We demonstrate waveguide-integrated silicon-germanium avalanche photodiodes with a maximum responsivity of 15.2 A/W at 16x avalanche gain, and 33 GHz bandwidth. Intensity-modulation-direct-detection (IMDD) and coherent channel reception test demonstrated the APD’s performance with higher-order formats, allowing 32 Gbaud PAM-4 and 40 Gbaud 16QAM channel reception without any digital signal processing conventionally used for receiver impairments mitigation.
© 2020 Optical Society of America under the terms of the OSA Open Access Publishing Agreement
The sustained growth in intra- and inter-data center trafﬁc has led to widespread deployment of high spectral efficiency modulation formats, such as multi-level pulse-amplitude modulation (PAM) and quadrature amplitude modulation (QAM). This development is motivated, among other requirements, by the need to scale throughputs while circumventing conventional power-consumption penalty associated with baudrate scaling, imposed by CMOS dynamic power . The use of multi-level formats, however, requires additional link power budget margin, provided by the combination of increased laser power on the transmitter side, reduced loss budget, or enhanced receiver sensitivity by elevating gain of the transimpedance amplifiers (TIA). Such link design approach consequently dictates higher cost and power consumption per data unit throughput, and lower link reach than binary channels. Such compromises impose barriers in system designs that are bound by strict power consumption envelope while demanding highest possible throughput: typical examples are represented by in-package interconnect designs for next generation switches and network interfaces.
Avalanche photodiodes (APDs) offer an alternative approach for advanced modulation format link designs in short-reach scenarios where optical power is limited. With internal gain generated by photocarriers-initiated impact ionization, APDs provide responsivity beyond quantum-limited external efﬁciency and enhance the sensitivity of thermal-noise limited receivers [2,3], thereby allowing for lower optical power into the photonic circuit, and reduced or even suppressed transimpedance gain requirement. Furthermore, advances made in low-defect Ge epitaxial growth on Si techniques allow integration of low dark-current APDs directly coupled with silicon photonic circuits, thus enabling low-cost manufacturing of APD-enabled systems using commercial silicon foundry processes.
A majority of demonstrations on waveguide-integrated Si-Ge APDs are based on separate-absorption-charge-multiplication (SACM) structure to reduce excess noise by confining carrier multiplication in Si [3–10]. The SACM approach, however, requires epitaxial silicon growth for field control and multiplication layer formation [3–9], which is incompatible with standard photonics processes optimized for dual-polarization operation . Another demonstration type , while requires no epitaxial silicon growth, shows limited bandwidth (10 GHz) due to reduced drift velocity resulted from the weak electric field in the Ge absorption layer and silicon charge layer. On the other hand, while Ge APDs are generally understood to suffer from high multiplication noise in germanium, these devices can achieve high gain using carrier multiplication close to avalanche breakdown , and noise could be effectively suppressed by manipulating the electric field and multiplication distribution [13–15]. Demonstrations with lateral PIN  and vertical PIN  structures all show improved sensitivity performances, however, at limited data rates of 10 Gbps.
Recognizing this limitation, we investigate another lateral PIN-based Ge-on-Si APDs fabricated in a standard foundry process capable of delivering high bandwidth and gain for multi-level PAM and QAM channel reception. The fabricated APD exhibits high primary responsivity of 0.95 A/W at 1550 nm, a 3-dB bandwidth of 33 GHz, and a multiplication gain of 16 at −20 dBm input power before reaching junction breakdown at 12.5V reverse bias. The APD is capable of receiving 64 Gbps (32 Gbaud) PAM−4 channel and 40 Gbaud 16QAM channel without receiver equalization. To the best of our knowledge, this is the first demonstration of Si-Ge APD for coherent detection.
2. Design and characterization
The Si-Ge waveguide APDs were fabricated at commercial foundry using 0.18 um silicon photonics process without Si epitaxial growth. The schematic of the designed APD with lateral PIN junction structure is shown in Fig. 1(a). A pair of shallow implants (p/n) with 1018 cm−3 peak concentration were used to form a p-i-n junction in silicon with 500 nm intrinsic width. The silicon p-i-n junction was optimized to generate strong electric field in Ge (shown in Fig. 1(d)) for initiating impact ionization, while simultaneously minimizing free carrier optical absorption by highly doped contact implants (p++ / n++). A 500-nm thick Ge optical absorption layer was subsequently deposited by selective epitaxial growth on the doped silicon slab.
Numerical model showed that the electric field (Fig. 1(d)) was concentrated to a 200-nm thick layer within the Ge-Si junction. Despite the observation that the peak electric field was located in the Si layer, cross-sectional integration of impact ionization generation rate showed that the carrier multiplication primarily occurred in Ge (96.5%), as Ge possesses an order of magnitude higher ionization coefficient within the operating field strength at 0.2–0.4 MV/cm , . Although the near-unity ratio between hole and electron ionization coefficient of Ge would have resulted in high excess noise , the localized electric field near Si-Ge boundary had largely confined impact ionization to within 200 nm of the implanted Si regions, which could be attributed to the reduced excess noise factor seen in the experimental characterization due to dead-space effect ,  . The APD (microscope image in Fig. 1(b)) was coupled to silicon waveguide via adiabatic taper, allowing efficient evanescent coupling of the input light to the 1 µm×50 µm Ge crystal. Subsequent dual-layer aluminum metallization provides electrical connection to exposed pads for characterization.
The dark / illuminated I-V characteristics at the room temperature (22°C), as well as the corresponding gain of the APD, are shown in Fig. 2(a). A continuous-wave laser at 1550 nm wavelength was coupled into the APD device via on-chip vertical grating couplers. The fiber-to-chip coupling loss, at 7 dB, was de-embedded from all optical power measurements. The in-waveguide power was −20 dBm. The dark current was 24 nA at < 4 V reverse bias, and increased to 100 µA at 12.5 V, beyond which the diode underwent junction breakdown. The primary responsivity at unity gain (bias < 2 V) was 0.95 A/W and a maximum gain of 16 was achieved before breakdown.
The frequency response at various bias voltages was measured with a calibrated electro-optic modulator and vector network analyzer (VNA), as shown in Fig. 2(b). The 3-dB opto-electrical (O/E) bandwidth at various multiplication gains is summarized in Fig. 2(c). The bandwidth increased with increasing bias voltage until reaching a maximum bandwidth of 33 GHz at 9.5 V bias. Further increase in reverse bias reduced bandwidth to 22 GHz at 12.5 V, due to increased ionization build-up time .
The noise characteristics of the APD were extracted from dark/light shot-noise measurements with −20 dBm in-waveguide optical power. The noise power spectral density (PSD) of the output current was measured using a low-noise amplifier and electrical signal analyzer. The measured noise PSD was compared against Eq. (1), which accounts for shot noise due to primary photocarrier generation and subsequent avalanche multiplication, as well as system thermal noise and laser relative intensity noise (RIN) :
In Eq. (1), q denotes the electron charge, ID and IL are the dark and photo current at unit gain, M is the corresponding gain, F is the excess noise factor, RL is the system impedance, and Δf is the bandwidth. Nthermal and NRIN are the thermal noise and laser RIN. To demonstrate that the design was able to leverage dead-space effect to reduce excess noise generation in Ge, the extracted excess noise factor F versus gain M was compared against theoretical values of McIntyre’s theory in terms of the ratio of the hole-electron ionization coefficients keff: 
The measured excess noise factor F (Fig. 3) showed that the effective ionization ratio keff was bounded within 0.15–0.25, far below the ionization ratio of bulk Ge (≈0.9) . The excess noise reduction is attributed to the localization of the ionization (Fig. 1(e)).
The linearity of the photoconductive response of the APD was evaluated by characterizing the total harmonic distortion (THD). A Mach-Zehnder modulator driven by a single-frequency signal was used to generate a stimulus at 5 GHz. The THD of the stimulus, measured using a reference photodiode (Discovery Semiconductor DSC-10H), was less than 0.7% at an input power up to −2 dBm. The THD of the APD, as shown in Fig. 4, shows that the device provided sufficiently linear response (< 2%) for PAM-4 and 16-QAM modulations, even at high gain and input power regimes.
3. PAM-4 channel reception
The performance of the APD for TIA-less PAM-4 channel reception was characterized using the setup shown in Fig. 5. The PAM-4 channel was generated by a 64 Gsamples/s digital-to-analog converter (DAC) with 16-GHz 3-dB bandwidth and was electrically amplified to drive a LiNbO3 Mach-Zehnder modulator (MZM) with a 3-dB bandwidth of 30 GHz. A 1549.3 nm external cavity laser with RIN < −140 dBc/Hz was used for optical carrier supply. The output of the MZM was amplified by an erbium-doped fiber amplifier (EDFA) and attenuated by a variable optical attenuator (VOA) for controlling incident power into the APD. Frequency response of the transmitter (DAC, modulator driver and modulator) as well as nonlinear distortions intrinsic to MZM and driver were measured using a reference 40-GHz photodetector (Discovery Semiconductor DSC-10H), and subsequently pre-compensated for in the digitized samples uploaded to the DAC. The photocurrent output of the APD was received by an equivalent-time oscilloscope with 50 GHz bandwidth via a GSG probe and a bias-tee, through which the desired reverse bias was applied. On-chip 50Ω shunt resistors across signal and ground pads of each photodiode were used to reduce RF mismatch loss in the RF probe and coaxial cables, thereby improving frequency response. However, the on-chip shunt termination inevitable decreased the amount of photocurrent reaching the oscilloscope, as half of the photocurrent was diverged to ground.
The PAM4 eye diagrams were captured at −3 dBm incident power to the APD under test, which corresponded to −9.2 dBm inner optical modulation amplitude (OMA). Figure 6 plots the Q-factor and BER results of the received eye against reverse bias from −3 V to −12 V, which corresponds to avalanche gain (M) of 1 to 2.6 at the −3 dBm input power level. The multiplication gain provided a maximum 1.7 dB increase in Q-factor over the thermal-noise limited value at M = 1. Absence of eye level distortion and inter-symbol interference (ISI) in the recorded eye diagram suggested that the APD response remained linear over the bandwidth of the channel (∼ 26 GHz).
4. 16-QAM coherent channel reception
Coherent transmission further enhances spectral efficiency by allowing higher cardinality modulations than intensity-modulation-direct-detection (IMDD) systems at the same SNR, thereby providing scalable path towards Tb/s/wavelength capacity. Leveraging internal gain provided by APDs, coherent receiver with APDs can improve the sensitivity with a lower optical power or electrical gain than traditional coherent receivers , thereby reducing power consumption of such systems.
The performance of the proposed APD in coherent channel detection was characterized using APDs integrated with silicon optical hybrid based on 4 × 4 multimode interferometric (MMI) coupler, as shown in Fig. 7. The test channel was created from an external cavity laser at 1549.3 nm with 5 kHz linewidth. A 16-QAM modulation at variable baud-rate was imprinted onto the carrier using a nested-MZM with > 30 GHz bandwidth, driven by quad-channel DAC with 64 Gsamples/s sampling rate. The data pattern was shaped by a raised-cosine filter with a roll-off factor of 0.1. Pre-compensation of the data pattern was further applied in order to remove the frequency response and nonlinear distortion of the transmitter (nested-MZM + modulator driver + DAC) and that of the real-time oscilloscope (RTO) serving as the digitizer. The frequency response and nonlinear distortion of the transmitter and RTO was characterized using a 40-GHz wide reference coherent receiver constructed out of discrete InP photodetectors (Finisar BPDV2120R) and optical hybrid. To allow control of channel optical signal-to-noise ratio (OSNR), a filtered amplified spontaneous emission (ASE) noise source with variable output power was coupled with the 16-QAM channel before entering the receiver, and the OSNR was monitored prior to the device under test. The channel was further filtered to 0.6 nm bandwidth and amplified before being coupled to the signal port of the APD-integrated coherent receiver. The LO was derived from the same laser source powering the test channel via a 50/50 coupler, and subsequently amplified to compensate for fiber-to-chip coupling loss. The photocurrents from the four APDs of the integrated coherent receiver were coupled to the input ports of the real-time oscilloscope (Tektronix DPO72004A), which captured the photocurrent (via the internal 50 Ω load) at 50 Gsamples/s. Channel deskew and carrier phase recovery were performed offline to recover the transmitted data. Subsequent characterizations were performed at an averaged signal / LO power of −9 dBm / −7 dBm respectively into each APD, of which the fiber-to-chip coupling loss of 7 dB and hybrid excess loss of 1 dB were taken into account.
Figure 8 depicts the sensitivity of the APD-integrated coherent receiver in terms of Q2 versus OSNR at 0.1 nm for a 40-Gbaud 16-QAM channel when biased at −3 V and −11 V (i.e., M = 1 and 2.2 respectively). The internal gain of the APDs provided a 2-dB improvement in Q-factor over the thermal-noise limited receiver at M = 1, and allowed reaching BER = 4.5×10−3 (Q2 = 8.8 dB) at 14.5 dB OSNR which corresponds to error-free reception with the staircase hard decision forward error-correction coding (HDFEC) . Near-theoretic sensitivity and distortion-free constellation further suggest that no discernible nonlinear distortion (compression and bandwidth modulation due to output current) was present in this coherent receiver.
We presented a waveguide-integrated Si-Ge APD fabricated in standard silicon photonics foundry process. A high primary responsivity of 0.95 A/W, a gain of 16, and a bandwidth of 26–33 GHz allowed signal reception of 32 Gbaud PAM-4 channel and 40 Gbaud 16QAM channel. The reported p-i-n design enables massive, low-cost integration of avalanche devices in standard silicon photonics platform without extra Si epitaxy steps.
Defense Advanced Research Projects Agency (DARPA).
The authors declare no conflicts of interest.
1. D. Thomson, A. Zilkie, J. E. Bowers, T. Komljenovic, G. T. Reed, L. Vivien, D. Marris-Morini, E. Cassan, L. Virot, J.-M. Fédéli, J.-M. Hartmann, J. H. Schmid, D.-X. Xu, F. Boeuf, P. O’Brien, G. Z. Mashanovich, and M. Z. Nedeljkovic, “Roadmap on silicon photonics,” J. Opt. 18(7), 073003 (2016). [CrossRef]
2. Payman Zarkesh-Ha, Robert Efroymson, Earl Fuller, Joe C. Campbell, and Majeed M. Hayat, “5.2 dB Sensitivity Enhancement in 25Gbps APD-Based Optical Receiver Using Dynamic Biasing,” Optical Fiber Communication Conference. Optical Society of America, 2020.
3. Z. Huang, C. Li, D. Liang, K. Yu, C. Santori, M. Fiorentino, W. Sorin, S. Palermo, and R. G. Beausoleil, “25 Gbps low-voltage waveguide Si–Ge avalanche photodiode,” Optica 3(8), 793–798 (2016). [CrossRef]
4. Y. Kang, H. D. Liu, M. Morse, M. J. Paniccia, M. Zadka, S. Litski, G. Sarid, A. Pauchard, Y. H. Kuo, H. W. Chen, W. S. Zaoui, J. E. Bowers, A. Beling, D. C. McIntosh, X. Zheng, and J. C. Campbell, “Monolithic germanium/silicon avalanche photodiodes with 340 GHz gain–bandwidth product,” Nat. Photonics 3(1), 59–63 (2009). [CrossRef]
5. M. Huang, Su. Li, C. Pengfei, H. Guanghui, S. Tzung-I, W. Chen, C.-y. Hong, and D. Pan, “Germanium on silicon avalanche photodiode,” IEEE J. Sel. Top. Quantum Electron. 24(2), 1–11 (2018). [CrossRef]
6. Sungbong Park, Yann Malinge, Olufemi Dosunmu, Gregory Lovell, Seth Slavin, Kelly Magruder, Yimin Kang, and Ansheng Liu, “50-Gbps receiver subsystem using Ge/Si avalanche photodiode and integrated bypass capacitor,” 2019 Optical Fiber Communications Conference and Exhibition (OFC). IEEE, 2019.
7. Binhao Wang, Zhihong Huang, Xiaoge Zeng, Di Liang, Marco Fiorentino, Wayne V. Sorin, and Raymond G. Beausoleil, “50 Gb/s PAM4 low-voltage Si-Ge avalanche photodiode,” CLEO: Science and Innovations. Optical Society of America, 2019.
8. Zhihong Huang, Binhao Wang, Yuan Yuan, Di Liang, Marco Fiorentino, and Raymond G. Beausoleil, “64Gbps PAM4 Modulation for a Low Energy Si-Ge Waveguide APD with Distributed Bragg Reflectors,” Optical Fiber Communication Conference. Optical Society of America, 2020.
9. Yuan Yuan, Zhihong Huang, Binhao Wang, Wayne Sorin, Di Liang, Joe C. Campbell, and Raymond G. Beausoleil, “Superior Temperature Performance of Si-Ge Waveguide Avalanche Photodiodes at 64Gbps PAM4 Operation,” 2020 Optical Fiber Communications Conference and Exhibition (OFC). IEEE, 2020.
10. N. J. D. Martinez, C. T. Derose, R. W. Brock, A. L. Starbuck, A. T. Pomerene, A. L. Lentine, D. C. Trotter, and P. S. Davids, “High performance waveguide-coupled Ge-on-Si linear mode avalanche photodiodes,” Opt. Express 24(17), 19072–19081 (2016). [CrossRef]
11. A. E.-J. Lim, J. Song, Q. Fang, C. Li, X. Tu, N. Duan, K. K. Chen, R. P.-C. T. Tern, and T. Y. Liow, “Review of silicon photonics foundry efforts,” IEEE J. Sel. Top. Quantum Electron. 20(4), 405–416 (2014). [CrossRef]
12. H. Melchior and W. T. Lynch, “Signal and noise response of high speed germanium avalanche photodiodes,” IEEE Trans. Electron Devices ED-13(12), 829–838 (1966). [CrossRef]
13. S. Assefa, F. Xia, and Y. A. Vlasov, “Reinventing germanium avalanche photodetector for nanophotonic on-chip optical interconnects,” Nature 464(7285), 80–84 (2010). [CrossRef]
14. L. Virot, P. Crozat, J.-M.. Fédéli, J.-M. Hartmann, D. Marris-Morini, E. Cassan, F. Boeuf, and L. Vivien, “Germanium avalanche receiver for low power interconnects,” Nat. Commun. 5(1), 4957 (2014). [CrossRef]
15. H. T. Chen, J. Verbist, P. Verheyen, P. D. Heyn, G. Lepage, J. D. Coster, P. Absil, X. Yin, J. Bauwelinck, J. Van Campenhout, and G. Roelkens, “High sensitivity 10Gb/s Si photonic receiver based on a low-voltage waveguide-coupled Ge avalanche photodetector,” Opt. Express 23(2), 815–822 (2015). [CrossRef]
16. R. Van Overstraeten and H. De Man, “Measurement of the ionization rates in diffused silicon pn junctions,” Solid-State Electron. 13(5), 583–608 (1970). [CrossRef]
17. D. R. Decker and C. N. Dunn, “Determination of germanium ionization coefficients from small-signal IMPATT diode characteristics,” IEEE Trans. Electron Devices 17(4), 290–299 (1970). [CrossRef]
18. M. M. Hayat, W. L. Sargeant, and B. E. Saleh, “Effect of dead space on gain and noise in Si and GaAs avalanche photodiodes,” IEEE J. Quantum Electron. 28(5), 1360–1365 (1992). [CrossRef]
19. M. M. Hayat, O.-H. Kwon, S. Wang, J. C. Campbell, B. E. A. Saleh, and M. C. Teich, “Boundary effects on multiplication noise in thin heterostructure avalanche photodiodes: theory and experiment [Al/sub 0.6/Ga/sub 0.4/As/GaAs],” IEEE Trans. Electron Devices 49(12), 2114–2123 (2002). [CrossRef]
20. M. Teich, K. Matsuo, and B. Saleh, “Excess noise factors for conventional and superlattice avalanche photodiodes and photomultiplier tubes,” IEEE J. Quantum Electron. 22(8), 1184–1193 (1986). [CrossRef]
21. R. J. McIntyre, “Multiplication noise in uniform avalanche diodes,” IEEE Trans. Electron Devices ED-13(1), 164–168 (1966). [CrossRef]
22. K. Wen, Y. Zhao, J. Gao, S. Zhang, and J. Tu, “Design of a coherent receiver based on InAs electron avalanche photodiode for free-space optical communications,” IEEE Trans. Electron Devices 62(6), 1932–1938 (2015). [CrossRef]
23. B. P. Smith, A. Farhood, A. Hunt, F. R. Kschischang, and J. Lodge, “Staircase codes: FEC for 100 Gb/s OTN,” J. Lightwave Technol. 30(1), 110–117 (2012). [CrossRef]