Germanium on Silicon Avalanche Photodiode - IEEE Xplore

IEEE JOURNAL OF SELECTED TOPICS IN QUANTUM ELECTRONICS, VOL. 24, NO. 2, MARCH/APRIL 2018

3800911

Germanium on Silicon Avalanche Photodiode Mengyuan Huang, Member, IEEE, Su Li, Pengfei Cai, Guanghui Hou, Tzung-I Su, Wang Chen, Ching-yin Hong, and Dong Pan (Invited Paper)

Abstract—Silicon photonics is considered as one of the promising technologies for high-speed optical fiber communications. Among various silicon photonic devices, germanium on silicon avalanche photodiodes (Ge/Si APDs) have attracted tremendous attention due to their the properties of high performance and low cost. The sensitivity of 10Gb/s APD reached −29.5dBm at 1550 nm with the bit error rate of 1 × 10−12 in 2015, and 25 Gb/s APD reached a highest sensitivity ever reported at about −23.5dBm at 1310 nm in 2016. Furthermore, a linear 28Gbaud APD receiver is found to have about 5dB more sensitivity than PIN solution for both multimode fiber (MMF) and single-mode fiber (SMF). Even with large defect density from lattice mismatch between Ge and Si substrate layer, Ge/Si APD devices demonstrate good reliability and pass all required qualification tests according to GR-468. In this paper, the structure, materials, process, performances, and reliability of APDs will be reviewed. Index Terms—Avalanche photodiode, germanium on silicon, silicon photonics, fiber communications.

I. INTRODUCTION VALANCHE photodiodes (APDs) are widely used in today’s fiber communication systems, and millions of high speed APD devices were implemented in last several years thanks to the rapid growth of access networks such as Fiber to Building/Home [1]. Traditionally, the majority of APD receivers are made by InP-based devices, which provide a 5–10 dB improvement to sensitivity compared to PIN PD receivers when operating data rate is below 10 Gb/s [2]. However, due to the exponentially increasing demand of data rate in applications such as high definition TV (HDTV) and 5th generation (5G) wireless communications, the bandwidth requirements of fiber communications today are reaching 100 Gb/s and will grow even further to 400 Gb/s in next three years [3]. For III-V APDs, the materials’ intrinsic characteristics such as low gain-bandwidth product limit their deployments in next generation fiber communication networks. Currently, the reported 3dB bandwidth of 25 Gb/s InP-based APD is 22 GHz at the gain (M) of 3.3 which drops to 18 GHz when M reaches 10 [4]. Such limited bandwidth is insufficient for today’s 100 Gb/s system and cannot meet bandwidth requirement of coming 400 Gb/s ap-

A

Manuscript received June 2, 2017; revised August 25, 2017; accepted August 30, 2017. Date of publication September 7, 2017; date of current version October 10, 2017. (Corresponding author: Dong Pan.) The authors are with SiFotonics Technologies Company Ltd., Woburn, MA 01801 USA (e-mail: [email protected]). Color versions of one or more of the figures in this paper are available online at http://ieeexplore.ieee.org. Digital Object Identifier 10.1109/JSTQE.2017.2749958

Fig. 1. Schematic diagram of SCAM Ge/Si APD and its electric field distribution.

plications. Many developments activities are on-going for new devices and components, among which Ge/Si APD stands out because of Si’s large gain-bandwidth product and low ionization coefficient ratio. In addition, the CMOS compatibility of Ge/Si devices provides the possibility of cost effective solution for high speed applications. CMOS foundries have the capability to develop outstanding high speed optical Ge/Si devices especially when they have already made the mass production of high speed GeSi ICs [5]. The main challenge of Ge/Si optical devices at CMOS foundries is to realize high quality process of a pure Ge layer grown on Si. A. Ge/Si APD Basic Structure and Its Operation Conditions The basic Ge/Si APD structure has a separate absorption, charge and multiplication (SCAM) design. It is a combination of these materials’ merits including Ge’s good absorption in telecommunication wavelengths and Si’s outstanding avalanche characteristics. For high speed applications, it is necessary to ensure that the photo-generated carriers reach their saturation velocity, which required the proper control over the electrical field in Ge absorption layer. To fully utilized the benefit of Si avalanche properties such as high gain bandwidth and low noise, it is essential to maintain a strong electrical field in the intrinsic Si laser to make sure avalanche happens in the Si layer. Fig. 1 presents the schematic structure of SCAM Ge/Si APD and its electric field distribution during operation. Ge/Si APD

1077-260X © 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications standards/publications/rights/index.html for more information.

3800911

Fig. 2.


Fig. 3.

SEM image of EPD results of 1.5 μm Ge on Si.

Fig. 4.

SEM image of EPD results of 0.5 μm Ge on Si.

Fig. 5.

Cross section TEM image of 1.5 μm Ge on Si.

Schematic diagram of Ge APD on Si and its electrical field distribution.

is different from Ge APD on Si (presented in Fig. 2 [6]). The latter one has a different film stack and utilizes Ge layer (rather than Si layer) as avalanche material. II. MATERIAL STUDY: GE GROWTH ON SI To deposit a pure Ge layer on Si with high quality is the first and principal challenge of Ge/Si APDs. Ge growth on Si have been realized in the last few decades by different tools such as MBE, UHV-CVD, LPCVD, RPCVD, etc. [7]–[10]. In order to improve the quality of Ge layer on Si for electronic and photonic applications, numerous efforts were involved including adding buffer layer, post growth annealing, etc. [11]–[13]. In following sections, we review some critical parameters of Ge growth on Si and their impacts on Ge/Si APDs. A. Ge Dislocation Distribution: Interface Region and Surface Region Because of the existence of 4.2% lattice mismatch between Ge and Si, the deposited Ge layer contains a huge amount of dislocations (both misfit and threading). There are several methods to measure the dislocation density in Ge layer such as etch pit density (EPD) [10], cross section TEM (X-TEM) [14], etc. We utilized both EPD and X-TEM to study the quality of our Ge film. Figs. 3 and 4 show our Ge film EPD results. From the EPD results, we found that threading dislocation density increases when the thickness of Ge film becomes thinner. The dislocation density becomes 10 times larger when the epitaxial layer’s thickness decreases from 1.5 to 0.5 μm. Moreover, the TEM cross section of Ge on Si film also indicates that the majority of the dislocations are trapped in the region near Ge/Si interface as shown in Fig. 5. The threading dislocation density increases exponentially with Ge thickness decreasing, which has also been reported by several other groups [14]–[17]. Fig. 6 summarizes the reported threading dislocation density dependence on Ge (GeSi) thicknesses. Though many reports have demonstrated that the dislocation density can be reduced to the level of 1 × 107 /cm2 at the

surface region of a thick epitaxial Ge layer (> 1 μm), the dislocation density is obviously larger (> 1 × 1010 /cm2 ) at the interface region of this thick Ge layer. Such high interface threading dislocation density results from the fact that the metastable critical thickness is less than 5 nm for Ge on Si, which means large numbers of misfit dislocations are generated to relax a thick Ge layer [18]. Given that the spacing of misfit dislocation is at the order of sub-micrometers [19], a huge amount of threading dislocation is inevitably formed at interfaces regions after Ge depositions.

HUANG et al.: GERMANIUM ON SILICON AVALANCHE PHOTODIODE

3800911

Fig. 7.

XRD results of Ge on Si film with and without annealing.

Fig. 8.

Calculated Ge on Si film absorption coefficient at C-bands.

Fig. 6. Threading dislocation density of Ge (GeSi) on Si with different thickness.

B. Ge Dislocations, Deep Level Defect States and Ge/Si APD Dark Current Ge epitaxial layer contains a large amount of threading dislocations, and the concentration of deep level defect states is proportional to the threading dislocation density [20]. The fundamental reason of leakage current is the generation and recombination of minority carriers in the depletion region from the dislocation related deep level states [21]. Moreover, according to Hurkx model, the leakage current is dominated by ShockleyRead-Hall (SRH) generation mechanism and trap-assisted tunneling under high electric field [22]: W (1 + Γ)qni σn υth ND NT D dx (1) J= 0

where J is the leakage current density, q is the electron charge, ni is the intrinsic carrier concentration, σn is the effective capture cross section, W is the depletion width, υth is the carrier thermal velocity, ND is the dislocation density, NT D is the defect state density, and Γ is the electric field enhance factor. Ge/Si APD has higher leakage current density than Ge/Si PIN devices due to its operation under high electric field while it depletes interface region with high threading dislocation density (1 × 1010 /cm2 ) [23]. C. Ge Absorption, Tensile Strain and Ge/Si APD Responsivity Although bulk Ge’s absorption wavelength ranges from visual lights (600 nm) to infrared (>2 μm), the direct bandgap of bulk Ge at room temperature is only 0.8 eV., and bulk Ge’s absorptions at fiber communication wavelengths are therefore weak especially with wavelengths beyond 1550 nm [24]. For Ge on Si, the 1550 nm-band absorptions become better because of the existence of an in-plane strain inside epitaxial Ge layer, which is caused by the different thermal expansion coefficients of Ge and Si [25]. Similar results were also observed on our Ge/Si epitaxial film. Fig. 7 shows XRD results of our Ge on Si films. According to the XRD results, after annealing, the Ge’s Kα 1 peak position is changed from 66.0768° to 66.0981°, which

means the in-plane strain of Ge film increasing from 0.137% to 0.175%, leading to a significant improvement on absorption coefficient at 1550 nm bands. Fig. 8 is the calculated results of absorption coefficient based on experimental XRD results of annealed samples and deformation potential theory [26]. From Fig. 8, we can see that compared to the reported bulk Ge’s poor 1550 nm absorption coefficient (< 500 cm−1 ), the Ge on Si film has a 1550 nm absorption coefficient exceeding 3200 cm−1 after annealing. This number is still smaller than bulk Ge’s 1310 nm-band coefficients [24] or InGaAs’s 1550 nmband ones [27], further improvements in responsivity have to be made by the optimization of device design including the employment of resonance cavity enhanced (RCE) structure [28] or waveguide structure [29] before realizing actual applications of Ge/Si APDs. III. RCE GE/SI APD A. RCE Ge/Si APD Structure The Ge/Si APDs presented in this work are based on a novel RCE design shown in Fig. 9. Fig. 10 shows the cross-section SEM image of an APD device. The two primary features of this novel design are: 1. we utilized a n-Si layer between heavily bottom contact layer and intrinsic Si layers [30]; 2. a backside

3800911

Fig. 9.


Schematic cross section of Ge/Si APD. Fig. 11.

OM image of a 10 Gb/s Ge/Si APD.

Fig. 10. SEM image of a Ge/Si APD (removed dielectric layers by dilute hydrofluoric acid).

metal reflector was implemented to form RCE structure for normal incidence APD design [31], [32]. The fabrication process of this Ge/Si APD started with an 8-in silicon-on-insulator (SOI) wafer. The top Si layer was implanted to form n-contact, followed by selective Si growth. After the depositions of intrinsic Si layer, a p-type implant process was conducted to form p-Si layer. Afterwards, an oxide layer was deposited and patterned to form windows for Ge selective growth. After that, a Ge layer was deposited by a commercial CVD tool. After Ge growth, an amorphous-Si layer was deposited and implanted to form p-contact. When device metallization was finished, dielectric films were deposited for passivation and anti-reflection. After front-side processing, the wafer was grinded to a target thickness, followed by etching and metal deposition processes for backside reflector formation. The whole fabrication process was completed in a standard CMOS commercial foundry. B. 10 Gb/s Ge/Si APD Fig. 11 is the optical microscopic (OM) image of our 10 Gb/s Ge/Si APD which has a Ge mesa diameter = 35 μm and was applied the RCE structure to enhance the responsivity. Fig. 12 presents typical curves of our 10 Gb/s APD 25 °C dark current, photo current, and responsivity at 1550 nm. The breakdown voltage (Vbr ) of Ge/Si APD is defined as the voltage applied when dark current is equal to 100 μA at 25 °C, and the typical value of Vbr is close to −28.5 V. The APD dark current is 3 μA at M = 12, and the corresponding dark current density is ∼ 26 mA/cm2 at unity gain, which is very close to the reported Ge/Si PIN devices’ results ∼ 10 mA/cm2 [33]. The additional leakage current is from interfacial dislocation and high electrical field. The punchthrough voltage (Vpt ) of Ge/Si APD is defined as the applied voltage when both Ge and Si layers are completely

Fig. 12. Photo current, dark current and 1550 nm responsivity curves of 10 Gb/s Ge/Si APD at 25 °C. (Dark and photo currents were measured under no input light and 10 μW 1550 nm light respectively).

depleted, of which the typical value is close to −21 V. Under punchthrough voltage, our measured device’s capacitance is 120 fF, which meets our calculated results when both Ge and Si layers are completely depleted. Also, under this Vpt , Ge layer has enough electric field to ensure the photo-generated carriers’ reaching saturation velocity and meeting high speed operation requirements. Besides this ultralow dark current density and capacitance, the room temperature 1550 nm responsivity is also high and reaches 0.9 A/W at unity gain after implementation of RCE structure. To further study the performance of our APD, we packaged the APD with a commercial 10 Gb/s transimpedance amplifier (TIA). Fig. 13 presents the schematic plan of APD receiver circuits. In APD biasing circuit, we added a 2kOhm serial resistor to control the voltage drop on APD devices when input powers have a huge change (e.g., from −27 to −5 dBm). When operating at low input optical power, the voltage drop on serial resistor is small and has almost no impact on APD bias. On the other hand, if the input power becomes very large, the voltage drop is more obvious and the voltage on APD device is reduced. When APD bias is smaller than Vpt , the Ge layer is not depleted and does not have enough electric field to support high speed


3800911

TABLE I GE/SI APD SENSITIVITY CALCULATIONS 25 °C Parameter Dark current @M = 12 (μA) Responsivity @M = 1 (A/W) Excess noise factor @M = 12 Bandwidth @M = 12 (GHz) 10 G TIA input RMS noise (μA) APD shot noise (μA) Receiver total RF noise (μA) 10 Gb/s Sensitivity (dBm)

Fig. 13.

Schematic diagram of APD receiver circuits.

Ge/Si APD 3 0.9 3.00 7 0.9 1.09 1.41 −29.5

that of commercial III-V counterparts. However, such relatively high dark current is not the dominant limitation of receiver sensitivity because dark current is only a part of total current during APD operation, and the rest is related to average photo current. When APD operates at input power equal to −29.5 dBm (1.12 μW), the average photo current can reach 13.44 μA under operating condition (gain = 12 and primary responsivity = 0.9 A/W), which is much larger than dark current and is the dominant factor of shot noise. The relationship of DC current and noise is described as following equation [35]: (2) iAPDshot = 2q(Idark + Iphoto )F M 2 B here q is electron charge, Idar k is dark current (at unity gain), Iphoto is average photo current (at unity gain), M is APD gain, B is receiver bandwidth, F is excess noise factor. The excess noise factor is related to APD gains (M) and ionization coefficient ratio (k) as equation [36]: F = kM + (2 − 1/M ) · (1 − k)

Fig. 14. 10 Gb/s Ge/Si APD receiver sensitivity and overload at 10.3 Gb/s, 1550 nm. (Test conditions: ER = 10 dB, 1550 nm, 10.3 Gb/s, PRBS = 2ˆ31–1, NRZ, 25 °C).

operation and therefore it causes some error. With proper controls of electrical in Ge layer, Ge/Si APD receiver demonstrates high performance on sensitivity and overload. At 10Gb/s, the receiver sensitivity is −29.5 dBm for a BER of 10−12 and the overload is −4 dBm as shown in Fig. 14. For our receiver, there is a 25.5 dB error free dynamic range (the gap between overload and sensitivity) which is sufficient for commercial 1550 nm module applications at different operation distances e.g., from back-to-back (large input power) to 80 km (weak input power) applications. The high sensitivity and overload of this Ge/Si APD are even better than the results of III-V APDs [34]. Compared to III-V APD, Ge/Si APD has a relatively large dark current under operating conditions especially at low data rate such as 10 Gb/s. Our 10 Gb/s Ge/Si APD dark current reaches 3 μA at M = 12, which is about 2 orders higher than

(3)

Under similar gain (M = 12), Si based APD has much lower k value (∼ 0.1) and that value of InP based APD is very large (> 0.5). The low excess noise factor of Ge/Si APDs is the fundamental reason that Ge/Si APD performs better sensitivity than III-V APDs. Besides APD shot noise, the white noise of TIA circuit (input RMS noise current) is another major contributor of the total noise of APD receiver. The total noise of APD receiver can be expressed as equation [37]: (4) iROSA = i2APDshot + i2TIA When TIA noise is much larger than APD shot noise, the receiver sensitivity is limited by TIA rather than APD. For 10 Gb/s commercial TIAs, their RMS noise current is close to 1 μA [38]. Table I provides the calculated sensitivity of 10 Gb/s Ge/Si receiver based on experimental data. The theoretical result matches measured sensitivity very well. Besides 1550 nm-band sensitivity data, we also evaluated our 10 Gb/s APD receiver at O and L bands performance for 10 G PON application. Ge’s O-bands’ absorption coefficients are large, thus our 10 Gb/s APD demonstrated an sensitivity of -29.5dBm at the bit error rate of 1 × 10-12 as shown in Fig. 15 and this result is better than reported best Ge/Si or III-V APDs measured data [23], [34].

3800911


TABLE II GE/SI APD PERFORMANCE AT DIFFERENT TEMPERATURE Conditions

−40 °C 25 °C 85 °C

1310 nm sensitivity (dBm), 1577 nm sensitivity (dBm), 10.3 Gb/s, ER = 10 dB, BER 10.3 Gb/s, ER = 7 dB, BER = −12 = 1 × 10 , PRBS = 2ˆ31–1, 1 × 10−3 , PRBS = 2ˆ31–1, −29.5 −29.5 −27.5

−32.8∗(with heating) −32.8 −31.3

Fig. 15. 10 Gb/s Ge/Si APD receiver sensitivity at 10.3 Gb/s, 1310 nm. (Test conditions: ER = 10 dB, 1310 nm, 10.3 Gb/s, PRBS = 2ˆ31–1, NRZ, 25 °C).

Fig. 17.

OM image of a 25 Gb/s Ge/Si APD.

Fig. 16. 10 Gb/s Ge/Si APD receiver sensitivity at 10.3 Gb/s, 1577.8 nm. (Test conditions: ER = 7 dB, 1577.8 nm, 10.3 Gb/s, PRBS = 2ˆ31–1, NRZ, 25 °C).

On the other hand, Ge’s absorption coefficients at L-band wavelengths are small, but with the benefits from tensile strains and RCE design, our 10 Gb/s APD receiver’s 1577 nm responsivity reaches 0.8 A/W at 25 °C. Such high responsivity ensures the good sensitivity of our APD as shown in Fig. 16. Under high sensitivity bias conditions (such as M = 12), the dark current of 10 Gb/s Ge/Si APD is 3 μA at 25 °C, and this dark current increases to ∼ 25 μA at 85 °C. The dark current increase causes 1.5 ∼ 2 dB sensitivity degradation at 85 °C. On the other hand, high temperature operation shrinks Ge’s direct bandgap and enhances the absorption of wavelengths beyond its bandgap. As a result, 10 Gb/s APD receiver’s 1577 nm performance degradation is controlled less than 1.5 dB. For 1577 nm operating at −40 °C, heating is necessary for device reaching high performance because of the significant change of Ge’s bandgap at low temperature. Table II summarized the Ge/Si APD solutions for 10 G PON wavelengths. Here, all 10 Gb/s 1577 nm sensitivity was measured under light source extinction ratio = 7 dB for meeting the actual module operations. Within the industrial temperature range from −40 to 85 °C, our Ge/Si APD sensitivity is better than −31.3 dBm with the bit error rate of 1 × 10−3 which is an outstanding performance with enough margins for mass pro-

Fig. 18. Photo current, dark current and 1310 nm responsivity curves of D20 Ge/Si APD at 25 °C. (Dark and photo currents were measured under no input light and 50 μW 1310 nm light respectively).

ductions of 10 G PON module. With the high performance of 10Gb/s Ge/Si APDs and potential cost benefits from CMOS platform, large scale 10G PON implementation is feasible. C. 25Gb/s Ge/Si APD Fig. 17 is OM image of our 25 Gb/s Ge/Si APD which has a Ge mesa diameter = 20 μm and was applied RCE structure to enhance the responsivity. Fig. 18 presents typical curves of our 25 Gb/s APD room temperature dark current, photo current, and responsivity at 1310 nm. The breakdown voltage (Vbr ) of 25 Gb/s Ge/Si APD is −18.3 V. When biased at Vbr − 1 (1 volt lower than Vbr ), the APD dark current at 25 °C is only 0.81 μA with a gain of 8. Fig. 19 presents the reported dark current density of 25 Gb/s waveguide and normal incident APDs under operating condi-


Fig. 19.

3800911

Reported 25 Gb/s APD dark current density at M = 8.

Fig. 21. Ge/Si APD receiver back-to-back sensitivity and overload at 25.78 Gb/s. (Test conditions: 25.78 Gb/s, 1310 nm, ER = 9.5 dB, NRZ, PRBS = 2ˆ31–1, 25 °C).

Fig. 20.

25 Gb/s Ge/Si APD S21 curves under different gains.

tions at M = 8. One can see that the dark current density of our APD is very close to (or the same as) that of InP APDs under operating bias, which demonstrates that lattice mismatched devices accomplishes similar low dark current density as devices with matched lattice. The punchthrough voltage of 25 Gb/s Ge/Si APD is −11 V, and the measured device’s capacitance is 55 fF under Vpt , which also meets our calculated results. In addition, the measured 1310 nm responsivity is up to 0.7 A/W under unity gain for 25 Gb/s APD with backside reflector. The maximum measured 3 dB bandwidth of 25 Gb/s Ge/Si APD is 34.5 GHz (at M = 3.5). This high bandwidth is measured at the bias slightly larger than Vpt . It is the outcome of photo-generated carriers’ reaching their saturation velocity in Ge layer and the avalanche build-up time being small under lower gain [42]. The APD device’s 3-dB bandwidth decreases at higher gains due to longer avalanche build-up time. Under normal operation conditions (M = 8 ∼ 12), our Ge/Si APD 3-dB bandwidth is in the range from 26.5 to 21 GHz as shown in Fig. 20. Fig. 21 presents the back-to-back sensitivity of our Ge/Si APD receiver packaged with a commercial 25 Gb/s transimpedance amplifier (TIA). At the data rate of 25.78 Gb/s, the receiver exhibits a sensitivity of −23.5 dBm at 10−12 BER and an overload of +0.2 dBm. Our 25 Gb/s APD dark current (0.81 μA) has

almost no impact on 25 Gb/s system sensitivity, and these sensitivity limitations are from average photo current’s shot noise and 25 Gb/s TIA RMS noise current = 2 μA [43]. Besides error observed at low input optical power that is below sensitivity, we can also observe some error happening at very large input optical power. It is related to the high input power. For example, 1mW causes large photo current 2.45 mA ( = minimum gain (3.5) × input optical power (0.2 dBm) × responsivity (0.7 A/W)). It is larger than the average overload current threshold of TIA circuits which is equal to 2 mA [43]. The dynamic range of our 25 Gb/s APD receiver is reaching 23.7 dB, it is suitable for commercial 1310 nm module applications at various operation distances up to 40 km. Furthermore, we studied our 25 Gb/s APD receiver performance under different temperatures. Based on the measured data, the sensitivity changes are very small (< 1.0 dB) within industrial operation temperature (−40 to 85 °C). Unlike 10 Gb/s devices, our 25 Gb/s APD dark current at 25 °C is very small and it only increases to ∼6 μA at 85 °C temperature operations. This dark current increment has no impact on 25 Gb/s sensitivity as shown in Fig. 22. The high sensitivity and overload meet requirements of IEEE 25GBASE-ER and 100GBASE-ER4-lite without a semiconductor optical amplifier (SOA). The implementation of our APDs can greatly reduce power dissipation, cost and footprint, potentially realizing a compact module (such as CFP4 and QSFP28) for 100/400 G communications. In addition to limiting TIA, we also developed a 28 Gbaud APD receiver with linear TIA. For linear operations, the linearity of both APD and TIA should be carefully verified for complex modulation applications. Fig. 23 is the gain and Fig. 24 is the S21 of the linear receiver during operations under different input optical powers. According to Fig. 23 and Fig. 24, the APD gain only changes slightly and the APD S21 curves are well consistent when input optical power is within the range from −20 to −10 dBm. The small drop of APD gain is related to the space charge effects under high input optical powers [44] which becomes more

3800911


Fig. 22. Ge/Si APD receiver back-to-back sensitivity under different operation temperature. (Test conditions: 25.78 Gb/s, 1310 nm, ER = 9.5 dB, NRZ, PRBS = 2ˆ31–1).

Fig. 23.

Ge/Si APD gain dependence on input power.

Fig. 25. 28 Gbaud Ge/Si APD linear receiver at the wavelength of 1310 nm sensitivity.

Fig. 26.

28 Gbaud Ge/Si APD linear receiver 850 nm sensitivity.

28 Gbaud linear APD receiver sensitivity data at both MMF and SMF respectively. Based on these measured data, APD linear receiver can provide an extra ∼5 dB and ∼5.7 dB than PIN PD solutions for 200 G and 400 G PAM4 applications (by using KP4 FEC) [45], [46]. Such high sensitivity APD devices can compensate insertion losses from multiple wavelengths Mux/DeMux, reduce launch powers of transmitter sides, and extend operation distance. This APD based solution is an ideal one for next generation data center and telecommunication. IV. RELIABILITY A. High Temperature Operations Test Fig. 24.

Ge/Si APD linear receiver RF responses vs. input power.

significant with input power larger than −7 dBm. A slight gain drop has no significant impact on BER because APD receiver has better BER at such high optical power (−7dBm). At lower power, receiver with such large bandwidth (3 dB bandwidth >26.5 GHz) and good linearity can support complex modulation applications such as PAM4. Figs. 25 and 26 include our

To evaluate the reliability of Ge/Si APDs and estimate their failure rate and lifetime, the high temperature operation test at 175 °C has been conducted. The samples used for the test are normal incident. After wafer dicing and standard production screening process that includes visual inspection and probing test, the Ge/Si APD chips are bonded into TO46 headers without TIAs. During the stress test at 175 °C, the reverse current on the APDs is set to be 100 μA. With a time interval of ∼ 500 hrs,


3800911

TABLE III SAMPLES FOR HIGH TEMPERATURE OPERATIONS TEST Batch #1 #2 #3 #4

TABLE IV RANDOM FAILURE RATE AND MTTF ESTIMATIONS

Data Rate

Sample qty (pcs)

Stress time (hrs)

Failures

10 Gb/s 10 Gb/s 10 Gb/s 25 Gb/s

30 20 30 29

3528 2640 2328 2016

0 0 0 0

Random failure Random failure rate @40 °C (FIT) MTTF (year)

60% CL

90% CL

63 1798

160 716

TABLE V WEAR-OUT FAILURE RATE AND LIFETIME ESTIMATIONS Wear-out failure Median Lifetime (year) 25 Years Wear-Out Failure Rate @ 40 °C (FIT)

Fig. 27. Breakdown voltage drift plot of 109 pieces (4 batches) of Ge/Si APDs subjected to the high temperature operations test at 175 °C.

Fig. 28. Dark current drift plot of 109 pieces (4 batches) of Ge/Si APDs subjected to the high temperature operations test at 175 °C.

the room-temperature breakdown voltage and dark current at operating voltage of the tested APDs were measured. The failure criteria are: the variance of breakdown voltage exceeds 0.8 V or dark current changes for more than 50% of the initial value. 109 pieces of APD chips from 4 different wafers have been used in the reliability test, the sample allocation and testing results of which are summarized in Table III. Figs. 27 and 28 show the room temperature breakdown voltage and dark current drift plots of 109 pieces (4 batches) of Ge/Si APDs stressed at 175 °C. No failures have been observed according to the failures criteria given above. Based on the stress test data, both random failure rate/lifetime and wear-out failure rate/lifetime of Ge/Si APDs can be estimated. The random failure rate λ of APDs can be estimated as in (5): λ=

χ2 1 × 109 = × 109 MTTF 2 × T DH × AF

(5)

σ = 0.5

σ=2

2757 1

54548 0.56

where MTTF is mean time to failure, χ2 is the chi-square value, which is determined by the confidence level and the degree of freedom (number of failures), TDH is the total device hours, AF is the acceleration factor, which can be calculated by (2): Ea 1 1 AF = exp − (6) k Top Tstress where Ea is the activation energy (use 0.35 eV for the random failure mode as suggested by Telcordia GR-468-CORE [47]), k is Boltzmann’s constant (8.6 × 10−5 eV/K)), Top is the operating temperature (40 °C), and Tstress is the temperature during stress test (175 °C). The random failure rate (FIT, which is failure in time, 1 FIT = 1 failure in 109 device hours) and MTTF of the Ge/Si APDs are estimated as shown in Table IV. The testing data of samples batch #1 is used to estimate the wear-out failure rate and lifetime of Ge/Si APDs. We assume 1 wear-out failure happens at 3528 hours (a worst-case assumption). An activation energy of 0.7 eV (as suggested in GR-468-CORE [47]) and lognormal distribution with a standard deviation (σ) of 0.5 and 2 are used. The estimated median lifetime and wear-out failure rate after 25 years of operations are shown in Table V. B. Damp Heat Tests The low-cost non-hermetic packaging technology has been widely adopted by Si- and Si/Ge-based integrated circuits (ICs) [48], [49]. However, to date, the non-hermetic packaging is not widely implemented in III-V based optical components due to reliability concerns [50]–[52]. To evaluate the reliability of Ge/Si APDs in non-hermetic packaging applications, both biased damp heat (BDH) and unbiased damp heat (uBDH) tests have been conducted. The stress conditions for the BDH and uBDH tests are: 85 °C and 85% RH, at a bias same as normal operating voltage for BDH and zero bias for uBDH, respectively. The tested APD chips are bonded into TO46 headers with non-hermetic plastic package. Their room temperature breakdown voltage and dark current were periodically measured, and checked with the failure criteria: breakdown voltage changes for

3800911


TABLE VI SAMPLES FOR BDH AND UBDH TESTS

test. The 10 Gb/s and 25 Gb/s Ge/Si APDs have passed all the qualification tests listed above.

Batch

Data rate

Bias

Sample qty (pcs)

Stress time (hrs)

Failures

#1 #2 #3

10 Gb/s 25 Gb/s 25 Gb/s

Yes Yes No

34 20 10

2784 2016 2016

0 0 0

V. CONCLUSION In this paper, we reviewed the material, structure, and performance of Ge/Si APDs. Ge growth on Si is the most critical process in Ge/Si device fabrication and we reviewed various Ge film’s characteristics (deep level defects states, tensile, etc.) and their impacts on APD performance. Based on the properties of Ge-on-Si film, we designed a Ge/Si APD. Our APD demonstrated world record sensitivity at 25 Gb/s NRZ and 56 Gb/s PAM4 applications. Besides the high performance, we conducted comprehensive study on Ge material and Ge/Si devices reliability, the measured data prove that even with such high dislocation density in depleted region of Ge absorber, our APD passed all qualification tests requested by GR-468. Such high performance and high reliable devices pave the road for current 100 Gb/s and future 400 Gb/s or 1 Tb/s applications.

Fig. 29. Breakdown voltage drift plot of 64 pieces (3 batches) of Ge/Si APDs subjected to the BDH and uBDH tests.

REFERENCES

Fig. 30. Dark current drift plot of 64 pieces (3 batches) of Ge/Si APDs subjected to the BDH and uBDH tests.

more than 0.8 V or dark current changes for more than 50% of the initial value. 64 pieces (3 batches) of APD chips have been used in the BDH and uBDH tests. Table VI shows the sample allocation and testing results. Figs. 29 and 30 show the room temperature breakdown voltage and dark current drift plots of the Ge/Si APDs subjected to the BDH and uBDH tests. No failures have been observed according to the failures criteria given above. The BDH and uBDH testing data shown above indicate that Ge/Si APDs demonstrate superior reliability in certain humidity environment, which meet the reliability standard in GR-468CORE for non-hermetic packaging and are highly promising to be used in the low-cost non-hermetic applications. C. Qualification Tests The qualification tests required by GR-468-CORE for photodiodes include: die shear and wire pull strength tests, mechanical shock and vibration tests, temperature cycling test (−40 ◦ C/ +85 ◦ C, 50 cycles) and high temperature operations

[1] J. Kunstler, “Transformation of wireline access networks requires smarter solutions,” Ovum TMT Intelligence Informa, 2016. [2] J. Campbell et al., “Recent advances in avalanche photodiodes,” IEEE J. Sel. Topics Quantum Electron., vol. 10, no. 4, pp. 777–786, Jul./Aug. 2004. [3] Standard for Ethernet Amendment: Media Access Control Parameters, Physical Layers and Management Parameters for 200 Gb/s and 400 Gb/s Operation, IEEE P802.3ab 3bs. 200Gb/s and 400Gb/s Ethernet Task Force. [4] M. Nada, Y. Muramoto, H. Yokoyama, T. Ishibashi, and H. Matsuzaki, “High power tolerant InAlAs avalanche photodiode for 25Gbit/s applications,” Electron. Lett., vol. 49, pp. 62–63, 2013. [5] J. S. Dunn et al., “Foundation of RF CMOS and SiGe BiCMOS technologies,” IBM J. Res. Develop., vol. 47, no 2.3, pp. 101–138, 2003. [6] H. Chen et al., “Sub-5V germanium waveguide avalanche photodiode based 25 Gb/s 1310 nm optical receiver,” in Proc. Asia Commun. Photon. Conf., 2015, Paper AM1B.4. [7] P. Bandaru et al., “The fabrication of p-Ge/n-Si photodetectors, compatible with back-end Si CMOS processing, by low temperature (< 400oC) molecular beam epitaxy and electron-beam evaporation,” in Proc. Mater. Res. Soc. Symp., vol. 796, 2004, pp. V2.8.1–V2.8.6. [8] O. Olubuyide, D. T. Danielson, L. C. Kimerling, and J. L. Hoyt, “Impact of seed layer on material quality of epitaxial germanium on silicon deposited by low pressure chemical vapor deposition,” Thin Solid Film, vol. 508, pp. 14–19, 2005. [9] J. M. Hartmann et al., “Reduced pressure-chemical vapor deposition of Ge thick layers on Si (001) for microelectronics and optoelectronics purposes,” in Proc. MRS, vol. 809, Cambridge Univ. Press, 2004pp. B4.3.1– B4.3.6. [10] H.-C. Luan et al., “High-quality Ge epilayers on Si with low threadingdislocation densities,” Appl. Phys. Lett., vol. 75, no. 19, pp. 2909–2911, 1999. [11] W. Y. Loh et al., “Impact of local strain from selective epitaxial germanium with thin Si/SiGe buffer on high-performance p-i-n photodetectors with a low thermal budget,” IEEE Electron Device Lett., vol. 28, no. 11, pp. 984–986, Nov. 2007. [12] H.-C. Luan et al., “Germanium photodetectors for silicon microphotonics by direct epitaxy on silicon,” in Proc. MRS, vol. 607, Cambridge Univ. Press, 1999, pp. 279–284. [13] Y. Yamamoto, G. Kozlowski, P. Zaumseil, and B. Tillack, “Low threading dislocation Ge on Si by combining deposition and etching,” Thin Solid Films, vol. 520, pp. 3216–3221, 2012. [14] Y. Yamamoto, P. Zaumseil, T. Arguirov, M. Kittler, and B. Tillack, “Low threading dislocation density Ge deposited on Si (1 0 0) using RPCVD,” Solid-State Electron., vol. 60, pp. 2–6, 2011. [15] K. Ismail, “Effect of dislocations in strained Si/SiGe on electron mobility,” J. Vacuum Sci. Technol. B: Microelectron. Nanometer Struct. Process. Meas. Phenomena, vol. 14, no. 4, pp. 2776–2779, 1996.


[16] G. Wang et al., “A model of threading dislocation density in strain-relaxed Ge and GaAs epitaxial films on Si (100),” Appl. Phys. Lett., vol. 94, 2009, Art. no. 102115. [17] J. M. Hartmann et al., “Reduced pressure-chemical vapor deposition of intrinsic and doped Ge layers on Si (001) for microelectronics and optoelectronics purposes,” J. Crystal Growth, vol. 274, no. 1, pp. 90–99, 2005. [18] E. Kasper, Ed., Properties of Strained and Relaxed Silicon Germanium. London, U.K.: INSPEC, 1995. [19] M. Kim, P. Hashemi, and J. L. Hoyt, “Increased critical thickness for high Ge-content strained SiGe-on-Si using selective epitaxial growth,” Appl. Phys. Lett., vol. 97, no. 26, 2010, Art. no. 262106. [20] L. Giovane, L. M. Giovane, H.-C. Luan, A. M. Agarwal, and L. C. Kimerling, “Correlation between leakage current density and threading dislocation density in SiGe p-i-n diodes grown on relaxed graded buffer layers,” Appl. Phys. Lett., vol. 78, 2001. [21] O. Ryuzan, “Current-voltage characteristics of plastically deformed germanium P-N junctions,” J. Phys. Soc. Jpn., vol. 16, pp. 2177–2187, 1961. [22] M. Gonzalez et al., “Electric field dependence of trap-assisted-tunneling current in strained SiGe source/drain junctions,” Appl. Phys. Lett., vol. 94, pp. 233507-1–233507-3, 2009. [23] Y. Kang et al., “Monolithic Ge/Si avalanche photodiodes with 340 GHz gain-bandwidth product,” Nature Photon., vol. 3, pp. 59–63, 2009. [24] W. C. Dash and R. Newman, “Intrinsic optical absorption in single-crystal germanium and silicon at 77 K and 300 K,” Phys. Rev., vol. 99, no. 4, pp. 1151–1153, 1955. [25] Y. Ishikawa et al., “Strain-induced band gap shrinkage in Ge grown on Si substrate,” Appl. Phys. Lett., vol. 82, 2003. [26] C. Herring and E. Vogt, “Transport and deformation-potential theory for many-valley semiconductors with anisotropic scattering,” Phys. Rev., vol. 101, no. 3, p. 944, 1956. [27] S. Adachi, “Optical functions of InGaAsP: Tables,” in Properties of Indium Phosphide (EMIS Data Review Series No. 6). London, U.K.: INSPEC, 1991, pp. 416–428. ¨ u and Samuel Strite, “Resonant cavity enhanced photonic de[28] M. S. Unl¨ vices,” J. Appl. Phys., vol. 78, no. 2, pp. 607–639, 1995. [29] Y. Kang et al., “High performance Ge/Si avalanche photodiodes development in intel,” in Proc. Opt. Fiber Commun. Conf. Expo., 2011, Paper OWZ1. [30] M. Huang et al., “Avalanche photodiode with special lateral doping concentration,” US patent 8 778 725. [31] M. Huang et al., “High performance GeSi avalanche photodiode operating beyond Ge bandgap limits,” US patent 8786043. [32] M. Huang et al., “High speed germanium on silicon avalanche photodiode,” US patent 9478689. [33] J. Liu, “Monolithically integrated Ge-on-Si active photonics,” Photonics, vol. 1, pp. 162–197, 2014. [34] B. Levine et al., “−29 dBm sensitivity, InAlAs APD-based receiver for 10 Gb/s long-haul (LR-2) applications,” in Proc. Opt. Fiber Commun. Conf. 2005, 2005, Paper OFM5. [35] Y. M. Blanter and M. B¨uttiker, “Shot noise in mesoscopic conductors,” Phys. Rep., vol. 336, no. 1, pp. 1–166, 2000. [36] R. McIntyre, “Multiplication noise in uniform avalanche diodes,” IEEE Trans. Electron Devices, vol. 13, no. 1, pp. 164–168, Jan. 1966. [37] J. Gao, Optoelectronic Integrated Circuit Design and Device Modeling. New York, NY, USA: Wiley, 2011, p. 235. [38] Datasheet of ONET8551T 11.3Gbps Limiting Transimpedance Amplifier with RSSI. [39] M. Nada, Y. Muramoto, H. Yokoyama, T. Ishibashi, and H. Matsuzaki, “Triple-mesa avalanche photodiode with inverted P-down structure for reliability and stability,” J. Lightw. Technol., vol. 32, no. 8, pp. 1543– 1548, Apr. 2014. [40] K. Shiba, T. Nakata, T. Takeuchi, K. Kasahara, and K. Makita, “Theoretical and experimental study on waveguide avalanche photodiodes with an undepleted absorption layer for 25-Gb/s operation,” J. Lightw. Technol., vol. 29, no. 2, pp. 153–161, Jan. 2011. [41] M. Nada et al., “Inverted InAlAs/InGaAs avalanche photodiode with low– high–low electric field profile,” Jpn. J. Appl. Phys., vol. 51, pp. 02BG031–02BG03-4, 2012. [42] R. Emmons, “Avalanche-photodiode frequency response,” J. Appl. Phys., vol. 38, pp. 3705–3714, 1967. [43] Datasheet of IN2841TA 28Gbps Transimpedance Amplifier/Limiting Amplifier. [44] H. C. Bowers, “Space-charge-induced negative resistance in avalanche diodes,” IEEE Trans. Electron Devices, vol. ED-15, no. 6, pp. 343–350, Jun. 1968.

3800911

[45] F. Chang, “First demonstration of PAM4 transmissions for record reach and high-capacity SWDM links over MMF using 40G/100G PAM4 IC chipset with real-time DSP,” in Proc. Opt. Fiber Commun. Conf. Opt. Soc. Amer., 2017, pp. 1–3. [46] K. Zhong et al., “Amplifier-less transmission of 56 Gbit/s PAM4 over 60 km using 25 Gbps EML and APD,” in Proc. Opt. Fiber Commun. Conf. Exhib., 2017, Paper Tu2D.1. [47] Generic Reliability Assurance Requirements for Optoelectronic Devices Used in Telecommunications Equipment, GR-468-CORE, Issue 2, 2004. [48] J. W. Blade, “The effectiveness of silicone gels for corrosion prevention of silicon circuits: The final report of the IEEE computer society computer packaging committee special task force,” IEEE Trans. Compon., Hybrids, Manuf. Technol., vol. 14, no. 2, pp. 352–365, Jun. 1991. [49] N. Sinnadurai, “Plastic packaging is highly reliable,” IEEE Trans. Rel., vol. 45, no. 2, pp. 184–193, Jun. 1996. [50] R. R. Sutherland, J. C. D. Stokol, C. P. Skimshire, B. M. MacDonald, and N. F. Sloan, “The reliability of planar InGaAs/InP PIN photodiodes with organic coatings for use in low cost receivers,” in Proc. SPIE Fiber Opt. Rel., vol. 1774, pp. 226–232, 1989. [51] W. S. Ring et al., “Reliability concerns and packaging issues of nonhermetic VCSEL and LED based transceivers for the enterprise market,” in Proc. 2004 54th Electron. Compon. Technol. Conf., vol. 1, 2004, pp. 186–191. [52] J. W. Osenbach and T. L. Evanosky, “Temperature-humidity-bias-behavior and acceleration model for InP planar PIN photodiodes,” J. Lightw. Technol., vol. 14, no. 8, pp. 1865–1881, Aug. 1996.

Mengyuan Huang (M’15) received the M.S. degree in electrical and computer engineering from Duke University in 2009. He joined SiFotonics Technologies in 2010. He is currently a Senior Technical Manager at SiFotonics Technologies on Ge/Si APD-related projects. He has coauthored more than 20 publications and been granted 11 U.S. patents about Ge/Si devices.

Su Li received the Ph.D. degree in materials science and engineering from Tsinghua University, Beijing, in 2011. In 2012, he joined SiFotonics Technologies, where he works on the material characteristics of Ge/Si epitaxy.

Pengfei Cai received the Ph.D. degree in optoelectronics from Tsinghua University, Beijing, in 2007. He joined SiFotonics Technologies in 2007 and he is the VP of technologies to develop silicon-based PD/APD detectors, waveguidebased devices, and related optical component products.

Guanghui Hou received the M.S. degree from the Chinese Academy of Sciences in 2007. He joined SiFotonics Technologies in 2007 and focused on various high-speed testing of silicon photonics devices.

Tzung-I Su received the M.S. degree in physics from National Cheng Kung University, Taiwan, in 2001. In 2013, he joined SiFotonics Technologies, where he worked on silicon photonics process design.

Wang Chen received the M.E. degree in electronics science and technology from Tsinghua University, Beijing, in 2006. In 2007, he joined SiFotonics Technologies, where he worked on germanium on silicon photodetector process development and characterization.

Ching-yin Hong received the Ph.D. degree in materials science and engineering from Massachusetts Institute of Technology in 2003. She is the Co-Founder of SiFotonics Technologies Co., Ltd.

Dong Pan received the Ph.D. degree from the Chinese Academy of Science. He was with the University of Virginia and Massachusetts Institute of Technologies. He is the Founder and CEO of SiFotonics Technologies Co., Ltd., which specializes in the commercialization of silicon photonics from Ge/Si devices to silicon photonics integration.