increased by elongating the DTF with use of a larger number of paired segments (DSF followed by STF of appropriate length) in the DTF, as was the case in Refs. 7 and 8. Similarly, increasing the power launched into the DTF beyond the 120 mW used in our experiment by employing a higher-power optical ampli- fier would result in a larger mark-to-space ratio.
Figure 5. (a) Optical spectrum and (b) autocorrelation trace of the pulse train at the output of the DTF. In (b), the dashed curve is a theoretical fit and the time scale on the abscissa is normalized to the pulse repetition period of 17.1 ps. Also in (b) the thin solid curve is the theoretical autocorrelation trace for a two-frequency laser source. oscillators that are commonly employed in the telecom- munication networks of today.
Figure 5. (a) Optical spectrum and (b) autocorrelation trace of the pulse train at the output of the DTF. In (b), the dashed curve is a theoretical fit and the time scale on the abscissa is normalized to the pulse repetition period of 17.1 ps. Also in (b) the thin solid curve is the theoretical autocorrelation trace for a two-frequency laser source. oscillators that are commonly employed in the telecom- munication networks of today.
HLAN 1s a_trame-based architecture, which is implemented on a unidirectional bus as shown in Figure 1. A headend generates frames of empty slots and puts them on the bus. Guaranteed bandwidth (GBW) traffic is transmitted on the GBW segment, bandwidth on demand (BOD) traffic is transmitted on the BOD segment and data is received on the RCV segment. Note that only one bus is required and that all receivers are downstream of the transmitters. Figure 1. HLAN topology
HLAN 1s a_trame-based architecture, which is implemented on a unidirectional bus as shown in Figure 1. A headend generates frames of empty slots and puts them on the bus. Guaranteed bandwidth (GBW) traffic is transmitted on the GBW segment, bandwidth on demand (BOD) traffic is transmitted on the BOD segment and data is received on the RCV segment. Note that only one bus is required and that all receivers are downstream of the transmitters. Figure 1. HLAN topology
Figure 2 shows a block diagram of some of the hardware components needed to implement this high-speed, yet simple ne block diagram shows t receiver node, but transmitter node are a components shown in th interface between the op the The the work architecture. he components in components in the most identical. The e shaded boxes, at the ical bus, and the node electronics, are high performance optical devices that mniet onerate at the on ical data rate. Figure 2. Block diagram of components in a user receiver node.
Figure 2 shows a block diagram of some of the hardware components needed to implement this high-speed, yet simple ne block diagram shows t receiver node, but transmitter node are a components shown in th interface between the op the The the work architecture. he components in components in the most identical. The e shaded boxes, at the ical bus, and the node electronics, are high performance optical devices that mniet onerate at the on ical data rate. Figure 2. Block diagram of components in a user receiver node.
Figure 1 Schematic illustration of bandwidth allocation in TDM, WDM, and CDMA optical networks.
Figure 1 Schematic illustration of bandwidth allocation in TDM, WDM, and CDMA optical networks.
Figure 4 Intensity cross-correlation measurements of input and output pulses from 2.5 km dispersion compensated fiber link. Left: ~250-fs input pulses. Right: ~470-fs input pulses.
Figure 4 Intensity cross-correlation measurements of input and output pulses from 2.5 km dispersion compensated fiber link. Left: ~250-fs input pulses. Right: ~470-fs input pulses.
Figure 1. Cascadeable WDM _ local access communications system, consisting of bit-interleaved WDM _ light source, sequentially emitting N wavelength channels, a cascadable distribution fabric consisting of amplifier/power-split units, and a single data-encoding TDM modulator for each WDM-PON. Bit- interleaved WDM data streams are shown at reference points © and © entering the distribution stage and after the modulator, respectively.
Figure 1. Cascadeable WDM _ local access communications system, consisting of bit-interleaved WDM _ light source, sequentially emitting N wavelength channels, a cascadable distribution fabric consisting of amplifier/power-split units, and a single data-encoding TDM modulator for each WDM-PON. Bit- interleaved WDM data streams are shown at reference points © and © entering the distribution stage and after the modulator, respectively.
Figure 2. Spectra of channel 8 transmitting PRBS, without (a) and with (b) distribution stage. Upper traces show spectra after modulator; lower traces show spectra after router port 8. Resolution bandwidth is 0.1 nm. BER vs. received power (c) without (dashed) and with (solid) distribution stage. In the first such measurement, a PRBS is transmitted on channel 8, at 1563.7 nm, while all other channels are blanked. The transmission spectra, Fig. 2, show the light transmitted into the distribution fiber (upper trace) and that received after router port 8 (lower trace) without (Fig. 2a) and with (Fig. 2b) amplifier/power-split distribution stage, respectively. The peaked output spectrum results from nonuniform gain in the optical amplifier. The BER vs. received power, Fig. 2c, shows that the received power
Figure 2. Spectra of channel 8 transmitting PRBS, without (a) and with (b) distribution stage. Upper traces show spectra after modulator; lower traces show spectra after router port 8. Resolution bandwidth is 0.1 nm. BER vs. received power (c) without (dashed) and with (solid) distribution stage. In the first such measurement, a PRBS is transmitted on channel 8, at 1563.7 nm, while all other channels are blanked. The transmission spectra, Fig. 2, show the light transmitted into the distribution fiber (upper trace) and that received after router port 8 (lower trace) without (Fig. 2a) and with (Fig. 2b) amplifier/power-split distribution stage, respectively. The peaked output spectrum results from nonuniform gain in the optical amplifier. The BER vs. received power, Fig. 2c, shows that the received power
Figure 3. Spectra of channels 1-15 transmitting, without (a) and with (b) distribution stage. Upper traces show spectra after modulator; lower traces show spectra after router port 8. Resolution bandwidth 0.i)nm. BER vs. received power (c) without (dashed) and with (solid) distribution stage.
Figure 3. Spectra of channels 1-15 transmitting, without (a) and with (b) distribution stage. Upper traces show spectra after modulator; lower traces show spectra after router port 8. Resolution bandwidth 0.i)nm. BER vs. received power (c) without (dashed) and with (solid) distribution stage.
Figure 1 schematically shows the setup of the four optical pu wavelength laser. Actively modelocked ses are generated from the laser diode by incorporating an intracavity spectral filter to define he individual spectral components. The end mirror M3 reflec 9ack to s the four selected spectral components he gain device to complete the four wavelength channel generation. A colinear composite four-wave ength pulse train is coupled out from the zeroth order grating reflection. Modelocking is made oy injecting ~1W rf sinusoidal signal at a frequency of 2.5GHz with 175mA of dc bias current into the liode chip. The position of the diode in the cavity is chosen to enhance the eighth harmonic modelocking frequency, ).5GHz. which in this case, is designed to be Figure 2 illustrates the output four wavelength spectra in active modelocking (upper) and in cw operation (lower). Stable four wavelength lasing was demonstrated at a pulse repetition rate of 600 MHz, 2.1 GHz, and 2.5 GHz. In comparison,
Figure 1 schematically shows the setup of the four optical pu wavelength laser. Actively modelocked ses are generated from the laser diode by incorporating an intracavity spectral filter to define he individual spectral components. The end mirror M3 reflec 9ack to s the four selected spectral components he gain device to complete the four wavelength channel generation. A colinear composite four-wave ength pulse train is coupled out from the zeroth order grating reflection. Modelocking is made oy injecting ~1W rf sinusoidal signal at a frequency of 2.5GHz with 175mA of dc bias current into the liode chip. The position of the diode in the cavity is chosen to enhance the eighth harmonic modelocking frequency, ).5GHz. which in this case, is designed to be Figure 2 illustrates the output four wavelength spectra in active modelocking (upper) and in cw operation (lower). Stable four wavelength lasing was demonstrated at a pulse repetition rate of 600 MHz, 2.1 GHz, and 2.5 GHz. In comparison,
Figure 2. Output spectra of four wavelengths.
Figure 2. Output spectra of four wavelengths.
igure 3. Autocorrelation traces of output pulses at each wavelength channel and composite four- wavelength channel.
igure 3. Autocorrelation traces of output pulses at each wavelength channel and composite four- wavelength channel.
Figure 4. Tuning of the four wavelength spectra. Figure 4(a) shows the of the composite four waveleng' spectral separation of ~Inm, an bandwidth of the SLA (>20nm) that as many as 19 wavelengt tuning characteristic h output, with a fixed d tuned over the gain . Figure 4(a) suggests hs separated by Inm could be supported from this device. In Figure 4(b), the tuning characteristic is individual wavelength separa ~0.8nm to ~2nm. Notice tha shown where the ion is -varied from in both cases, the spectral intensities decrease with large detuning, showing the influence of the gain spectrum of the SLA.
Figure 4. Tuning of the four wavelength spectra. Figure 4(a) shows the of the composite four waveleng' spectral separation of ~Inm, an bandwidth of the SLA (>20nm) that as many as 19 wavelengt tuning characteristic h output, with a fixed d tuned over the gain . Figure 4(a) suggests hs separated by Inm could be supported from this device. In Figure 4(b), the tuning characteristic is individual wavelength separa ~0.8nm to ~2nm. Notice tha shown where the ion is -varied from in both cases, the spectral intensities decrease with large detuning, showing the influence of the gain spectrum of the SLA.
Figure 7. Interferometric autocorrelation of temporal beat signal of two wavelengths. in order to obtain a Clearer understanding Of the possible spectral phase correlation, the multiwavelength laser was configured to operate with only two wavelengths. This would imply that the temporal beat signal would be sinusoidal. We then measured the interferometric autocorrelation, which is shown in Fig. 7, with the corresponding spectrum as the inset.
Figure 7. Interferometric autocorrelation of temporal beat signal of two wavelengths. in order to obtain a Clearer understanding Of the possible spectral phase correlation, the multiwavelength laser was configured to operate with only two wavelengths. This would imply that the temporal beat signal would be sinusoidal. We then measured the interferometric autocorrelation, which is shown in Fig. 7, with the corresponding spectrum as the inset.
Figure 1 shows a schematic of our transmitter. Fig. 1 Schematic of the 206-channel chirped-pulse WDM transmitter. Transmitter Set-Up In this paper, we demonstrate that CPWDM can be scaled to a record 206 WDM channels, with independent data transmitted on each channel. To our knowledge, this is the largest number of channels ever reported for a WDM source. The bit rate on each channel is 36.7 Mb/s, and the channel spacing is about 37 GHz (0.3 nm), with wavelengths ranging from 1535.3 nm to 1596.3 nm.
Figure 1 shows a schematic of our transmitter. Fig. 1 Schematic of the 206-channel chirped-pulse WDM transmitter. Transmitter Set-Up In this paper, we demonstrate that CPWDM can be scaled to a record 206 WDM channels, with independent data transmitted on each channel. To our knowledge, this is the largest number of channels ever reported for a WDM source. The bit rate on each channel is 36.7 Mb/s, and the channel spacing is about 37 GHz (0.3 nm), with wavelengths ranging from 1535.3 nm to 1596.3 nm.
polarization rotation, and the so-called “rejection port” of the laser is used to couple as much as 40 mW of average power in a single mode fiber [12]. Pulses with a bandwidth in excess of 70 nm are extracted from this laser. The spectrum is mapped onto the time axis by propagation through a single- -mode fiber with total dispersion oe v7 mM rr rr. a
polarization rotation, and the so-called “rejection port” of the laser is used to couple as much as 40 mW of average power in a single mode fiber [12]. Pulses with a bandwidth in excess of 70 nm are extracted from this laser. The spectrum is mapped onto the time axis by propagation through a single- -mode fiber with total dispersion oe v7 mM rr rr. a
Figure 3: Optical spectrum and pulse train at the output of the chirping fiber.
Figure 3: Optical spectrum and pulse train at the output of the chirping fiber.
Figure 4: Feed-forward equalization set-up. used to provide the feed-forward signal. The extracted pulses are fed to a second (reference) electro-absorption modulator with absorption characteristics similar to those of the data encoding modulator. Only a DC bias is applied to the reference modulator, and its output is detected with a 100 MHz inverting receiver. After proper RF delay, the output of the receiver is finally combined with the 9.94 Gb/s TDM data stream using a resistive power combiner.
Figure 4: Feed-forward equalization set-up. used to provide the feed-forward signal. The extracted pulses are fed to a second (reference) electro-absorption modulator with absorption characteristics similar to those of the data encoding modulator. Only a DC bias is applied to the reference modulator, and its output is detected with a 100 MHz inverting receiver. After proper RF delay, the output of the receiver is finally combined with the 9.94 Gb/s TDM data stream using a resistive power combiner.
Figure 6: Optical spectrum at the output of the transmitter when the same TDM word “11001100...11” is encoded on every pulse. The 52 peaks seen here imply 206 WDM channels within the 3 dB bandwidth. transmitter when the TDM word “11001100...11” is repetitively encoded onto every pulse.
Figure 6: Optical spectrum at the output of the transmitter when the same TDM word “11001100...11” is encoded on every pulse. The 52 peaks seen here imply 206 WDM channels within the 3 dB bandwidth. transmitter when the TDM word “11001100...11” is repetitively encoded onto every pulse.
Figure 7: Bit-error-rate curves for channel 92 in broadcast and multiplex modes. A 2'! pseudo-random bit stream was used in both cases. The inset shows an eye diagram for this channel in multiplex mode.
Figure 7: Bit-error-rate curves for channel 92 in broadcast and multiplex modes. A 2'! pseudo-random bit stream was used in both cases. The inset shows an eye diagram for this channel in multiplex mode.
Fig, 2 Reflected and transmitted optical pulses from the NOLM at 4 = 1554 nm. Fig. 2 shows output pulses from the transmitted and reflected ports of the NOLM. These measurements are taken using only one data channel. The transmitted signal is a coded pulse train at 1 GHz and 4 = 1554 nm. The reflected pulse train shows that one out of every 33 pulses is switched. The switching energy, measured after 1 km loop, is approximately 4 pJ, which is close to the predicted value of 4.2 pJ. We noticed that insufficient switching in the reflected pulses is a consequence of the jitter of gain-switched laser and subsequent signal averaging in the scope.
Fig, 2 Reflected and transmitted optical pulses from the NOLM at 4 = 1554 nm. Fig. 2 shows output pulses from the transmitted and reflected ports of the NOLM. These measurements are taken using only one data channel. The transmitted signal is a coded pulse train at 1 GHz and 4 = 1554 nm. The reflected pulse train shows that one out of every 33 pulses is switched. The switching energy, measured after 1 km loop, is approximately 4 pJ, which is close to the predicted value of 4.2 pJ. We noticed that insufficient switching in the reflected pulses is a consequence of the jitter of gain-switched laser and subsequent signal averaging in the scope.
Fig. 1 Experimental set-up for all-optical coding demonstration.
Fig. 1 Experimental set-up for all-optical coding demonstration.
Fig. 3 Multiplexed channels for a simulated WDM-to-TDN conversion with a total capacity of 33 Gb/s.
Fig. 3 Multiplexed channels for a simulated WDM-to-TDN conversion with a total capacity of 33 Gb/s.
technique that periodically alternates the fast and slow axes of the fiber to achieve repeated collisions between the contro! and signal pulses [11]. These collisions result in an accumulation of the net nonlinear collisional phase shift. The cross-splice is achieved by rotating one fiber’s birefringence axis 90° with respect to the other fiber before fusion splicing. Figure 1. Experimental setup for soliton collisions in a nonlinear optical loop mirror (NOLM). OPO is an optical parametric oscillator, 50/50 indicates a 3-dB PM fiber coupler, PBS indicates a polarizing beam splitter, 4/2 indicates a half-wave plate, and ® indicates a cross splice.
technique that periodically alternates the fast and slow axes of the fiber to achieve repeated collisions between the contro! and signal pulses [11]. These collisions result in an accumulation of the net nonlinear collisional phase shift. The cross-splice is achieved by rotating one fiber’s birefringence axis 90° with respect to the other fiber before fusion splicing. Figure 1. Experimental setup for soliton collisions in a nonlinear optical loop mirror (NOLM). OPO is an optical parametric oscillator, 50/50 indicates a 3-dB PM fiber coupler, PBS indicates a polarizing beam splitter, 4/2 indicates a half-wave plate, and ® indicates a cross splice.
Figure 2. Switching contrast ratio versus the relative delay of the control soliton. The solid line corresponds to the experimental data and the dashed line indicates the result from a numerical simulation.
Figure 2. Switching contrast ratio versus the relative delay of the control soliton. The solid line corresponds to the experimental data and the dashed line indicates the result from a numerical simulation.
Figure 3. Autocorrelation traces of the switched signal pulse measured at the transmission port of the 3-dB fiber coupler. The dashed line: control pulse absent. The dots: control pulse present. The autocorrelation of the original OPO pulses (solid line) is presented for comparison.
Figure 3. Autocorrelation traces of the switched signal pulse measured at the transmission port of the 3-dB fiber coupler. The dashed line: control pulse absent. The dots: control pulse present. The autocorrelation of the original OPO pulses (solid line) is presented for comparison.
Figure 1. Block diagram of experimental arrangement “a The Raman effect on ultrashort optical pulses propagating in silica fibers and the resulting soliton self frequency shift effect have been quite well investigated [2,3]. An optical intensity dependent switch using the soliton self-frequency shift has also been reported [4]. Although the soliton self frequency shift effect is considered a detrimental effect in high bit rate communication systems since it is a potential source of jitter, its pulse-width and intensity dependent frequency conversion characteristics can be exploited to perform a new function, namely thresholding, for an ultrashort pulse CDMA receiver. A block diagram of the receiver is shown in fig. 1. Input pulses to the experiment are generated by a stretched pulse modelocked fiber ring laser [5] and are spectrally filtered to yield pulse durations of
Figure 1. Block diagram of experimental arrangement “a The Raman effect on ultrashort optical pulses propagating in silica fibers and the resulting soliton self frequency shift effect have been quite well investigated [2,3]. An optical intensity dependent switch using the soliton self-frequency shift has also been reported [4]. Although the soliton self frequency shift effect is considered a detrimental effect in high bit rate communication systems since it is a potential source of jitter, its pulse-width and intensity dependent frequency conversion characteristics can be exploited to perform a new function, namely thresholding, for an ultrashort pulse CDMA receiver. A block diagram of the receiver is shown in fig. 1. Input pulses to the experiment are generated by a stretched pulse modelocked fiber ring laser [5] and are spectrally filtered to yield pulse durations of
Figure 3. Contrast ratio of the nonlinear receiver for different cut-off wavelengths of the long wavelength pass filter as a function of average signal power in the thresholder fiber. Figure 2. (a) power spectrum of the coded signal at the input of the nonlinear thresholder, (b) power spectrum of the uncoded signal at the input of the nonlinear thresholder. (c) and (d) power spectrum at the output of the thresholder corresponding to (a) and (b) respectively. Average power in the nonlinear thresholder fiber is 1.84 mW for each case.
Figure 3. Contrast ratio of the nonlinear receiver for different cut-off wavelengths of the long wavelength pass filter as a function of average signal power in the thresholder fiber. Figure 2. (a) power spectrum of the coded signal at the input of the nonlinear thresholder, (b) power spectrum of the uncoded signal at the input of the nonlinear thresholder. (c) and (d) power spectrum at the output of the thresholder corresponding to (a) and (b) respectively. Average power in the nonlinear thresholder fiber is 1.84 mW for each case.
Figure 1: Pictorial representation of NRZ and impulsively encoded optical data streams. T is the bit period. Non-return-to-zero (NRZ) coding, in which optical pulses occupy a full bit period, is the most widely used format in optical fiber communication. The popularity of NRZ stems, in part, from its minimum use of electronic bandwidth to modulate the output of a continuous-wave laser at a given bit rate. New transmitters using modelocked lasers are however being considered for which “impulsive coding” is amore natural format. Contrary to NRZ, the optical power of an impulsively encoded pulse train occupies only a small fraction of the bit periods (Fig. 1). In this paper, we show that impulsive coding results in better receiver sensitivities compared to NRZ. We report impulsive sensitivity improvements as large as 4.7 dB and 5.8 dB for two pin-FET front-ends [1]. These values are in good agreement with a model of the response and noise characteristics of the receivers. They also confirm a
Figure 1: Pictorial representation of NRZ and impulsively encoded optical data streams. T is the bit period. Non-return-to-zero (NRZ) coding, in which optical pulses occupy a full bit period, is the most widely used format in optical fiber communication. The popularity of NRZ stems, in part, from its minimum use of electronic bandwidth to modulate the output of a continuous-wave laser at a given bit rate. New transmitters using modelocked lasers are however being considered for which “impulsive coding” is amore natural format. Contrary to NRZ, the optical power of an impulsively encoded pulse train occupies only a small fraction of the bit periods (Fig. 1). In this paper, we show that impulsive coding results in better receiver sensitivities compared to NRZ. We report impulsive sensitivity improvements as large as 4.7 dB and 5.8 dB for two pin-FET front-ends [1]. These values are in good agreement with a model of the response and noise characteristics of the receivers. They also confirm a
Figure 3: Input photocurrent and corresponding input voltage for NRZ and impulsive coding. B is the bit rate. Although qualitatively correct, the RC model presented above overestimates the impulsive signal enhancement factor by 60 to 70 percent for standard pin-FET transimpedance receivers. To obtain better quantitative agreement of our model with experiment, we consider the transimpedance circuit of Figure 4.
Figure 3: Input photocurrent and corresponding input voltage for NRZ and impulsive coding. B is the bit rate. Although qualitatively correct, the RC model presented above overestimates the impulsive signal enhancement factor by 60 to 70 percent for standard pin-FET transimpedance receivers. To obtain better quantitative agreement of our model with experiment, we consider the transimpedance circuit of Figure 4.
Figure 6: Proportionality factor y appearing in Eq. (3), as a function of the “ringing” parameter Xp .
Figure 6: Proportionality factor y appearing in Eq. (3), as a function of the “ringing” parameter Xp .
Figure 7: Measured and modeled output voltage noise power spectral densities (PSD) for a QDFB-100 front-end, The transfer function parameters are chosen as in Fig. 5, and we use the constants ya =75pA/VHz and Vb =67fA/VHz/ MHz. The agreement is excellent, suggesting that our simple noise model is adequate for the frequency range shown. constants ya = 7.5pA/VHz and vb =67fA/VHz/ MHz.
Figure 7: Measured and modeled output voltage noise power spectral densities (PSD) for a QDFB-100 front-end, The transfer function parameters are chosen as in Fig. 5, and we use the constants ya =75pA/VHz and Vb =67fA/VHz/ MHz. The agreement is excellent, suggesting that our simple noise model is adequate for the frequency range shown. constants ya = 7.5pA/VHz and vb =67fA/VHz/ MHz.
Figure 8: Experimental set-up to measure impulsive and NRZ sensitivities. The receiver consists of a front-end and a low-pass output filter. The validity of our model is tested experimentally using the set-up of Figure 8.
Figure 8: Experimental set-up to measure impulsive and NRZ sensitivities. The receiver consists of a front-end and a low-pass output filter. The validity of our model is tested experimentally using the set-up of Figure 8.
Figure 9 also compares the bit-error rate curves obtained with this front-end using either types of coding. As can be seen, an impulsive sensitivity improvement of 5.8 dB is found in this case. The PRBS pattern length used for these measurements is gy, Table 1 summarizes our experimental results and shows good agreement with our model.
Figure 9 also compares the bit-error rate curves obtained with this front-end using either types of coding. As can be seen, an impulsive sensitivity improvement of 5.8 dB is found in this case. The PRBS pattern length used for these measurements is gy, Table 1 summarizes our experimental results and shows good agreement with our model.
Figure 9: NRZ and impulsive eye diagrams obtained with the QDFB-100 front-end for an optical power of -46 dBm. The inset on the NRZ panel shows the same eye diagram with a 10X resolution. 22 MHz and 300 MHz low-pass filters are used at the output of the receiver front-end for NRZ and impulsive coding respectively. ‘The lower panel compares the bit-error rata curves for both types of coding. average power of -46 dBm for NRZ and for impulsive coding. These data clearly show a larger eye opening in the case of impulsive coding.
Figure 9: NRZ and impulsive eye diagrams obtained with the QDFB-100 front-end for an optical power of -46 dBm. The inset on the NRZ panel shows the same eye diagram with a 10X resolution. 22 MHz and 300 MHz low-pass filters are used at the output of the receiver front-end for NRZ and impulsive coding respectively. ‘The lower panel compares the bit-error rata curves for both types of coding. average power of -46 dBm for NRZ and for impulsive coding. These data clearly show a larger eye opening in the case of impulsive coding.
Figure 10: Sensitivity of the STZ-01 receiver front-end (87 MHz bandwidth) for NRZ and impulsive coding as a function of the output low-pass filter bandwidth. The curves are predictions from our model and the squares and circles are actual measurements for NRZ and impulsive coding respectively. affecting the signal, leading to better sensitivities. When the output filter bandwidth falls below the bit rate, the output voltage for isolated “1” and “0” cannot quite reach the top and bottom rails, leading to inter-symbol interference. The reduction of the noise bandwidth is then increasingly offset by a reduction in signal, leading to a minimum in the sensitivity curve. The filter bandwidth at which this minimum occurs depends on the exact frequency dependence of the noise. It is 37 % of the bit rate when the corner noise frequency is zero, and 68 % of the bit rate when the quadratic noise term in eq.(4) is absent [1].
Figure 10: Sensitivity of the STZ-01 receiver front-end (87 MHz bandwidth) for NRZ and impulsive coding as a function of the output low-pass filter bandwidth. The curves are predictions from our model and the squares and circles are actual measurements for NRZ and impulsive coding respectively. affecting the signal, leading to better sensitivities. When the output filter bandwidth falls below the bit rate, the output voltage for isolated “1” and “0” cannot quite reach the top and bottom rails, leading to inter-symbol interference. The reduction of the noise bandwidth is then increasingly offset by a reduction in signal, leading to a minimum in the sensitivity curve. The filter bandwidth at which this minimum occurs depends on the exact frequency dependence of the noise. It is 37 % of the bit rate when the corner noise frequency is zero, and 68 % of the bit rate when the quadratic noise term in eq.(4) is absent [1].
Figure 1: The dependence of optimized gain of fiber-optical amplifier on distance pumping radiation. The vector NSE describing parallel process in nonlinear media is also considered [2]. Soliton solutions in planar waveguides display interest- ing signal dynamics. Interaction of these soli- tons can be used as multiplexers and demulit- plexers in a number of potential soliton commu- nication applications. A large class of optimiza- tion problems in optimal control theory, and its applications to the signal processing and image processing is considered. Application of opti- mal control theory to eliminate timing jitter effect [9], [10], {11].The new results have inter- disciplinary character and include the soliton theory, theory of optimal control of distributed media and numerous applications to the math- ematical models of distributed systems, such as parallel and distributed computing.
Figure 1: The dependence of optimized gain of fiber-optical amplifier on distance pumping radiation. The vector NSE describing parallel process in nonlinear media is also considered [2]. Soliton solutions in planar waveguides display interest- ing signal dynamics. Interaction of these soli- tons can be used as multiplexers and demulit- plexers in a number of potential soliton commu- nication applications. A large class of optimiza- tion problems in optimal control theory, and its applications to the signal processing and image processing is considered. Application of opti- mal control theory to eliminate timing jitter effect [9], [10], {11].The new results have inter- disciplinary character and include the soliton theory, theory of optimal control of distributed media and numerous applications to the math- ematical models of distributed systems, such as parallel and distributed computing.
Figure 2: Soliton propagation in fiber-optical amplifier with optimized gain
Figure 2: Soliton propagation in fiber-optical amplifier with optimized gain
Figure 1. Conventional direct sequence (DS) modem, shaded regions indicate areas of potential performance limitations for large bandwidth systems. Front end correlation gives us the opportunity to utilize another property of the photoconductive switch, large dynamic range. With no other non-linear components (mixers, etc.) appearing before the sampling gate, the system dynamic range is set by the dynamic range of the gate. Large dynamic range is necessary in a burst mode system where peak-power will exceed average power by a factor of approximately the processing gain. Furthermore, the large breakdown voltage of the photoconductive switch serves to isolate the following stages from large transient signals.
Figure 1. Conventional direct sequence (DS) modem, shaded regions indicate areas of potential performance limitations for large bandwidth systems. Front end correlation gives us the opportunity to utilize another property of the photoconductive switch, large dynamic range. With no other non-linear components (mixers, etc.) appearing before the sampling gate, the system dynamic range is set by the dynamic range of the gate. Large dynamic range is necessary in a burst mode system where peak-power will exceed average power by a factor of approximately the processing gain. Furthermore, the large breakdown voltage of the photoconductive switch serves to isolate the following stages from large transient signals.
However by use of small (<10 pm) photoconductive sampling gaps, laser pulse energies on the order of nJ can be used to perform sampling. Hence, the system can be driven by laser diodes, resulting in much higher data rates and a much smaller package. Figure 3. Test message sent between computers over opto- electronic spread-spectrum link.
However by use of small (<10 pm) photoconductive sampling gaps, laser pulse energies on the order of nJ can be used to perform sampling. Hence, the system can be driven by laser diodes, resulting in much higher data rates and a much smaller package. Figure 3. Test message sent between computers over opto- electronic spread-spectrum link.
Figure 1. Schematic cross section of high speed VCSEL structure with superposed equivalent circuit. High modulation bandwidths were achieved with an oxide confined VCSEL structure modified to decrease parasitic circuit elements. Figure 1 shows a schematic cross section of the VCSEL with a corresponding small signal equivalent circuit. Coplanar waveguide pads designed for on wafer probing were placed on a 5 ym thick polyimide to reduce the capacitance between the pad and the conducting substrate to approximately 50 fF. The device capacitance was further reduced by implanting the mesa area lying outside the active region. This was necessary due to the high capacitance of the thin oxide layer. The non-radiative recombination associated with the deep implant damage may also reduce the charge storage associated with diffusion underneath the oxide region. Finally, the sheet resistance of the upper mirror layer was reduced to ~30Q/square by making it n-type instead of the conventional p-up structure.
Figure 1. Schematic cross section of high speed VCSEL structure with superposed equivalent circuit. High modulation bandwidths were achieved with an oxide confined VCSEL structure modified to decrease parasitic circuit elements. Figure 1 shows a schematic cross section of the VCSEL with a corresponding small signal equivalent circuit. Coplanar waveguide pads designed for on wafer probing were placed on a 5 ym thick polyimide to reduce the capacitance between the pad and the conducting substrate to approximately 50 fF. The device capacitance was further reduced by implanting the mesa area lying outside the active region. This was necessary due to the high capacitance of the thin oxide layer. The non-radiative recombination associated with the deep implant damage may also reduce the charge storage associated with diffusion underneath the oxide region. Finally, the sheet resistance of the upper mirror layer was reduced to ~30Q/square by making it n-type instead of the conventional p-up structure.
The small signal response of VCSELs as a function of bias current was measured using a calibrated vector network analyzer. Air coplanar waveguide probes were used to drive the lasers and the output was collected into a cone-tipped 50 jum core diameter fiber. A New Focus photodector specified to have a -3dB response >30 GHz was connected through approximately 4 m of the multimode fiber. The modulation response of this system is plotted in Figure 3 for one laser at various bias currents. The -3dB frequency increases rapidly to 10 GHz at 1 mA and 15 GHz at 1.6 mA. The maximum observed -3dB bandwidth was 21.5 GHz at 4.8 mA as shown in the figure. The majority of the devices in this region of the wafer exhibited maximum bandwidths in excess of 19 GHz. For low biases, there is a droop of up to 2dB in the response at low frequencies (< 5 Ghz). The droop decreases at higher bias currents. Contributions to this droop may include the detector response, current crowding in the laser, or other bias dependent circuit effects. traditional damped resonator model modified Co Inciude a LOW Tequeley GIUUP See pees improve the fit. The resonant frequency, equivalent damping frequency (y/27), and -3dB bandwidth are plotted in Figure 4. At low bias currents, the bandwidth and resonant frequency increase in proportion to the square root of the current above threshold as expected from the conventional rate equation analysis. The rate of increase in this region yields a modulation current efficiency factor (MCEF) of 14.2 GHz/\mA which is slightly lower than the highest value we previously reported for oxide confined VCSELs with InGaAs quantum wells[1]. The resonant frequency increases steadily to 15 GHz at 2.7 mA and then becomes nearly constant. While multimode operation could limit the photon density in the fundamental mode and thus the resonant frequency of the fundamental, this is not expected to occur until 4 mA based on observation of the laser output spectra. Other possibilities include higher junction temperatures reducing the differential gain and current crowding reducing the overlap of the modulated carrier density and the mode.
The small signal response of VCSELs as a function of bias current was measured using a calibrated vector network analyzer. Air coplanar waveguide probes were used to drive the lasers and the output was collected into a cone-tipped 50 jum core diameter fiber. A New Focus photodector specified to have a -3dB response >30 GHz was connected through approximately 4 m of the multimode fiber. The modulation response of this system is plotted in Figure 3 for one laser at various bias currents. The -3dB frequency increases rapidly to 10 GHz at 1 mA and 15 GHz at 1.6 mA. The maximum observed -3dB bandwidth was 21.5 GHz at 4.8 mA as shown in the figure. The majority of the devices in this region of the wafer exhibited maximum bandwidths in excess of 19 GHz. For low biases, there is a droop of up to 2dB in the response at low frequencies (< 5 Ghz). The droop decreases at higher bias currents. Contributions to this droop may include the detector response, current crowding in the laser, or other bias dependent circuit effects. traditional damped resonator model modified Co Inciude a LOW Tequeley GIUUP See pees improve the fit. The resonant frequency, equivalent damping frequency (y/27), and -3dB bandwidth are plotted in Figure 4. At low bias currents, the bandwidth and resonant frequency increase in proportion to the square root of the current above threshold as expected from the conventional rate equation analysis. The rate of increase in this region yields a modulation current efficiency factor (MCEF) of 14.2 GHz/\mA which is slightly lower than the highest value we previously reported for oxide confined VCSELs with InGaAs quantum wells[1]. The resonant frequency increases steadily to 15 GHz at 2.7 mA and then becomes nearly constant. While multimode operation could limit the photon density in the fundamental mode and thus the resonant frequency of the fundamental, this is not expected to occur until 4 mA based on observation of the laser output spectra. Other possibilities include higher junction temperatures reducing the differential gain and current crowding reducing the overlap of the modulated carrier density and the mode.
Figure 2. Band vending due to space charge. degrade proper high-field carrier transport. Figure 2
Figure 2. Band vending due to space charge. degrade proper high-field carrier transport. Figure 2
yie aq Do widegap diffusion block diffi da potential gradient can be incorporated to generate uasi-field to effectively accelerate electrons (Fig. 1). ping-grading can also be used for this purpose. The ayer is placed to avoid the back usion of electrons toward the anode. Compared to conventional pin-PDs, the carrier traveling distance over ing layers is considerably larger the absorption and collec UTC-PDs. Actually, a s. The average elec col ecting layer is twice delay in electron injection ron traveling distance in hat in a conventional struc Thus, to achieve fast photoresponse in UTC-PDs, electron mobility in the absorption layer and velocity overshoot in the collection layer both play essential ro from absorption layer to the collector does exist in UTC- the ure. high es.
yie aq Do widegap diffusion block diffi da potential gradient can be incorporated to generate uasi-field to effectively accelerate electrons (Fig. 1). ping-grading can also be used for this purpose. The ayer is placed to avoid the back usion of electrons toward the anode. Compared to conventional pin-PDs, the carrier traveling distance over ing layers is considerably larger the absorption and collec UTC-PDs. Actually, a s. The average elec col ecting layer is twice delay in electron injection ron traveling distance in hat in a conventional struc Thus, to achieve fast photoresponse in UTC-PDs, electron mobility in the absorption layer and velocity overshoot in the collection layer both play essential ro from absorption layer to the collector does exist in UTC- the ure. high es.
Figure 3. Calculated 3 dB bandwidth.
Figure 3. Calculated 3 dB bandwidth.
Figure 4. Fabricated InP/InGaAs UTC-PD structure. this V, value, the diode was found to operate at average current densities up to as high as 400 kA/cm2. The FWHM of the output signal is 3.4 ps and the Fourier transform of the pulse response gives an f3gp of 80 GHz. This f3gg3 value is slightly greater than the theoretical prediction in Fig. 3. Probable reasons are higher minority electron mobility and/or a self-biasing effect on the absorption layer in the actual device. Device Fabrication this V, value, the diode was found to operate at average BAVA 2 ers Bea We fabricated InP/InGaAs UTC-PDs with uniform- bandgap absorption layers. The epitaxial layers were grown on semi-insulating InP substrates by MOVPE. A schematic cross-section and a band diagram are shown in Fig. 4. The absorption layer is p-Ing,53Gao.47As doped at 2.5 x 1018 /cm3 (220-nm thick) . The collection layer consists of undoped InP (10 nm), n-InP (n = 2.0 x 1018 /cm3, 15 nm) and undoped InP (200 nm). Here, n-InP was inserted to lower the potential spike produced at the absorption layer/collecting layer interface. The diffusion- block layer is pseudomorphic p*-Ing 4gGag.54As (p = 2.0 x 1019 /em3, 10 nm). The contact layers (pt-Ing 53Gag.47As and nt-InP) were heavily doped to over p = 2.0 x 1019 /cm3 to minimize contact resistance. The dopants were C and Si, respectively, for p and n-type layers.
Figure 4. Fabricated InP/InGaAs UTC-PD structure. this V, value, the diode was found to operate at average current densities up to as high as 400 kA/cm2. The FWHM of the output signal is 3.4 ps and the Fourier transform of the pulse response gives an f3gp of 80 GHz. This f3gg3 value is slightly greater than the theoretical prediction in Fig. 3. Probable reasons are higher minority electron mobility and/or a self-biasing effect on the absorption layer in the actual device. Device Fabrication this V, value, the diode was found to operate at average BAVA 2 ers Bea We fabricated InP/InGaAs UTC-PDs with uniform- bandgap absorption layers. The epitaxial layers were grown on semi-insulating InP substrates by MOVPE. A schematic cross-section and a band diagram are shown in Fig. 4. The absorption layer is p-Ing,53Gao.47As doped at 2.5 x 1018 /cm3 (220-nm thick) . The collection layer consists of undoped InP (10 nm), n-InP (n = 2.0 x 1018 /cm3, 15 nm) and undoped InP (200 nm). Here, n-InP was inserted to lower the potential spike produced at the absorption layer/collecting layer interface. The diffusion- block layer is pseudomorphic p*-Ing 4gGag.54As (p = 2.0 x 1019 /em3, 10 nm). The contact layers (pt-Ing 53Gag.47As and nt-InP) were heavily doped to over p = 2.0 x 1019 /cm3 to minimize contact resistance. The dopants were C and Si, respectively, for p and n-type layers.
Figure 5. Photoresponse of an InP/InGaAs UTC-PD. Output Saturation Behavior The output peak voltage Vp is shown in Fig. 6 as a function of input power with the bias voltage Vpias as a parameter. The input-output curves clearly indicate saturation of the output voltage (= current), where V,'s change first linearly prior to the saturation. We also confirmed that f3gpn values decrease rapidly when V,'s saturated. The saturation behavior is basically dominated by two mechanisms: The first is the shift in the operating point to lower reverse bias. The
Figure 5. Photoresponse of an InP/InGaAs UTC-PD. Output Saturation Behavior The output peak voltage Vp is shown in Fig. 6 as a function of input power with the bias voltage Vpias as a parameter. The input-output curves clearly indicate saturation of the output voltage (= current), where V,'s change first linearly prior to the saturation. We also confirmed that f3gpn values decrease rapidly when V,'s saturated. The saturation behavior is basically dominated by two mechanisms: The first is the shift in the operating point to lower reverse bias. The
Figure 7. Output current at saturation point. Figure 6. Input-output characteristics.
Figure 7. Output current at saturation point. Figure 6. Input-output characteristics.
Figure 1. Microphotograph of the fabricated RCI photodiode The epitaxial layers are grown by a solid-source Mb on semi-insulating GaAs substrates. We fabricated the epitaxial wafers using a monolithic microwave-compatible fabrication process. A microphotograph of the fabricated photodiodes is shown in Figure 1. First, ohmic contacts to the N+ layers were formed by a recess etch through the 0.3 micron N- layer. This was followed by a self-aligned Au- Ge-Ni liftoff and a rapid thermal anneal. The semi- transparent Schottky contact was formed by deposition of 200 A Au. Using an isolation mask, we etched away all of the epilayers except the active areas. Then, we evaporated Ti/Au interconnect metal which formed coplanar waveguide (CPW) transmission lines on top of the semi- insulating substrate. The next step was the deposition and patterning of a 2000 A thick silicon nitride layer. The thickness of the nitride layer was chosen to act as an antireflection coating for the RCE Schottky photodiode at the design wavelength. Besides passivation and protection of the surface, the nitride was also used as the dielectric of the metal-insulator-metal bias capacitors. Finally, 1.5 micron thick Au layer was used as an airbridge to connect the center of the CPW to the top Schottky metal. The resulting Schottky diodes had breakdown voltages larger than 12 V. The dark-current of a 150x150 lum device at - 1V bias was 30 nA. Using the forward current-voltage characteristics, we measured the barrier height of the Schottky junction to be 0.83 eV.
Figure 1. Microphotograph of the fabricated RCI photodiode The epitaxial layers are grown by a solid-source Mb on semi-insulating GaAs substrates. We fabricated the epitaxial wafers using a monolithic microwave-compatible fabrication process. A microphotograph of the fabricated photodiodes is shown in Figure 1. First, ohmic contacts to the N+ layers were formed by a recess etch through the 0.3 micron N- layer. This was followed by a self-aligned Au- Ge-Ni liftoff and a rapid thermal anneal. The semi- transparent Schottky contact was formed by deposition of 200 A Au. Using an isolation mask, we etched away all of the epilayers except the active areas. Then, we evaporated Ti/Au interconnect metal which formed coplanar waveguide (CPW) transmission lines on top of the semi- insulating substrate. The next step was the deposition and patterning of a 2000 A thick silicon nitride layer. The thickness of the nitride layer was chosen to act as an antireflection coating for the RCE Schottky photodiode at the design wavelength. Besides passivation and protection of the surface, the nitride was also used as the dielectric of the metal-insulator-metal bias capacitors. Finally, 1.5 micron thick Au layer was used as an airbridge to connect the center of the CPW to the top Schottky metal. The resulting Schottky diodes had breakdown voltages larger than 12 V. The dark-current of a 150x150 lum device at - 1V bias was 30 nA. Using the forward current-voltage characteristics, we measured the barrier height of the Schottky junction to be 0.83 eV.
Figure 2. (a) Measured, and (b) simulated photoresponse of the RCE photodiode. multi-mode fiber, and the other end of the fiber. was brought in close proximity to the wafer, where a circular region with ~ 75 pm diameter was illuminated. The spectral response of the photocurrent obtained from this set-up was corrected by first measuring the optical power coming out of the fiber with a power meter. For photospectral measurements, we used a 150x150 wm photodiode biased at -2.0 Volts.
Figure 2. (a) Measured, and (b) simulated photoresponse of the RCE photodiode. multi-mode fiber, and the other end of the fiber. was brought in close proximity to the wafer, where a circular region with ~ 75 pm diameter was illuminated. The spectral response of the photocurrent obtained from this set-up was corrected by first measuring the optical power coming out of the fiber with a power meter. For photospectral measurements, we used a 150x150 wm photodiode biased at -2.0 Volts.
Figure 3. Pulse response of the RCE photodiode This research is partially supported by the Office of Naval Research under grant No. NO0014-96-1-0652 and the Turkish Scientific and Technical Research Council under Project No. EEEAG-156. We also acknowledge a National Science Foundation International Collaborative Research funding grant (No. INT-9601770).
Figure 3. Pulse response of the RCE photodiode This research is partially supported by the Office of Naval Research under grant No. NO0014-96-1-0652 and the Turkish Scientific and Technical Research Council under Project No. EEEAG-156. We also acknowledge a National Science Foundation International Collaborative Research funding grant (No. INT-9601770).
We demonstrate the first monolithic PIC for generation of mm-wave signals. This new PIC integrates a passively mode- locked semiconductor ring laser, optical amplifier and high-speed photodiode for generation, amplification and detection of an optical pulse train with 30 to 90 GHz pulse-repetition frequency. Output is an electrical signal, generated by the photodiode, whose fundamental frequency is the pulse repetition rate of the mode-locked laser. The circuit uses a novel waveguide photodiode (WGPD) integrated with a mm-wave transmission line specifically designed for high-speed operation and signal extraction on a heavily-doped GaAs substrate. Figure 1. Schematic of actual mm-wave generation circuit. The mm-wave signal generation PIC design for 60 and 90 GHz signals is shown in Fig. 1. A closed-ring GaAs/AlGaAs diode laser with a separately contacted, reverse-biased saturable absorber is passively mode locked generating a continuous train of short (1 to 10 ps) optical pulses in the ring cavity. Output of the ring laser is coupled to optical waveguide amplifier using a multi-mode Y-junction. A direct-waveguide photodetector (WGPD) converts the optical pulse train emerging from the amplifier to a mm-wave electrical output. Finally, a 50-Ohm transmission line is integrated with the WGPD for removal of the electrical signal. A WGPD was chosen rather than a velocity- matched TWPD so that the PIC requires only one epitaxial growth sequence. The pn-junction design is optimized for laser and amplifier operation and does not have a large enough depletion width to accommodate a _ velocity-matched photodetector. An alternate configuration was used for 30 GHz
We demonstrate the first monolithic PIC for generation of mm-wave signals. This new PIC integrates a passively mode- locked semiconductor ring laser, optical amplifier and high-speed photodiode for generation, amplification and detection of an optical pulse train with 30 to 90 GHz pulse-repetition frequency. Output is an electrical signal, generated by the photodiode, whose fundamental frequency is the pulse repetition rate of the mode-locked laser. The circuit uses a novel waveguide photodiode (WGPD) integrated with a mm-wave transmission line specifically designed for high-speed operation and signal extraction on a heavily-doped GaAs substrate. Figure 1. Schematic of actual mm-wave generation circuit. The mm-wave signal generation PIC design for 60 and 90 GHz signals is shown in Fig. 1. A closed-ring GaAs/AlGaAs diode laser with a separately contacted, reverse-biased saturable absorber is passively mode locked generating a continuous train of short (1 to 10 ps) optical pulses in the ring cavity. Output of the ring laser is coupled to optical waveguide amplifier using a multi-mode Y-junction. A direct-waveguide photodetector (WGPD) converts the optical pulse train emerging from the amplifier to a mm-wave electrical output. Finally, a 50-Ohm transmission line is integrated with the WGPD for removal of the electrical signal. A WGPD was chosen rather than a velocity- matched TWPD so that the PIC requires only one epitaxial growth sequence. The pn-junction design is optimized for laser and amplifier operation and does not have a large enough depletion width to accommodate a _ velocity-matched photodetector. An alternate configuration was used for 30 GHz
Figure 3. Electron micrograph view of complete all-optical mm- wave signal generation PIC. Total length of the circuit is less than 2 mm. Figure 2. Cross-section views of the three principle circuit elements comprising the all-optical mm-wave signal generator. (a) Laser and amplifier active waveguide. (b) Waveguide photodiode. (c) Millimeter-wave transmission line. Key dimensions are: W = 12 um, S=5 pm and G= 10 pm.
Figure 3. Electron micrograph view of complete all-optical mm- wave signal generation PIC. Total length of the circuit is less than 2 mm. Figure 2. Cross-section views of the three principle circuit elements comprising the all-optical mm-wave signal generator. (a) Laser and amplifier active waveguide. (b) Waveguide photodiode. (c) Millimeter-wave transmission line. Key dimensions are: W = 12 um, S=5 pm and G= 10 pm.
Figure 4. Measured performance of three different mm-wave signal generation PICs. Large, medium and small rings are 860, 430 and 290 um diameter respectively. amplifier stage while using the second stage in reverse bias. Since intra-band carrier relaxation times are similar to the mode- locked pulse length, measurement of CW gain is an approximation of the actual gain under mode-locked operation.
Figure 4. Measured performance of three different mm-wave signal generation PICs. Large, medium and small rings are 860, 430 and 290 um diameter respectively. amplifier stage while using the second stage in reverse bias. Since intra-band carrier relaxation times are similar to the mode- locked pulse length, measurement of CW gain is an approximation of the actual gain under mode-locked operation.
Figure 2, The transverse and the longitudinal electric field at the position of 80 mm Figure 1. I - V characteristic for the photoconductive switch
Figure 2, The transverse and the longitudinal electric field at the position of 80 mm Figure 1. I - V characteristic for the photoconductive switch
[6] T. Itatani, T. Nakagawa, F. Kano, K. Ohta and Y. Sugiyama, Trans. IEICE vol. E78-C pp.73(1995).
[6] T. Itatani, T. Nakagawa, F. Kano, K. Ohta and Y. Sugiyama, Trans. IEICE vol. E78-C pp.73(1995).
Fig. 1. Schematic representation of the waveguide in cross section (upper part) and of the experimental setup for the electrooptic on-wafer pulse propagation measurements (lower part). Our TFMSLs (see Fig. 1) are fabricated on low- resistivity (5-8 Qcm) Si substrates. First, an 800- nm-thick Al ground-conductor metallization is de- posited onto the Si wafer by electron beam evapora- tion. Then, the insulator material, commercially avail- able Cyclotene 3022-46 from Dow Chemicals, is spin— deposited. Cyclotene is a resin based on Bisbenzocy- clobutene (BCB) momomers developed for applications in microelectronics such as passivation coatings and di- electrics for inter—chip connects in multi-chip modules [8]. It is cured for 60 minutes at 250°C in a N2 ambi- ent. After polymerization, the signal conductor with an integrated photoconductive (PC) switch is fabri- cated. First, a 500x500 ym? piece of 500-nm-thick lift-off LT-GaAs (LT: low-temperature-grown) is van— der-Waals bonded onto the polymer at the position in- tended for the PC switch (see below). After removal
Fig. 1. Schematic representation of the waveguide in cross section (upper part) and of the experimental setup for the electrooptic on-wafer pulse propagation measurements (lower part). Our TFMSLs (see Fig. 1) are fabricated on low- resistivity (5-8 Qcm) Si substrates. First, an 800- nm-thick Al ground-conductor metallization is de- posited onto the Si wafer by electron beam evapora- tion. Then, the insulator material, commercially avail- able Cyclotene 3022-46 from Dow Chemicals, is spin— deposited. Cyclotene is a resin based on Bisbenzocy- clobutene (BCB) momomers developed for applications in microelectronics such as passivation coatings and di- electrics for inter—chip connects in multi-chip modules [8]. It is cured for 60 minutes at 250°C in a N2 ambi- ent. After polymerization, the signal conductor with an integrated photoconductive (PC) switch is fabri- cated. First, a 500x500 ym? piece of 500-nm-thick lift-off LT-GaAs (LT: low-temperature-grown) is van— der-Waals bonded onto the polymer at the position in- tended for the PC switch (see below). After removal
Fig. 2. Electric signals probed at several positions z; (given in mm) on the 35-Q TFMSL (upper graph) and on the 70-0 TFMSL (lower graph). The signals are normalized to the maxi- mum of the first pulse. the electromagnetic signal with lattice vibrations whose propagation direction is oriented perpendicular to the wafer surface [13]. Reflection at the topside of the elec- trooptic crystal results in a standing-wave resonance.
Fig. 2. Electric signals probed at several positions z; (given in mm) on the 35-Q TFMSL (upper graph) and on the 70-0 TFMSL (lower graph). The signals are normalized to the maxi- mum of the first pulse. the electromagnetic signal with lattice vibrations whose propagation direction is oriented perpendicular to the wafer surface [13]. Reflection at the topside of the elec- trooptic crystal results in a standing-wave resonance.
Fig. 3. Frequency dependence of the effective permittivity of both TFMSLs. crease of €,,-¢7 with rising f when the cut-off frequency for the next highest mode is approached [2], [5]. The rise of €,er¢ for f — O results from the finite conductiv- ity of the metallization (skin effect) and is well known in the analysis of waveguides including low-—dispersive
Fig. 3. Frequency dependence of the effective permittivity of both TFMSLs. crease of €,,-¢7 with rising f when the cut-off frequency for the next highest mode is approached [2], [5]. The rise of €,er¢ for f — O results from the finite conductiv- ity of the metallization (skin effect) and is well known in the analysis of waveguides including low-—dispersive
Fig. 4. Frequency dependence of the attenuation of both TFM- SLs. Fig. 4 presents the attenuation a(f) as a function of frequency for the two TFMSLs. In both cases, a(f)
Fig. 4. Frequency dependence of the attenuation of both TFM- SLs. Fig. 4 presents the attenuation a(f) as a function of frequency for the two TFMSLs. In both cases, a(f)
Fig. 5. Pulse propagation experiment on the CPW: A freely po- sitionable probe injects electric pulses that are detected with an EO crystal. Because the CPW does not contain an integrated PC switch, electric-pulse generation is performed with the help of a freely positionable PC probe as illus- trated in Fig. 5. We employ a second-generation probe that has been improved compared to earlier silicon—on—
Fig. 5. Pulse propagation experiment on the CPW: A freely po- sitionable probe injects electric pulses that are detected with an EO crystal. Because the CPW does not contain an integrated PC switch, electric-pulse generation is performed with the help of a freely positionable PC probe as illus- trated in Fig. 5. We employ a second-generation probe that has been improved compared to earlier silicon—on—
Fig. 7. Frequency—dependent wave propagation characteristics of the CPW. tained by external injection with a probe. The Fourier transform of the pulse, displayed in the inset of Fig. 6, shows that the pulse has an extremely large bandwidth with useful frequency components up to 1.5 THz.
Fig. 7. Frequency—dependent wave propagation characteristics of the CPW. tained by external injection with a probe. The Fourier transform of the pulse, displayed in the inset of Fig. 6, shows that the pulse has an extremely large bandwidth with useful frequency components up to 1.5 THz.
Fig. 6. Electrooptically detected transient on a 50-§2 CPW after pulse injection with a freely positionable PC probe. Inset: Fourier spectrum. pulses that might interfere with each other. The ampli- tude of the detected signal is 800 mV (bias on the PC gap: 15 V), the full width at half maximum is 750 fs. To our knowledge, this is the shortest pulse ever ob-
Fig. 6. Electrooptically detected transient on a 50-§2 CPW after pulse injection with a freely positionable PC probe. Inset: Fourier spectrum. pulses that might interfere with each other. The ampli- tude of the detected signal is 800 mV (bias on the PC gap: 15 V), the full width at half maximum is 750 fs. To our knowledge, this is the shortest pulse ever ob-
Figure 1. Photoresponse transient measured at T = 50 K. The inset shows the sample configuration and the electro- optic sampling beams.
Figure 1. Photoresponse transient measured at T = 50 K. The inset shows the sample configuration and the electro- optic sampling beams.
Figure 2. Solution of the Rothwarf-Taylor equations used to model the photoresponse. The solid lines are the excess quasi- particle and phonon densities, and the dashed line is the quasi- particle generation term.
Figure 2. Solution of the Rothwarf-Taylor equations used to model the photoresponse. The solid lines are the excess quasi- particle and phonon densities, and the dashed line is the quasi- particle generation term.
Figure 3. Measured transient (dots) at T = 20 K and the results from the simulation (solid line).
Figure 3. Measured transient (dots) at T = 20 K and the results from the simulation (solid line).
Figure |. The SEM image of the polysilicon surface after preferential etching at grain boundaries reveals grain sizes on the order of 30 nm. In order to independently confirm the polySi grain size and obtain preliminary data on absorption depth, a Perkin- Elmer Lambda 9 spectrophotometer was used to measure the transmission of the polySi-on-silica sample (see Fig. 3). The interpolated data, corrected for thin-film etalon fringes, show a broadened absorption band edge and a small but
Figure |. The SEM image of the polysilicon surface after preferential etching at grain boundaries reveals grain sizes on the order of 30 nm. In order to independently confirm the polySi grain size and obtain preliminary data on absorption depth, a Perkin- Elmer Lambda 9 spectrophotometer was used to measure the transmission of the polySi-on-silica sample (see Fig. 3). The interpolated data, corrected for thin-film etalon fringes, show a broadened absorption band edge and a small but
Figure 3. The solid line shows transmission spectra of polySi sample, after accounting for the thin-film etalon effects (dashed line). Figure 2. An SEM cross-section image showing some columnarity in the 2.3-ym polySi layer.
Figure 3. The solid line shows transmission spectra of polySi sample, after accounting for the thin-film etalon effects (dashed line). Figure 2. An SEM cross-section image showing some columnarity in the 2.3-ym polySi layer.
A 34-GHz (10-ps internal rise time) sampling oscilloscope measurement setup, shown in Fig. 4, allowed convenient measurement of relative quantum efficiencies for various laser wavelengths, since our EO sampling system operated up to only near-infrared wavelengths and did not extend into the fiber optic communication wavelengths. The oscilloscope
A 34-GHz (10-ps internal rise time) sampling oscilloscope measurement setup, shown in Fig. 4, allowed convenient measurement of relative quantum efficiencies for various laser wavelengths, since our EO sampling system operated up to only near-infrared wavelengths and did not extend into the fiber optic communication wavelengths. The oscilloscope
Figure 5. The photoresponse at 800-nm illumination wavelength measured for different switch bias levels. The inset plots a linear dependence on gap voltage bias. Figure 4. Sampling oscilloscope measurement setup. Inset shows the interdigitated switch geometry. Switch dimensions 2 mm x 2 mm; finger separation 100 ym.
Figure 5. The photoresponse at 800-nm illumination wavelength measured for different switch bias levels. The inset plots a linear dependence on gap voltage bias. Figure 4. Sampling oscilloscope measurement setup. Inset shows the interdigitated switch geometry. Switch dimensions 2 mm x 2 mm; finger separation 100 ym.
Figure 7. The photoresponse FWHM at 800-nm wavelength and 40 nJ/cm2 fluence versus the switch bias. Figure 6. The photoresponse at 800 nm illumination wavelength measured for different pulse average power levels. Inset shows linear response versus illumination fluence.
Figure 7. The photoresponse FWHM at 800-nm wavelength and 40 nJ/cm2 fluence versus the switch bias. Figure 6. The photoresponse at 800 nm illumination wavelength measured for different pulse average power levels. Inset shows linear response versus illumination fluence.
Our switch configuration was not designed for EO measurements [3]. Nevertheless, we managed to get preliminary results by connecting the switch as a meander- type slot line. The signal was generated and measured entirely on the face of the switch structure. It propagated in
Our switch configuration was not designed for EO measurements [3]. Nevertheless, we managed to get preliminary results by connecting the switch as a meander- type slot line. The signal was generated and measured entirely on the face of the switch structure. It propagated in
Figure 8. The photoresponse at 1.55-ym illumination wavelength. The quantum efficiency was less than 1074.
Figure 8. The photoresponse at 1.55-ym illumination wavelength. The quantum efficiency was less than 1074.
Figure 9. Electro-optically measured signal from a polySi switch connected as a meander-type slot line.
Figure 9. Electro-optically measured signal from a polySi switch connected as a meander-type slot line.
Figure 1 shows a schematic diagram of the cross section of a GaAs heterojunction field-effect transistor (HFET), which is one of the devices investigated in this study. Complementary HFET technology represents a low- power alternative to MESFET technology that maintains the high speed and total dose immunity of GaAs that is required for space-based digital applications. Ion strikes in the high-field region between the gate and drain typically give rise to the largest current pulses (also referred to as photocurrents) on the device terminals. We have
Figure 1 shows a schematic diagram of the cross section of a GaAs heterojunction field-effect transistor (HFET), which is one of the devices investigated in this study. Complementary HFET technology represents a low- power alternative to MESFET technology that maintains the high speed and total dose immunity of GaAs that is required for space-based digital applications. Ion strikes in the high-field region between the gate and drain typically give rise to the largest current pulses (also referred to as photocurrents) on the device terminals. We have
Figure 3. 3 MeV a-Particle-induced drain charge-collection transients measured for an n-channel HFET device with and without the LT GaAs buffer layer (into a 50 Q load). Vg = 0 V; Vs =0V; Vp =4 V for the LT device; and Vp = 1.5 V for the conventional device. - Figure 2 shows a typical charge-collection transient measured for an n-channel HFET device induced with 610
Figure 3. 3 MeV a-Particle-induced drain charge-collection transients measured for an n-channel HFET device with and without the LT GaAs buffer layer (into a 50 Q load). Vg = 0 V; Vs =0V; Vp =4 V for the LT device; and Vp = 1.5 V for the conventional device. - Figure 2 shows a typical charge-collection transient measured for an n-channel HFET device induced with 610
Figure 4. Time-integrated charge-collection measurements performed as a function of gate bias for irradiation by 3 Mea particles. Vp = 2V; Vs = OV.
Figure 4. Time-integrated charge-collection measurements performed as a function of gate bias for irradiation by 3 Mea particles. Vp = 2V; Vs = OV.
Figure 5. Schematic diagram of the enhancement mode MESFET structure investigated in this study.
Figure 5. Schematic diagram of the enhancement mode MESFET structure investigated in this study.
Figure 6. Drain charge-collection transients calculated as a function of the carrier recombination time, T,, in the 1 pm thick buffer layer. L, = 0.3 um; Vp = 2.5V; Vg = 0.05 V; Vs = OV.
Figure 6. Drain charge-collection transients calculated as a function of the carrier recombination time, T,, in the 1 pm thick buffer layer. L, = 0.3 um; Vp = 2.5V; Vg = 0.05 V; Vs = OV.
Figure 7. Constant electron density contours (cm) at 10 ps after the laser pulse for structures (a) without and (b) with a 0.5 um thick buffer layer with a 1 ps lifetime. L, = 0.3 um; Vp = 2.5 V; Vg = 0.05 V; and Vs = 0 V.
Figure 7. Constant electron density contours (cm) at 10 ps after the laser pulse for structures (a) without and (b) with a 0.5 um thick buffer layer with a 1 ps lifetime. L, = 0.3 um; Vp = 2.5 V; Vg = 0.05 V; and Vs = 0 V.
Figure 8 Constant electron density contours (cnr?) at 100 ps after the laser pulse for structures (a) without and (b) with a 0.5 pm thick buffer layer with a 1 ps lifetime. L, = 0.3 um; Vp = 2.5 V: Vg = 0.05 V; and Vs =0V.
Figure 8 Constant electron density contours (cnr?) at 100 ps after the laser pulse for structures (a) without and (b) with a 0.5 pm thick buffer layer with a 1 ps lifetime. L, = 0.3 um; Vp = 2.5 V: Vg = 0.05 V; and Vs =0V.
Fig. 1 Normalized time-resolved reflectivity of (a) as-implanted (200 keV and 1043 ions/em2), (b) RTA-annealed (600 for 30 s), and (C) furnaceannealed (600 for 30 minutes )GaAs:As+ samples.
Fig. 1 Normalized time-resolved reflectivity of (a) as-implanted (200 keV and 1043 ions/em2), (b) RTA-annealed (600 for 30 s), and (C) furnaceannealed (600 for 30 minutes )GaAs:As+ samples.
Dark current characteristics of an Auston-type PCS fabricated on (a) S. |. GaAs substrate (dotted line), (b) furnace-annealed (solid line), and (c) RTA-annealed (dashed line) low-dose-implanted GaAs:As* samples. The inset is a magnified view of traces (a) and (b).
Dark current characteristics of an Auston-type PCS fabricated on (a) S. |. GaAs substrate (dotted line), (b) furnace-annealed (solid line), and (c) RTA-annealed (dashed line) low-dose-implanted GaAs:As* samples. The inset is a magnified view of traces (a) and (b).
advanced laboratory technologies are required to achieve this data rate with the speed-critical circuits, as long as the other specifications are not essentially relaxed. Recent measurement results of mounted Si-bipolar chips are shown in Fig. 1 [5-7]. The circuits, again developed in cooperation between RUB and Siemens, are listed in the last column of Table 1. They were fabricated either with an improved laboratory version of the B6HF [8] (DEMUX), a former epitaxial-base technology (decision circuit), or a recent SiGe technology [9] (rest). All these ICs are used in an experimental 20 Gb/s TDM transmis- sion system [2]. Seong a For generating the eye diagrams in Fig. 1, the circuits were driven by a pseudo-random bit sequence generator with a word length of 2'°-1 bits [6]. The 60 Gb/s of the MUX, presented in [6], is the highest data rate ever gen- erated by an IC in any technology and the DEMUX [5], too, is far above the required data rate, despite of the slower (implanted base) technology applied. The excellent retiming capability of the DEMUX at the record input data rate of 46 Gb/s is demonstrated in Fig. 1b for an intentionally degraded input eye diagram. The output eye diagrams of the speed-critical circuits, transimpedance preamplifier and modulator driver, are given at 20 Gb/s. Here, the photo diode, which drives the amplifier, is modeled by an electrical network realized on the measu- ring substrate [10]. The high gain (58 dBQ) and low noise (estimated jy ~ 12 pA/ Hz ) of the transimpedance pre- amplifier (due to the high shunt feedback resistance of 900 2 in the first stage) as well as the comparatively high output voltage swing of the modulator driver (maximum values: 2.3 V for single-ended and 4.6 V for differential operation) are record values for 20 Gb/s Si ICs [7].
advanced laboratory technologies are required to achieve this data rate with the speed-critical circuits, as long as the other specifications are not essentially relaxed. Recent measurement results of mounted Si-bipolar chips are shown in Fig. 1 [5-7]. The circuits, again developed in cooperation between RUB and Siemens, are listed in the last column of Table 1. They were fabricated either with an improved laboratory version of the B6HF [8] (DEMUX), a former epitaxial-base technology (decision circuit), or a recent SiGe technology [9] (rest). All these ICs are used in an experimental 20 Gb/s TDM transmis- sion system [2]. Seong a For generating the eye diagrams in Fig. 1, the circuits were driven by a pseudo-random bit sequence generator with a word length of 2'°-1 bits [6]. The 60 Gb/s of the MUX, presented in [6], is the highest data rate ever gen- erated by an IC in any technology and the DEMUX [5], too, is far above the required data rate, despite of the slower (implanted base) technology applied. The excellent retiming capability of the DEMUX at the record input data rate of 46 Gb/s is demonstrated in Fig. 1b for an intentionally degraded input eye diagram. The output eye diagrams of the speed-critical circuits, transimpedance preamplifier and modulator driver, are given at 20 Gb/s. Here, the photo diode, which drives the amplifier, is modeled by an electrical network realized on the measu- ring substrate [10]. The high gain (58 dBQ) and low noise (estimated jy ~ 12 pA/ Hz ) of the transimpedance pre- amplifier (due to the high shunt feedback resistance of 900 2 in the first stage) as well as the comparatively high output voltage swing of the modulator driver (maximum values: 2.3 V for single-ended and 4.6 V for differential operation) are record values for 20 Gb/s Si ICs [7].
Table 1. High-speed ICs for optical- fiber systems in Si-bipolar technologies. The data rates in the second column were achieved with production technologies, those in the last column with Siemens laboratory technologies.
Table 1. High-speed ICs for optical- fiber systems in Si-bipolar technologies. The data rates in the second column were achieved with production technologies, those in the last column with Siemens laboratory technologies.
Fig. 2. Scheme of a 40 Gb/s optical-fiber TDM system
Fig. 2. Scheme of a 40 Gb/s optical-fiber TDM system
Fig. 3. 40 Gb/s limiting transimpedance amplifier; gain 66 dBQ (2 kQ), AVo = 0.6 V: a) Simplified block diagram (offset control not shown), b) Simulated eye diagram of the output voltage at 40 Gb,
Fig. 3. 40 Gb/s limiting transimpedance amplifier; gain 66 dBQ (2 kQ), AVo = 0.6 V: a) Simplified block diagram (offset control not shown), b) Simulated eye diagram of the output voltage at 40 Gb,
Fig. 4. 40 Gb/s power MUX; AVg = 2 V: a) Simplified block diagram, b) Simulated eye diagram across the EAM quantum wells at 40 Gb/s.
Fig. 4. 40 Gb/s power MUX; AVg = 2 V: a) Simplified block diagram, b) Simulated eye diagram across the EAM quantum wells at 40 Gb/s.
Figure 2. Block diagram of 4:1 multiplexer circuit High speed MUX circuits are important for both Time Division Multiplexing (TDM) and Wavelength Division Multiplexing (WDM). We have demonstrated a 4:1 multiplexer that operates at 40 Gb/s using our hybrid digital/microwave process. A block diagram of this circuit is shown in Fig. 2. The architecture used requires only the last 2:1 MUX stage to operate at the maximum bit rate. The rest of the circuit operates at half the bit rate. Another outstanding feature of this design is the low device count and low power consumption compared to multiplexing with four MS-DFFs. The circuit was tested on-wafer, using RF probes. The test data signals were generated by a 10 Gb/s bit-error- rate tester. The 4:1 multiplexer operated up to a maximum bit rate of 40 Gb/s, with data input levels of 125 mV,» per channel, and a single-ended clock input level of 500 mV, ». A microphotograph of this 4:1 multiplexer is shown in Fig. 3, the output eye diagram and output waveforms when operating at 40 Gb/s are shown in Fig, 4a. and Fig. 4b. To our knowledee. this is the fastest 4:1 MUX reported (3).
Figure 2. Block diagram of 4:1 multiplexer circuit High speed MUX circuits are important for both Time Division Multiplexing (TDM) and Wavelength Division Multiplexing (WDM). We have demonstrated a 4:1 multiplexer that operates at 40 Gb/s using our hybrid digital/microwave process. A block diagram of this circuit is shown in Fig. 2. The architecture used requires only the last 2:1 MUX stage to operate at the maximum bit rate. The rest of the circuit operates at half the bit rate. Another outstanding feature of this design is the low device count and low power consumption compared to multiplexing with four MS-DFFs. The circuit was tested on-wafer, using RF probes. The test data signals were generated by a 10 Gb/s bit-error- rate tester. The 4:1 multiplexer operated up to a maximum bit rate of 40 Gb/s, with data input levels of 125 mV,» per channel, and a single-ended clock input level of 500 mV, ». A microphotograph of this 4:1 multiplexer is shown in Fig. 3, the output eye diagram and output waveforms when operating at 40 Gb/s are shown in Fig, 4a. and Fig. 4b. To our knowledee. this is the fastest 4:1 MUX reported (3).
Figure 5. Output eye diagrams (data and inverted data), and divided clock output at 30 Gb/s. Vertical scale:400mV/div. The 4:1 MUX design was also fabricated at Rockwell Semiconductor Systems GaAs production facility in Newbury Park, CA. (5). Operation beyond 30 Gb/s was measured, with the resulting data eye diagrams and divided clock output shown in Figure. 5.
Figure 5. Output eye diagrams (data and inverted data), and divided clock output at 30 Gb/s. Vertical scale:400mV/div. The 4:1 MUX design was also fabricated at Rockwell Semiconductor Systems GaAs production facility in Newbury Park, CA. (5). Operation beyond 30 Gb/s was measured, with the resulting data eye diagrams and divided clock output shown in Figure. 5.
Figure 3. Microphotograph of 4:1 multiplexer chip. Figure 4. Output eye diagrams (data and inverted data) at 40 Gb/s. Vertical scale:400mV/div.
Figure 3. Microphotograph of 4:1 multiplexer chip. Figure 4. Output eye diagrams (data and inverted data) at 40 Gb/s. Vertical scale:400mV/div.
Figure 6. 1:4 DMUX schematic block diagram enn inane ee EES The 1:4 demultiplexer was designed with similar multifunctional circuits as the 4:1 MUX (3,4). Figure 6. shows the circuit principle, which features a novel two-stage architecture (6). The master-slave (MS) flipflops are used in the initial 1:2 demultiplexing to regenerate and stabilize the incoming data. A slave {S} D-type latch is added to the lower master- slave D-type flip-flop (MS-DFF) for bit alignment. The final stage of demultiplexing is accomplished with six freeze-type {F} latches (7). The input clock CK1 (with a frequency half that of the data rate) is used to time all latches (D and F type). CK2 is used as a select signal by the F latches, and is delayed relative to CK1 by the gate delay of the frequency divider. Because only one phase of the divided clock (CK2) is used, a dynamic frequency divider may be used in place of the conventional MS static frequency divider with reduced power dissipation and potentially increased circuit operating speed. However, in this implementation a MS-DFF divider was used to generate CK2, to maximize the operating frequency range of the circuit. The simulated output eye diagrams (15 Gb/s), with 60 Gb/s input data, are shown in Figure 7.
Figure 6. 1:4 DMUX schematic block diagram enn inane ee EES The 1:4 demultiplexer was designed with similar multifunctional circuits as the 4:1 MUX (3,4). Figure 6. shows the circuit principle, which features a novel two-stage architecture (6). The master-slave (MS) flipflops are used in the initial 1:2 demultiplexing to regenerate and stabilize the incoming data. A slave {S} D-type latch is added to the lower master- slave D-type flip-flop (MS-DFF) for bit alignment. The final stage of demultiplexing is accomplished with six freeze-type {F} latches (7). The input clock CK1 (with a frequency half that of the data rate) is used to time all latches (D and F type). CK2 is used as a select signal by the F latches, and is delayed relative to CK1 by the gate delay of the frequency divider. Because only one phase of the divided clock (CK2) is used, a dynamic frequency divider may be used in place of the conventional MS static frequency divider with reduced power dissipation and potentially increased circuit operating speed. However, in this implementation a MS-DFF divider was used to generate CK2, to maximize the operating frequency range of the circuit. The simulated output eye diagrams (15 Gb/s), with 60 Gb/s input data, are shown in Figure 7.
Figure 7. Simulated output eye diagrams for 1:4 DMUX
Figure 7. Simulated output eye diagrams for 1:4 DMUX
Figure 8. Input test signal to 1:4 DMUX at 30 Gb/s. Vertical scale:160 mV/div
Figure 8. Input test signal to 1:4 DMUX at 30 Gb/s. Vertical scale:160 mV/div
The circuit was tested on-wafer using a 30 Gb/s test signal generated by connecting an on-wafer 4:1 MUX and 1:4 DMUX with 0.8m of semirigid cable. The input test signal is shown in Fig. 8. The resulting demultiplexed output eye diagrams (CH1-3) and divided clock outputs are shown ‘in Fig. 9. The divided clock output was designed with a falling edge, which is independent of data rate, in its relative position to the output data channel eyes, and is used to time subsequent circuits.
The circuit was tested on-wafer using a 30 Gb/s test signal generated by connecting an on-wafer 4:1 MUX and 1:4 DMUX with 0.8m of semirigid cable. The input test signal is shown in Fig. 8. The resulting demultiplexed output eye diagrams (CH1-3) and divided clock outputs are shown ‘in Fig. 9. The divided clock output was designed with a falling edge, which is independent of data rate, in its relative position to the output data channel eyes, and is used to time subsequent circuits.
Figure 10. Die microphotograph of 1:4 DMUX
Figure 10. Die microphotograph of 1:4 DMUX
We have designed a multifunctional circuit, fabricated in a baseline heterojunction bipolar transistor (HBT) technology, for-use.in laboratory system experiments (8). The IC, when combined with an external transimpedance amplifier, and a high-Q filter (PLL-or dielectric resonator) can be used as a regenerator for non-return to zero (NRZ) lightwave systems. Figure 11. Block diagram of the data regeneration/clock extraction chip, including extemal! PLL as narrow band filter.
We have designed a multifunctional circuit, fabricated in a baseline heterojunction bipolar transistor (HBT) technology, for-use.in laboratory system experiments (8). The IC, when combined with an external transimpedance amplifier, and a high-Q filter (PLL-or dielectric resonator) can be used as a regenerator for non-return to zero (NRZ) lightwave systems. Figure 11. Block diagram of the data regeneration/clock extraction chip, including extemal! PLL as narrow band filter.
Figure 12. Limiting amplifier circuit detail. The decision circuit (data regeneration and retiming) was implemented in a master-slave configuration with static D- type latches, with two emitter followers in the feedback path for maximum speed. The decision circuit operated up to 22 Gbit/s, under a full rate clock and data input.
Figure 12. Limiting amplifier circuit detail. The decision circuit (data regeneration and retiming) was implemented in a master-slave configuration with static D- type latches, with two emitter followers in the feedback path for maximum speed. The decision circuit operated up to 22 Gbit/s, under a full rate clock and data input.
Figure 13. Timing recovery circuit detail. Timing recovery used nonlinear processing (an on-chip differentiate-and-rectify circuit), to generate a clock spectral component at the bit rate (Fig. 13). The differentiator was a differential pair with its emitters connected by BE and BC junction transistor capacitors (11). The capacitance was adjustable, with an externally applied reverse bias. Its operation was very insensitive to the applied voltage. Rectification was implemented with two __ transistors connected at both the collectors and the emitters. A series 39Q resistor matched this output (single-ended) to 50 Q. An optional differential output was generated with an on-chip single-ended to differential converter with open collector output drivers. The converter acted as an amplifier, and bandlimited the differential output to about 15 GHz. The clock input signal is on-chip terminated into 50 ©, and is amplified by two stages of limiting amplification (similar to the data path), before being applied to the input of the on-chip MSDFF. If the recovered clock, after external high-Q filtering, is placed through a 1:N frequency divider, then the output of the MSDFF will be one of the N constituent data streams multiplexed together in the input data. In our case N=4 (the PLL provided an internal divide- by-two, together with an external 1:2 divider).
Figure 13. Timing recovery circuit detail. Timing recovery used nonlinear processing (an on-chip differentiate-and-rectify circuit), to generate a clock spectral component at the bit rate (Fig. 13). The differentiator was a differential pair with its emitters connected by BE and BC junction transistor capacitors (11). The capacitance was adjustable, with an externally applied reverse bias. Its operation was very insensitive to the applied voltage. Rectification was implemented with two __ transistors connected at both the collectors and the emitters. A series 39Q resistor matched this output (single-ended) to 50 Q. An optional differential output was generated with an on-chip single-ended to differential converter with open collector output drivers. The converter acted as an amplifier, and bandlimited the differential output to about 15 GHz. The clock input signal is on-chip terminated into 50 ©, and is amplified by two stages of limiting amplification (similar to the data path), before being applied to the input of the on-chip MSDFF. If the recovered clock, after external high-Q filtering, is placed through a 1:N frequency divider, then the output of the MSDFF will be one of the N constituent data streams multiplexed together in the input data. In our case N=4 (the PLL provided an internal divide- by-two, together with an external 1:2 divider).
Figure 14. Output waveform of on-chip d/dt+ rectify. The output of the on-chip differentiate and rectify circuit at 24 Gb/s is shown in Fig. 14., with a 2-1 pseudorandom bit sequence (PRBS) at the input to the MUX. The recovered spectral component at 24 Gb/s is shown in Figure 15, with a 2°41 PRBS input to the MUX. The 24 Gbit/s spectrum was measured, with a down conversion mixer (20 dB conversion loss, 10 dBm LO at 19.63 GHz), to have a power level of -32.27 dBm (a front-end down conversion mixer was used due to spectrum analyzer limitations). The spurious clock spectral component generated from nondata sources (primarily MUX clock feedthrough) was measured to be - 47.35 dBm at 24 Gb/s. At 30 Gb/s, the signal and spurious levels were measured to be -47.33 and -53.76 dBm, respectively. No measures were taken to reduce the magnitude of the spurious component.
Figure 14. Output waveform of on-chip d/dt+ rectify. The output of the on-chip differentiate and rectify circuit at 24 Gb/s is shown in Fig. 14., with a 2-1 pseudorandom bit sequence (PRBS) at the input to the MUX. The recovered spectral component at 24 Gb/s is shown in Figure 15, with a 2°41 PRBS input to the MUX. The 24 Gbit/s spectrum was measured, with a down conversion mixer (20 dB conversion loss, 10 dBm LO at 19.63 GHz), to have a power level of -32.27 dBm (a front-end down conversion mixer was used due to spectrum analyzer limitations). The spurious clock spectral component generated from nondata sources (primarily MUX clock feedthrough) was measured to be - 47.35 dBm at 24 Gb/s. At 30 Gb/s, the signal and spurious levels were measured to be -47.33 and -53.76 dBm, respectively. No measures were taken to reduce the magnitude of the spurious component.
Figure 15. Output spectrum of d/dt+rectify. The recovered clock spectral component generated on-chip was high-Q filtered with an off-chip PLL constructed from discrete microwave components. The PLL consisted of a third-order loop with a YIG oscillator VCO, a microwave mixer, power divider, and a loop filter using a commercial operational amplifier. The loop bandwidth was 300 kHz. Because the PLL required an input signal of between -10 dBm to +10 dBm, the differentiate-and-rectify output of the IC was amplified by microwave amplifiers. The output power of the VCO was +13 dBm. A frequency doubler within the PLL loop allows for an inherent frequency division by two.
Figure 15. Output spectrum of d/dt+rectify. The recovered clock spectral component generated on-chip was high-Q filtered with an off-chip PLL constructed from discrete microwave components. The PLL consisted of a third-order loop with a YIG oscillator VCO, a microwave mixer, power divider, and a loop filter using a commercial operational amplifier. The loop bandwidth was 300 kHz. Because the PLL required an input signal of between -10 dBm to +10 dBm, the differentiate-and-rectify output of the IC was amplified by microwave amplifiers. The output power of the VCO was +13 dBm. A frequency doubler within the PLL loop allows for an inherent frequency division by two.
Figure 16. Output 7.5 Gb/s demultiplexed eye diagram with 30 Gb/s input data.
Figure 16. Output 7.5 Gb/s demultiplexed eye diagram with 30 Gb/s input data.
Figure 17. Microphotograph of VGA and simplified circuit diagram of TA/TI stage.
Figure 17. Microphotograph of VGA and simplified circuit diagram of TA/TI stage.
Q3-Q6 and the Ry’s form the TI portion. Variation of I, by an external gain controlled voltage changes the transimpedance (2m) of the TA amplifier which in turn varies the overall gain of the amplifier. Q5 and Q6 in the parallel feedback path of the TI amplifier buffer the output from the feedback network which results in higher gain, wider bandwidth, and constant output DC voltage independent of the gain-controlled voltages (15). The emitter followers between the stages provide level shifting and improve the input/output impedance matching between the output TI portion of the first amplifier stage and the input TA portion of the second amplifier stage. The measured performance of this VGA is shown in Fig. 18. Note that this circuit has a gain variation of only + 1 dB, constant group delay within the passband, and a gain controlled range of 10-16 dB. This performance is suitable for use in > 30 Gb/s fiber-optic TDM transmission systems (16). The AGC amplifier was fabricated with a transistor structure which exhibited a f, of 60 GHz and fhx of 111 GHz at the amplifier bias point. The peak f, was 95 GHz at 1.2x10° A/cm’ with finx of 130 GHz
Q3-Q6 and the Ry’s form the TI portion. Variation of I, by an external gain controlled voltage changes the transimpedance (2m) of the TA amplifier which in turn varies the overall gain of the amplifier. Q5 and Q6 in the parallel feedback path of the TI amplifier buffer the output from the feedback network which results in higher gain, wider bandwidth, and constant output DC voltage independent of the gain-controlled voltages (15). The emitter followers between the stages provide level shifting and improve the input/output impedance matching between the output TI portion of the first amplifier stage and the input TA portion of the second amplifier stage. The measured performance of this VGA is shown in Fig. 18. Note that this circuit has a gain variation of only + 1 dB, constant group delay within the passband, and a gain controlled range of 10-16 dB. This performance is suitable for use in > 30 Gb/s fiber-optic TDM transmission systems (16). The AGC amplifier was fabricated with a transistor structure which exhibited a f, of 60 GHz and fhx of 111 GHz at the amplifier bias point. The peak f, was 95 GHz at 1.2x10° A/cm’ with finx of 130 GHz
Figure 21. Transimpedance and phase for transimpedance amplifier Conclusion
Figure 21. Transimpedance and phase for transimpedance amplifier Conclusion
Figure 20. Schematic diagram of pre-amplifier circuit
Figure 20. Schematic diagram of pre-amplifier circuit
Table 1. Epitaxial layer structures for the InP/InGaAs DHBT.
Table 1. Epitaxial layer structures for the InP/InGaAs DHBT.
Fig. 1. Schematic cross-sectional view of the fabricated DHBTs The DHBT epitaxial layer structure shown in Table 1 was grown by low-pressure metalorganic vapor phase epitaxy (MOVPE). The Zn doped p*-InxGaj.xAs compositionally graded base layer structure (In-As composition r, anging from 0.53 at the B/C interface to 0.47 at E/B one) was used to reduce the electron transit time and to lower t he extrinsic base contact resistivity. The potential drop across the graded base layer is 34 mV [9]. The InGaAs/InP comp tor layer structures are designed to improve t breakdown behavior without any significant cu ing effects, and to reduce the transit time [10]. thick n+-InGaAs layer was inserted to lower t contact resistivity. The 400-nm-thick InP subco osite collec- he collector rrent block- The 20-nm- he collector Hector layer was employed on behalf of the conventional InGaAs subcollector to lower the junction temperature because InP has over-one-order-higher thermal conductivity than InGaAs (0.68 versus 0.05 W/cm-K at 300 K) [11]. Thermal design is important for achieving high-density ICs.
Fig. 1. Schematic cross-sectional view of the fabricated DHBTs The DHBT epitaxial layer structure shown in Table 1 was grown by low-pressure metalorganic vapor phase epitaxy (MOVPE). The Zn doped p*-InxGaj.xAs compositionally graded base layer structure (In-As composition r, anging from 0.53 at the B/C interface to 0.47 at E/B one) was used to reduce the electron transit time and to lower t he extrinsic base contact resistivity. The potential drop across the graded base layer is 34 mV [9]. The InGaAs/InP comp tor layer structures are designed to improve t breakdown behavior without any significant cu ing effects, and to reduce the transit time [10]. thick n+-InGaAs layer was inserted to lower t contact resistivity. The 400-nm-thick InP subco osite collec- he collector rrent block- The 20-nm- he collector Hector layer was employed on behalf of the conventional InGaAs subcollector to lower the junction temperature because InP has over-one-order-higher thermal conductivity than InGaAs (0.68 versus 0.05 W/cm-K at 300 K) [11]. Thermal design is important for achieving high-density ICs.
The small-scale InP/InGaAs DHBTs were fabricated by self-aligned processing as reported previously [4,5]. A schematic cross-sectional view of the fabricated DHBTs is shown in Fig. 1. Using the lift-off Ti/Pt/Au emitter electrode metal as a mask, the E/B mesa-etching was carried out by a combination of Cly-based ECR-RIE and selective wet etch- ing (citric acid:H2O2:H20 and HCI:H3POq for the InGaAs and InP layers, respectively). The Pt/Ti/Pt/Au base electrode metal was then evaporated on the area including the emitter metal. The TLM measurements of the extrinsic base layers exhibited good sheet resistance of 616 Q2/square and specific contact resistivity of 2.4 x 10-7 Qcem2.
The small-scale InP/InGaAs DHBTs were fabricated by self-aligned processing as reported previously [4,5]. A schematic cross-sectional view of the fabricated DHBTs is shown in Fig. 1. Using the lift-off Ti/Pt/Au emitter electrode metal as a mask, the E/B mesa-etching was carried out by a combination of Cly-based ECR-RIE and selective wet etch- ing (citric acid:H2O2:H20 and HCI:H3POq for the InGaAs and InP layers, respectively). The Pt/Ti/Pt/Au base electrode metal was then evaporated on the area including the emitter metal. The TLM measurements of the extrinsic base layers exhibited good sheet resistance of 616 Q2/square and specific contact resistivity of 2.4 x 10-7 Qcem2.
Fig. 3. Common-emitter collector I-V characteristics of DHBTs showing the load line. Figure 3 shows typical common-emitter/collector I-V char- acteristics for a DHBT with a 6-1m2 emitter metal with an emitter width of 1.2 um, corresponding to the emitter dimen- sions in the internal circuits. The current gain is over 50. For DHBTs in internal circuits, Ic ranges from 0.3 to 3.5 mA as a Vcg ranges from 1.6 to 1.1 V to reduce the power dissi- pation of the IC. It is found that these DHBT characteristics indicate a good collector breakdown behavior at least within the bias conditions for the IC operation. This is due to the InP/InGaAs composite collector layer structure and InP subcollector layer. Average and standard deviations of cur- rent gains at a collector current density Jc of 1x105 A/cm? across 2-inch wafers were 60.1 and 4.8, respectively.
Fig. 3. Common-emitter collector I-V characteristics of DHBTs showing the load line. Figure 3 shows typical common-emitter/collector I-V char- acteristics for a DHBT with a 6-1m2 emitter metal with an emitter width of 1.2 um, corresponding to the emitter dimen- sions in the internal circuits. The current gain is over 50. For DHBTs in internal circuits, Ic ranges from 0.3 to 3.5 mA as a Vcg ranges from 1.6 to 1.1 V to reduce the power dissi- pation of the IC. It is found that these DHBT characteristics indicate a good collector breakdown behavior at least within the bias conditions for the IC operation. This is due to the InP/InGaAs composite collector layer structure and InP subcollector layer. Average and standard deviations of cur- rent gains at a collector current density Jc of 1x105 A/cm? across 2-inch wafers were 60.1 and 4.8, respectively.
Fig. 4. Ic-Vcg characteristics of an InP-subcollector (solid line) and an InGaAs-subcollector (dotted line) DHBTs with 20- jim2 emitter.
Fig. 4. Ic-Vcg characteristics of an InP-subcollector (solid line) and an InGaAs-subcollector (dotted line) DHBTs with 20- jim2 emitter.
Fig. 8. Histograms of fr's and fmax's for the 6- and 20-um2 emitter DHBTs. Fig. 7. fy and fmax dependences on Vcg for the DHBT.
Fig. 8. Histograms of fr's and fmax's for the 6- and 20-um2 emitter DHBTs. Fig. 7. fy and fmax dependences on Vcg for the DHBT.
Fig. 6. fp and fmax dependences on Ic for the DHBT.
Fig. 6. fp and fmax dependences on Ic for the DHBT.
Ic at Vcg of 1.3 V for a DHBT with a 6-um? emitter. In spite of using the 420-nm-thick InP/InGaAs composite collector layers, the peak fy of 183 GHz is quite high, due to the 45- nm-thick In,Gaj-xAs graded base structure. Also, the de- pendences of fy and fmax on Vce at Ic of 3 mA and 6 mA for the same DHBT are shown in Fig. 7. fr's rapidly increases from alow Vcg of 0.6 V. The peak fmax is as high as 235 GHz at Ic of 6 mA since the extrinsic base resistivity of the In,Gay.xAs graded base layer is very low [9]. A further nar- rowing of the emitter width will enable our DHBTs to en- hance fmax values [12].
Ic at Vcg of 1.3 V for a DHBT with a 6-um? emitter. In spite of using the 420-nm-thick InP/InGaAs composite collector layers, the peak fy of 183 GHz is quite high, due to the 45- nm-thick In,Gaj-xAs graded base structure. Also, the de- pendences of fy and fmax on Vce at Ic of 3 mA and 6 mA for the same DHBT are shown in Fig. 7. fr's rapidly increases from alow Vcg of 0.6 V. The peak fmax is as high as 235 GHz at Ic of 6 mA since the extrinsic base resistivity of the In,Gay.xAs graded base layer is very low [9]. A further nar- rowing of the emitter width will enable our DHBTs to en- hance fmax values [12].
Figure 9 shows a block diagram of a regenerative receiver IC. The IC consists of a transimpedance-type preamplifier, cascode-type postamplifier, automatic offset controller (AOC), 90-degree delay, phase detector, lag-lead low-pass filter (LPF), hysterisis-comparator-type voltage-controlled os- cillator (VCO), master-slave D-type flip-flop (D-F/F). and output driver [13]. The VCO was designed to oscillate in the frequency range from 19 to 21 GHz. The reshaping is per- formed by a preamplifier, postamplifier, and AOC. The retiming is accomplished by a phase-locked loop (PLL) to extract clock signal. The D-F/F regenerates the NRZ data using the clock signal extracted from the PLL. Fig. 9. Block diagram of the regenerative receiver IC.
Figure 9 shows a block diagram of a regenerative receiver IC. The IC consists of a transimpedance-type preamplifier, cascode-type postamplifier, automatic offset controller (AOC), 90-degree delay, phase detector, lag-lead low-pass filter (LPF), hysterisis-comparator-type voltage-controlled os- cillator (VCO), master-slave D-type flip-flop (D-F/F). and output driver [13]. The VCO was designed to oscillate in the frequency range from 19 to 21 GHz. The reshaping is per- formed by a preamplifier, postamplifier, and AOC. The retiming is accomplished by a phase-locked loop (PLL) to extract clock signal. The D-F/F regenerates the NRZ data using the clock signal extracted from the PLL. Fig. 9. Block diagram of the regenerative receiver IC.
Fig. 11. Eye diagrams of 20-Gbit/s data from the IC (upper trace) and demultiplexed 10-Gbit/s data from the GaAs MESFET D-F/F (lower trace). Fig. 10. Microphotograph of the fabricated IC (1.6 x 1.6 mm in size).
Fig. 11. Eye diagrams of 20-Gbit/s data from the IC (upper trace) and demultiplexed 10-Gbit/s data from the GaAs MESFET D-F/F (lower trace). Fig. 10. Microphotograph of the fabricated IC (1.6 x 1.6 mm in size).
demultiplexed 10-Gbit/s data from the D-F/F [8]. Error-free operation is obtained for an input dynamic range of 13 dB. The power dissipation is 0.6 W at a supply voltage Veg of -5 VO
demultiplexed 10-Gbit/s data from the D-F/F [8]. Error-free operation is obtained for an input dynamic range of 13 dB. The power dissipation is 0.6 W at a supply voltage Veg of -5 VO
Fig. 12. Performance comparison between our regencrative receiver and other receivers.
Fig. 12. Performance comparison between our regencrative receiver and other receivers.
TW-FETs are fully-distributed devices that use the FET’s electrodes not only as the electrical contacts but also as the transmission lines[2}. They have been mainly employed as broadband amplifiers in MMICs. Each electrode line is terminated with the matched impedance for the suppression of signal reflection. And the signal propagation velocity on the gate line is set equal to that on the drain line to yield the broadband amplification through the in-phase superposition of the infinitesimally small waves induced along the drain line (See Fig. 1). However. ihey are regarded as unpractical devices because of the large losses of the gate electrodes. We have newly developed a potential ultrafast digital circuit in which the traveling-wave field cffect transistors (TW-FETs) are employed as the core elements in order to fully utilize the distribution effects of the active devices in the circuits.
TW-FETs are fully-distributed devices that use the FET’s electrodes not only as the electrical contacts but also as the transmission lines[2}. They have been mainly employed as broadband amplifiers in MMICs. Each electrode line is terminated with the matched impedance for the suppression of signal reflection. And the signal propagation velocity on the gate line is set equal to that on the drain line to yield the broadband amplification through the in-phase superposition of the infinitesimally small waves induced along the drain line (See Fig. 1). However. ihey are regarded as unpractical devices because of the large losses of the gate electrodes. We have newly developed a potential ultrafast digital circuit in which the traveling-wave field cffect transistors (TW-FETs) are employed as the core elements in order to fully utilize the distribution effects of the active devices in the circuits.
(I) When the signal propagates in the c mode (the 1 mode) only, termination with the c mode (the 7 mode) characteristic impedance results in the complete suppression of the reflection and the velocity matching condition is automatically satisfied. Figure 3 shows the supporting calculated results.
(I) When the signal propagates in the c mode (the 1 mode) only, termination with the c mode (the 7 mode) characteristic impedance results in the complete suppression of the reflection and the velocity matching condition is automatically satisfied. Figure 3 shows the supporting calculated results.
OE OE IES EOI Various logic functions can be per formed by the combination of the TW-FETs. Typical examples of the TW-logic circuits are shown in Fig. 5, where (a) is a NOR and (b) is a set-reset latching circuit. Figure 5. Examples of traveling-wave logic circuits. (a) A NOR circuit and (b) a set-reset latching circuit.
OE OE IES EOI Various logic functions can be per formed by the combination of the TW-FETs. Typical examples of the TW-logic circuits are shown in Fig. 5, where (a) is a NOR and (b) is a set-reset latching circuit. Figure 5. Examples of traveling-wave logic circuits. (a) A NOR circuit and (b) a set-reset latching circuit.
Figure 4. Termination for suppression of multiple reflections. where Zee and Zr are the c and the m mode impedance on the gate line and Le, Cz, La, gm. x and We are the gate inductance, the gate capacitance, the drain inductance, the transconductance, the inductive coupling coefficient and the gate width, respectively. This equation suggests the coupled TW-FET works as a phase shifter with gain. That is, it recovers its original function.
Figure 4. Termination for suppression of multiple reflections. where Zee and Zr are the c and the m mode impedance on the gate line and Le, Cz, La, gm. x and We are the gate inductance, the gate capacitance, the drain inductance, the transconductance, the inductive coupling coefficient and the gate width, respectively. This equation suggests the coupled TW-FET works as a phase shifter with gain. That is, it recovers its original function.
heterojunction field effect transistor (HFET) having a It of 80 GHz. The terminal resistances were monolithically integrated and were connected with the TW-FET throughout the coplanar waveguides (CPWs).The characteristic impedance of the TW-FET is 5 Q for the c mode and 90 Q for the 1 mode. The calculated fraction of the c mode to the 7 mode was large, and the effective characteristic impedance of the TW-FET was considered to be very small. Therefore, we designed the characteristic impedance of the CPWs to be 25 Q, which was the lowest value our process could produce maturely. A conventional DCFL inverter was also fabricated on the same wafer for comparison. Figure 6. Microphotograph of a fabricated traveling-wave inverter.
heterojunction field effect transistor (HFET) having a It of 80 GHz. The terminal resistances were monolithically integrated and were connected with the TW-FET throughout the coplanar waveguides (CPWs).The characteristic impedance of the TW-FET is 5 Q for the c mode and 90 Q for the 1 mode. The calculated fraction of the c mode to the 7 mode was large, and the effective characteristic impedance of the TW-FET was considered to be very small. Therefore, we designed the characteristic impedance of the CPWs to be 25 Q, which was the lowest value our process could produce maturely. A conventional DCFL inverter was also fabricated on the same wafer for comparison. Figure 6. Microphotograph of a fabricated traveling-wave inverter.
Figure 9 Response of ther traveling-wave inverter. (a) Measured and calculated waveforms and (b) frequency dependence of ther transfer function.
Figure 9 Response of ther traveling-wave inverter. (a) Measured and calculated waveforms and (b) frequency dependence of ther transfer function.
Figure 8 Transition of the signal on the drain linc. Hence. the rise and fall time become steeper and the amplitude becomes larger when the signal propagates forward on the drain line. We can sce both of these fundamental characteristics of the TW-FET in Fig. 8. To our knowledge, this is the first ever observation of the traveling-wave operation of a FET in the time-domain.
Figure 8 Transition of the signal on the drain linc. Hence. the rise and fall time become steeper and the amplitude becomes larger when the signal propagates forward on the drain line. We can sce both of these fundamental characteristics of the TW-FET in Fig. 8. To our knowledge, this is the first ever observation of the traveling-wave operation of a FET in the time-domain.
transistors on a semi-insulating InP substrate, self-aligned base contacts for high speed o peration, thin film TaN resistors and high resistivity epitaxial resistors, Siz3N4 and polyimide capacitors, and two levels of interconnects. The IC fabrication process constructs he individual HBTs first (typically 2x2 jm? through 2x20 pm?) followed by the passive elements. Then the wa fers are planarized with polyimide and vias are formed t hrough the polyimide to expose buried electrodes which are contacted with second level interconnects (overlay metal). The transistor characteristics of HBTs with 2x2 pm? emitters include common emitter current gain in excess of 40. RF characteristics include peak unity current gain cutoff frequency, fr, of 80 GHz.
transistors on a semi-insulating InP substrate, self-aligned base contacts for high speed o peration, thin film TaN resistors and high resistivity epitaxial resistors, Siz3N4 and polyimide capacitors, and two levels of interconnects. The IC fabrication process constructs he individual HBTs first (typically 2x2 jm? through 2x20 pm?) followed by the passive elements. Then the wa fers are planarized with polyimide and vias are formed t hrough the polyimide to expose buried electrodes which are contacted with second level interconnects (overlay metal). The transistor characteristics of HBTs with 2x2 pm? emitters include common emitter current gain in excess of 40. RF characteristics include peak unity current gain cutoff frequency, fr, of 80 GHz.
Figure 4. ADC quantize analog signal into discrete voltage and time intervals. Figure 5. Frequency spectrum of a quantized signal. Another limit to ADC performance is circuit generated noise. The equivalent input-referred thermal noise of an ADC is given by Another limit to ADC performance is circuit generated noise. The equivalent input-referred thermal noise of an ADC is given by
Figure 4. ADC quantize analog signal into discrete voltage and time intervals. Figure 5. Frequency spectrum of a quantized signal. Another limit to ADC performance is circuit generated noise. The equivalent input-referred thermal noise of an ADC is given by Another limit to ADC performance is circuit generated noise. The equivalent input-referred thermal noise of an ADC is given by
Figure 7. An example of a 3-bit flash ADC.
Figure 7. An example of a 3-bit flash ADC.
Figure 8. Die photo of 3-bit flash ADC implemented in InP HBT IC technology.
Figure 8. Die photo of 3-bit flash ADC implemented in InP HBT IC technology.
from 0 to Fs/2, oversampling (averaging) can improve the SNR by 3dB (0.5 bits) for every doubling of the clock rate. Using this technique, however, requires very high oversampling rates. For example, to achieve 10 bit resolution on a 50 MHz signal with a single bit ADC requires a 12.5 THz sample rate. A more practical solution to oversampling with a single bit ADC is to add a loop filter inside a feedback loop. This is the basic concept behind delta-sigma modulation [5].
from 0 to Fs/2, oversampling (averaging) can improve the SNR by 3dB (0.5 bits) for every doubling of the clock rate. Using this technique, however, requires very high oversampling rates. For example, to achieve 10 bit resolution on a 50 MHz signal with a single bit ADC requires a 12.5 THz sample rate. A more practical solution to oversampling with a single bit ADC is to add a loop filter inside a feedback loop. This is the basic concept behind delta-sigma modulation [5].
Figure 9. 1 order AD modulator loop and the resulting noise power spectrum of its output. modulator (Figure 10). Figure 11 is a die photo of the InP HBT 2™ order AZ modulator. The chip operates from + 5 volt supplies and dissipates 1W. Ata sample rate of 3.2 GHz and a signal bandwidth of 50 MHz (OSR of 32 and 100 MSPS Nyquist rate) the modulator demonstrates a SFDR of 71 dB (12-bits dynamic range and the ideal signal-to-noise ratio of 55 dB for a second-order modulator at an oversampling ratio of 32. For an oversampling ratio of 256, the modulator achieves a peak SNR of 70 dB. ).
Figure 9. 1 order AD modulator loop and the resulting noise power spectrum of its output. modulator (Figure 10). Figure 11 is a die photo of the InP HBT 2™ order AZ modulator. The chip operates from + 5 volt supplies and dissipates 1W. Ata sample rate of 3.2 GHz and a signal bandwidth of 50 MHz (OSR of 32 and 100 MSPS Nyquist rate) the modulator demonstrates a SFDR of 71 dB (12-bits dynamic range and the ideal signal-to-noise ratio of 55 dB for a second-order modulator at an oversampling ratio of 32. For an oversampling ratio of 256, the modulator achieves a peak SNR of 70 dB. ).
Figure 11. Die photograph of the InP HBT 2™ order AZ modulator.
Figure 11. Die photograph of the InP HBT 2™ order AZ modulator.
Figure 12. Bandpass AX modulator loop and the resulting noise power spectrum of its output. The region of surpressed quantization noise can be tuned from DC to IF by replacing the integrator with a bandpass filter. Figure 13 shows a bandpass AZ modulator topology with the resulting noise power spectrum. The primary motivation of bandpass converters is the simplicity they impart to systems dealing with narrow-band signals. The use of a bandpass AZ modulator permits a direct conversion of an analog signal to digital form at IF for RF communication and radar systems.
Figure 12. Bandpass AX modulator loop and the resulting noise power spectrum of its output. The region of surpressed quantization noise can be tuned from DC to IF by replacing the integrator with a bandpass filter. Figure 13 shows a bandpass AZ modulator topology with the resulting noise power spectrum. The primary motivation of bandpass converters is the simplicity they impart to systems dealing with narrow-band signals. The use of a bandpass AZ modulator permits a direct conversion of an analog signal to digital form at IF for RF communication and radar systems.
Figure 15. The output spectrum of the tunable 2" order bandpass AZ modulator for various center frequencies. Measurements on this bandpass modulator yield SNR values ranging from 89 dB over a 366 kHz bandwidth to 42 dB over a 62.6 MHz bandwidth. Figure 15 shows the output spectrum for various center frequencies and demonstrates notch tunability. A peak SNR of 84.9 dB, 81.2 dB and 80.5 dB was obtained corresponding to notch positions of 62.5 MHz, 41.4 MHz and 24.4 MHz respectively.
Figure 15. The output spectrum of the tunable 2" order bandpass AZ modulator for various center frequencies. Measurements on this bandpass modulator yield SNR values ranging from 89 dB over a 366 kHz bandwidth to 42 dB over a 62.6 MHz bandwidth. Figure 15 shows the output spectrum for various center frequencies and demonstrates notch tunability. A peak SNR of 84.9 dB, 81.2 dB and 80.5 dB was obtained corresponding to notch positions of 62.5 MHz, 41.4 MHz and 24.4 MHz respectively.
Acknowledgment Figure 13. The architecture of a fully differential realization of a second order 1-bit output bandpass AZ modulator.
Acknowledgment Figure 13. The architecture of a fully differential realization of a second order 1-bit output bandpass AZ modulator.
Figure 2: Transferred-Substrate HBT; cross-section, SEM, and performance
Figure 2: Transferred-Substrate HBT; cross-section, SEM, and performance
Figure 3: 650 GHz SRTD oscillators; device, oscillator, and array
Figure 3: 650 GHz SRTD oscillators; device, oscillator, and array
Figure 1. Polarization-gate setup. Our scheme employs a polarization-gate beam geometry identical to that commonly used for optical switching using third-order optical-Kerr media [11]; however, a second-order type II BBO crystal replaces the third-order nonlinear-optical medium (Figure 1). The optical switch demonstrated here yields the same effect as the third-order Kerr gate — rotating the input beam polarization when a 45-degree-polarized gating pulse is present — and therefore shares its advantages of simple arrangement, collinear input and signal beams, and non-interferometric interaction. It is distinguished from the third-order optical-Kerr effect, however, by a strong angular dependence of the switching efficiency (we found an acceptance angle of
Figure 1. Polarization-gate setup. Our scheme employs a polarization-gate beam geometry identical to that commonly used for optical switching using third-order optical-Kerr media [11]; however, a second-order type II BBO crystal replaces the third-order nonlinear-optical medium (Figure 1). The optical switch demonstrated here yields the same effect as the third-order Kerr gate — rotating the input beam polarization when a 45-degree-polarized gating pulse is present — and therefore shares its advantages of simple arrangement, collinear input and signal beams, and non-interferometric interaction. It is distinguished from the third-order optical-Kerr effect, however, by a strong angular dependence of the switching efficiency (we found an acceptance angle of
Figure 2a. Schematic of the two second-order processes contributing to the switching process at relatively low intensities. For very high peak intensities of the gate beam, however, a different CSN process amplifies the signal produced by the above CSN process (Figure 2b), which allows switching efficiencies even larger than 100% by parametric amplification. In this process, at first horizontally polarized second-harmonic light is produced with type II phase matching solely by the two polarization components of the gate beam. Subsequently, a self-diffracted beam is generated at the fundamental frequency by down conversion of the second-harmonic light and the vertically polarized signal light from the cascaded second-order process described above. Finally, the self-diffracted beam interacts with second-harmonic light and produces more vertically polarized light at the fundamental frequency and in the direction of the signal by down conversion.
Figure 2a. Schematic of the two second-order processes contributing to the switching process at relatively low intensities. For very high peak intensities of the gate beam, however, a different CSN process amplifies the signal produced by the above CSN process (Figure 2b), which allows switching efficiencies even larger than 100% by parametric amplification. In this process, at first horizontally polarized second-harmonic light is produced with type II phase matching solely by the two polarization components of the gate beam. Subsequently, a self-diffracted beam is generated at the fundamental frequency by down conversion of the second-harmonic light and the vertically polarized signal light from the cascaded second-order process described above. Finally, the self-diffracted beam interacts with second-harmonic light and produces more vertically polarized light at the fundamental frequency and in the direction of the signal by down conversion.
Figure 2b. Schematic of the three second-order processes contributing to the amplified switching process at high peak intensities of the gate beam.
Figure 2b. Schematic of the three second-order processes contributing to the amplified switching process at high peak intensities of the gate beam.
Figure 3. Measured switching efficiency vs. gate-beam intensity for two different crystal lengths. Figure 3 shows the switching efficiency 1, defined as the ratio between the intensity of the vertically polarized signal beam and the intensity of the horizontally polarized probe beam, as a function of the peak intensity of the gate beam. For small gate intensities (first two data points), a dependence of 1 on he square of the gate intensity is observed, characteristic for second-order processes. For higher gate intensities (next two data points) the increase of n is at first slower than expected from a square dependence, which can be attributed to depletion effects in the crystal. Finally, for very large intensities of the gate beam (last point), a strong increase of the switching efficiency is observed, which is caused by the self-diffraction process with parametric amplification described above. The highest observed efficiency was 320% for a crystal length of 1.0 mm and 60% for a crystal length of 0.5 mm, using a gate-pulse peak intensity of 800 GW/cm*. The 1-kHz train pulses were obtained from a regeneratively-amplified Ti:Sapphire laser operating near 850 nm. The pulses had a length of about 250 fs and were focussed to a spot size (FWHM) of about 200 um in the BBO crystal.
Figure 3. Measured switching efficiency vs. gate-beam intensity for two different crystal lengths. Figure 3 shows the switching efficiency 1, defined as the ratio between the intensity of the vertically polarized signal beam and the intensity of the horizontally polarized probe beam, as a function of the peak intensity of the gate beam. For small gate intensities (first two data points), a dependence of 1 on he square of the gate intensity is observed, characteristic for second-order processes. For higher gate intensities (next two data points) the increase of n is at first slower than expected from a square dependence, which can be attributed to depletion effects in the crystal. Finally, for very large intensities of the gate beam (last point), a strong increase of the switching efficiency is observed, which is caused by the self-diffraction process with parametric amplification described above. The highest observed efficiency was 320% for a crystal length of 1.0 mm and 60% for a crystal length of 0.5 mm, using a gate-pulse peak intensity of 800 GW/cm*. The 1-kHz train pulses were obtained from a regeneratively-amplified Ti:Sapphire laser operating near 850 nm. The pulses had a length of about 250 fs and were focussed to a spot size (FWHM) of about 200 um in the BBO crystal.
Figure 5. PG-FROG trace measured by spectrally resolving the signal in the polarization-gate geometry. (Contour lines at 5%, 10%, 20%, 40%, 60%, and 80% of the maximum intensity.) pulse obtained from both geometries, showing the good agreement between the two independent measurements
Figure 5. PG-FROG trace measured by spectrally resolving the signal in the polarization-gate geometry. (Contour lines at 5%, 10%, 20%, 40%, 60%, and 80% of the maximum intensity.) pulse obtained from both geometries, showing the good agreement between the two independent measurements
Figure 6. Spectral intensity and phase of the amplified laser pulses measured by spectrally resolving the signal in the polarization-gate setup (PG) and the related self-diffracted signal (SD).
Figure 6. Spectral intensity and phase of the amplified laser pulses measured by spectrally resolving the signal in the polarization-gate setup (PG) and the related self-diffracted signal (SD).
Figure 1: Illustrative calculation of MI-FROG trace. Time is in units of pump pulse widths with the probe two units wide, the gate 0.33 units wide, and A = 6. No phase differ- ence exists between the two individual probe pulses.
Figure 1: Illustrative calculation of MI-FROG trace. Time is in units of pump pulse widths with the probe two units wide, the gate 0.33 units wide, and A = 6. No phase differ- ence exists between the two individual probe pulses.
Figure 2: Illustrative calculation of MI-FROG trace. Time is in units of pump pulse widths with the probe two units wide, the gate 0.33 units wide, and A = 6. A linear phase difference, with negative slope, exists between the two indi- vidual probe pulses, resulting in a blueshift of the trailing pulse.
Figure 2: Illustrative calculation of MI-FROG trace. Time is in units of pump pulse widths with the probe two units wide, the gate 0.33 units wide, and A = 6. A linear phase difference, with negative slope, exists between the two indi- vidual probe pulses, resulting in a blueshift of the trailing pulse.
Figure 3: Illustrative calculation of MI-FROG trace. Time is in units of pump pulse widths with the probe two units wide, the gate 0.33 units wide, and A = 6. A cross-phase modulation from a copropagating pump pulse exists be- tween the two individual probe pulses, resulting in a red- shift /blueshift on the leading/trailing edge the second pulse.
Figure 3: Illustrative calculation of MI-FROG trace. Time is in units of pump pulse widths with the probe two units wide, the gate 0.33 units wide, and A = 6. A cross-phase modulation from a copropagating pump pulse exists be- tween the two individual probe pulses, resulting in a red- shift /blueshift on the leading/trailing edge the second pulse.
Figure 5: Raw MI-FROG (bottom row, central feature only) and trailing-pulse TREEFROG traces measured with no phase shift (no pump) for +100 fs (left column), 0 fs (center column), and -100 fs (right column) pump- probe delay.
Figure 5: Raw MI-FROG (bottom row, central feature only) and trailing-pulse TREEFROG traces measured with no phase shift (no pump) for +100 fs (left column), 0 fs (center column), and -100 fs (right column) pump- probe delay.
Figure 4: Raw CCD frame containing the power spec- trum of input pulse (far left), and MI-FROG trace. The power spectrum of the input pulse, measured in the same spectrometer independently from and simultane- ously with the FROG/MI-FROG traces, is used in the interative phase recovery analysis. Note the fringes on the individual side features of the MI-FROG trace.
Figure 4: Raw CCD frame containing the power spec- trum of input pulse (far left), and MI-FROG trace. The power spectrum of the input pulse, measured in the same spectrometer independently from and simultane- ously with the FROG/MI-FROG traces, is used in the interative phase recovery analysis. Note the fringes on the individual side features of the MI-FROG trace.
Figure 6: Raw MI-FROG (bottom row, central feature only) and trailing-pulse TREEFROG traces measured in the case of cross-phase modulation in fused silica between a 400 nm pump pulse and 800 nm probe for +100 fs (left column), 0 fs (center column), and -100 fs (right column) pump-probe delay.
Figure 6: Raw MI-FROG (bottom row, central feature only) and trailing-pulse TREEFROG traces measured in the case of cross-phase modulation in fused silica between a 400 nm pump pulse and 800 nm probe for +100 fs (left column), 0 fs (center column), and -100 fs (right column) pump-probe delay.
Figure 7: Results of MI-FROG interferogram analy- sis and iterative phase recovery. For each time delay, the intensity profiles of the unpumped, |P(¢t ? , and pumped, |Q(t |? , probe pulses are plotted, along with their relative phase difference, Ad(t). The intensity profile of the gate pulse, iG(t)|’, is, for clarity, shown only for the zero delay case. Intensity profiles are unit normalized and the centroid of |P(t)|? is used to define t = 0. No other adjustable parameters are used.
Figure 7: Results of MI-FROG interferogram analy- sis and iterative phase recovery. For each time delay, the intensity profiles of the unpumped, |P(¢t ? , and pumped, |Q(t |? , probe pulses are plotted, along with their relative phase difference, Ad(t). The intensity profile of the gate pulse, iG(t)|’, is, for clarity, shown only for the zero delay case. Intensity profiles are unit normalized and the centroid of |P(t)|? is used to define t = 0. No other adjustable parameters are used.
Figure 8: Time-resolved coherent spectroscopy of ul- trafast ionization in room air. Greyscale in this figure represents the magnitude of the frequency shift, with black representing +0.5 THz shift. The dark diagonal band (corresponding to the ionization front) and the area above it is a region of blueshift, while the region in the lower left, be ow the diagonal, are redshifts.
Figure 8: Time-resolved coherent spectroscopy of ul- trafast ionization in room air. Greyscale in this figure represents the magnitude of the frequency shift, with black representing +0.5 THz shift. The dark diagonal band (corresponding to the ionization front) and the area above it is a region of blueshift, while the region in the lower left, be ow the diagonal, are redshifts.
Figure 1. (a) Principle of optical signal measurement based on electro-absorption sampling. (b) Relationship between optical absorption (light output) and applied voltage.
Figure 1. (a) Principle of optical signal measurement based on electro-absorption sampling. (b) Relationship between optical absorption (light output) and applied voltage.
Figure 2. Characteristics of the EA modulator module used for the experiments. (a) Photograph of the device module. (b) Transmission characteristics. (c) Frequency response.
Figure 2. Characteristics of the EA modulator module used for the experiments. (a) Photograph of the device module. (b) Transmission characteristics. (c) Frequency response.
Figure 4. Experimental setup for optical signal measurement using electro-absorption modulator as a sampling gate.
Figure 4. Experimental setup for optical signal measurement using electro-absorption modulator as a sampling gate.
Figure 3. (a) Set up for gating time measurement. (b) Measured gating characteristics of the EA modulator. This includes photodetector and sampling oscilloscope responses.
Figure 3. (a) Set up for gating time measurement. (b) Measured gating characteristics of the EA modulator. This includes photodetector and sampling oscilloscope responses.
Figure 5. Measured optical pulse pattern signals at 10 Gb/s. (a) fixed (1010...) signal, (b) RZ eye- diagram signal, (c) NRZ eye-diagram signal.
Figure 5. Measured optical pulse pattern signals at 10 Gb/s. (a) fixed (1010...) signal, (b) RZ eye- diagram signal, (c) NRZ eye-diagram signal.
Figure 6. Scheme to improve the bandwidth. Effective gating time Tg can be controlled by adjusting the electrical delay 1.
Figure 6. Scheme to improve the bandwidth. Effective gating time Tg can be controlled by adjusting the electrical delay 1.
We consider the example in which a square wave is applied to the slave laser PZT at a scanning frequency of fs. Then the cavity length mismatch is Figure 1. Diagram of scanning dual-laser system. Scanning is achieved by summing the stabilization signal from the PLL with the output of a signal generator.
We consider the example in which a square wave is applied to the slave laser PZT at a scanning frequency of fs. Then the cavity length mismatch is Figure 1. Diagram of scanning dual-laser system. Scanning is achieved by summing the stabilization signal from the PLL with the output of a signal generator.
Figure 2. Master and slave laser pulses during scanning of 4.6 MHz dual laser system with Tax = 10 nsec, fi = 30 Hz. (1-sec exposure). The scanning laser system discussed here consists of a pair of passively modelocked, Erbium- doped fiber lasers. Each has a repetition frequency of Vo = 4.6 MHz (we have also constructed a dual laser system with vo = 50 MHz). In order to minimize the relative jitter and timing drift, these lasers were wound on the same spool.® The pulses from the slave laser are amplified in an Erbium- doped fiber amplifier (EDFA). Various scan ranges and scan frequencies can be easily obtained by adjusting the signal generator in a straightforward way. Figure 2 shows pulses from the master and slave lasers during scanning. With the chosen scanning voltage applied, a scan range of Tinar = 10 nsec was obtained at a scan frequency of f, = 30 Hz. For clarity, the center of the scan range has been adjusted so that the pulses in Fig. 2 never overlap. However, it is a simple matter to adjust the RF phase setting so that pulse coincidence (ATp{t) = 0) occurs
Figure 2. Master and slave laser pulses during scanning of 4.6 MHz dual laser system with Tax = 10 nsec, fi = 30 Hz. (1-sec exposure). The scanning laser system discussed here consists of a pair of passively modelocked, Erbium- doped fiber lasers. Each has a repetition frequency of Vo = 4.6 MHz (we have also constructed a dual laser system with vo = 50 MHz). In order to minimize the relative jitter and timing drift, these lasers were wound on the same spool.® The pulses from the slave laser are amplified in an Erbium- doped fiber amplifier (EDFA). Various scan ranges and scan frequencies can be easily obtained by adjusting the signal generator in a straightforward way. Figure 2 shows pulses from the master and slave lasers during scanning. With the chosen scanning voltage applied, a scan range of Tinar = 10 nsec was obtained at a scan frequency of f, = 30 Hz. For clarity, the center of the scan range has been adjusted so that the pulses in Fig. 2 never overlap. However, it is a simple matter to adjust the RF phase setting so that pulse coincidence (ATp{t) = 0) occurs
Figure 3 Applied waveform (dashed) and resulting time delay waveforms (solid) as measured by monitoring the phase detector error voltage (V..). Scan frequency is 100 Hz.
Figure 3 Applied waveform (dashed) and resulting time delay waveforms (solid) as measured by monitoring the phase detector error voltage (V..). Scan frequency is 100 Hz.
Figure 4. Pump-probe measurement using scanning laser system and timing calibration method. A Fabry-Perot etalon (FP) generates the pulse train for timing calibration.
Figure 4. Pump-probe measurement using scanning laser system and timing calibration method. A Fabry-Perot etalon (FP) generates the pulse train for timing calibration.
Figure 5. Transient absorption near the band-edge of intrinsic InGaAs. The 40 psec interval near time-zero (inset) shows an initial fast transient (~1 psec). while simultaneously accumulating and averaging data from the pump-probe experiment on the Y- channel. In our simple demonstration experiment, the first cross-correlation peak is used to trigger the oscilloscope. Figure 5 is a data trace from such a measurement over a 10 nsec time interval (scanned at 30 Hz).
Figure 5. Transient absorption near the band-edge of intrinsic InGaAs. The 40 psec interval near time-zero (inset) shows an initial fast transient (~1 psec). while simultaneously accumulating and averaging data from the pump-probe experiment on the Y- channel. In our simple demonstration experiment, the first cross-correlation peak is used to trigger the oscilloscope. Figure 5 is a data trace from such a measurement over a 10 nsec time interval (scanned at 30 Hz).
Figure 1. Scanning electron microscope picture of the epitaxial lift-off probe. The probe is 130 by 230 pum (at its widest points) by | um and consists of a Ti/Au interdigital MSM switch with fingersize and spacing of 1.5 um. The total switch area is about 30x30 um?. bulk substrate required a metallization process and several etching steps [2]. The teardrop shape of the LT GaAs layer (Fig.1) was formed by a phosphoric acid mesa wet-etch through the LT GaAs layer. A lift-off process was then used to deposit 500 A 3000 A of Ti/Au for the interdigital metal-semiconductor-metal (MSM) PC switch, electrode and tip. Lapping and backside etching, were used to release the probe from the GaAs bulk substrate. Ammonium hydroxide and hydrofluoric acid were used to etch the GaAs substrate and Alos;GajsAs etch-stop layer, respectively, while the front side was protected by black wax. Xylene was then used to dissolve the black wax separating the probes. A 5 x 5 mm’ die area produced 500 probes with a yield of 90%.
Figure 1. Scanning electron microscope picture of the epitaxial lift-off probe. The probe is 130 by 230 pum (at its widest points) by | um and consists of a Ti/Au interdigital MSM switch with fingersize and spacing of 1.5 um. The total switch area is about 30x30 um?. bulk substrate required a metallization process and several etching steps [2]. The teardrop shape of the LT GaAs layer (Fig.1) was formed by a phosphoric acid mesa wet-etch through the LT GaAs layer. A lift-off process was then used to deposit 500 A 3000 A of Ti/Au for the interdigital metal-semiconductor-metal (MSM) PC switch, electrode and tip. Lapping and backside etching, were used to release the probe from the GaAs bulk substrate. Ammonium hydroxide and hydrofluoric acid were used to etch the GaAs substrate and Alos;GajsAs etch-stop layer, respectively, while the front side was protected by black wax. Xylene was then used to dissolve the black wax separating the probes. A 5 x 5 mm’ die area produced 500 probes with a yield of 90%.
Figure 2. Photoconductive sampling setup with the probe contacting a coplanar transmission line. The JFET source follower is connected to the probe via silver epoxy. contact simply by observing the reflection from the bending probe tip.
Figure 2. Photoconductive sampling setup with the probe contacting a coplanar transmission line. The JFET source follower is connected to the probe via silver epoxy. contact simply by observing the reflection from the bending probe tip.
Figure 3. Temporal waveform resolved using the LT GaAs PC probe with integrated JFET source follower. The noise level measured at the baseline near t=0 ps is 15 nV/(Hz)!”. amplifier. This capacitance limits the modulation frequency of the beam directed at the transmission line gap to about | to 2 kHz. Typically, the sensitivity of photoconductive sampling below 1-kHz modulation frequency is several uV/(Hz)'. By introducing the amplifier with a smaller 3- pF capacitance, the charging time was reduced, increasing the allowed modulation frequency to tens of kHz. The important advantage of a high modulation frequency is that the 1/f noise due to gating laser fluctuations can be reduced [5]. However, above a specific frequency value (modulation frequency bandwidth) the measured signal falls below the actual signal under test.
Figure 3. Temporal waveform resolved using the LT GaAs PC probe with integrated JFET source follower. The noise level measured at the baseline near t=0 ps is 15 nV/(Hz)!”. amplifier. This capacitance limits the modulation frequency of the beam directed at the transmission line gap to about | to 2 kHz. Typically, the sensitivity of photoconductive sampling below 1-kHz modulation frequency is several uV/(Hz)'. By introducing the amplifier with a smaller 3- pF capacitance, the charging time was reduced, increasing the allowed modulation frequency to tens of kHz. The important advantage of a high modulation frequency is that the 1/f noise due to gating laser fluctuations can be reduced [5]. However, above a specific frequency value (modulation frequency bandwidth) the measured signal falls below the actual signal under test.
Figure 4. Lumped-element circuit diagram for sampling through passivation. o+ and o- correspond to the positive and negative induced charges.
Figure 4. Lumped-element circuit diagram for sampling through passivation. o+ and o- correspond to the positive and negative induced charges.
Figure 5. Normalized time-domain wavetorms measured through a passivation layer. The PC probe used a 2-mW gating laser power. The a-o modulator was set to 50 kHz. By deconvolving the measured and transmission line signals the response of the probe was found to be 3.5 ps. At 50 kHz the measured signal by the lock-in amplifier is about half of the actual signal on the transmission line. The sensitivity or the noise is 15 nV/(Hz)'” which is found by taking the root mean square value calculated from the baseline. Therefore, the minimum detectable signal is 30 nV/(Hz)'?
Figure 5. Normalized time-domain wavetorms measured through a passivation layer. The PC probe used a 2-mW gating laser power. The a-o modulator was set to 50 kHz. By deconvolving the measured and transmission line signals the response of the probe was found to be 3.5 ps. At 50 kHz the measured signal by the lock-in amplifier is about half of the actual signal on the transmission line. The sensitivity or the noise is 15 nV/(Hz)'” which is found by taking the root mean square value calculated from the baseline. Therefore, the minimum detectable signal is 30 nV/(Hz)'?
While near-field scanning optical microscopy (NSOM) has attracted much attention for localized probing[6-8], sub- wavelength confinement of visible light was preceded by microwave demonstrations[9], and a wide range of contrast mechanisms in semiconducting, superconducting and biological materials and systems awaits exploration at the level of nanometer spatial resolution in the spectrum lying between DC and visible light[10]. Figure 1: 100 100 pm topographical scan of NLTL output section with the time-domain waveform of one 1.400 GHz cycle measured at the diode indicated by the arrow; the 10- 90 % falltime of the steepest edge is 30 ps, and the calibrated voltage scale is on the right. Incomplete contact metal liftoff can also be seen; this is otherwise difficult to detect under optical illumination. Reproduced with permission from Ref. [11}. Copyright 1997, American Institute of Physics.
While near-field scanning optical microscopy (NSOM) has attracted much attention for localized probing[6-8], sub- wavelength confinement of visible light was preceded by microwave demonstrations[9], and a wide range of contrast mechanisms in semiconducting, superconducting and biological materials and systems awaits exploration at the level of nanometer spatial resolution in the spectrum lying between DC and visible light[10]. Figure 1: 100 100 pm topographical scan of NLTL output section with the time-domain waveform of one 1.400 GHz cycle measured at the diode indicated by the arrow; the 10- 90 % falltime of the steepest edge is 30 ps, and the calibrated voltage scale is on the right. Incomplete contact metal liftoff can also be seen; this is otherwise difficult to detect under optical illumination. Reproduced with permission from Ref. [11}. Copyright 1997, American Institute of Physics.
Using a micromanipulator with a sharp-tipped tool, excess gold at the base of the cantilever can be removed. While this reduces the area for attaching the output wire, making attachment more difficult, it substantially reduces the capacitance between the metal and the conducting silicon body, in this case to ~ 30 fF, the capacitance due to the half-cylinder of metal on the cantilever alone. Using silver epoxy, 0.86 mm (outer diameter) semi-rigid coaxial cable[22] is anchored at the side of the probe body, minimizing its profile. An 18pm diameter gold bond wire is epoxied to the coax center conductor, then to the gold-coated base of the cantilever, then looped back on itself to the coax center conductor again, halving its inductance (Fig. 2). Approximately 50 mm of the coaxial cable extends from the mounting point to facilitate connection to a longer but lower-loss cable running to the signal measurement equipment. yo
Using a micromanipulator with a sharp-tipped tool, excess gold at the base of the cantilever can be removed. While this reduces the area for attaching the output wire, making attachment more difficult, it substantially reduces the capacitance between the metal and the conducting silicon body, in this case to ~ 30 fF, the capacitance due to the half-cylinder of metal on the cantilever alone. Using silver epoxy, 0.86 mm (outer diameter) semi-rigid coaxial cable[22] is anchored at the side of the probe body, minimizing its profile. An 18pm diameter gold bond wire is epoxied to the coax center conductor, then to the gold-coated base of the cantilever, then looped back on itself to the coax center conductor again, halving its inductance (Fig. 2). Approximately 50 mm of the coaxial cable extends from the mounting point to facilitate connection to a longer but lower-loss cable running to the signal measurement equipment. yo
The sample, a NLiL on GaAs, 1S a nign-impedance coplanar waveguide transmission line periodically loaded with reverse-biased Schottky diodes (inset, Fig. 3) to compress an input sinusoid into a sawtooth waveform having a pico- or sub-picosecond edge—here a falling edge due to the polarity of the diodes on the structure. To approximate the switching waveforms expected from advanced computational structures in the low GHz regime, we drove the NLTL with 1.400-2.000 GHz signals from a Hewlett-Packard 83620A microwave synthesizer, amplified by an HP 83020A power amplifier. The waveform was measured on an HP54750A 50GHz — sampling oscilloscope, triggered by another 83620A synthesizer because it cannot trigger on the high-frequency NLTL drive. The oscilloscope would trigger at lower-frequency drives (down to 200MHz, the low-end limit of the 83020A) but waveform compression on the NLTL was minimal.
The sample, a NLiL on GaAs, 1S a nign-impedance coplanar waveguide transmission line periodically loaded with reverse-biased Schottky diodes (inset, Fig. 3) to compress an input sinusoid into a sawtooth waveform having a pico- or sub-picosecond edge—here a falling edge due to the polarity of the diodes on the structure. To approximate the switching waveforms expected from advanced computational structures in the low GHz regime, we drove the NLTL with 1.400-2.000 GHz signals from a Hewlett-Packard 83620A microwave synthesizer, amplified by an HP 83020A power amplifier. The waveform was measured on an HP54750A 50GHz — sampling oscilloscope, triggered by another 83620A synthesizer because it cannot trigger on the high-frequency NLTL drive. The oscilloscope would trigger at lower-frequency drives (down to 200MHz, the low-end limit of the 83020A) but waveform compression on the NLTL was minimal.
As shown in Fig. 3, we measured the expected spatial nonlinearity along the NLTL, with a progressively steeper falling edge moving along the line; while three sample traces are shown, the spatial resolution of the field could be observed through measurable variations in the waveform at < 0.1 jim steps of the probe moving along the line. A new observation afforded by this measurement was that, although the diodes are approximately equidistant from one another, the first one (Diode A) had a much slower waveform than its successors, probably due to a signal reflection from the 90° bend in the transmission line just before it. To verify this, one could perform a high- resolution scanning field measurement of the bend area and compare it with a full-wave field simulation: Fourier- transforming the time-domain measurements or simply connecting a microwave spectrum analyzer to the probe would quickly collect this information. Another new observation was that the metal liftoff process used to make diode contacts on the NLTL was incomplete on this sample (Fig. 1). This is difficult to see under normal illumination in a high-power optical microscope since the gold contacts give nearly uniform reflection. Such direct local diagnostics of very fast, very dense IC’s are now possible with this new probing technique. Figure 4: Fourier transform of waveform from Diode C (Fig. 3) and equivalent-circuit model of tip, cantilever, bond wires and output cabling. Approximate values are Cp = 40 fF, Ream = 102, Com = 30 fF, L = 0.5 nH, Coan = Coapo = 10 pF, Ry, = 3 @.
As shown in Fig. 3, we measured the expected spatial nonlinearity along the NLTL, with a progressively steeper falling edge moving along the line; while three sample traces are shown, the spatial resolution of the field could be observed through measurable variations in the waveform at < 0.1 jim steps of the probe moving along the line. A new observation afforded by this measurement was that, although the diodes are approximately equidistant from one another, the first one (Diode A) had a much slower waveform than its successors, probably due to a signal reflection from the 90° bend in the transmission line just before it. To verify this, one could perform a high- resolution scanning field measurement of the bend area and compare it with a full-wave field simulation: Fourier- transforming the time-domain measurements or simply connecting a microwave spectrum analyzer to the probe would quickly collect this information. Another new observation was that the metal liftoff process used to make diode contacts on the NLTL was incomplete on this sample (Fig. 1). This is difficult to see under normal illumination in a high-power optical microscope since the gold contacts give nearly uniform reflection. Such direct local diagnostics of very fast, very dense IC’s are now possible with this new probing technique. Figure 4: Fourier transform of waveform from Diode C (Fig. 3) and equivalent-circuit model of tip, cantilever, bond wires and output cabling. Approximate values are Cp = 40 fF, Ream = 102, Com = 30 fF, L = 0.5 nH, Coan = Coapo = 10 pF, Ry, = 3 @.
Figure |. Schematic of system setup. One laser beam generates electrical pulse train. The other beam turns on optical switch on the probe. Figure | shows the system setup. The probe is supported by piezoelectric actuators in the x-, y-, and z-directions. The probe has a photo-conductive semiconductor switch (PCSS) on its thin-film cantilever[9]. One of the two electrodes of the PCSS is a sharp probe tip made of Pt, and
Figure |. Schematic of system setup. One laser beam generates electrical pulse train. The other beam turns on optical switch on the probe. Figure | shows the system setup. The probe is supported by piezoelectric actuators in the x-, y-, and z-directions. The probe has a photo-conductive semiconductor switch (PCSS) on its thin-film cantilever[9]. One of the two electrodes of the PCSS is a sharp probe tip made of Pt, and
Figure 2. Top view of sample device. Two rectangles show the area ohserved hv SFOEM The cantilever is made of low temperature grown GaAs (LT-GaAs) and the material in the gap region between the two electrodes works as the PCSS [9]. An ultra-short optical pulse train from a mode locked Ti sapphire laser is used to turn the PCSS on and off. Since photo-excited carriers in LT-GaAs have an extremely short life (10-11), the response time of the PCSS is about I ps or shorter.
Figure 2. Top view of sample device. Two rectangles show the area ohserved hv SFOEM The cantilever is made of low temperature grown GaAs (LT-GaAs) and the material in the gap region between the two electrodes works as the PCSS [9]. An ultra-short optical pulse train from a mode locked Ti sapphire laser is used to turn the PCSS on and off. Since photo-excited carriers in LT-GaAs have an extremely short life (10-11), the response time of the PCSS is about I ps or shorter.
Figure 3. Bottom view of probe. Cantilever and pyramidal protrusion are made of LT-GaAs thin film In the first step of the probe fabrication, a Si pyramidal protrusion was made by anisotropic etching with a KOH solution under masking with a square SiQ2 pattern. Since the etching rate of a Si (111) surface is very slow, a pyramidal protrusion surrounded by (111) and its equivalent surfaces are automatically fabricated. At this stage the tip radius was smaller than 100 nm. Although a normal SFM
Figure 3. Bottom view of probe. Cantilever and pyramidal protrusion are made of LT-GaAs thin film In the first step of the probe fabrication, a Si pyramidal protrusion was made by anisotropic etching with a KOH solution under masking with a square SiQ2 pattern. Since the etching rate of a Si (111) surface is very slow, a pyramidal protrusion surrounded by (111) and its equivalent surfaces are automatically fabricated. At this stage the tip radius was smaller than 100 nm. Although a normal SFM
the PCSS becomes comparable above several hundreds GHz,
the PCSS becomes comparable above several hundreds GHz,
Figure 5. Simultaneously bserved images of interdigit coupler. Left is topographic image. Right is potential image. In the potential image, smal] waves with a short period have no meaning because they are due to 50 Hz noise from a power line. The upper electrode has a higher potentia and the lower one has a lower potential. In the GaAs substrate area between metal fingers, the measured potentia has no meaning. Because the substrate is an insulator, the charge sampled at the previous metallic point is maintained for a short time. In this measurement, the probe scans from left to right. Therefore, the substrate region of the potential image in Figure 5 seems as if it has the same potential as the left metal finger. The SFOEM is able to move the probe position anywhere in the observed area. We measured the time evolution of the potential at the hree points labeled A, B, and C in Figure 5. Figure 6 shows the results. The B curve seems to include the differential component of curve A. This is reasonable because the two electrodes have capacitive coupling. The C curve is almost similar to B. However, C is slightly
Figure 5. Simultaneously bserved images of interdigit coupler. Left is topographic image. Right is potential image. In the potential image, smal] waves with a short period have no meaning because they are due to 50 Hz noise from a power line. The upper electrode has a higher potentia and the lower one has a lower potential. In the GaAs substrate area between metal fingers, the measured potentia has no meaning. Because the substrate is an insulator, the charge sampled at the previous metallic point is maintained for a short time. In this measurement, the probe scans from left to right. Therefore, the substrate region of the potential image in Figure 5 seems as if it has the same potential as the left metal finger. The SFOEM is able to move the probe position anywhere in the observed area. We measured the time evolution of the potential at the hree points labeled A, B, and C in Figure 5. Figure 6 shows the results. The B curve seems to include the differential component of curve A. This is reasonable because the two electrodes have capacitive coupling. The C curve is almost similar to B. However, C is slightly
Figure 6. Time evolution of pulse voltage measured at fixed point. Each label corresponds to the position with the same letter in Fig. 5.
Figure 6. Time evolution of pulse voltage measured at fixed point. Each label corresponds to the position with the same letter in Fig. 5.
high speed avalanche photodiode used exhibits a maximum sensitivity. The microwave synthesizer, the modelocker synthesizer of the laser system and the spectrum analyzer are phase stabilized via a phase locked loop (PLL). For the phase measurements the spectrum analyzer is replaced by a lock-in amplifier and a reference synthesizer. Figure 1. Experimental setup, (a) for the frequency domain measurements, (b) for the time domain measurements
high speed avalanche photodiode used exhibits a maximum sensitivity. The microwave synthesizer, the modelocker synthesizer of the laser system and the spectrum analyzer are phase stabilized via a phase locked loop (PLL). For the phase measurements the spectrum analyzer is replaced by a lock-in amplifier and a reference synthesizer. Figure 1. Experimental setup, (a) for the frequency domain measurements, (b) for the time domain measurements
Figure 2. Sketch of a nonlinear transmission line
Figure 2. Sketch of a nonlinear transmission line
“igure 3. Electro-optic signal of the fundamental microwave at 15 GHz and of its higher harmonics along the center conductor of an NLTL from input to output diodes. As can be seen the amplitude of the fundamental signal decreases in the direction of propagation whereas the amplitude of the higher harmonics increase indicating that they are generated along the transmission line. The obvious standing wave patterns are caused by phase mismatching of the harmonics and additionally by an impedance mismatch at the end of the line.
“igure 3. Electro-optic signal of the fundamental microwave at 15 GHz and of its higher harmonics along the center conductor of an NLTL from input to output diodes. As can be seen the amplitude of the fundamental signal decreases in the direction of propagation whereas the amplitude of the higher harmonics increase indicating that they are generated along the transmission line. The obvious standing wave patterns are caused by phase mismatching of the harmonics and additionally by an impedance mismatch at the end of the line.
Figure 5. Reference coplanar waveguide on semiinsulating substrate; (a) metallization structure with scanned area; (b) result of 2D field mappings at 7 GHz
Figure 5. Reference coplanar waveguide on semiinsulating substrate; (a) metallization structure with scanned area; (b) result of 2D field mappings at 7 GHz
‘igure 6. Transversal distribution of electro-optic signal al positions (a) 1 and (b) 2 as marked in Figs. 4(a) and 5(a) at 7 GHz
‘igure 6. Transversal distribution of electro-optic signal al positions (a) 1 and (b) 2 as marked in Figs. 4(a) and 5(a) at 7 GHz
Figure 7. NLTL consisting of 10 Schottky diodes in a CPW; (a) metallization structure; results of 2D field mappings at (b) 6 GHz, (c) 12 GHz and (d) 18 GHz.
Figure 7. NLTL consisting of 10 Schottky diodes in a CPW; (a) metallization structure; results of 2D field mappings at (b) 6 GHz, (c) 12 GHz and (d) 18 GHz.
Fig. 1. Schematic of the planar 7-GHz slot transmitter.
Fig. 1. Schematic of the planar 7-GHz slot transmitter.
Fig. 2. Photograph of a 76.5-GHz SIMMWIC dipole transmitter. The chip size is 2 mm x 3 mm (x/y).
Fig. 2. Photograph of a 76.5-GHz SIMMWIC dipole transmitter. The chip size is 2 mm x 3 mm (x/y).
Fig. 4. Time-resolved electric waveform at 20 um probing height above the IMPATT diode. The solid line is a sine fit to the measured data (circles) indicating periodic signals at 24.42 GHz and 73.26 GHz. The bias current (I) through the IMPATT diode is varied.
Fig. 4. Time-resolved electric waveform at 20 um probing height above the IMPATT diode. The solid line is a sine fit to the measured data (circles) indicating periodic signals at 24.42 GHz and 73.26 GHz. The bias current (I) through the IMPATT diode is varied.
Fig. 5. Amplitudes of the EO signal contribution of the fun- damental synchronization frequency at 24.42 GHz, the parasitic 2nd harmonic at 48.84 GHz, and the desired resonance frequency at 73.26 GHz as a function of the bias current applied to the IMPATT diode.
Fig. 5. Amplitudes of the EO signal contribution of the fun- damental synchronization frequency at 24.42 GHz, the parasitic 2nd harmonic at 48.84 GHz, and the desired resonance frequency at 73.26 GHz as a function of the bias current applied to the IMPATT diode.
Fig. 6. Electrooptically measured amplitude of the electric field distribution in x-direction for 20 um probing height at 24.42 GHz (a) and 73.26 GHz (b).
Fig. 6. Electrooptically measured amplitude of the electric field distribution in x-direction for 20 um probing height at 24.42 GHz (a) and 73.26 GHz (b).
Fig. 7. Amplitude (upper part) and phase (lower part) of the electric wave at 73.26 GHz radiated into free-space and detected with a photoconductive 75-GHz dipole antenna at 2 mm probing height. The 2 mm x 3 mm chip area of the 76.5-GHz dipole transmitter is indicated by the black box in the upper part of the figure.
Fig. 7. Amplitude (upper part) and phase (lower part) of the electric wave at 73.26 GHz radiated into free-space and detected with a photoconductive 75-GHz dipole antenna at 2 mm probing height. The 2 mm x 3 mm chip area of the 76.5-GHz dipole transmitter is indicated by the black box in the upper part of the figure.
Figure 1. Interferometric imager. crystal surface contacting the DUT is highly reflective. A reference mirror is on piezoelectric actuator (I) in the reference “leg”. Reflections from the reference mirror and EO crystal interact with the wedged window a second time, pass through polarizing filter (K) and relay lens (L) to camera (M). An interference pattern is formed by the interaction of both beams in the plane of the CCD.
Figure 1. Interferometric imager. crystal surface contacting the DUT is highly reflective. A reference mirror is on piezoelectric actuator (I) in the reference “leg”. Reflections from the reference mirror and EO crystal interact with the wedged window a second time, pass through polarizing filter (K) and relay lens (L) to camera (M). An interference pattern is formed by the interaction of both beams in the plane of the CCD.
Figure 2. Polarimetric imager.
Figure 2. Polarimetric imager.
- a4 — Pixel sensitivity variations are not truly noise, and can be digitally corrected by calibrating the camera. CCD thermal noise can be reduced by cooling the sensor. The detrimental effects of the remaining noise sources can be reduced by use of modulation.
- a4 — Pixel sensitivity variations are not truly noise, and can be digitally corrected by calibrating the camera. CCD thermal noise can be reduced by cooling the sensor. The detrimental effects of the remaining noise sources can be reduced by use of modulation.
Equation (4a) describes the fraction of electrons in each pixel attributed to the EO effect, Eq. (4b) describes the interferometer's quality, and (4c) defines the system performance as a function of static bias Ty and for a given quality p. The optimum optical bias point at which function f is maximized, T'y°"', is found by optimizing Eq. (4c) with respect to T:
Equation (4a) describes the fraction of electrons in each pixel attributed to the EO effect, Eq. (4b) describes the interferometer's quality, and (4c) defines the system performance as a function of static bias Ty and for a given quality p. The optimum optical bias point at which function f is maximized, T'y°"', is found by optimizing Eq. (4c) with respect to T:
Figure 5. Optimum optical bias point (operating point) To? and sensitivity factor f °" can be found, given quality p. This factor, also shown in Fig. 5, allows us to estimate the maximum attainable system sensitivity. Analysis of the optical system shows numerous sources that degrade interferometer quality (see Ref. [8]). In the remaining discussion, the values calculated for the system in Fig. 1 are p(a=0.94, b=0.085) = 1.044.
Figure 5. Optimum optical bias point (operating point) To? and sensitivity factor f °" can be found, given quality p. This factor, also shown in Fig. 5, allows us to estimate the maximum attainable system sensitivity. Analysis of the optical system shows numerous sources that degrade interferometer quality (see Ref. [8]). In the remaining discussion, the values calculated for the system in Fig. 1 are p(a=0.94, b=0.085) = 1.044.
Figure 7. Maximum system sensitivity occurs when the compensator retardation ~ -7/8 (= -A/16) relative to an intensity null. Note the asymmetry of this plot. One feature evident in Fig. 6 is the presence of image noise. Obvious noise sources include non-uniform iNumination (i.e., diffraction-rings from contaminated optics) and pixel-pixel sensitivity variations (i.e., local peaks and valleys). Modulation of the bias (only) was used during acquisition. Sixteen images were acquired and averaged in approximately 8 seconds to produce this result. No other forms of modulation, sensor cooling, or digital correction were used.
Figure 7. Maximum system sensitivity occurs when the compensator retardation ~ -7/8 (= -A/16) relative to an intensity null. Note the asymmetry of this plot. One feature evident in Fig. 6 is the presence of image noise. Obvious noise sources include non-uniform iNumination (i.e., diffraction-rings from contaminated optics) and pixel-pixel sensitivity variations (i.e., local peaks and valleys). Modulation of the bias (only) was used during acquisition. Sixteen images were acquired and averaged in approximately 8 seconds to produce this result. No other forms of modulation, sensor cooling, or digital correction were used.
Figure 6. 2-D image of coplanar structure (bottom); gap size (dark stripes) = 200m. The super-image is the observed transverse E-field map with a bias of 33 VDC. Our initial experiments use the polarimeter of Fig. 2. Figure 6 shows a 2-D image of a simple coplanar structure (bottom), with gap size = 200 um. The super-image is the observed electric-field map with an applied bias of 33 V DC, and with T, optimized.
Figure 6. 2-D image of coplanar structure (bottom); gap size (dark stripes) = 200m. The super-image is the observed transverse E-field map with a bias of 33 VDC. Our initial experiments use the polarimeter of Fig. 2. Figure 6 shows a 2-D image of a simple coplanar structure (bottom), with gap size = 200 um. The super-image is the observed electric-field map with an applied bias of 33 V DC, and with T, optimized.
Figure 1: Schematic drawing of the coplanar waveguide transmission line fabricated on an ion implanted silicon- on-sapphire wafer. The device consists of a center line with 10 jm wide channels on each side. The black circles show where the pump and probe beams impinge on the device.
Figure 1: Schematic drawing of the coplanar waveguide transmission line fabricated on an ion implanted silicon- on-sapphire wafer. The device consists of a center line with 10 jm wide channels on each side. The black circles show where the pump and probe beams impinge on the device.
Figure 2: Temporal waveform obtained using second harmonic generation. The upper curve corresponds to the generation of an electrical transient by the application of a bias across the pump photoconductor gap. The lower curve is a reference scan obtained in the absence of an where A, B, and C are constants that include all angular dependencies, xy” designates the complex conjugate of x, and L and Q designate linear and quadratic dependencies on the transient electric field, respectively. The SH intensity, therefore, consists of a quiescent term, be, that is constant for a fixed experimental arrangement, a first- order signal term, AI? , that varies linearly with the magnitude of the subpicosecond electrical pulse, and a second-order signal term, AI”, that varies quadratically with the magnitude of the subpicosecond electrical pulse. We note that the magnitude of E>" can be varied by simply rotating the polarization of the obliquely incident probe beam.
Figure 2: Temporal waveform obtained using second harmonic generation. The upper curve corresponds to the generation of an electrical transient by the application of a bias across the pump photoconductor gap. The lower curve is a reference scan obtained in the absence of an where A, B, and C are constants that include all angular dependencies, xy” designates the complex conjugate of x, and L and Q designate linear and quadratic dependencies on the transient electric field, respectively. The SH intensity, therefore, consists of a quiescent term, be, that is constant for a fixed experimental arrangement, a first- order signal term, AI? , that varies linearly with the magnitude of the subpicosecond electrical pulse, and a second-order signal term, AI”, that varies quadratically with the magnitude of the subpicosecond electrical pulse. We note that the magnitude of E>" can be varied by simply rotating the polarization of the obliquely incident probe beam.
Figure 3: The normalized temporal response of the THz field measured by electric field-induced second harmonic generation. For clarity, the contribution of the static SH radiation has been subtracted.
Figure 3: The normalized temporal response of the THz field measured by electric field-induced second harmonic generation. For clarity, the contribution of the static SH radiation has been subtracted.
electro-optic sampling via the linear electro-optic effect (Pockels effect) offers a flat frequency response over an ultrawide bandwidth. Because field detection is purely an electro-optic process, the system bandwidth is mainly limited by either the pulse duration of the probe laser or the lowest TO phonon frequency of the sensor crystal. Further, since electro-optic sampling is purely an optical technique, it does not require electrode contact or wiring on the sensor crystal. Figure 2 shows the details of the sampling setup. Simple tensor analysis indicates that a <110> oriented zincblende crystal as a sensor gives the best sensitivity. The po parallel to the [i,-1 birefringence of the sensor crystal, via an applied electric fie optical probe beam passing through the crystal. The ellipticity modulation of the optical beam can then be po amplitude and phase of the applied electric field. The de arization of the THz beam and optical probe beam are 0] crystal direction. Modulating the d (THz), will modulate the polarization ellipticity of the arization analyzed to provide information on both the ection system will analyze a polarization change from the electro-optic crystal and correlate it with the amplitude
electro-optic sampling via the linear electro-optic effect (Pockels effect) offers a flat frequency response over an ultrawide bandwidth. Because field detection is purely an electro-optic process, the system bandwidth is mainly limited by either the pulse duration of the probe laser or the lowest TO phonon frequency of the sensor crystal. Further, since electro-optic sampling is purely an optical technique, it does not require electrode contact or wiring on the sensor crystal. Figure 2 shows the details of the sampling setup. Simple tensor analysis indicates that a <110> oriented zincblende crystal as a sensor gives the best sensitivity. The po parallel to the [i,-1 birefringence of the sensor crystal, via an applied electric fie optical probe beam passing through the crystal. The ellipticity modulation of the optical beam can then be po amplitude and phase of the applied electric field. The de arization of the THz beam and optical probe beam are 0] crystal direction. Modulating the d (THz), will modulate the polarization ellipticity of the arization analyzed to provide information on both the ection system will analyze a polarization change from the electro-optic crystal and correlate it with the amplitude
Fig. 3: Temporal electro-optic signal of a 1 ps THz pulse measured by a ZnTe sensor. , Figure 3 plots the temporal electro-optic waveform of a | ps THz pulse as measured by a balanced detector after both he THz pulse and optical probe pulse co-propagate hrough the <110> ZnTe sensor crystal. The time delay is provided by changing the relative length of the optical beam path between the THz pulses and the optical probe pulses. Detection sensitivity is significantly improved by increasing the interaction length of pulsed field and optical probe beam within the crystal. The signal-to-noise ratio can exceed 1,000:1 using unfocused beams, 10,000:1 using unamplified focused beams (using a laser oscillator as an optical source), and 100,000:1 using focused amplified beams (using a regenerative amplified laser source) with a ZnTe sensor crystal. Fig. 4 plots the signal and noise spectra, with a SNR>50,000 from 0.1 to 1.2 THz.
Fig. 3: Temporal electro-optic signal of a 1 ps THz pulse measured by a ZnTe sensor. , Figure 3 plots the temporal electro-optic waveform of a | ps THz pulse as measured by a balanced detector after both he THz pulse and optical probe pulse co-propagate hrough the <110> ZnTe sensor crystal. The time delay is provided by changing the relative length of the optical beam path between the THz pulses and the optical probe pulses. Detection sensitivity is significantly improved by increasing the interaction length of pulsed field and optical probe beam within the crystal. The signal-to-noise ratio can exceed 1,000:1 using unfocused beams, 10,000:1 using unamplified focused beams (using a laser oscillator as an optical source), and 100,000:1 using focused amplified beams (using a regenerative amplified laser source) with a ZnTe sensor crystal. Fig. 4 plots the signal and noise spectra, with a SNR>50,000 from 0.1 to 1.2 THz.
where Pp is the output optical probe power with zero field applied to the crystal, and Vz is the half-wave-voltage of the sensor crystal. By measuring Pox from a calibrated voltage source as a function of time delay between the THz pulse and optical probe pulse, the time-resolved sign and amplitude of V can be obtained, and a numerical FFT provides frequency information.
where Pp is the output optical probe power with zero field applied to the crystal, and Vz is the half-wave-voltage of the sensor crystal. By measuring Pox from a calibrated voltage source as a function of time delay between the THz pulse and optical probe pulse, the time-resolved sign and amplitude of V can be obtained, and a numerical FFT provides frequency information.
Fig. 5: Temporal electro-optic signal of 200 fs THz pulse measured by a ZnTe sensor.
Fig. 5: Temporal electro-optic signal of 200 fs THz pulse measured by a ZnTe sensor.
Fig. 7: Electro-optic signal versus optical probe average power. A linear response in both the generation and detection of the THz pulses is crucial. Figures 7 and 8 plot the electro- optic signal versus average optical probe power and peak optical pump power, respectively. Excellent linearity is achieved versus probe power whereas pump-beam saturation has been observed. By increasing the optically illuminated area of the photoconductor on the THz emitter, the total emitted THz power can be made to scale up linearly with illumination area (assuming a nonsaturating optical fluence).
Fig. 7: Electro-optic signal versus optical probe average power. A linear response in both the generation and detection of the THz pulses is crucial. Figures 7 and 8 plot the electro- optic signal versus average optical probe power and peak optical pump power, respectively. Excellent linearity is achieved versus probe power whereas pump-beam saturation has been observed. By increasing the optically illuminated area of the photoconductor on the THz emitter, the total emitted THz power can be made to scale up linearly with illumination area (assuming a nonsaturating optical fluence).
has a pulse duration of 45 fs, and currently the THz signal bandwidth is limited by the GaAs emitter (TO phonon at 8 THz). Fig. 10 is the corresponding frequency spectrum. It shows a 3-dB bandwidth of 3.6 THz and a cutoff frequency near 7 THz. By using GaP as both emitter and sensor, we expect a cutoff frequency greater than 10 THz.
has a pulse duration of 45 fs, and currently the THz signal bandwidth is limited by the GaAs emitter (TO phonon at 8 THz). Fig. 10 is the corresponding frequency spectrum. It shows a 3-dB bandwidth of 3.6 THz and a cutoff frequency near 7 THz. By using GaP as both emitter and sensor, we expect a cutoff frequency greater than 10 THz.
Fig. 10: Amplitude spectrum of a temporal signal shown in Fig. 9. This is the widest bandwidth of a coherent detector ever reported. an ae Fig. 9: Temporal signal from optical rectification measured by a GaP sensor. 53 fs (10% to 90% risetime in ring oscillation) can be well resolved.
Fig. 10: Amplitude spectrum of a temporal signal shown in Fig. 9. This is the widest bandwidth of a coherent detector ever reported. an ae Fig. 9: Temporal signal from optical rectification measured by a GaP sensor. 53 fs (10% to 90% risetime in ring oscillation) can be well resolved.
Fig. 13: Amplitude spectrum of a temporal signal shown in Fig. 13. Fig. 12: Signal from one of the 288x384 CCD pixels (100 time-resolved frames).
Fig. 13: Amplitude spectrum of a temporal signal shown in Fig. 13. Fig. 12: Signal from one of the 288x384 CCD pixels (100 time-resolved frames).
Fig. 11: Setup for the conversion of a THz image into an optical image.
Fig. 11: Setup for the conversion of a THz image into an optical image.
Fig. 15: An image of a focused THz beam at 0.57 THz. Figures 15 and 16 are plots of the spatial distribution (6x6 mm’) of the THz field amplitude imaged near the focal point at 0.57 THz and 1.08 THz, respectively. The pixel size is 22 um, and the focal plane imaging is l-to-1. In the diffraction limit, as can be seen in the figures, components of THz field at a higher frequencies (here 1.08 THz) focus more tightly than those at the low frequencies (here 0.57 THz). Further, the largest THz induced photomodulation was over 50%,
Fig. 15: An image of a focused THz beam at 0.57 THz. Figures 15 and 16 are plots of the spatial distribution (6x6 mm’) of the THz field amplitude imaged near the focal point at 0.57 THz and 1.08 THz, respectively. The pixel size is 22 um, and the focal plane imaging is l-to-1. In the diffraction limit, as can be seen in the figures, components of THz field at a higher frequencies (here 1.08 THz) focus more tightly than those at the low frequencies (here 0.57 THz). Further, the largest THz induced photomodulation was over 50%,
Fig. 14: Temporal signal from 288 pixels. converted into a 2D optical intensity distribution after the readout beam passes through a crossed polarizer, and the optical image is then recorded by a linear diode array or a digital CCD camera. Figures 12 and 13 plot temporal signal and its corresponding FFT from a single CCD pixel. The data in Fig. 12 was taken from 100 sequential frames with a temporal step of 66.6 fs. Fig. 14 shows the temporal signal from a linear diode array (288 pixels).
Fig. 14: Temporal signal from 288 pixels. converted into a 2D optical intensity distribution after the readout beam passes through a crossed polarizer, and the optical image is then recorded by a linear diode array or a digital CCD camera. Figures 12 and 13 plot temporal signal and its corresponding FFT from a single CCD pixel. The data in Fig. 12 was taken from 100 sequential frames with a temporal step of 66.6 fs. Fig. 14 shows the temporal signal from a linear diode array (288 pixels).
Fig. 17: A focal-plane image of the THz field from a quadrupole.
Fig. 17: A focal-plane image of the THz field from a quadrupole.
Fig. 16: An image of a focused THz beam at 1.08 THz. We have imaged the field distribution from a planar quadrupole THz emitter. Figure 17 shows the quadrupole emitter where the center electrode is biased with two side electrodes grounded and the corresponding radiation pattern. The center electrode is | mm wide and has a length over 1 cm. The gaps between the side electrodes are 1.5 mm and 2 mm, respectively. This quadrupole geometry generates two unbalanced dipoles with opposite polarity. The peak field distributions are plotted in 2-D and 3-D. Figure 17 clearly indicates the radiation pattern and the polarity of two opposite dipoles.
Fig. 16: An image of a focused THz beam at 1.08 THz. We have imaged the field distribution from a planar quadrupole THz emitter. Figure 17 shows the quadrupole emitter where the center electrode is biased with two side electrodes grounded and the corresponding radiation pattern. The center electrode is | mm wide and has a length over 1 cm. The gaps between the side electrodes are 1.5 mm and 2 mm, respectively. This quadrupole geometry generates two unbalanced dipoles with opposite polarity. The peak field distributions are plotted in 2-D and 3-D. Figure 17 clearly indicates the radiation pattern and the polarity of two opposite dipoles.
Fig. 18: One of the focal-plane images with a rod sweeping across the THz beam path in the focal plane.
Fig. 18: One of the focal-plane images with a rod sweeping across the THz beam path in the focal plane.
Figure 1: Phase matching angle versus THz frequency for far-infrared generation by collinear type II difference frequency generation in LiNbO,,. occur even when optical dispersion is included (using a visible wavelength pump). Thus, this approach is clearly not optimized for the generation of broadband THz radiation.
Figure 1: Phase matching angle versus THz frequency for far-infrared generation by collinear type II difference frequency generation in LiNbO,,. occur even when optical dispersion is included (using a visible wavelength pump). Thus, this approach is clearly not optimized for the generation of broadband THz radiation.
For 4nie, we utilize the wide bandgap nature of the semiconductor to adjust the amount of optical dispersion in the near infrared. Assuming an optical pump source centered at 800 nm, we have calculated the coherence length for difference frequency mixing and electro-optic sampling. Figure 3 shows this parameter versus THz frequency, with and without consideration of optical dispersion. The peak structure in the solid trace (with consideration of optical dispersion) is due to the optimal matching of the relevant optical and THz velocities. Beyond 2 THz, the coherence length decreases rapidly due to the dispersion of the ZnTe crystals in the FIR. We should note that the frequency dependent coherence length can be readily altered by simply tuning the optical wavelength. Figure 2: Coherence length versus THz frequency for the poled polymer using optical excitation at 740 nm. The dotted line, calculated using Eq. (2), neglects the dispersion at optical frequencies. The solid line, calculated using Eq. (4), includes the effects of dispersion at optical frequencies.
For 4nie, we utilize the wide bandgap nature of the semiconductor to adjust the amount of optical dispersion in the near infrared. Assuming an optical pump source centered at 800 nm, we have calculated the coherence length for difference frequency mixing and electro-optic sampling. Figure 3 shows this parameter versus THz frequency, with and without consideration of optical dispersion. The peak structure in the solid trace (with consideration of optical dispersion) is due to the optimal matching of the relevant optical and THz velocities. Beyond 2 THz, the coherence length decreases rapidly due to the dispersion of the ZnTe crystals in the FIR. We should note that the frequency dependent coherence length can be readily altered by simply tuning the optical wavelength. Figure 2: Coherence length versus THz frequency for the poled polymer using optical excitation at 740 nm. The dotted line, calculated using Eq. (2), neglects the dispersion at optical frequencies. The solid line, calculated using Eq. (4), includes the effects of dispersion at optical frequencies.
Figure 3: Coherence length versus THz frequency for ZnTe using optical excitation at 800 nm. The dotted line, calculated using Eq. (2), neglects the dispersion at optical frequencies. The solid line, calculated using Eq. (4), includes the effects of dispersion at optical frequencies. ZnTe crystal. Using a mode-locked Ti:sapphire laser producing 130 fs pulses at 800 nm, the THz electromagnetic transient produced by optical rectification in the ZnTe crystal was imaged into an identical ZnTe. The total separation between the emitter and detector was 50 cm. The measured field full width at half maximum (FWHM) pulsewidth of the temporal waveform of the THz electric field is approximately 270 fs. The corresponding amplitude spectrum, shown in Figure 4, exhibits a 3 dB bandwidth of ~ 2 THz. This is consistent with the data in Figure 3, where we find that the maximum frequency for obtaining a coherence length exceeding the thickness of the ZnTe crystals is ~2.25 THz. Thus, we would expect a rolloff in sensitivity for higher frequencies from the reduced interaction length. In addition to the limitations imposed by phase matching, the absorption caused by the dominant low-lying vibrational resonance in ZnTe at 5.4 THz [8] is expected to attenuate the high frequency components of the THz radiation.
Figure 3: Coherence length versus THz frequency for ZnTe using optical excitation at 800 nm. The dotted line, calculated using Eq. (2), neglects the dispersion at optical frequencies. The solid line, calculated using Eq. (4), includes the effects of dispersion at optical frequencies. ZnTe crystal. Using a mode-locked Ti:sapphire laser producing 130 fs pulses at 800 nm, the THz electromagnetic transient produced by optical rectification in the ZnTe crystal was imaged into an identical ZnTe. The total separation between the emitter and detector was 50 cm. The measured field full width at half maximum (FWHM) pulsewidth of the temporal waveform of the THz electric field is approximately 270 fs. The corresponding amplitude spectrum, shown in Figure 4, exhibits a 3 dB bandwidth of ~ 2 THz. This is consistent with the data in Figure 3, where we find that the maximum frequency for obtaining a coherence length exceeding the thickness of the ZnTe crystals is ~2.25 THz. Thus, we would expect a rolloff in sensitivity for higher frequencies from the reduced interaction length. In addition to the limitations imposed by phase matching, the absorption caused by the dominant low-lying vibrational resonance in ZnTe at 5.4 THz [8] is expected to attenuate the high frequency components of the THz radiation.
sensitivity beyond 3 THz. The experimental results’ are consistent with the calculations. References:
sensitivity beyond 3 THz. The experimental results’ are consistent with the calculations. References:
Figure 1. The THz ranging system.
Figure 1. The THz ranging system.
Figure 2. (a) Incident pulse on the target, measured by replacing target with a 25.4 mm dia. metal sphere. (b) Amplitude spectrum of pulse in Fig. 2a. (c) Returned signal with no target in beam path. Our procedure to characterize the frequency dependent cross-section of an actual target is as follows. The pulse
Figure 2. (a) Incident pulse on the target, measured by replacing target with a 25.4 mm dia. metal sphere. (b) Amplitude spectrum of pulse in Fig. 2a. (c) Returned signal with no target in beam path. Our procedure to characterize the frequency dependent cross-section of an actual target is as follows. The pulse
Figure 3. (a) Measured time domain reflection (upper curve) compared to calculated reflection (ower curve) from a 1.96 mm dia. fused quartz cylinder at the beam waist (51 cm distance from the deflecting mirrors) of the silicon lens. The orientation of the cylinder axis is perpendicular to the incident electric field polarization. (b) Comparison of measurements (dots) and theory (solid line) on an expanded time scale for pulses B and C. (c) Corresponding measured (open circles) and calculated (solid line) normalized cross section. of the solid diclectric cylinder may be calculated exactly [9], and is shown as the solid line in Fig. 3c. Agreement is good from 0.2 to greater than 0.8 THz. In this figure the calculated cross section has a spectral resolution of 3.6
Figure 3. (a) Measured time domain reflection (upper curve) compared to calculated reflection (ower curve) from a 1.96 mm dia. fused quartz cylinder at the beam waist (51 cm distance from the deflecting mirrors) of the silicon lens. The orientation of the cylinder axis is perpendicular to the incident electric field polarization. (b) Comparison of measurements (dots) and theory (solid line) on an expanded time scale for pulses B and C. (c) Corresponding measured (open circles) and calculated (solid line) normalized cross section. of the solid diclectric cylinder may be calculated exactly [9], and is shown as the solid line in Fig. 3c. Agreement is good from 0.2 to greater than 0.8 THz. In this figure the calculated cross section has a spectral resolution of 3.6
Figure 4. (a) Measured time domain reflection (upper curve) compared to calculated reflection (lower curve) from a 2.82 mm dia. silicon sphere at the beam waist of the silicon lens. (b) Comparison of measurements (dots) and theory (solid line) on an expanded time scale for pulses B anc C. (c) Corresponding measured (open circles) and calculated (solid line) normalized cross section. The scattered radiation from a crystalline silicon sphere 2.82 mm in diameter is shown as the upper curve
Figure 4. (a) Measured time domain reflection (upper curve) compared to calculated reflection (lower curve) from a 2.82 mm dia. silicon sphere at the beam waist of the silicon lens. (b) Comparison of measurements (dots) and theory (solid line) on an expanded time scale for pulses B anc C. (c) Corresponding measured (open circles) and calculated (solid line) normalized cross section. The scattered radiation from a crystalline silicon sphere 2.82 mm in diameter is shown as the upper curve
A schematic of the apparatus used for tomographic T-ray imaging is shown in figure 2. Here, the THz beam is incident on the sample at nearly normal incidence, and comes to a focus at the sample surface. The reflected beam is re-collimated, then captured by a pick-off mirror which directs it to the receiver antenna. For an object with multiple reflecting internal surfaces, the waveform returned from the object consists of a series of replicas of the input pulse, of varying magnitude, polarity, and temporal distortion. This is illustrated using the example of a ball- point pen, with the input and reflected THz waveforms from a single point on the pen shown in figure 3. Figure 2. Schematic of the apparatus used for tomographic T-ray imaging.
A schematic of the apparatus used for tomographic T-ray imaging is shown in figure 2. Here, the THz beam is incident on the sample at nearly normal incidence, and comes to a focus at the sample surface. The reflected beam is re-collimated, then captured by a pick-off mirror which directs it to the receiver antenna. For an object with multiple reflecting internal surfaces, the waveform returned from the object consists of a series of replicas of the input pulse, of varying magnitude, polarity, and temporal distortion. This is illustrated using the example of a ball- point pen, with the input and reflected THz waveforms from a single point on the pen shown in figure 3. Figure 2. Schematic of the apparatus used for tomographic T-ray imaging.
Figure 3. THz waveforms (a) incident on, and (b) reflected from a bic pen. The small oscillations following the main pulse in (a) are a result of residual water vapor in the beam path, and do not significantly affect the measurement. A reflected pulse from each of the eight dielectric interfaces of the outside casing and the inner cylinder holding the ink can be observed in (b).
Figure 3. THz waveforms (a) incident on, and (b) reflected from a bic pen. The small oscillations following the main pulse in (a) are a result of residual water vapor in the beam path, and do not significantly affect the measurement. A reflected pulse from each of the eight dielectric interfaces of the outside casing and the inner cylinder holding the ink can be observed in (b).
Figure 4. THz tomographic image of a bic pen. The vertical axis of the figure is parallel to the long (cylinder) axis of the pen, while the horizontal axis of the figure corresponds to depth into the pen. The THz pulse is incident from the left. The eight dark bands correspond to the eight dielectric interfaces (either from plastic to air or air to plastic) encountered by the pulse. Figure 4 shows a tomographic cross-section of the pen, obtained by evaluating waveforms such as the one in Fig. 3b for points along the length of the ball-point pen. Only very rudimentary signal processing was performed in this example, including the squaring of each waveform and conversion of the result to a grayscale. In figure 4, the vertical axis is parallel to the long (cylinder) axis of the pen, while the horizontal axis corresponds to depth inside the pen. Each interface clearly shows up as a dark stripe.
Figure 4. THz tomographic image of a bic pen. The vertical axis of the figure is parallel to the long (cylinder) axis of the pen, while the horizontal axis of the figure corresponds to depth into the pen. The THz pulse is incident from the left. The eight dark bands correspond to the eight dielectric interfaces (either from plastic to air or air to plastic) encountered by the pulse. Figure 4 shows a tomographic cross-section of the pen, obtained by evaluating waveforms such as the one in Fig. 3b for points along the length of the ball-point pen. Only very rudimentary signal processing was performed in this example, including the squaring of each waveform and conversion of the result to a grayscale. In figure 4, the vertical axis is parallel to the long (cylinder) axis of the pen, while the horizontal axis corresponds to depth inside the pen. Each interface clearly shows up as a dark stripe.
the waveforms returned from its front and back surfaces cannot be distinguished, and appear as a single distorted waveform. Figure Sc shows the waveform of figure 5b, after numerical processing of the input waveform. This involves a filtered Fourier deconvolution of the instrument response (figure 5a), followed by a wavelet filtering to remove noise. This procedure produces a sharp spike at a time delay corresponding to the position of every reflecting interface. Thus, it helps to determine more accurately the positions of the various interfaces. In contrast to figure 5b, the front and back surfaces of the thin (~120 um) magnetic recording material are clearly resolved in the deconvolved data. This is consistent with the expected resolution of L,/2, where L, = 200 lum is the coherence length of the THz pulse in the intervening material. In contrast, when no other reflections are nearby, we find that the position of a reflecting surface can be determined with a precision of only a few microns, as discussed in section 2 above. Figure 5. THz waveforms incident on (a), and reflected from (b), a 3.5” floppy disk. The two surfaces of the front and rear covers are visible, as are the reflections from the thin magnetic recording material. Curve (c) (in arbitrary units) represents the reflected waveform after signal processing. Curve (b) has been scaled up by a factor of four, and the three curves have been offset for clarity.
the waveforms returned from its front and back surfaces cannot be distinguished, and appear as a single distorted waveform. Figure Sc shows the waveform of figure 5b, after numerical processing of the input waveform. This involves a filtered Fourier deconvolution of the instrument response (figure 5a), followed by a wavelet filtering to remove noise. This procedure produces a sharp spike at a time delay corresponding to the position of every reflecting interface. Thus, it helps to determine more accurately the positions of the various interfaces. In contrast to figure 5b, the front and back surfaces of the thin (~120 um) magnetic recording material are clearly resolved in the deconvolved data. This is consistent with the expected resolution of L,/2, where L, = 200 lum is the coherence length of the THz pulse in the intervening material. In contrast, when no other reflections are nearby, we find that the position of a reflecting surface can be determined with a precision of only a few microns, as discussed in section 2 above. Figure 5. THz waveforms incident on (a), and reflected from (b), a 3.5” floppy disk. The two surfaces of the front and rear covers are visible, as are the reflections from the thin magnetic recording material. Curve (c) (in arbitrary units) represents the reflected waveform after signal processing. Curve (b) has been scaled up by a factor of four, and the three curves have been offset for clarity.
indicated by the dashed line in figure 6a. The amplitude of he reflected T-Ray waveforms for each horizontal (x) position is displayed as a function of delay in this omographic image. The reflected waveforms are processed in similar fashion to the data of figure 5c, and the amplitude of the resulting waveform is translated into a gray scale. The vertical axis of figure 6b (optical depth) is proportional to the time delay of the waveform. Converting his axis to physical depth requires knowledge of the sample’s refractive index profile, which can be obtained from the amplitudes of the reflected pulses, as described below. Figure 6. (a) THz reflection image of a floppy disk, in which the metallic hub is visible. (b) Tomographic image generated along the dashed line in (a). The vertical axis corresponds to depth into the object.
indicated by the dashed line in figure 6a. The amplitude of he reflected T-Ray waveforms for each horizontal (x) position is displayed as a function of delay in this omographic image. The reflected waveforms are processed in similar fashion to the data of figure 5c, and the amplitude of the resulting waveform is translated into a gray scale. The vertical axis of figure 6b (optical depth) is proportional to the time delay of the waveform. Converting his axis to physical depth requires knowledge of the sample’s refractive index profile, which can be obtained from the amplitudes of the reflected pulses, as described below. Figure 6. (a) THz reflection image of a floppy disk, in which the metallic hub is visible. (b) Tomographic image generated along the dashed line in (a). The vertical axis corresponds to depth into the object.
Figure 7 (a) impulse response function from figure 5c (in arbitrary units) (b) refractive index profile derived from this function (although not ideal) approximation to the actual profile of the sample. This result is shown in figure 7. The refractive indices of the materials are in reasonable agreement with measured values and the temporal positions are correct, but the overall background va ue rises with increasing delay. This effect, in which the index does not return to n= | after the sample, is a cumu lative result of the various approximations made in t he derivation of equations (3) and (4). A more sophisticated analysis will be required to overcome these difficulties.
Figure 7 (a) impulse response function from figure 5c (in arbitrary units) (b) refractive index profile derived from this function (although not ideal) approximation to the actual profile of the sample. This result is shown in figure 7. The refractive indices of the materials are in reasonable agreement with measured values and the temporal positions are correct, but the overall background va ue rises with increasing delay. This effect, in which the index does not return to n= | after the sample, is a cumu lative result of the various approximations made in t he derivation of equations (3) and (4). A more sophisticated analysis will be required to overcome these difficulties.
Figure 2a shows THz time-domain data of pulses propagated through free space, i.e. the standard imaging set-up without aperture, and pulses transmitted through the tapered metal tip with the smallest aperture. The corresponding Fourier spectra, normalized to the same maximum value, are shown in Fig. 2b. According to the maximum peak-to-peak "amplitude of the time-resolved data, the amplitude transmission of the aperture is approximately 1/130, however, the spectra indicate that this
Figure 2a shows THz time-domain data of pulses propagated through free space, i.e. the standard imaging set-up without aperture, and pulses transmitted through the tapered metal tip with the smallest aperture. The corresponding Fourier spectra, normalized to the same maximum value, are shown in Fig. 2b. According to the maximum peak-to-peak "amplitude of the time-resolved data, the amplitude transmission of the aperture is approximately 1/130, however, the spectra indicate that this
Figure2: a) THz waveforms before and after transmission through the near-field aperture. b) Normalized Fourier spectra of the data shown in a). factor is strongly frequency dependent. Both the larger focal diameter for the low-frequency components and their larger diffraction losses after passing the aperture lead to an effective spectral narrowing and a low-frequency cut-off at approximately 0.4 THz. Nevertheless there is significant signal amplitude for frequencies above 0.5 THz, corresponding to a wavelength of 12x dip. We also note that the cutoff is less sharp and at a much lower frequency than for a long waveguide with comparable diameter.[6] The spectral high-pass filtering of the signal also accounts for the changes of the THz waveform from a well-defined single-cycle pulse towards a more oscillating structure. In addition, both waveforms are slightly modulated due to residual water vapor absorption in the beam path.
Figure2: a) THz waveforms before and after transmission through the near-field aperture. b) Normalized Fourier spectra of the data shown in a). factor is strongly frequency dependent. Both the larger focal diameter for the low-frequency components and their larger diffraction losses after passing the aperture lead to an effective spectral narrowing and a low-frequency cut-off at approximately 0.4 THz. Nevertheless there is significant signal amplitude for frequencies above 0.5 THz, corresponding to a wavelength of 12x dip. We also note that the cutoff is less sharp and at a much lower frequency than for a long waveguide with comparable diameter.[6] The spectral high-pass filtering of the signal also accounts for the changes of the THz waveform from a well-defined single-cycle pulse towards a more oscillating structure. In addition, both waveforms are slightly modulated due to residual water vapor absorption in the beam path.
Figure 3: Near-field (a) and diffraction-limited standard (b) T-ray image of two quadratic gold pads.
Figure 3: Near-field (a) and diffraction-limited standard (b) T-ray image of two quadratic gold pads.
Figure 4: THz line-scans over a razor blade. compared to 0.92 THz). However, this small difference can clearly not account for the observed improvement of the spatial resolution. To prove this, we also scanned a razor blade across the near-field tip and through the focus of the imaging set-up. The corresponding data are shown in Fig.
Figure 4: THz line-scans over a razor blade. compared to 0.92 THz). However, this small difference can clearly not account for the observed improvement of the spatial resolution. To prove this, we also scanned a razor blade across the near-field tip and through the focus of the imaging set-up. The corresponding data are shown in Fig.
Figure 5: Near-field (a) and diffraction-limited (b) T-ray image of a resolution test pattern. Figure 5a shows a near-field image of the central part (group 2 and 3) of the test pattern obtained with ‘the smallest aperture and by scanning the sample in mechanical contact with the near-field tip. For comparison, a diffraction-limited standard image of the same sample area is shown in Fig. Sb.
Figure 5: Near-field (a) and diffraction-limited (b) T-ray image of a resolution test pattern. Figure 5a shows a near-field image of the central part (group 2 and 3) of the test pattern obtained with ‘the smallest aperture and by scanning the sample in mechanical contact with the near-field tip. For comparison, a diffraction-limited standard image of the same sample area is shown in Fig. Sb.
Figure 6: THz line-scans over group 1 of the resolution target for various tip-sample separations. The curves are offset for clarity. The THz amplitude was integrated over a frequency range of 0.6-2.3 THz in the near-field image and 1.4-2.3 THz in the standard image, corresponding to average wavelengths of 220 um and 180 ym, respectively. The smallest resolvable feature in the near-field image are the vertical lines of group 3, element 3, which have a line width and separation of 50 um.
Figure 6: THz line-scans over group 1 of the resolution target for various tip-sample separations. The curves are offset for clarity. The THz amplitude was integrated over a frequency range of 0.6-2.3 THz in the near-field image and 1.4-2.3 THz in the standard image, corresponding to average wavelengths of 220 um and 180 ym, respectively. The smallest resolvable feature in the near-field image are the vertical lines of group 3, element 3, which have a line width and separation of 50 um.
Fig. 2: (a) Measured THz signal in a 60um dipole, 5um gap; see Fig. 1 for reference; (b) THz intensity as a function of dipole length and gap. pulse derived from the same Ti-sapphire laser. The emitter is then scanned two dimensionally with a 0.4m spatial resolution and the THz waveforms for each coordinate are recorded by scanning the delay of the beam exciting the antenna.
Fig. 2: (a) Measured THz signal in a 60um dipole, 5um gap; see Fig. 1 for reference; (b) THz intensity as a function of dipole length and gap. pulse derived from the same Ti-sapphire laser. The emitter is then scanned two dimensionally with a 0.4m spatial resolution and the THz waveforms for each coordinate are recorded by scanning the delay of the beam exciting the antenna.
Fig. 3: Two dimensional scans of the THz emission intensity for several singular electric field offset dipole emitters with gaps of (a) 5, (b) 10, (c) 30, (4) 90m. The smail-gap emitters in (a) and (b) show perfectly symmetrical THz emission. A 30um This is consistent with the well known current surge model. In the offset dipole structure, the
Fig. 3: Two dimensional scans of the THz emission intensity for several singular electric field offset dipole emitters with gaps of (a) 5, (b) 10, (c) 30, (4) 90m. The smail-gap emitters in (a) and (b) show perfectly symmetrical THz emission. A 30um This is consistent with the well known current surge model. In the offset dipole structure, the
Fig. 4: The polarization of the THz emission from stripline and offset dipole emitters is related to the directions of the carriers’ flow in the semiconductor. The THz radiation from the offset dipole has a considerable (~25%) orthogonal component. gap (c), shows a slight asymmetry , and the THz emission from large gap emitters (d) is clearly asymmetrical.
Fig. 4: The polarization of the THz emission from stripline and offset dipole emitters is related to the directions of the carriers’ flow in the semiconductor. The THz radiation from the offset dipole has a considerable (~25%) orthogonal component. gap (c), shows a slight asymmetry , and the THz emission from large gap emitters (d) is clearly asymmetrical.
carriers’ flow is tilted. In such offset structure, while the THz wave is generated in the semiconductor, it has components in both polarization directions, but only one direction is enhanced by the dipole formed by the straight electrode lines. Fig. 5: The THz signal detected using (a) offset dipole singular electric field detector; (b) regular dipole detector. The offset singular detector yields
carriers’ flow is tilted. In such offset structure, while the THz wave is generated in the semiconductor, it has components in both polarization directions, but only one direction is enhanced by the dipole formed by the straight electrode lines. Fig. 5: The THz signal detected using (a) offset dipole singular electric field detector; (b) regular dipole detector. The offset singular detector yields
Figure 1, The cw-tuning characteristics of a Ti:sapphire laser with the SBR were compared to the case without folding. An SBR was placed in the cavity for the beam-path folding at the same incidence angle as used in the mode-locked laser.
Figure 1, The cw-tuning characteristics of a Ti:sapphire laser with the SBR were compared to the case without folding. An SBR was placed in the cavity for the beam-path folding at the same incidence angle as used in the mode-locked laser.
Figure 4. THz radiation spectra obtained by Fourier transformation of the autocorrelation from a Polarizing Michelson interferometer. tricarbocyanine Iodide) in ethylene glycol. The Ti:sapphire laser cavity consists of a six-mirror cavity with an additional focus for a dye-jet, a 1% output coupler for monitoring femtosecond pulse formation, a single-plate birefringent filter as a tuning element, and a pair of high- dispersion SF6 Brewster prisms with 35-cm separation. A cw all-line Ar laser pumping source was operated at 12.5 Ww. ee EN Be Se Eee The autocorrelation trace and the spectrum were monitored with a rapid-scanning autocorrelator and an optical spectrum analyzer. 180-fsec pulses assuming a sech? shape were obtained with 3.9-nm spectral width at 768 nm, and these yielded a nearly transform-limited time and bandwidth product of 0.356. The average and peak output powers inside the cavity were 5.4W and 375 kW (80-MHz repetition rate). The peak power a the optical pulse irradiating SBR exceeded 8.3 MW/cm2, assuming a 1 mm beam diameter and considering the 80 degree incidence angle. The THz radiation was emitted in the transmitted direction and reflection direction. The transmitted THz radiation was picked off by a flat Al- mirror. The spectra of the THz radiation were obtained by Fourier transformation of the autocorrelation of the radiation from a Polarizing Michelson interferometer (Graseby Specac) as shown in Fig. 1. The interferometer was evacuated to avoid the possible absorption of water- vapor in the air. A liquid-helium-temperature cooled InSb bolometer (QMC model QFI/2) with subnanowatt sensitivity was provided for detection. For the lock-in detection, a mechanical chopper operated at 206 Hz was inserted in the cavity of the mode-locked laser. For this chopping frequency, a pulse-formation time of much less than 100 msec is short enough to reach the steady state. [11] Therefore the THz radiation from this laser can be considered to be modulated with a duty cycle of almost 50%. The average power of the radiation coupled to the interferometer was calibrated to be 2-3 nW in one beam.
Figure 4. THz radiation spectra obtained by Fourier transformation of the autocorrelation from a Polarizing Michelson interferometer. tricarbocyanine Iodide) in ethylene glycol. The Ti:sapphire laser cavity consists of a six-mirror cavity with an additional focus for a dye-jet, a 1% output coupler for monitoring femtosecond pulse formation, a single-plate birefringent filter as a tuning element, and a pair of high- dispersion SF6 Brewster prisms with 35-cm separation. A cw all-line Ar laser pumping source was operated at 12.5 Ww. ee EN Be Se Eee The autocorrelation trace and the spectrum were monitored with a rapid-scanning autocorrelator and an optical spectrum analyzer. 180-fsec pulses assuming a sech? shape were obtained with 3.9-nm spectral width at 768 nm, and these yielded a nearly transform-limited time and bandwidth product of 0.356. The average and peak output powers inside the cavity were 5.4W and 375 kW (80-MHz repetition rate). The peak power a the optical pulse irradiating SBR exceeded 8.3 MW/cm2, assuming a 1 mm beam diameter and considering the 80 degree incidence angle. The THz radiation was emitted in the transmitted direction and reflection direction. The transmitted THz radiation was picked off by a flat Al- mirror. The spectra of the THz radiation were obtained by Fourier transformation of the autocorrelation of the radiation from a Polarizing Michelson interferometer (Graseby Specac) as shown in Fig. 1. The interferometer was evacuated to avoid the possible absorption of water- vapor in the air. A liquid-helium-temperature cooled InSb bolometer (QMC model QFI/2) with subnanowatt sensitivity was provided for detection. For the lock-in detection, a mechanical chopper operated at 206 Hz was inserted in the cavity of the mode-locked laser. For this chopping frequency, a pulse-formation time of much less than 100 msec is short enough to reach the steady state. [11] Therefore the THz radiation from this laser can be considered to be modulated with a duty cycle of almost 50%. The average power of the radiation coupled to the interferometer was calibrated to be 2-3 nW in one beam.
Fig. 1. Experimental setup for the measurement of the electro- magnetic pulses in forward (a) and backward (b) direction. Opti- cal excitation and gating is achieved by femtosecond laser pulses (A & 810 nm). tical excitation as a function of time.
Fig. 1. Experimental setup for the measurement of the electro- magnetic pulses in forward (a) and backward (b) direction. Opti- cal excitation and gating is achieved by femtosecond laser pulses (A & 810 nm). tical excitation as a function of time.
Fig. 2. Schematic view of the absorption and reflection of the generated pulse within the thin film (indicated for backward di- rection).
Fig. 2. Schematic view of the absorption and reflection of the generated pulse within the thin film (indicated for backward di- rection).
Fig. 3. Optical response of a YBCO thin film after excitation with fs laser pulses for several temperatures. The inset shows the quasiparticle recombination times derived by fits with Rothwarf- Taylor equations.
Fig. 3. Optical response of a YBCO thin film after excitation with fs laser pulses for several temperatures. The inset shows the quasiparticle recombination times derived by fits with Rothwarf- Taylor equations.
Fig. 5. Dependence of the emitted field intensity on the bias current. The solid line is the dependence expected from theory. The dependence of the emitted field intensity on the bias current J, is shown in Fig. 5. From Eq. 1, a linear dependence of the field amplitude on J, is expected, or equivalently, a quadratic behavior of the intensity, proving the validity of Eq. 1.
Fig. 5. Dependence of the emitted field intensity on the bias current. The solid line is the dependence expected from theory. The dependence of the emitted field intensity on the bias current J, is shown in Fig. 5. From Eq. 1, a linear dependence of the field amplitude on J, is expected, or equivalently, a quadratic behavior of the intensity, proving the validity of Eq. 1.
Fig. 4. (a) Time-domain THz-radiation signal in forward direc- tion at a temperature of 8 K. (b) Fourier transform of the time- domain signal in (a). For comparison, the spectrum of a signal at higher temperature is shown.
Fig. 4. (a) Time-domain THz-radiation signal in forward direc- tion at a temperature of 8 K. (b) Fourier transform of the time- domain signal in (a). For comparison, the spectrum of a signal at higher temperature is shown.
After calculating E(t), the limited spectral band- width of the THz receiver antenna (~ 1.5 THz) is taken into account by convolution of the Fourier spectra of E(t) with the receiver spectral function [7]. Fig. 6. Comparison of the temperature dependence of the peak frequency (a) and the signal amplitudes (b) as obtained from the THz measurements (circles and triangles) and from calculations with the data of the optical experiments (full squares).
After calculating E(t), the limited spectral band- width of the THz receiver antenna (~ 1.5 THz) is taken into account by convolution of the Fourier spectra of E(t) with the receiver spectral function [7]. Fig. 6. Comparison of the temperature dependence of the peak frequency (a) and the signal amplitudes (b) as obtained from the THz measurements (circles and triangles) and from calculations with the data of the optical experiments (full squares).
In this case, Eq. (2) has to be considered and the entire temporal signal has to be recorded. We treat the Fabry- Pérot effect as a perturbation. We start the calculation with a fairly good approximation of 7, which is derived as follows: we consider that the dephasing arg(T(@)) is mostly induced by the real part of 7 and that the modulus of T(@) depends mainly on the absorption x. Thus we obtain a first pair (n,«), with which we calculate the Fabry-Pérot term (summation in (1)). We divide the transmission coefficient T(@) by this term, getting a curve with a strongly reduced contribution to internal reflections. Therefore we can apply the above method (thick sample case) for deriving a better value of 7. Then, we repeat this procedure until the value of T,)- doesn't vary from one step of computation to another one by more than 0.1 of the experimental uncertainty. Usually, the method converges over the full range of practical values 7 <n $5,0 <x 10) only after a few iterations.
In this case, Eq. (2) has to be considered and the entire temporal signal has to be recorded. We treat the Fabry- Pérot effect as a perturbation. We start the calculation with a fairly good approximation of 7, which is derived as follows: we consider that the dephasing arg(T(@)) is mostly induced by the real part of 7 and that the modulus of T(@) depends mainly on the absorption x. Thus we obtain a first pair (n,«), with which we calculate the Fabry-Pérot term (summation in (1)). We divide the transmission coefficient T(@) by this term, getting a curve with a strongly reduced contribution to internal reflections. Therefore we can apply the above method (thick sample case) for deriving a better value of 7. Then, we repeat this procedure until the value of T,)- doesn't vary from one step of computation to another one by more than 0.1 of the experimental uncertainty. Usually, the method converges over the full range of practical values 7 <n $5,0 <x 10) only after a few iterations.
Figure 3 : Refractive index of BK7 glass derived from the data of Figure 2. Reproduced with permission from Ref. 2, Copyright 1997, IEEE. Figure 2 : THz pulse recorded with and without a BK7 glass slide (d=1.07 mm). Reproduced with permission from Ref. 2, Copyright 1997, IEEE.
Figure 3 : Refractive index of BK7 glass derived from the data of Figure 2. Reproduced with permission from Ref. 2, Copyright 1997, IEEE. Figure 2 : THz pulse recorded with and without a BK7 glass slide (d=1.07 mm). Reproduced with permission from Ref. 2, Copyright 1997, IEEE.
We show here results obtained with a /.07-mm-thick BK7 glass slide, for which the first internal reflection is detected far after the first pulse and thus is not seen in Fig. 2. The material parameters, derived with our method, are shown in Fig, 3. The calculation, leading to the entire curves of Fig. 3 (one calculated value every 5 GHz over the 1000 GHz range) takes 35 ms with a Pentium 90 PC computer.
We show here results obtained with a /.07-mm-thick BK7 glass slide, for which the first internal reflection is detected far after the first pulse and thus is not seen in Fig. 2. The material parameters, derived with our method, are shown in Fig, 3. The calculation, leading to the entire curves of Fig. 3 (one calculated value every 5 GHz over the 1000 GHz range) takes 35 ms with a Pentium 90 PC computer.
Figure 4 : THz pulses transmitted through a bare Si wafer and through the same wafer covered by a Si-porous film (i.e. before and after chemical treatment). Reproduced with permission from Ref. 2, Copyright 1997, IEEE. by the Si wafer without (reference) and with the film (Fig. 4). As the refractive index of the porous-Si layer is smaller than the bulk Si value, the THz pulse transmitted by the wafer with the film reaches the detector before the pulse transmitted by the wafer before the chemical treatment. Moreover, the film, with a low index of refraction, acts as an anti-reflection coating. Thus the pulse transmitted trough the substrate and the film is stronger that the pulse transmitted only through the substrate. The index of refraction is shown in Fig. 5, as determined using the method presented in this paper. Also on the same figure are the values we obtained without taking into account the Fabry-Pérot effect. These values differ strongly from the real index. This behavior is also observed for the absorption (see Fig. 6), which becomes unphysical (negative) without taking into account the Fabry-Pérot effect. This definitively proves that even if no internal reflections appear in the pulse temporal shape, one has always to take into account the Fabry-Pérot effect in the extraction procedure.
Figure 4 : THz pulses transmitted through a bare Si wafer and through the same wafer covered by a Si-porous film (i.e. before and after chemical treatment). Reproduced with permission from Ref. 2, Copyright 1997, IEEE. by the Si wafer without (reference) and with the film (Fig. 4). As the refractive index of the porous-Si layer is smaller than the bulk Si value, the THz pulse transmitted by the wafer with the film reaches the detector before the pulse transmitted by the wafer before the chemical treatment. Moreover, the film, with a low index of refraction, acts as an anti-reflection coating. Thus the pulse transmitted trough the substrate and the film is stronger that the pulse transmitted only through the substrate. The index of refraction is shown in Fig. 5, as determined using the method presented in this paper. Also on the same figure are the values we obtained without taking into account the Fabry-Pérot effect. These values differ strongly from the real index. This behavior is also observed for the absorption (see Fig. 6), which becomes unphysical (negative) without taking into account the Fabry-Pérot effect. This definitively proves that even if no internal reflections appear in the pulse temporal shape, one has always to take into account the Fabry-Pérot effect in the extraction procedure.
Figure 6 : Absorption coefficient (energy) of porous-Si, with and without taking into account the Fabry-Pérot effect in the thin film. Reproduced with permission from Ref. 2, Copyright 1997, IEEE. Figure 5 : Index of refraction of porous-Si, with and without taking into account the Fabry-Pérot effect in the thin film. Reproduced with permission from Ref. 2, Copyright 1997, IEEE.
Figure 6 : Absorption coefficient (energy) of porous-Si, with and without taking into account the Fabry-Pérot effect in the thin film. Reproduced with permission from Ref. 2, Copyright 1997, IEEE. Figure 5 : Index of refraction of porous-Si, with and without taking into account the Fabry-Pérot effect in the thin film. Reproduced with permission from Ref. 2, Copyright 1997, IEEE.
Figure 8 : Index of refraction of p-doped silicon together with standard deviation. Reproduced with permission from Ref. 2, Copyright 1997, IEEE. Figure 7 : THz pulse transmitted by a 225 yum thick p- doped Si wafer. Reproduced with permission from Ref. 2, Copyright 1997, IEEE.
Figure 8 : Index of refraction of p-doped silicon together with standard deviation. Reproduced with permission from Ref. 2, Copyright 1997, IEEE. Figure 7 : THz pulse transmitted by a 225 yum thick p- doped Si wafer. Reproduced with permission from Ref. 2, Copyright 1997, IEEE.
Figure 1: Terahertz waveforms from YBazCu307 at @ = 50°, ¢ = O° for a) S-polarized incident and S-polarized detected radiation, b) S-polarized incident, P-pol. detected, c) P-pol. incident, S-pol. detected, and d) P-pol. incident and P-pol. detected.
Figure 1: Terahertz waveforms from YBazCu307 at @ = 50°, ¢ = O° for a) S-polarized incident and S-polarized detected radiation, b) S-polarized incident, P-pol. detected, c) P-pol. incident, S-pol. detected, and d) P-pol. incident and P-pol. detected.
Figure 2: Terahertz waveforms from YBajCu30O7 at 9 = 50°, é = 22° for a) S-polarized incident and S-polarized detected radiation, b) S-polarized incident, P-pol. detected, c) P-pol. incident, S-pol. detected, and d) P-pol. incident and P-pol. detected.
Figure 2: Terahertz waveforms from YBajCu30O7 at 9 = 50°, é = 22° for a) S-polarized incident and S-polarized detected radiation, b) S-polarized incident, P-pol. detected, c) P-pol. incident, S-pol. detected, and d) P-pol. incident and P-pol. detected.
Figure 3: The radiated THz field depends linearly on the incident 800 nm power. The spot size on the film had a diameter of ~ 2 mm.
Figure 3: The radiated THz field depends linearly on the incident 800 nm power. The spot size on the film had a diameter of ~ 2 mm.
Figure 4: Radiated THz field is plotted as a function of incident angle (—@,) in the top figure and as a func- tion of the internal angle, (6;), in the bottom figure. The solid circles represent data taken with P-polarized 800 nm incident and detecting P-polarized THz. The solid curves are fits, see text for more information.
Figure 4: Radiated THz field is plotted as a function of incident angle (—@,) in the top figure and as a func- tion of the internal angle, (6;), in the bottom figure. The solid circles represent data taken with P-polarized 800 nm incident and detecting P-polarized THz. The solid curves are fits, see text for more information.
Figure 1: Schematics of the terahertz time-domain spectrometer used for gas analysis.
Figure 1: Schematics of the terahertz time-domain spectrometer used for gas analysis.
To demonstrate chemical recognition of gases and gas mixtures we have studied the polar gases H2O, NH3, HCI, and CH;CN in the pressure range between 0.3 and 13 kPa. Common for these species is that they have strong perma- nent dipole moments and hence they interact strongly with the terahertz field. Figure 2 shows a sample replica of a terahertz. waveform measured after propagating through NH; vapor at a pressure of 12.8 kPa. The impulse seen at zero delay excites some of the lowest lying transitions in the rotation-inversion spectrum of NH; [14]. Pronounced oscillations, due to the FID of the excited resonances, are seen at later time delays. Figure 2: Terahertz waveform resulting from the propagation through NH; vapor. The waveform is an average of 1000 real- time scans. The occurrences at 29 and 51 ps are due to reflections in the setup.
To demonstrate chemical recognition of gases and gas mixtures we have studied the polar gases H2O, NH3, HCI, and CH;CN in the pressure range between 0.3 and 13 kPa. Common for these species is that they have strong perma- nent dipole moments and hence they interact strongly with the terahertz field. Figure 2 shows a sample replica of a terahertz. waveform measured after propagating through NH; vapor at a pressure of 12.8 kPa. The impulse seen at zero delay excites some of the lowest lying transitions in the rotation-inversion spectrum of NH; [14]. Pronounced oscillations, due to the FID of the excited resonances, are seen at later time delays. Figure 2: Terahertz waveform resulting from the propagation through NH; vapor. The waveform is an average of 1000 real- time scans. The occurrences at 29 and 51 ps are due to reflections in the setup.
Figure 3: Estimated versus the measured pressure from a single- species-recognition of HCI. The analysis is based on terahertz waveforms obtained by standard lock-in techniques.
Figure 3: Estimated versus the measured pressure from a single- species-recognition of HCI. The analysis is based on terahertz waveforms obtained by standard lock-in techniques.
Figure 4: Analysis of NH3 and H,O mixture. For each mixture we plot a single point representing the partial pressure of the two gas species. The analysis bases on terahertz waveforms obtained by lock-in techniques.
Figure 4: Analysis of NH3 and H,O mixture. For each mixture we plot a single point representing the partial pressure of the two gas species. The analysis bases on terahertz waveforms obtained by lock-in techniques.
Figure 5: Real-time gas sensing of a mixture of NH; and H,0. Each data point is based on the average of 10 waveforms and solid lines represent the adjacent average of 15 data points. The vertical, dashed lines mark experimental events.
Figure 5: Real-time gas sensing of a mixture of NH; and H,0. Each data point is based on the average of 10 waveforms and solid lines represent the adjacent average of 15 data points. The vertical, dashed lines mark experimental events.
Figure 6: The power spectrum of H,O estimated from an IIR filter using M=50 poles (thick solid line) and from a Fourier analysis (dashed line). Thin solid line and dotted line are the integrated power spectral density for the LPC and the Fourier analysis, re- spectively. A comparison between the power spectrum esti- mated from the IIR filter and the power spectrum estimated from Fourier transformation is shown in Fig. 6. The com- parison serves as a confidence measure for the LPC analy- sis but is not used elsewhere in the chemical recognition.
Figure 6: The power spectrum of H,O estimated from an IIR filter using M=50 poles (thick solid line) and from a Fourier analysis (dashed line). Thin solid line and dotted line are the integrated power spectral density for the LPC and the Fourier analysis, re- spectively. A comparison between the power spectrum esti- mated from the IIR filter and the power spectrum estimated from Fourier transformation is shown in Fig. 6. The com- parison serves as a confidence measure for the LPC analy- sis but is not used elsewhere in the chemical recognition.
dimensional vector space. Vectors representing known gas species are stored in a codebook, and by successive meas- urements and subsequent LPC analysis for a variety of dif- ferent gases, the codebook is built. The final codebook contains a set of linearly independent vectors {a,}, i = 1, 2,.... p and p<M, that span a subspace of the M- dimensional vector space. A_ single-species-recognition procedure consists in a full search through the codebook to find the best match between the trial vector, i.e., the LPC coefficients of an unknown species, and the vectors in the codebook. The minimum Euclidean distance is used as classification criteria. A gas mixture analysis uses the fact hat the LPC vector representing the mixture, a, is the vec- tor sum of coded vectors for the individual species weighted by their mole fractions. In practice, an orthogonal basis, {b;}, is constructed and is used to calculate the or- hogonal ‘projections. A linear transformation establishes he relation between the two bases {a;} and {b;}. These principles are illustrated in Figure 7 using a binary mixture as example. Figure 7: Two dimensional representation of a codebook. The vectors a, and a, represent two different gas species. From these vectors two orthogonal vectors b; and 2 are constructed. The vector a corresponds to a mixture that contains the fraction x, of species 1 and x, of species 2.
dimensional vector space. Vectors representing known gas species are stored in a codebook, and by successive meas- urements and subsequent LPC analysis for a variety of dif- ferent gases, the codebook is built. The final codebook contains a set of linearly independent vectors {a,}, i = 1, 2,.... p and p<M, that span a subspace of the M- dimensional vector space. A_ single-species-recognition procedure consists in a full search through the codebook to find the best match between the trial vector, i.e., the LPC coefficients of an unknown species, and the vectors in the codebook. The minimum Euclidean distance is used as classification criteria. A gas mixture analysis uses the fact hat the LPC vector representing the mixture, a, is the vec- tor sum of coded vectors for the individual species weighted by their mole fractions. In practice, an orthogonal basis, {b;}, is constructed and is used to calculate the or- hogonal ‘projections. A linear transformation establishes he relation between the two bases {a;} and {b;}. These principles are illustrated in Figure 7 using a binary mixture as example. Figure 7: Two dimensional representation of a codebook. The vectors a, and a, represent two different gas species. From these vectors two orthogonal vectors b; and 2 are constructed. The vector a corresponds to a mixture that contains the fraction x, of species 1 and x, of species 2.
| Figure |: Simplified Four-wave mixing set-up. Two applications of four-wave mixing have emerged in recent years. The first is to use the amplitude versus detuning frequency information provided by the two new mixing products to characterize ultra-fast carrier dynamics and the second is to use the phase conjugate relationship of the new waves to perform all-optical signal processing. Both application areas will be discussed [1]. In the first part of this presentation, we will review the motivations for applying four- wave mixing elements in optical telecommunications systems. Specifically their application to carrier spectral inversion for fiber dispersion compensation and to spectral translation of optical carriers in WDM systems will be discussed. Recent achievements in each of these areas will be reviewed including the current signal-to- -noise performance of FWM mixers. Results from record conversion systems experiments will be presented. These include the results in figure 2 showing the wavelength conversion of 10 Gb/s data over spans as large as 18 nm [2]. The figure inset shows the eye diagram for the maximum shift of 18 nm.
| Figure |: Simplified Four-wave mixing set-up. Two applications of four-wave mixing have emerged in recent years. The first is to use the amplitude versus detuning frequency information provided by the two new mixing products to characterize ultra-fast carrier dynamics and the second is to use the phase conjugate relationship of the new waves to perform all-optical signal processing. Both application areas will be discussed [1]. In the first part of this presentation, we will review the motivations for applying four- wave mixing elements in optical telecommunications systems. Specifically their application to carrier spectral inversion for fiber dispersion compensation and to spectral translation of optical carriers in WDM systems will be discussed. Recent achievements in each of these areas will be reviewed including the current signal-to- -noise performance of FWM mixers. Results from record conversion systems experiments will be presented. These include the results in figure 2 showing the wavelength conversion of 10 Gb/s data over spans as large as 18 nm [2]. The figure inset shows the eye diagram for the maximum shift of 18 nm.
Figure 2: FWM down-shift replicas. Inset: eye diagram for 18 nm shift at 10 Gb/s PRBS.
Figure 2: FWM down-shift replicas. Inset: eye diagram for 18 nm shift at 10 Gb/s PRBS.
LIOTTA IOUS UIE PU Pe Mt More recently, we have considered the application of four-wave mixing to measurement of interwell carrier equilibration in multi quantum-well active layers [3]. For this work we have used the polarization selection rules associated with tensile and compressively strained quantum wells to induce spatially selective mixing. Transport of this mixing to neighboring wells is then probed selectively using again the polarization selection rules.
LIOTTA IOUS UIE PU Pe Mt More recently, we have considered the application of four-wave mixing to measurement of interwell carrier equilibration in multi quantum-well active layers [3]. For this work we have used the polarization selection rules associated with tensile and compressively strained quantum wells to induce spatially selective mixing. Transport of this mixing to neighboring wells is then probed selectively using again the polarization selection rules.
Figure 3 shows the input wave configuration for the experiment. TM components mix only in the tensile wells of the structures causing carrier density modulation that is spatially localized to only these wells. Conventional four-wave mixing occurs as the input waves scatter from this modulation. In addition, however, some carrier density modulation is coupled by way of inter-well transport into neighboring compressively strained quantum wells. Once coupled into these wells, scattering of TE polarized waves is possible through the TE dipole component of the compressive wells. By measuring the detuning frequency dependence of both the TE and TM components of four-wave mixing, it is possible to directly measure the interwell carrier equilibration rate. In particular, the ratio of the TE component to the TM component will decrease with increasing detuning frequency since interwell transport of higher frequency modulations will be less efficient. Ultimately, this ratio will flatten reflecting the ratio of TE to TM conventional four-wave mixing that occurs only within the tensile wells. These predictions are verified in figure 4 which shows the ratio of FWM TE to TM powers in an eight well amplifier (four tensile and four compressive).
Figure 3 shows the input wave configuration for the experiment. TM components mix only in the tensile wells of the structures causing carrier density modulation that is spatially localized to only these wells. Conventional four-wave mixing occurs as the input waves scatter from this modulation. In addition, however, some carrier density modulation is coupled by way of inter-well transport into neighboring compressively strained quantum wells. Once coupled into these wells, scattering of TE polarized waves is possible through the TE dipole component of the compressive wells. By measuring the detuning frequency dependence of both the TE and TM components of four-wave mixing, it is possible to directly measure the interwell carrier equilibration rate. In particular, the ratio of the TE component to the TM component will decrease with increasing detuning frequency since interwell transport of higher frequency modulations will be less efficient. Ultimately, this ratio will flatten reflecting the ratio of TE to TM conventional four-wave mixing that occurs only within the tensile wells. These predictions are verified in figure 4 which shows the ratio of FWM TE to TM powers in an eight well amplifier (four tensile and four compressive).
Figure 3. Optical set up for all-optical gate switching using spin-dependent optical nonlinearity. We thank Drs. T. Nishimura, H. Yoshida and Y. Matsui for their useful discussions. We also acknowledge Drs. T. Sakurai, Y. Katayama, and F. Saitoh for encouragement. This work was supported by the New Energy and Industrial Technology Development Organization in the frame of the Femtosecond Technology Project.
Figure 3. Optical set up for all-optical gate switching using spin-dependent optical nonlinearity. We thank Drs. T. Nishimura, H. Yoshida and Y. Matsui for their useful discussions. We also acknowledge Drs. T. Sakurai, Y. Katayama, and F. Saitoh for encouragement. This work was supported by the New Energy and Industrial Technology Development Organization in the frame of the Femtosecond Technology Project.
WUE UY LE ANAS dd ytlo it Ue POM Ua UEP, Femtosecond reflectivity of InP/InGaAs NBR also depends on this absorption-limited multiple reflection as explained in the next section. Fig. 1. Measured reflectance and group-delay spectra.
WUE UY LE ANAS dd ytlo it Ue POM Ua UEP, Femtosecond reflectivity of InP/InGaAs NBR also depends on this absorption-limited multiple reflection as explained in the next section. Fig. 1. Measured reflectance and group-delay spectra.
Fig. 2. Calculated reflectance and group-delay spectra. ine measured reliectance and group-delay spectra were compared with the results from a model calculation based on matrix formalism of light propagation through Bragg reflector. In the calculation, absorption in the InGaAs layers were taken into account by using the absorption spectrum of a bulk InGaAs layer measured independently. Boundary conditions are set so as to keep both complex electric field and its derivative continuous at each interface between InP and InGaAs. The calculated reflectance and group-delay spectra are plotted in Fig. 2. The calculated reflectance spectrum yields the features observed in the measurement such as the reflectance minimum and the disappearance of fringes at wavelengths shorter than the absorption-edge. The calculated group delay peaks at the wavelength of the minimum reflectance as the measured group delay does. The measured spectra, therefore, reflect the multiple reflection and the light propagation in an absorptive Bragg reflector.
Fig. 2. Calculated reflectance and group-delay spectra. ine measured reliectance and group-delay spectra were compared with the results from a model calculation based on matrix formalism of light propagation through Bragg reflector. In the calculation, absorption in the InGaAs layers were taken into account by using the absorption spectrum of a bulk InGaAs layer measured independently. Boundary conditions are set so as to keep both complex electric field and its derivative continuous at each interface between InP and InGaAs. The calculated reflectance and group-delay spectra are plotted in Fig. 2. The calculated reflectance spectrum yields the features observed in the measurement such as the reflectance minimum and the disappearance of fringes at wavelengths shorter than the absorption-edge. The calculated group delay peaks at the wavelength of the minimum reflectance as the measured group delay does. The measured spectra, therefore, reflect the multiple reflection and the light propagation in an absorptive Bragg reflector.
Fig. 3. Transient reflectivity measured at wavelengths between 1.48 um and 1.52 pm. In Fig. 3, transient reflectivity obtained at wavelengths between 1.48 um and 1.52 ym is plotted in a logarithmic scale. The dots and the solid lines correspond to the experimental data and fitting curves, respectively. The wavelength range in the measurements lies in the long-wavelength edge of the stop band. The pump and probe pulses overlap at a delay time ~ 1.4 ps. The transient reflectivity increases when the hot carriers are photoexcited by the pump pulses incident on the surface. The decay of the transient reflectivity consists of the three processes as indicated. Two of them yield exponential decay with fast and slow relaxation time while the rest contributes to constant background in the time scale in Fig. 3.
Fig. 3. Transient reflectivity measured at wavelengths between 1.48 um and 1.52 pm. In Fig. 3, transient reflectivity obtained at wavelengths between 1.48 um and 1.52 ym is plotted in a logarithmic scale. The dots and the solid lines correspond to the experimental data and fitting curves, respectively. The wavelength range in the measurements lies in the long-wavelength edge of the stop band. The pump and probe pulses overlap at a delay time ~ 1.4 ps. The transient reflectivity increases when the hot carriers are photoexcited by the pump pulses incident on the surface. The decay of the transient reflectivity consists of the three processes as indicated. Two of them yield exponential decay with fast and slow relaxation time while the rest contributes to constant background in the time scale in Fig. 3.
Fig. 4. Illustration of pulse propagation modulated by photoexcited carries in InP/InGaAs NBR. reflection. The reflectivity increase occurs instantaneously as far as the group-delay dispersion is negligible. The experimental rise time of the transient reflectivity in Fig. 3 is limited by the duration of pump and probe pulses. The change in group delay is less than 10 fs over the pulse-spectral range as expected from the group-delay spectrum in Fig. 1. The reflectivity decay, on the other hand, is the reverse process induced by the absorption recovery during carrier thermalisation. The carrier thermalisation is limited by carrier relaxation processes.
Fig. 4. Illustration of pulse propagation modulated by photoexcited carries in InP/InGaAs NBR. reflection. The reflectivity increase occurs instantaneously as far as the group-delay dispersion is negligible. The experimental rise time of the transient reflectivity in Fig. 3 is limited by the duration of pump and probe pulses. The change in group delay is less than 10 fs over the pulse-spectral range as expected from the group-delay spectrum in Fig. 1. The reflectivity decay, on the other hand, is the reverse process induced by the absorption recovery during carrier thermalisation. The carrier thermalisation is limited by carrier relaxation processes.
Fig. 5. Magnitude ratio of fast (R,,,,) to slow (R components as a function of wavelength. iow?!
Fig. 5. Magnitude ratio of fast (R,,,,) to slow (R components as a function of wavelength. iow?!
Figure 1. Experimental setup for time-resolved electroabsorption. The conventional pump-probe transmission setup is modified with additional bias modulation on the sample. The sample is placed in a closed circuit cryostat at 10 K. D1 and D2 are InGaAs-photodetectors. unambiguously disentangled. For the incoherent transport a differential transmission technique with bias modulation is employed, which allows to derive the pure EA signal without contribution of bleaching induced by the pump pulse. On the coherent time scale we observe Bloch oscillations in the transmission signal as reported in a previous study [6] in this material system. As a results of dynamical screening due to incoherent transport we observe that the period of the decaying Bloch oscillations is chirped.
Figure 1. Experimental setup for time-resolved electroabsorption. The conventional pump-probe transmission setup is modified with additional bias modulation on the sample. The sample is placed in a closed circuit cryostat at 10 K. D1 and D2 are InGaAs-photodetectors. unambiguously disentangled. For the incoherent transport a differential transmission technique with bias modulation is employed, which allows to derive the pure EA signal without contribution of bleaching induced by the pump pulse. On the coherent time scale we observe Bloch oscillations in the transmission signal as reported in a previous study [6] in this material system. As a results of dynamical screening due to incoherent transport we observe that the period of the decaying Bloch oscillations is chirped.
Figure 2. Field-induced transient transmission changes at 1.49 um (0.831 eV) at 10 K for different reverse bias voltage (V) at an average optical power of 2.5 mW. The curves are shifted with respect to each other for clarity. The dashed line at -1.05 V is a linear approximation of the transmission change around {=1 ps. Figure 2 shows the induced differential transmission change at on-off reverse bias voltages from -0.8 V to -1.10 V at an average optical excitation power of 2.5 mW.
Figure 2. Field-induced transient transmission changes at 1.49 um (0.831 eV) at 10 K for different reverse bias voltage (V) at an average optical power of 2.5 mW. The curves are shifted with respect to each other for clarity. The dashed line at -1.05 V is a linear approximation of the transmission change around {=1 ps. Figure 2 shows the induced differential transmission change at on-off reverse bias voltages from -0.8 V to -1.10 V at an average optical excitation power of 2.5 mW.
where i(V) is cw photocurrent which is proportional to the absorption. We compared the voltage dependence of transmission signal with the cw photocurrent change averaged over the spectral width of the fs-pulses of 20 meV at a photon energy of 0.831 eV. Figure 3 depicts the amplitude of Fig.2 at a time delay of 3.5 ps as a function of bias voltage. Figure 3. Amplitude of the data in Fig.2 at 3.5 ps as a function of bias voltage. The dashed line (slope) is used to estimate the voltage drop due to drift field screening during the dephasing of Bloch oscillations.
where i(V) is cw photocurrent which is proportional to the absorption. We compared the voltage dependence of transmission signal with the cw photocurrent change averaged over the spectral width of the fs-pulses of 20 meV at a photon energy of 0.831 eV. Figure 3 depicts the amplitude of Fig.2 at a time delay of 3.5 ps as a function of bias voltage. Figure 3. Amplitude of the data in Fig.2 at 3.5 ps as a function of bias voltage. The dashed line (slope) is used to estimate the voltage drop due to drift field screening during the dephasing of Bloch oscillations.
Figure 4. The negative differential cw photocurrent as function of bias voltage integrated over 20 meV at 0.831 eV.
Figure 4. The negative differential cw photocurrent as function of bias voltage integrated over 20 meV at 0.831 eV.
Figure 5. (a) Extracted Bloch oscillations at V=-1.05 V (dashed line). The solid line is a numerical fit with a linear chirp év/61 at a frequency of 3.7 THz. (b) Normalized Fourier power spectra of the experimental data in (a) at different time intervals.
Figure 5. (a) Extracted Bloch oscillations at V=-1.05 V (dashed line). The solid line is a numerical fit with a linear chirp év/61 at a frequency of 3.7 THz. (b) Normalized Fourier power spectra of the experimental data in (a) at different time intervals.
and idler waves from one of the OPAs are mixed in a AgGaS, crystal to produce femtosecond mid-infrared pulses by difference frequency generation (DFG). Thus, we have two independently tunable, synchronized sources of ultrashort pulses: one gives us access to the spectral range of the valence to conduction band, interband, transitions and the other gives us access to the n=1 to n=2, conduction band derived, intersubband transition. The overall temporal resolution of our experiments is approximately 200 fs. Figure 1. Schematic diagram of the laser system. Sample Properties The MQW sample used in our experiments was grown by molecular beam epitaxy on an InP semi-insulating substrate. The structure has a total of 140 periods, each period consisting of a 5 nm wide Ing 5;Gap47As well and an 8 nm wide Inp5,Alp4gAs barrier. The well layers were selectively doped with Si atoms at a concentration of ~2x10'* cm®, corresponding to an electron sheet density of ~ 1x10"? cm?. The optical transition between the n=1 and
and idler waves from one of the OPAs are mixed in a AgGaS, crystal to produce femtosecond mid-infrared pulses by difference frequency generation (DFG). Thus, we have two independently tunable, synchronized sources of ultrashort pulses: one gives us access to the spectral range of the valence to conduction band, interband, transitions and the other gives us access to the n=1 to n=2, conduction band derived, intersubband transition. The overall temporal resolution of our experiments is approximately 200 fs. Figure 1. Schematic diagram of the laser system. Sample Properties The MQW sample used in our experiments was grown by molecular beam epitaxy on an InP semi-insulating substrate. The structure has a total of 140 periods, each period consisting of a 5 nm wide Ing 5;Gap47As well and an 8 nm wide Inp5,Alp4gAs barrier. The well layers were selectively doped with Si atoms at a concentration of ~2x10'* cm®, corresponding to an electron sheet density of ~ 1x10"? cm?. The optical transition between the n=1 and
Figure 3. Transmission in the neighborhood of the interband resonances. Figure 2. Transmission in the neighborhood of the intersubband resonance at 4.7 um.
Figure 3. Transmission in the neighborhood of the interband resonances. Figure 2. Transmission in the neighborhood of the intersubband resonance at 4.7 um.
n=2 conduction subbands gives rise to an absorption line centered at 4.7 um, [6] corresponding to a photon energy of 264 meV. The infrared transmission spectrum, measured near Brewster's angle, is shown in Figure 2 [7]. This transition energy is in good agreement with envelope function approximation [EFA] calculations [8]. The normal incidence, interband transmission, shown in Figure 3, shows the characteristic steps of a quasi-two- dimensional system [9]. Doping the MQW sample n-type partially bleaches and broadens the absorption edge for the n=1 (valence) to n=1 (conduction) subband transition [10]. The n=2 absorption edge is much less affected due to the negligible population in the n=2 subband; the Fermi energy is estimated to be approximately 60 meV above the bottom of the n=1 subband.
n=2 conduction subbands gives rise to an absorption line centered at 4.7 um, [6] corresponding to a photon energy of 264 meV. The infrared transmission spectrum, measured near Brewster's angle, is shown in Figure 2 [7]. This transition energy is in good agreement with envelope function approximation [EFA] calculations [8]. The normal incidence, interband transmission, shown in Figure 3, shows the characteristic steps of a quasi-two- dimensional system [9]. Doping the MQW sample n-type partially bleaches and broadens the absorption edge for the n=1 (valence) to n=1 (conduction) subband transition [10]. The n=2 absorption edge is much less affected due to the negligible population in the n=2 subband; the Fermi energy is estimated to be approximately 60 meV above the bottom of the n=1 subband.
Figure 4. Degenerate pump-probe signals for two different intensities.
Figure 4. Degenerate pump-probe signals for two different intensities.
Figure 5. Time resolved transmission change for select wavelengths. Some results of non-degenerate experiments for probe wavelengths in the interband absorption range are presented in Figures 5 and 6. In these experiments the pump is tuned to resonantly excite the n=l to n=2 intersubband transition. The dynamics of the n=1 subband electrons at various energies are investigated by tuning the probe beam between 1.2 and 1.54 um. Since we are probing between a full valence band and an excited conduction subband, the signal we measure is only dependent on the changes in the population of the n=1 conduction subband. Using the fact that the conduction subbands are nearly parallel, we expect a uniform depopulation of the n=1 subband. This leads us to expect that the initial behavior of the pump-probe signal will be induced absorption, when probing initially populated states. This is indeed what we find, but by studying the subsequent dynamics we learn a great deal more about the relaxation processes.
Figure 5. Time resolved transmission change for select wavelengths. Some results of non-degenerate experiments for probe wavelengths in the interband absorption range are presented in Figures 5 and 6. In these experiments the pump is tuned to resonantly excite the n=l to n=2 intersubband transition. The dynamics of the n=1 subband electrons at various energies are investigated by tuning the probe beam between 1.2 and 1.54 um. Since we are probing between a full valence band and an excited conduction subband, the signal we measure is only dependent on the changes in the population of the n=1 conduction subband. Using the fact that the conduction subbands are nearly parallel, we expect a uniform depopulation of the n=1 subband. This leads us to expect that the initial behavior of the pump-probe signal will be induced absorption, when probing initially populated states. This is indeed what we find, but by studying the subsequent dynamics we learn a great deal more about the relaxation processes.
Figure 6. Spectrally resolved transmission change for select time delays. The lines are intended to guide the eye. decay, presented only for qualitative comparison. When we tune the probe wavelength shorter than 1.25 um we observe induced absorption followed by bleaching. The behavior of the pump-probe signals shown in Figure 5 can be qualitatively understood by considering two effects. First, electrons excited to the n=2 subband will undergo intersubband electron-phonon scattering back to the n=1 subband with ~200 meV of excess energy where they interact with the wnexcited (ignoring free carrier absorption), but nonequilibrium, n=1 population. Second, the unexcited n=1 population will also undergo electron- electron and electron-phonon intrasubband scattering, while seeking a new equilibrium. These two process both conspire to produce the transient overshoot in the n=1 subband population as well as drive the return to equilibrium. Another way to visualize this process is to spectrally resolve the relaxation.
Figure 6. Spectrally resolved transmission change for select time delays. The lines are intended to guide the eye. decay, presented only for qualitative comparison. When we tune the probe wavelength shorter than 1.25 um we observe induced absorption followed by bleaching. The behavior of the pump-probe signals shown in Figure 5 can be qualitatively understood by considering two effects. First, electrons excited to the n=2 subband will undergo intersubband electron-phonon scattering back to the n=1 subband with ~200 meV of excess energy where they interact with the wnexcited (ignoring free carrier absorption), but nonequilibrium, n=1 population. Second, the unexcited n=1 population will also undergo electron- electron and electron-phonon intrasubband scattering, while seeking a new equilibrium. These two process both conspire to produce the transient overshoot in the n=1 subband population as well as drive the return to equilibrium. Another way to visualize this process is to spectrally resolve the relaxation.
Figure 1. LTG-Be:InGaAs/InAlAs MQW p-i-n structure with mesa-etch profile and ohmic contacts.
Figure 1. LTG-Be:InGaAs/InAlAs MQW p-i-n structure with mesa-etch profile and ohmic contacts.
Figure 2. I-V characteristic for a LTG-MQW p-i-n modulator.
Figure 2. I-V characteristic for a LTG-MQW p-i-n modulator.
Figure 3. Linear absorption spectra of the STG and LTG MQW p-i-n structures. Excitation wavelengths used in the nonlinear transmission measurements are indicated by triangles (W). Figure 3 shows the linear absorption spectra of the unbiased STG and LTG p-i-n structures as measured at room-temperature. Note that the first heavy-hole-to-
Figure 3. Linear absorption spectra of the STG and LTG MQW p-i-n structures. Excitation wavelengths used in the nonlinear transmission measurements are indicated by triangles (W). Figure 3 shows the linear absorption spectra of the unbiased STG and LTG p-i-n structures as measured at room-temperature. Note that the first heavy-hole-to-
Figure 4. Pump/probe measurement configuration.
Figure 4. Pump/probe measurement configuration.
Figure 5. Nonlinear differential transmission of the STG (solid) and LTG (dotted) MQW p-i-n structures.
Figure 5. Nonlinear differential transmission of the STG (solid) and LTG (dotted) MQW p-i-n structures.
Figure 6. Effect of Be-doping and 500°C post-growth anneal on nonlinear absorption recovery. Pulse energy density = 40 fI/um2. The low-excitation (40 fJ/um*) normalized differential transmission data of Figure 6 reveal that the combination of LTG and Be-incorporation can be used to engineer the absorption recovery time in the MQW material. The STG material exhibits a 1/e-recovery time of ~125 ps. Material grown at 200°C and Be-doped with a concentration of 5x10"” cm? exhibits a reduced 1/e-recovery time of 17 ps. Increasing the Be doping to 2x10"* cm® further decreases the recovery time to 600 fs. We note that this recovery- time estimate is not significantly reduced by PIA effects since the free-carrier lifetime in this material was independently determined to be sub-ps by far-infrared THz measurements using the technique described in [10]. Following a post-growth anneal (500°C for 30 min), the low- and high-doped recovery times increase to 24 and 2.5 ps, respectively. The apparent resolution difference of the
Figure 6. Effect of Be-doping and 500°C post-growth anneal on nonlinear absorption recovery. Pulse energy density = 40 fI/um2. The low-excitation (40 fJ/um*) normalized differential transmission data of Figure 6 reveal that the combination of LTG and Be-incorporation can be used to engineer the absorption recovery time in the MQW material. The STG material exhibits a 1/e-recovery time of ~125 ps. Material grown at 200°C and Be-doped with a concentration of 5x10"” cm? exhibits a reduced 1/e-recovery time of 17 ps. Increasing the Be doping to 2x10"* cm® further decreases the recovery time to 600 fs. We note that this recovery- time estimate is not significantly reduced by PIA effects since the free-carrier lifetime in this material was independently determined to be sub-ps by far-infrared THz measurements using the technique described in [10]. Following a post-growth anneal (500°C for 30 min), the low- and high-doped recovery times increase to 24 and 2.5 ps, respectively. The apparent resolution difference of the
Figure 8. Initial electroabsorptive differential transmission results (dotted) for LTG-Be:InGaAs/InAlAs MQW p-i-n structure. The linear absorption spectrum (solid) is provided for reference.
Figure 8. Initial electroabsorptive differential transmission results (dotted) for LTG-Be:InGaAs/InAlAs MQW p-i-n structure. The linear absorption spectrum (solid) is provided for reference.
Figure 7. Linear absorption (solid) and nonlinear differential absorption (™@) for the LTG Be-doped InGaAs/InAlIAs MQWs: (a) unbiased p-i-n structure, and (b) undoped cladding structure.
Figure 7. Linear absorption (solid) and nonlinear differential absorption (™@) for the LTG Be-doped InGaAs/InAlIAs MQWs: (a) unbiased p-i-n structure, and (b) undoped cladding structure.
Table 1. Fitting parameters times agree well with those reported in other references. However, the trapped carrier lifetimes are surprising be- cause they increase for the faster, lower-temperature ma- The electron trapping times, tr = 1/7, and the trapped carrier lifetimes, 7 = 1/+, derived from the curve fits are shown in Table 1. The electron trapping
Table 1. Fitting parameters times agree well with those reported in other references. However, the trapped carrier lifetimes are surprising be- cause they increase for the faster, lower-temperature ma- The electron trapping times, tr = 1/7, and the trapped carrier lifetimes, 7 = 1/+, derived from the curve fits are shown in Table 1. The electron trapping
Figure 1. The differential signal and the fit curves for the 220°C grown GaAs sample with pump energies of 80 nJ, 40 nJ, 8 nJ and 0.8 nJ. The inset shows the 40 nJ curves on a log scale. where f, is the fraction of injected electrons left in the conduction band, g(t) is the electron generation term (pump pulse), N, is the initial carrier density, pre g(t), fy: is the fraction of traps which are occupied, 7. and ‘Yt are as mentioned previous and the a coefficient is related y, 7 is the ratio of (N,/N:), o the Auger process which is only relevant for the 210 °C sample. The model has three free parameters, neglecting the auger term. For each sample, we choose 7, by fitting the low-pump-energy data with a single exponential decay. Then we choose 7; so that the only parameter w hich needs to be adjusted to fit all the curves for a given sample is 7 (which is propor- tional to the number of injected electrons). This simple model results in very good fits over almost three orders of magnitude for all incident pump energies as demon- strated in Fig. 1.
Figure 1. The differential signal and the fit curves for the 220°C grown GaAs sample with pump energies of 80 nJ, 40 nJ, 8 nJ and 0.8 nJ. The inset shows the 40 nJ curves on a log scale. where f, is the fraction of injected electrons left in the conduction band, g(t) is the electron generation term (pump pulse), N, is the initial carrier density, pre g(t), fy: is the fraction of traps which are occupied, 7. and ‘Yt are as mentioned previous and the a coefficient is related y, 7 is the ratio of (N,/N:), o the Auger process which is only relevant for the 210 °C sample. The model has three free parameters, neglecting the auger term. For each sample, we choose 7, by fitting the low-pump-energy data with a single exponential decay. Then we choose 7; so that the only parameter w hich needs to be adjusted to fit all the curves for a given sample is 7 (which is propor- tional to the number of injected electrons). This simple model results in very good fits over almost three orders of magnitude for all incident pump energies as demon- strated in Fig. 1.
Figure 3. The differential signal and the fit curves for the 210°C-grown GaAs sample with pump-pulse energies of 1 nJ, 12 nJ and 40 nJ. The inset shows more clearly the increased initial decay rate in the high-fluence measurement. The Auger rate in pure crystals is strongly limited by mo- mentum conservation requirements. [5] From our data we estimate the Auger coefficient to be on the order of 107?° cm® s~!, several orders of magnitude larger than the es- timated values for normal GaAs. We believe momentum conservation is relaxed in the 210°C-grown GaAs by the large concentration of defects that reduce the crystalinity. This relaxation creates a much wider range of momentum states into which the Auger-excited electrons may scat- ter, resulting in a drastically increased Auger rate.
Figure 3. The differential signal and the fit curves for the 210°C-grown GaAs sample with pump-pulse energies of 1 nJ, 12 nJ and 40 nJ. The inset shows more clearly the increased initial decay rate in the high-fluence measurement. The Auger rate in pure crystals is strongly limited by mo- mentum conservation requirements. [5] From our data we estimate the Auger coefficient to be on the order of 107?° cm® s~!, several orders of magnitude larger than the es- timated values for normal GaAs. We believe momentum conservation is relaxed in the 210°C-grown GaAs by the large concentration of defects that reduce the crystalinity. This relaxation creates a much wider range of momentum states into which the Auger-excited electrons may scat- ter, resulting in a drastically increased Auger rate.
where cris the conductivity, @p is the plasma frequency and I is the momentum relaxation time. Figure 1: Terahertz setup for the pump-probe and absorption spectroscopy.
where cris the conductivity, @p is the plasma frequency and I is the momentum relaxation time. Figure 1: Terahertz setup for the pump-probe and absorption spectroscopy.
Figure 2: Typical sample structure grown for the study. A typical sample structure is shown in Fig. 2.
Figure 2: Typical sample structure grown for the study. A typical sample structure is shown in Fig. 2.
Figure 3: Transient THz transmission of 350°C as-grown GaAs (0.25% excess As) following near band-edge excitation. (a) Sub- picosecond electron capture is shown for the 3pm thick LTG layer. Similarly fast carrier capture is observed in the 1pm thick layer (inset) with a slow contribution from the substrate. (b) Transient response of the 1m thick layer annealed at 600°C and 900°C. The solid lines depict the results of our model. The slower.response of the annealed material is accompanied by a ~50% mobility enhancement :
Figure 3: Transient THz transmission of 350°C as-grown GaAs (0.25% excess As) following near band-edge excitation. (a) Sub- picosecond electron capture is shown for the 3pm thick LTG layer. Similarly fast carrier capture is observed in the 1pm thick layer (inset) with a slow contribution from the substrate. (b) Transient response of the 1m thick layer annealed at 600°C and 900°C. The solid lines depict the results of our model. The slower.response of the annealed material is accompanied by a ~50% mobility enhancement :
Figure 4: Transient THz transmission of 300°C grown GaAs (0.50% excess As) for both 600°C and 900°C anneal. The relative response is to be compared with the data of Fig. 3(b). The capture times are faster than similarly annealed 350°C grown material. The solid lines depict the results of our model.
Figure 4: Transient THz transmission of 300°C grown GaAs (0.50% excess As) for both 600°C and 900°C anneal. The relative response is to be compared with the data of Fig. 3(b). The capture times are faster than similarly annealed 350°C grown material. The solid lines depict the results of our model.
Figure 5: Normalized transient response of lm thick as grown LTG-GaAs layers grown at various temperatures: from 350°C to 500°C. The change in the initial fast decay of the signal with increasing growth temperature is quite apparent. In Fig. 5 the normalized response curves obtained for all the lpm thick as-grown LTG-GaAs epilayer at various growth temperatures are shown. The initial fast decay vanishes as the growth temperature is increased from 350°C to 500°C.
Figure 5: Normalized transient response of lm thick as grown LTG-GaAs layers grown at various temperatures: from 350°C to 500°C. The change in the initial fast decay of the signal with increasing growth temperature is quite apparent. In Fig. 5 the normalized response curves obtained for all the lpm thick as-grown LTG-GaAs epilayer at various growth temperatures are shown. The initial fast decay vanishes as the growth temperature is increased from 350°C to 500°C.
Figure 6: Time resolved transient spectroscopy of 350°C grown GaAs. For the experimental conditions AT is proportional to the conductivity as is shown to be Drude-like within Ips of excitation. From the Drude model fit to the above data, the observed momentum relaxation time (ty) of the excited carriers is much smaller (~180fs+50fs) than the capture time. This is in excellent agreement with the value reported earlier in the literature [19].
Figure 6: Time resolved transient spectroscopy of 350°C grown GaAs. For the experimental conditions AT is proportional to the conductivity as is shown to be Drude-like within Ips of excitation. From the Drude model fit to the above data, the observed momentum relaxation time (ty) of the excited carriers is much smaller (~180fs+50fs) than the capture time. This is in excellent agreement with the value reported earlier in the literature [19].
TABLE I. Electron capture time (in ps) as obtained from our best fit for the 1pm thick LTG layer. The relative electron mobility also extracted from the model is shown in [] brackets.
TABLE I. Electron capture time (in ps) as obtained from our best fit for the 1pm thick LTG layer. The relative electron mobility also extracted from the model is shown in [] brackets.
where », is the electron density, 7, is the filled trap density, Nop is the total defect density, , and 7, are hole and electron recombination time and 7z,,, is the decay time in where », is the electron density, n, is the filled trap density, The rate equations for the electron capture are written below assuming a single defect level, which lies within the energy gap.
where », is the electron density, 7, is the filled trap density, Nop is the total defect density, , and 7, are hole and electron recombination time and 7z,,, is the decay time in where », is the electron density, n, is the filled trap density, The rate equations for the electron capture are written below assuming a single defect level, which lies within the energy gap.

Related papers

<title>Picosecond silicon metal-semiconductor-metal photodiode</title>
Thomas Hsiang

Photodetectors and Power Meters, 1993

The view, opinions end/or findings contained in this report are those of the ar (s) and should not be construed as an official Departmient. of the Army 1008 g volc. or decision, unless so des anted by other documentation. Ift 1151WTIO I AA"MUY STTEMET 1b. DISTRIUTION COol Approved for public release; distribution unlimmited. Theulrafstchaacerstis f cysaflnesiicon metal-semiconductor-meta (MSM) phokodloes with finger widths and spacings donto 200 nMmsbjected to fe1 oe_ n otca-e .1. excitations, was elci-picsmln system. Electricalrspne wit ful-width at hafnxmmas short as 3.7 ps, at a corsodn 31 dEbnwdt f10 Hwr generated by violet-light excitation. T7hese diodes are the fastest silicon photod-2etectors reported to date. Detailed bias and light-intensilty dependenc of the diode response has been measured These resuts are used to obtain the veloci~ reltin of electron in silicon And to demonstrate the ideal transit-time-limited response of the= 94-21055 TERMSD W1 111110111140P PAWIS silconvmealemicoductor-mWt phosodiode, eletr-oWi smpling systm W 1_ CmO or TIN PA OF~r3 A I an.

View PDFchevron_right
Transmission study of picosecond photocarrier dynamics in free-standing porous silicon
Petr Malý

Solid State Communications, 1994

View PDFchevron_right
Mixing frequency achievement with a photoinduced load in silicon substrate
jumana boussey

Microwave and Optical Technology Letters, 2001

View PDFchevron_right
A simple high-speed Si Schottky photodiode
Steven Brueck

IEEE Photonics Technology Letters, 2000

Design, fabrication , and UV-heterodyne characterization of Ni-Si-Ni metal-semiconductor-metal Schottky barrier photodetectors is reported. Planar detectors were fabricated, with a simple 3-level lithography process on bulk Si, in both simple-gap and interdigitated geometries with gap dimensions from 1 to 5 pm. Frequency response of these devices was characterized using a CW-laser heterodyne system at 334.5 nm. For a 4.5-pm interdigitated device, a 3-dB response of 16 GHz is measured, giving 22 GHz when deconvolved from the package/eonnector. A detailed theoretical model of the photodiode response incorporating effects of camer transport and device geometry is in excellent agreement with the measurement.

View PDFchevron_right
A 75 GHz silicon metal-semiconductor-metal Schottky photodiode
Thomas Hsiang

Applied Physics Letters, 1993

The ultrafast characteristics of crystalline-silicon metal-semiconductor-metal (MSM) photodiodes with 300 nm finger width and spacing were measured with a subpicosecond electro-optic sampling system. Electrical responses with full width at half maximum as short as 5.5 and 11 ps, at corresponding 3 dB bandwidths of 75 and 38 GHz, were generated by violet and red photons, respectively. The difference is attributed to the photon penetration depth which is much larger than the diode finger spacing at red, but smaller at violet. Light-intensity dependence was also examined at different wavelengths, indicating a linear relation and a higher sensitivity in the violet. These results not only demonstrated the fastest silicon photodetector reported to date, but also pinpointed the dominant speed-limiting factor of silicon MSM photodiodes. A configuration is suggested to improve the speed of these detectors at long wavelengths.

View PDFchevron_right

Related papers

On Ultrafast Photoconductivity Dynamics and Crystallinity of Black Silicon
Saydulla Persheyev

Ieee Transactions on Terahertz Science and Technology, 2013

View PDFchevron_right
Silicon photodiode-waveguide coupling—Two-dimensional modelling, software simulation and experiments
Florin Caldararu

Solid-State Electronics, 1996

View PDFchevron_right
Photosensor and optical wave wide coupling in silicon technology
J. Calderer

Sensors and Actuators a-Physical, 1997

ARROW-type optical waveguides are designed for implementation on silicon using the materials (silicon dioxide and silicon nitride) and techniques (CVD, RIE) of CMOS integrated-circuit technology. Light is detected by a photodiode buried in the silicon substrate, which is made following the same process. The steps of this process are described and their influence on the optical properties of the guides is analysed. The optical signal attenuation in the waveguide-photodiode coupling region and the cut-off frequency of the system are measured in test devices. The advantages of the technological compatibility with CMOS circuits are discussed. © 1997 Elsevier Science S.A.

View PDFchevron_right
Nonlinearity Measurements of Silicon Photodetectors
Petri Kärhä

Applied Optics, 1998

Nonlinearities of the responsivity of various types of silicon photodetectors have been studied. These detectors are based on photodiodes with two sizes of the active area ͑10 ϫ 10 mm 2 and 18 ϫ 18 mm 2 ͒. The detector configurations investigated include single photodiodes, two reflection trap detectors, and a transmission trap detector. For all devices, the measured nonlinearity was less than 2 ϫ 10 Ϫ4 for photocurrents up to 200 A. The diameter of the measurement beam was found to have an effect on the nonlinearity. The measured nonlinearity of the trap detectors depends on the polarization state of the incident beam. The responsivity of the photodetectors consisting of the large-area photodiodes reached saturation at higher photocurrent values compared with the devices based on the photodiodes with smaller active area.

View PDFchevron_right
Theoretical study of a nonlinear fast silicon photoconductive switch
daniel pascal

The European Physical Journal Applied Physics, 1999

Silicon-On-Insulator waveguide can be used as photoconductor, the light being coupled in the silicon film by a diffraction grating. Nonlinear effects are induced by the photogeneration of electronhole pairs. This leads to photocurrent variations which are much faster than in linear regime and also much steeper than the incident light pulse. A model is presented which takes into account the refractive index variations arising from both the excess carrier density and the temperature rise induced by carrier recombination and Joule effect. The photocurrent is calculated for various values of the incident light power, incidence angle and pulse duration.

View PDFchevron_right
A New Paradigm in High-Speed and High-Efficiency Silicon Photodiodes for Communication—Part I: Enhancing Photon–Material Interactions via Low-Dimensional Structures
Ekaterina Ponizovskaya Devine

IEEE Transactions on Electron Devices

Photodetectors (PDs) used in communication systems require ultrafast response, high efficiency, and low noise. PDs with such characteristics are increasingly in demand for data centers, metro data links, and longhaul optical networks. In a surface-illuminated PD, high speed and high efficiency are often a tradeoff, since a high-speed device needs a thin absorption layer to reduce the carrier transit time, whereas a high-efficiency device needs a thick absorption layer to compensate for the low absorption coefficient of some semiconductors such as Si and Germanium (Ge) or SiGe alloys at wavelengths near the bandgap. In this part of this review, we present the recent efforts in enhancing the photon-material interactions by using low-dimensional structures that can control light for more interaction with the photoabsorbing materials, slow down the propagation group velocity and reduce surface reflection. We present recent demonstrations of high-speed PDs based on nanostructures enabled by both synthetic bottom-up or transformative top-down processing methods. In particular, we detail a CMOS-compatible ultrafast surfaceilluminated Si PD with 30-ps full-width at half-maximum, and >50% efficiency at 850 nm. A complementary discussion on device challenges and the integration of low-dimensional structures will be presented in the part II of this review.

View PDFchevron_right
Nonlinear Optics in Silicon
Ozdal Boyraz

Series in Optics and Optoelectronics, 2013

View PDFchevron_right
Photocarrier Excitation and Transport in Hyperdoped Planar Silicon Devices
Michael J Aziz

MRS Proceedings, 2011

We report an experimental study of photocarrier lifetime, transport, and excitation spectra in silicon-on-insulator doped with sulfur far above thermodynamic saturation. The spectral dependence of photocurrent in coplanar structures is consistent with photocarrier generation throughout the hyperdoped and undoped sub-layers, limited by collection of holes transported along the undoped layer. Holes photoexcited in the hyperdoped layer are able to diffuse to the undoped layer, implying

View PDFchevron_right
Picosecond GaAs photoconductors on silicon substrates for local integration with silicon devices and circuits
R. Mariella

IEEE Electron Device Letters, 2000

A fabrication procedure for local integration of GaAs photoconductive devices with processed silicon circuits has been investigated. The process described allows isolated regions of GaAs to be epitaxially grown by MBE at temperatures which are compatible with already processed silicon circuits with first-level metdliiation. GaAs photoconductors with 15-pm gap lengths fabricated on silicon substrates have exhibited > 16-mA sampling oscilloscope limited responses, with electrical pulse widths less than 20 ps determined by autocorrelation measurements.

View PDFchevron_right
Nonlinear change in refractive index and transmission coefficient of silicon at long-pulse, mJ-range, 1.54-μm excitation
Alexander Kir'yanov

Laser Physics, 2011

An experimental study and theoretical modeling of the nonlinear changes in transmission coeffi cient and refractive index of mono crystalline Silicon (Si) at long pulse, mJ range, single beam Z scan probing at 1.54 µm wavelength are reported. It is shown experimentally that at increasing pulse energy density the photo induced darkening permanently increases in Si while its photo refraction properties demonstrate a more complicate character, being a product of various type nonlinearities. A theoretical analysis based on simple assumptions of a square shape pulse in the time domain and Gaussian spatial distribution of the probe beam allows fitting of a whole of the experimentally measured open and closed aperture Z scans through an account of the main contributions in the light induced absorption and refractive index nonlinearities. These are revealed to originate from non direct two photon absorption and Kerr effect, induced absorption and dis persion of light generated free carriers, and light induced thermal lensing.

View PDFchevron_right