Open Access
Add to BibSonomyAbstract
This paper experimentally demonstrates a design optimization of an evanescently-coupled waveguide germanium-on-silicon photodetector (PD) towards high-speed (> 30 Gb/s) applications. The resulting PD provides a responsivity of 1.09 A/W at 1550 nm, a dark current of 3.5 µA and bandwidth of 42.5 GHz at 2 V reverse-bias voltage. To optimize the PD, the impact of various design parameters on performance is investigated. A novel optimization methodology for the PD’s responsivity based on the required bandwidth is developed. The responsivity of the PD is enhanced by enlarging its geometry and using off-centered contacts on top of the germanium, while an integrated peaking inductor mitigates the inherent bandwidth reduction from the responsivity optimization. The performance of the optimized PD and the conventional, smaller size non-optimized PD is compared to validate the optimization methodology. The sensitivity of the optimized PD improves by 3.2 dB over a smaller size non-optimized PD. The paper further discusses the impact of top metal contacts on the photodetector’s performance.
© 2016 Optical Society of America
1. Introduction
Silicon Photonics (SiP) has attracted significant attention in the last decade due to the possibility of integration using mature standard semiconductor fabrication techniques. Such integration enables more complex designs in a small footprint making SiP a prospective technology solution for the rapid growth of bandwidth required in today’s communication systems, especially for short-link applications. In the SiP platform, the waveguide germanium (Ge) photodetector (PD) has been the subject of important research in the last decade. In short-reach applications, the receiver noise is the dominant noise source. In this context, a PD with larger responsivity improves the receiver sensitivity and further reduces the required electrical gain of the downstream circuits. Consequently, more power-efficient optical interconnections are achieved, essential in such applications. Significant effort has been made to improve the performance of Ge-PDs by optimizing the fabrication process to reduce the dark current and increase the PD’s responsivity [1]. Recently, it has been shown that the responsivity of a Ge-PD improves considerably by reducing the optical loss caused by the metal contact directly above the Ge area. Three effective ways are proposed in the literature to reduce metal absorption loss: 1) removing the metal contact above the Ge area and changing the vertical p-i-n diode’s structure to a lateral one [2–4], 2) using a small size off-centered top contact above the Ge [5, 6], and 3) using multi-finger connections instead of large contacts [7]. Some of these techniques may be limited by the design rules of a specific fabrication process, such as minimum feature size, exclusion and inclusion distances. For instance, to fabricate a PD without having the metal contact above the Ge region, the width of the Ge would need to be as small as 0.5 µm to reduce the carrier transient time [3]. This is a feature size that is not provided in the fabrication run in which the PDs in this paper were fabricated. Further, the required minimum feature size of a contact via and the minimum distance between two contacts challenges the use of multi-finger compared to an off-centered top contact. Consequently, an off-centered top contact remains a more practical solution to reduce metal absorption.
The responsivity of a photodetector can be enhanced by increasing the photodetector’s geometry at the cost of lower bandwidth from inherently larger junction capacitance. However, inductive peaking can be used to mitigate that loss in bandwidth [8–10]. This work presents an optimization of a PD’s design geometry and the size and location of the top metal contact on the PD’s performance. Further, a novel methodology is detailed to optimize the responsivity and bandwidth of a PD for a given data rate. The proposed optimization process investigates design geometry for larger responsivity of a PD and considers modifying the top metal contacts within the fabrication constraints to reduce the optical loss caused by metal connection while an integrated peaking inductor mitigates the bandwidth tradeoff. To validate the optimization process, a common design from a foundry process design kit (PDK) of a small size photodetector with a germanium size of 8 × 8 µm2 is the reference PD. The optimized design with integrated peaking inductor has a germanium area of 8 × 20 µm2 and two off-centered electrodes on top of the germanium that provides similar bandwidth with lower reverse-bias voltage in comparison to the reference PD. The optimized design has a responsivity of 1.09 A/W at 1550 nm, a dark current of 3.5 µA and a bandwidth of 42.5 GHz at 2 V reverse-bias voltage. Comparatively, the non-optimized reference PD has a responsivity of 0.66 A/W at 1550 nm, a dark current of 1.05 µA and a bandwidth of 35 GHz at 2 V reverse-bias voltage. At a 4 V reverse-bias voltage, the reference PD has a bandwidth of 42 GHz, but with an increased dark current of 10 µA. The sensitivity of the optimized PD improves by 3.2 dB over the reference PD for a bit-error rate (BER) of 10−12 for a 30 Gb/s 231-1 pseudo random bit sequence (PRBS-31) NRZ-OOK data. The bias condition is set so that both PDs have the same bandwidth. The paper is organized into four sections. Section 2 presents the simulation results followed by the design flow of the optimized PD and the proposed optimization methodology. Section 3 presents the experimental results of the optimized and reference PDs followed by results analysis. Finally, the conclusion is presented in section 4.
2. Simulation and design methodology
This section demonstrates the numerical simulation results obtained using technology computer-aided design (TCAD) tools investigating the impact of top metal contact geometry and germanium-region dimension on the performance of an evanescently-coupled waveguide germanium-on-silicon PD with a vertical p-i-n diode. The innovative methodology details the optimization process for the PD’s responsivity based on the bandwidth requirement of a target application. It also discusses the design of an appropriate peaking inductor based on the small-signal model of a PD.
Figure 1 illustrates the side-view (left figures) and top-view (right figures) of the two approaches to the design of PDs discussed in this paper. Figures 1(a) and 1(b) show respectively the side-view and top-view of a Ge-PD with a single centered electrode on top of the n-doped germanium area (dark green). Figures 1(c) and 1(d) show respectively the side-view and top-view of a Ge-PD with two electrodes on top of the n-doped germanium area. The length (LGe) and width (WGe) of the germanium area, the length (Ln-doped) and width (Wn‑doped) of the n-doped germanium, and the length (Lseg) and width (Wseg) of the metal contacts are indicated in the figures. The two other contacts on top of the p ++ doped silicon area (dark red) represent the anode connection and have only a trivial degradation effect on the bandwidth of the PD. The highly doped silicon (Si-p ++ ) and germanium (Ge-n ++ ) layers form low resistance contacts with the metal layer.
Fig. 1 (a) Side-view and (b) top-view of a Ge-PD with a single electrode on top of the n-doped germanium (Ge-n ++ in dark green). (c) Side-view and (d) top view of a Ge-PD with two electrodes on top of the n-doped germanium.
2.1 Impact of top metal contact and PD dimension
Commercial software (Lumerical FDTD and DEVICE [11]) is used to study the degrading effect of the top metal contact on the responsivity of the germanium-on-silicon PD. The three-dimensional (3D) FDTD simulation software is used to calculate the electric field of the light evanescently coupling from the silicon to the germanium, as well as the absorbed optical power per unit volume [12]. Figure 2 shows the simulated results for the electric field propagation in a PD with a Ge area of 25 µm × 8 µm (indicated by the white contour box). Figures 2(a), 2(c), and 2(e) show the top view of the electric field, 10 nm below the top interface of the Ge area for a PD without the top metal contact, with a 1 µm wide centered metal contact, and with two off-centered 1 µm wide metal contacts, respectively. Figures 2(b), 2(d), and 2(f) show the corresponding side-views. Note that the large size of the Ge leads to a multimode detector. Figure 2(a) shows that the electric field is stronger in the center of Ge area. A centered top metal contact leads to more optical power loss due to metal absorption. Thus, Fig. 2(e) and 2(f) using two off-centered contacts located where the electric field is weaker would lead to less optical power loss. Increasing the gap between the two metal contacts further decreases optical power loss with better responsivity.
Fig. 2 The simulated electric field in a Ge-PD with a length of 25 µm, a width of 8 µm, and a thickness of 500 nm (linear scale). The Ge area is indicated by the white contour box. Left-side figures show the top view near the top interface of Ge (a) without top contact, (c) with one centered contact, (e) with two off-centered contacts. Right-side figures show the side view at the middle width of the Ge area for a PD (b) without top contact, (d) with one centered contact, (f) with two off-centered contacts.
For the simulation, an absorption coefficient of 1800 cm−1 at a wavelength of 1550 nm is used for the thin film Ge at room temperature [13]. The results show that a top centered aluminum (Al) contact with a width of 3.6 µm reduces the responsivity by 19% due to metal absorption compared to the simulation result not considering the effect of metal loss. With two off-centered smaller size contacts with a gap distance of 2 µm, the responsivity degradation is only around 1%. Responsivity, dark current, and bandwidth are further simulated for various widths and lengths of the Ge area. Figures 3(a) and 3(b) show the responsivity for various widths and lengths of the Ge area, respectively. In these simulations, the width of the n-doped Ge is half of the Ge width (WGe), and its length is the same as the length of the Ge area. The metal loss is ignored to reduce the simulation time. The simulations confirm that the responsivity depends more on the length of the PD than the width. Figures 4(a) and 4(b) show the dark current for various widths and lengths of the doped Ge area, respectively, at 2 V reverse-bias voltage. Assuming the bulk current is dominant, the dark current increases linearly with Ge area [14]. The measured dark currents in section 3.2 further support the assumption that the dark current is dominated by the bulk current. Figures 4(c) and 4(d) show the RF bandwidth for various widths and lengths of the Ge area, respectively, at 2 V reverse-bias voltage. TCAD software models the impact of carrier transit time as well as the RC time constants in the bandwidth calculation. Although the junction capacitor of the p-i-n PD decreases with reduced size, the smaller width of the n-doped Ge area leads to larger series resistance reducing the bandwidth of the PD. This can be observed in Fig. 4(c) for width less than 8 µm where the bandwidth does not improve. For width larger than 8 µm, the increased junction capacitor degrades the bandwidth of the PD. The tradeoffs lead to an optimum bandwidth for a width of 8 µm. A length increase of the PD increases the junction capacitor reducing the PD’s bandwidth. In this type of photodetector, the Ge thickness is 500 nm. Therefore, the Ge intrinsic thickness (t) is smaller than the 500 nm due to the n ++ doped Ge profile. Considering the carrier velocity-saturation (νs) in Ge of 0.6 × 107 cm/s, the calculated transit time bandwidth from ftr = 0.38 × νs/t is approximately 60 GHz [12]. Since the simulated bandwidth is area dependent, the photodetector bandwidth is mainly RC-limited.
Fig. 3 Responsivity variation of the PD with various sizes of the Ge area. (a) For a PD with a constant length of 10 µm and various widths. (b) For a PD with a constant width of 8 µm and various lengths.
Fig. 4 Dark current and bandwidth of the PD for various size of the Ge area. (a) The dark current of a PD with a constant length of 10 µm for various widths. (b) The dark current of a PD with a constant width of 8 µm for various lengths. (c) The bandwidth of a PD with a constant length of 10 µm for various widths, and (d) bandwidth of a PD with a constant width of 8 µm for various lengths.
2.2 Optimization methodology
The required bandwidth for a given data rate generally specifies the photodetector dimensions. Therefore in a photodetector with RC-limited bandwidth, the PD will usually be small at the expense of reduced responsivity. In the proposed methodology presented in Fig. 5, the photodetector is first designed for high responsivity and then an integrated peaking inductor enhances the bandwidth to reach the required bandwidth. The peaking inductor is designed for minimum settling time to provide maximum eye opening.
Fig. 5 Flowchart of the optimization methodology with target bandwidth BWT, calculated bandwidth BWL, and bandwidth peaking factor α
As an initial step, the peaking inductor is assumed to enhance the bandwidth by α = 1.5 times. The accuracy of the enhancement factor value α is based on the electrical model of the PD and the peaking inductor. The reduction in bandwidth (α) provides design margin towards larger responsivity. The responsivity can be maximized through appropriate dimensions of the PD, and size and location of the metal contact. For PDs longer than 20 µm, the responsivity improvement becomes negligible while the linear increase in dark current degrades the sensitivity from the increased shot noise. In such cases, instead of increasing the length of the PD, the optimization uses the peaking inductor to lower the required reverse-bias voltage thereby reducing the dark current. A photodetector with a Ge area of 8 × 8 µm2 is selected as a reference PD with a bandwidth of 45 GHz for a 50 Gb/s application. As α = 1.5, the bandwidth of the optimized PD without a peaking inductor is 1.5 times smaller than the reference PD. Consequently, the length of the optimized PD can be increased to 20 µm leading to better responsivity with a reduced bandwidth of 30.1 GHz (67% of 45.2 GHz). A more accurate optimized-length is calculated at a later stage based on the electrical model of the PD and the optimized peaking inductor.
2.3 Peaking inductor design
The design methodology for the integrated peaking inductor is now detailed. The green box in Fig. 6(a) shows the equivalent small-signal circuit of a photodetector. The current source IPD models the photocurrent, and CjPD is the PD junction capacitor. The PD resistance, RPD, is the series combination of the contact resistances and the resistance of the n-doped Ge and the p-doped silicon of the p-i-n photodetector. The red box in Fig. 6(a) is the lumped model of a non-ideal integrated inductor [8]. Lpeak models the peaking inductor in series with the small resistance of the spiral-shaped inductor (Rpar_ind) and in parallel with the parasitic capacitance of the inductor Cpar_ind. RL in Fig. 6(a) is the load resistance seen by the PD such as the input resistance of the transimpedance amplifier (TIA) or the 50 Ω terminated test and measurement equipment. CL is the combination of the parasitic capacitance of the pads and the parasitic capacitance of the circuit loading the PD. Designing a peaking inductor to minimize the rise and settling times requires determination of the small-signal circuit parameters. Cadence software is used to simulate the frequency and transient responses. The PD resistance (RPD) is estimated based on the sheet resistance of the doped Si and doped Ge, and the contact resistance reported by the foundry. The effective dimension of the doped semiconductor is calculated based on the path that the generated photocurrent follows to reach the anode and cathode pads. The PD’s capacitance is estimated from a reported measured capacitor per unit area (fF/µm2) of a fabricated PD in the similar fabrication process presented in [9]. ANSYS HFSS software calculates the parameters of the inductor and the parasitic capacitor of the pads. Table 1 summarizes the simulation parameters. Figure 6(b) shows the frequency response of the reference 8 × 8 µm2 PD (red line), as well as a long 8 × 20 µm2 PD without peaking inductor (dashed blue line), with an optimized peaking inductor (360 pH, blue line), and with a peaking inductor larger than the optimum value (540 pH, green dashed line). The optimized peaking inductor enhances the bandwidth by 1.503 times.
Fig. 6 (a) Equivalent small-signal circuit of PD with peaking inductor. (b) The simulated frequency response of an 8 × 8 µm2 reference PD and a 20 × 8 µm2 PD with and without peaking inductor.
Table 1. The parameters for the equivalent small-signal circuit of the reference PD and the longer PD with optimum peaking inductor and the longer PD with peaking inductor larger than optimum.
Figures 7(a) and 7(b) compare the simulated 50 Gb/s eye diagram of the reference PD (in red) to a long PD with optimum peaking inductor (in blue), and to a long PD with peaking inductor larger than the optimum value (in green), respectively. Interestingly, for a similar bandwidth of 45 GHz, the optimized PD with optimum peaking inductor (in blue) has slightly wider eye opening compared to the reference PD eye due to reduced intersymbol interference (ISI). Although the optimum peaking inductor for the design is 360 pH, the fabricated inductor is 540 pH following an overestimation of the PD’s junction capacitor reported in reference [15]. Using a peaking inductor larger than the optimum value generates peaking in the frequency response and reduces the optimum bandwidth. It also generates larger overshoot and undershoot in the transient response. However, the eye diagram has an eye-opening similar to that of the reference eye diagram.
Fig. 7 Simulated 50 Gb/s eye diagram of the 8 × 8 µm2 reference PD in comparison with the 20 × 8 µm2 PD (a) with the optimum peaking inductor (350 pH), and (b) with peaking inductor larger than the optimum value (540 pH).
3. Fabrication and the experimental results
This section presents the fabrication process used for the designed PDs. The experimental results of the reference PD and the optimized PD are compared to validate the proposed optimization methodology.
3.1 Fabrication process
The photodetectors are fabricated at the Institute of Microelectronics IME-A*Star in Singapore through a process detailed in Fig. 8(b). To build a photodetector, a 500 nm thick Ge was deposited through a selective epitaxial growth [1, 16]. Figure 8(a) shows the top view of the optimized PD with an integrated peaking inductor. The process provides two metal layers with a thickness of t1 = 0.75 µm and t2 = 2 µm separated by 1.5 µm giving the opportunity to design an integrated inductor. The left half of Table 2 summarizes the design parameters of the fabricated PDs. PD-A is the reference PD with a centered top contact, and PD-D is the optimized PD with two off-centered top contacts and an integrated peaking inductor. Two additional PDs are designed to separate the effect of using two off-centered contacts from the effect of increased size PD with peaking inductor: 1) PD-B has the same size as the reference PD but with two off-centered contacts, 2) PD-C has the same size as the optimized PD (PD-D) but with a centered contact. Further, it has a peaking inductor to enhance its bandwidth.
Fig. 8 (a) Fabricated photodetector with an integrated peaking spiral inductor. (b) Fabrication process layers.
Table 2. Summary of the photodetector’s dimensions and their measurements results.
3.2 DC and small-signal results
This subsection presents the measured dark current and responsivity of the PDs along with the small-signal measurement to determine the PD’s bandwidth. Using two connected grating couplers (GC) in proximity to each input GC of the photodetector, the GC loss is measured to calculate responsivity. The dark current of the PDs for various reverse-bias voltages is first measured with a precise ammeter (Keithley, sensitivity: 0.1 pA). Then, the responsivity is calculated from the measured photocurrent for an injected continuous wave (CW) at 1550 nm. The optical-electrical (OE) bandwidths of PDs are calculated from S-parameter (S21) measured by a 50 GHz Lightwave Component Analyzer (LCA) (Agilent N4373C). The right half of Table 2 summarizes the experimental results of the fabricated PDs. Additional experimental results of various modified PDs are shown in Appendix A which show the performance impact of different design modifications.
For a similar bandwidth of 42 GHz, the optimized photodetector with an integrated peaking inductor (PD-D) provides 65% larger responsivity at lower biasing voltage leading to smaller dark current compared to the reference PD (PD-A). Using one of the optimization methods, the responsivity improves by 36% and 30% in PD-B (same size as the reference PD but with two off-centered contacts) and in PD-C (same size as the optimized PD with peaking inductor but with a centered contact), respectively. Figure 9(a) shows the OE frequency response (S21) of the reference PD (PD-A) at 4 V reverse-bias voltage and the optimized PD (PD-D) at 2 V reverse-bias voltage, with their respective small-signal model simulation for parameters reported in Table 1. The measured S21 curves are normalized to the S21 value at 10 MHz. Figure 9(b) shows the corresponding reflection frequency response (S22). Although the simulation results for S22 are in good agreement with the measurement results, the simulation results for S21 show differences. The measured S21 of the reference PD exhibits small peaking (< 0.21 dB) below 1 GHz. The frequency response linearly decreases afterward (zoomed-in view in Fig. 9(a)). While the small-signal model (Fig. 6(a)) does not model this low-frequency roll-off behavior, it predicts the high-frequency behavior with the simulated 3-dB bandwidth of 45 GHz which is close to the measured value of 42 GHz for the reference PD. In Fig. 9(a) for frequencies larger than 10 GHz, the S21 simulated results of the reference PD have the same declining slope compared to the measurement results. On the other hand, by subtracting a DC offset (black arrow in zoomed-in view in Fig. 9(a)) from the simulated S21, the effect of the low-frequency roll-off is removed, and the simulation agrees with the measurements for frequencies larger than 10 GHz. The low-frequency roll-off also affects the measured S21 of the optimized PD with its peaking inductor (PD-D). However, the simulated S21 is in good agreement with the measurement result at high frequencies. Disagreement between simulation and measurements also relates to the inaccurately estimated parameters. The doping of the Ge and Si areas are assumed to be constant with uniform doping depth independent to the PDs’ dimensions. In fact, the fabrication process exhibits non-uniform doping density which impacts both parasitic resistance and capacitance. Further, the variation of the Ge thickness changes the parasitic capacitance. Finally, the larger difference for the PD with peaking inductor suggests that the inductor model needs to account for the parasitic capacitance between the integrated spiral inductor and the substrate [17].
Fig. 9 (a) Measured S21 OE frequency response of the optimized PD (PD-D) at 2 V reverse-bias voltage and reference PD (PD-A) at 4 V reverse-bias voltage with their respective small-signal simulations, (b) Measured S22 (reflection) of both PDs with their respective small-signal simulations.
3.3 Large-signal measurements
This subsection presents the BER measurements of the fabricated PDs and compares the measured BER and eye diagram of the optimized PD with the reference PD. Figure 10 illustrates the test bed for these measurements. The bit pattern generator (BPG) generates PRBS-31 data, and the error analyzer (EA) measures the error rate of the signal detected by the PD. The continuous wave (CW) generated by a DFB laser from EMCORE with 16 dBm optical power at 1550 nm is injected through a polarization controller (PC) with 0.5 dB insertion loss. A 40 Gb/s Mach-Zehnder modulator (MZM) with insertion loss of 8 dB at 1550 nm is driven by the baseband signal from the output of the BPG. The modulated optical carrier is injected into a variable optical attenuator (VOA) with an insertion loss of 2 dB. The VOA is used to reduce the received signal-to-noise ratio (SNR) in the BER measurement. Since the input grating coupler (GC) is polarization sensitive, the modulated data is then injected to another PC. The optical power at the input GC of the device (point E in Fig. 10) is 5 dBm. The modulated optical signal is launched to the GC, and the PD converts it to a photocurrent. The 50 Ω terminated measurement devices (EA or Sampling Scope) convert the photocurrent into voltage. The photodetector is biased through a 65 GHz bias tee. Although the bandwidth of the measured PDs is large enough to support 50 Gb/s data, the BER comparison is presented at 30 Gb/s. Beyond that data rate, the test bed exhibits a noise floor. A fair comparison to validate the proposed methodology is required where the difference in performance is not due to the test bed link budget.
The measured BER at 30 Gb/s for the reference photodetector (PD-A) is shown in Fig. 11 along with BER for the same size PD with two off-centered contacts (PD-B), and 20 µm length PD with a centered contact and peaking inductor (PD-C). Those three PDs are compared to separate the effect of using two off-centered contacts from the effect of increased size PD with peaking inductor. The bias condition for each PD is selected to match the PDs bandwidth (4 V for PD-A (42 GHz bandwidth) and PD-B (41.4 GHz bandwidth), and 2 V for PD-C (42.6 GHz bandwidth)). By extrapolation to a BER of 10−12, the sensitivity of the long PD with a centered contact and peaking inductor (PD-C) is 1.9 dB better than the reference PD while using two off-centered contacts exhibit similar improvement. Figure 11 also shows the measured BER for the optimized photodetector (PD-D) for various bias conditions. For matched bandwidth (2 V for optimized PD with a bandwidth of 42.5 GHz), the sensitivity of the optimized photodetector (PD-D) is 3.2 dB better than the reference photodetector (PD-A) at the extrapolated BER of 10−12. Increasing the reverse-bias voltage of the optimized PD to 4 V increases its bandwidth from 42.5 GHz to 45 GHz which in turn improves the sensitivity by 0.6 dB.
Fig. 11 (a) BER performance at 30 Gb/s for the reference PD (PD-A), the same size PD as the reference PD with two off-centered top contacts (PD-B), and 20 µm length PD with a centered top contact and peaking inductor (PD-C). (b) BER performance at 30 Gb/s for optimized PD (PD-D).
Figure 12 shows the output electrical eye diagrams of the reference PD (PD-A) with a bandwidth of 42 GHz at 4 V reverse-bias and the optimized PD (PD-D) with a bandwidth of 42.5 GHz at 2 V reverse-bias at various data rates. The optimized PD has larger eye opening amplitudes due to 65% larger responsivity. As expected from simulation results, the eye diagram of the optimized PD has less ISI than the reference PD.
Fig. 12 (top) Output electrical eye diagrams of the reference PD (PD-A) at various data rates. (bottom) output electrical eye diagram of optimized PD (PD-D) at corresponding data rates.
4. Conclusion
This paper presents an investigation into the impact of various design parameters of a Germanium-on-Silicon photodetector along with an optimization of the metal contact on the PD’s performance. Further, it proposes a methodology to maximize the responsivity of a PD based on the bandwidth requirement of a given data rate application. In the proposed optimization process, increasing the PD’s design geometry and using two off-centered small size contacts on top of the germanium enhances the responsivity of the PD. An integrated peaking inductor is designed to mitigate the bandwidth reduction trade-off of the responsivity optimization. To validate the optimization process, a reference design from a Silicon Photonic foundry process design kit (PDK) is selected. The small size photodetector with an 8 × 8 µm2 germanium area is used as a reference. The optimized PD design with an integrated peaking inductor has an 8 × 20 µm2 germanium area and two off-centered electrodes on top of the germanium. The optimized design provides a responsivity of 1.09 A/W at 1550 nm, a dark current of 3.5 µA and a bandwidth of 42.5 GHz at 2 V reverse-bias voltage. The reference PD has a similar bandwidth of 42 GHz at 4 V reverse-bias voltage with a responsivity of 0.66 A/W and an increased dark current of 10 µA at this bias voltage. The sensitivity of the optimized PD exhibits a 3.2 dB improvement compared to the small size PD for a bit-error rate (BER) of 10−12 at 30 Gb/s.
Appendix A
This section presents the measurement results of several photodetectors with various design parameters to study the performance impact of different design modifications. Table 3 summarizes the design properties of the fabricated devices and their measurement results.
Table 3. Summary of device dimensions of fabricated PDs and their measurements results.
Figure 13(a) shows the responsivity for various lengths of the photodetector with one centered contact and two off-centered contacts. The measurements are compared to simulation results (Fig. 3). The responsivity of PDs with one centered electrode is consistently lower than 0.9 A/W due to the metal absorption of the contact on top of the Ge area (green line). Using two off-centered top contacts with 1.6 µm distance between them, the responsivity improves by ~0.24 A/W (red line) for the same length of PD. The simulated responsivity without the effect of metal loss (blue line) is larger because the distance between the two contacts in two off-centered top contacts is 1.6 µm, and a small effect of the metal loss remains. For the PD with a length of 8 µm, increasing the gap distance to 2 µm in device no. 3 and 2.7 µm in a device no. 4 lead to a responsivity increase to 0.93 and 0.98 A/W, respectively, due to lower metal loss. Those results are in line with the simulation results of 1.01 A/W. Figure 13(b) shows the dark for various lengths of the photodetector with one centered contact and two off-centered contacts. The measurements are compared to simulation results (Fig. 4). The linear increase of the measured dark current (green line) with the length of the photodetector shows the dominance of the bulk current. The reasons for the differences between simulated dark current and measurement results are the following: 1) due to a limitation in the simulation, the Ge-Si, and Ge-metal interface models are not included in this simulation, 2) the electrical properties of the thin film Ge model are not accurate, and 3) the exact doping density levels of n-doped Ge remains unknown.
Fig. 13 Measured and simulated responsivity (a), and measured and simulated dark current (b) for various PD lengths with one centered top contact and with two off-centered top contacts. Results are compared to simulation results from Figs. 3 and 4.
The OE frequency response (S21) of various fabricated PDs is shown in Fig. 14. Each curve is normalized to its S21 value at 10 MHz frequency. The measured S21 of all PDs show the low-frequency roll-off. All photodetectors with a peaking inductor have a bandwidth larger than 39 GHz at 2 V reverse-bias voltage. The largest bandwidth reached for photodetectors without a peaking inductor is 35 GHz (no. 1).
The BER performance of various other fabricated devices is shown in Fig. 15. All devices with larger responsivity compared to the reference PD (no. 1) have better sensitivity. For instance, PD no. 6 with a longer Ge (15 µm) has slightly lower bandwidth than the reference PD (no. 1) but exhibits an improved sensitivity by 0.7 dB at an extrapolated BER of 10−12. Interestingly for PD no. 7 (Ge length of 15 µm with two top off-centered contacts) with lower bandwidth, the sensitivity improves by 2.8 dB at BER of 10−12 from the 56% increase in responsivity from the two off-centered contacts optimization and a longer Ge area.
Funding
We are gratefully acknowledging the financial support from the Natural Sciences and Engineering Research Council (NSERC) of Canada, in particular, the CREATE Silicon Electronic-Photonic Integrated Circuits (Si-EPIC) program.
Acknowledgments
We acknowledge the contribution and technical support of Canadian Microelectronics Corporation (CMC). The authors thank Prof. Plant’s group from McGill University for their help with the BER system and the LCA.
References and links
1. T. Y. Liow, K. W. Ang, Q. Fang, J. F. Song, Y. Z. Xiong, M. B. Yu, G. Q. Lo, and D. L. Kwong, “Silicon modulators and germanium photodetectors on SOI: monolithic integration, compatibility, and performance optimization,” IEEE J. Sel. Top. Quantum Electron. 16(1), 307–315 (2010). [CrossRef]
2. Y. Zhang, S. Yang, Y. Yang, M. Gould, N. Ophir, A. E. Lim, G. Q. Lo, P. Magill, K. Bergman, T. Baehr-Jones, and M. Hochberg, “A high-responsivity photodetector absent metal-germanium direct contact,” Opt. Express 22(9), 11367–11375 (2014). [CrossRef] [PubMed]
3. H. Chen, P. Verheyen, P. De Heyn, G. Lepage, J. De Coster, S. Balakrishnan, P. Absil, W. Yao, L. Shen, G. Roelkens, and J. Van Campenhout, “−1 V bias 67 GHz bandwidth Si-contacted germanium waveguide p-i-n photodetector for optical links at 56 Gbps and beyond,” Opt. Express 24(5), 4622–4631 (2016). [CrossRef]
4. S. Lischke, D. Knoll, C. Mai, L. Zimmermann, A. Peczek, M. Kroh, A. Trusch, E. Krune, K. Voigt, and A. Mai, “High bandwidth, high responsivity waveguide-coupled germanium p-i-n photodiode,” Opt. Express 23(21), 27213–27220 (2015). [CrossRef] [PubMed]
5. S. Liao, N. N. Feng, D. Feng, P. Dong, R. Shafiiha, C. C. Kung, H. Liang, W. Qian, Y. Liu, J. Fong, J. E. Cunningham, Y. Luo, and M. Asghari, “36 GHz submicron silicon waveguide germanium photodetector,” Opt. Express 19(11), 10967–10972 (2011). [CrossRef] [PubMed]
6. Y. Painchaud, M. Poulin, F. Pelletier, C. Latrasse, J. F. Gagné, S. Savard, G. Robidoux, M. Picard, S. Paquet, C. Davidson, M. Pelletier, M. Cyr, C. Paquet, M. Guy, M. Morsy-Osman, M. Chagnon, and D. V. Plant, “Silicon-based products and solutions,” SPIE OPTO, International Society for Optics and Photonics, 89880L–89880L (2014).
7. G. Li, Y. Luo, X. Zheng, G. Masini, A. Mekis, S. Sahni, H. Thacker, J. Yao, I. Shubin, K. Raj, J. E. Cunningham, and A. V. Krishnamoorthy, “Improving CMOS-compatible Germanium photodetectors,” Opt. Express 20(24), 26345–26350 (2012). [CrossRef] [PubMed]
8. M. Gould, T. Baehr-Jones, R. Ding, and M. Hochberg, “Bandwidth enhancement of waveguide-coupled photodetectors with inductive gain peaking,” Opt. Express 20(7), 7101–7111 (2012). [CrossRef] [PubMed]
9. A. Novack, M. Gould, Y. Yang, Z. Xuan, M. Streshinsky, Y. Liu, G. Capellini, A. E. Lim, G. Q. Lo, T. Baehr-Jones, and M. Hochberg, “Germanium photodetector with 60 GHz bandwidth using inductive gain peaking,” Opt. Express 21(23), 28387–28393 (2013). [CrossRef] [PubMed]
10. G. Chen, Y. Yu, S. Deng, L. Liu, and X. Zhang, “Bandwidth improvement for germanium photodetector using wire bonding technology,” Opt. Express 23(20), 25700–25706 (2015). [CrossRef] [PubMed]
11. Lumerical Computation Solutions (TCAD software), https://www.lumerical.com/
12. L. Chrostowski and M. Hochberg, Silicon Photonics Design: from Devices to Systems (Cambridge U. 2015).
13. V. Sorianello, L. Colace, N. Armani, F. Rossi, C. Ferrari, L. Lazzarini, and G. Assanto, “Low-temperature germanium thin films on silicon,” Opt. Mater. Express 1(5), 856–865 (2011). [CrossRef]
14. M. Takenaka, K. Morii, M. Sugiyama, Y. Nakano, and S. Takagi, “Dark current reduction of Ge photodetector by GeO2 surface passivation and gas-phase doping,” Opt. Express 20(8), 8718–8725 (2012). [CrossRef] [PubMed]
15. T. Baehr-Jones, R. Ding, A. Ayazi, T. Pinguet, M. Streshinsky, N. Harris, J. Li, L. He, M. Gould, Y. Zhang, A. E. Lim, T. Y. Liow, S. H. Teo, G. Q. Lo, and M. Hochberg, “A 25 Gb/s silicon photonics platform,” arXiv preprint arXiv:1203.0767 (2012).
16. K. W. Ang, S. Y. Zhu, J. Wang, K. T. Chua, M. B. Yu, G. Q. Lo, and D. L. Kwong, “Novel silicon-carbon (Si: C) schottky barrier enhancement layer for dark-current suppression in Ge-on-SOI MSM photodetectors,” IEEE Electron Device Lett. 29(7), 704–707 (2008). [CrossRef]
17. J. R. Long and M. A. Copeland, “The modeling, characterization, and design of monolithic inductors for silicon RF IC’s,” IEEE J. Solid-State Circuits 32(3), 357–369 (1997). [CrossRef]
