High-speed Ge photodetector monolithically integrated with large cross-section silicon-on-insulator waveguide Open Access

The substantial growth in web based multimedia and social networking applications is putting a significantly larger amount of data traffic on the current telecom and data infrastructure. This is leading to significant connectivity bottlenecks. It is widely accepted that silicon photonics, due to its electronics integration capability, proven manufacturing record and price-volume curve, will be the platform of choice for the next generation interconnect and communication solutions to address the connectivity bottlenecks. This has fueled significant research and development work in this area in the past few years.^1–11 Many silicon-based active photonics components, such as high-speed modulators and Ge photodetectors^4–10 have been demonstrated on submicron waveguides. However, submicron SOI waveguides still suffer from high fiber coupling loss, high polarization dependent loss, and large waveguide birefringence and phase noise. As a result they do not provide a satisfactory platform for implementation of passive and wavelength-division-multiplexing devices. On the other hand, many high performance optical components have been demonstrated for large cross-section SOI waveguides, such as arrayed waveguide gratings, Echelle gratings, and Triplexer filters^12–14 exhibiting low fiber coupling loss, low waveguide propagation loss, low polarization dependent loss, and low polarization dependent frequency shift. However, a compact, high-speed photodetector has never been demonstrated using large cross-section SOI waveguides. In this letter, we propose and demonstrate a compact, high speed Ge photodetector efficiently butt-coupled with a large cross-section SOI waveguide in which the Ge p-i-n junction is placed in the horizontal direction. A high speed Ge photodetector with greater than 32 GHz optical bandwidth is demonstrated. The device operates with a responsivity of 1.1 A/W at a wavelength of 1550 nm. In addition, the demonstrated Ge detector has a low fiber coupling loss $(<1.2dB)$ and low polarization dependent responsivity $(<0.3dB)$⁠. The demonstrated Ge photodetector enables integration with high performance, polarization independent demultiplexers made using compatible large cross-section SOI waveguides, to form monolithically integrated silicon photonics receivers for multichannel terabit data transmission applications.

A possible solution enabling high-speed Ge photodetector functionality within a large cross-section silicon waveguide platform can be based on a vertical p-i-n junction with a Ge waveguide grown over a large cross-section SOI waveguide, so that light is evanescently coupled to the Ge region. The coupling length between the large cross-section waveguide to the Ge waveguide in this configuration will be much longer than the alternative submicron SOI waveguide based devices. The excessive length of such a device will result in a large RC time delay and thus limit the speed of the detector. Another possible structure is a vertical p-i-n junction with Ge waveguide butt-coupled to a large cross-section SOI waveguide. In this case, although the device size can be small, the speed is still limited by the long transit time in the thicker intrinsic Ge region. To overcome the above limitations, a Ge photodetector is proposed and demonstrated here. The photodetector and the cross-section of the active p-i-n diode region are schematically shown in Figs. 1(a) and 1(b), respectively. As the light propagates from the ridge SOI waveguide to the photodetector region, the light is butt-coupled and absorbed in the active Ge region where photocurrent is generated. The Ge length only needs to be $10μm$ to absorb over 98% of the light as its absorption coefficient is around $4000cm−1$⁠.⁹ This horizontal p-i-n configuration enables a very narrow intrinsic Ge region, hence reducing the transit time in this region. The transit time limited bandwidth is given by¹⁵ 

where $υsat=6.0×106cm/s$ is the saturation drift velocity in Ge. For a $WGe=0.65μm$⁠, a transit time limited speed $>40GHz$ is calculated. The Ge sidewalls and slabs were doped to form the p and n regions. Doping levels of $5×1019cm−3$ for p and n regions were selected to give good Ohmic contacts to the Ge slabs. Around $0.075μm$ depth of shallow sidewall doping on Ge was used to minimize the free carrier absorption loss and photo-generated carries loss in the doped regions. The SOI waveguide was $3μm$ wide to ensure low coupling loss with a lensed fiber. The Ge waveguide has a narrower width of $0.8μm$ and is connected to the SOI waveguide by a horizontal taper.

The photodetector was fabricated on six inch SOI wafers with $0.375μm$ thick buried oxide (BOX) and a $3μm$ thick silicon epitaxial layer. First, the silicon recess region was formed by etching the silicon layer down to $0.6μm$ residual thickness above the BOX layer. Then, a 100 nm thick Ge buffer layer was selectively grown at $400°C$⁠, followed by $4μm$ thick Ge growth at $670°C$ inside the silicon recess region. The Ge film was intentionally over grown, then thinned down and planarized with a chemical-mechanical polishing (CMP) step. A final Ge thickness of $2.4μm$ was achieved after CMP. The wafers then underwent a Ge anneal to reduce the threading dislocation density in the Ge film. The silicon ridge waveguides, Ge waveguides, and the silicon horizontal tapers were formed by the same etching step. As the etching rate of Ge is greater than that of silicon, additional etching in the silicon region was performed, forming a silicon ridge waveguide with a $0.6μm$ thick slab and a Ge ridge waveguide with a $0.2μm$ thick slab. Boron and phosphorus were implanted in the sidewalls and slabs of the Ge waveguide to form a horizontal p-i-n junction and p-type and n-type Ohmic contact areas. After rapid thermal annealing dopant activation, the Ti/Al metal stack was deposited and patterned to form p-type and n-type metal contacts. Finally, oxide and nitride films were deposited as waveguide cladding and passivation layers. A cross-section scanning electron microscopy image of the final fabricated device is shown in Fig. 2.

The dark current at −0.5 and −1 V reverse biases was measured at 0.24 and $1.3μA$⁠, respectively, for the device with active area of $0.8μm×10μm$⁠. A lensed fiber with a spot size of $3.3μm$ was used to couple light into the SOI waveguide. The optical power from the lensed fiber was $100μW$⁠. In order to determine the responsivity of the photodetector, the fiber to waveguide coupling loss and the loss due to the SOI ridge waveguide were measured for the neighboring waveguides without the Ge detector. The measured fiber to waveguide coupling loss was 1.2 dB, and loss from the SOI waveguide was 0.2 dB, so the optical power that reached the Ge detector was about $72.4μW$⁠. The measured photocurrent for transverse electric (TE) and transverse magnetic (TM) polarized light were 77 and $80μA$⁠, respectively. The responsivity is calculated as 1.06 and 1.1 A/W at 1550 nm wavelength for TE and TM polarizations, respectively. The fiber-accessed responsivity defined by photocurrent divided by optical power from the lensed fiber is calculated at about 0.8 A/W. The responsivity of the above device over the wavelength range from 1260 to 1640 nm was also measured, and is shown in Fig. 3. The ideal responsivity with 100% quantum efficiency is plotted on the same graph. This demonstrates satisfactory operation over a 300 nm wide wavelength range. The measured responsivity follows the quantum limit curve up to 1560 nm. The responsivity was measured at 0.47 A/W even at the longer wavelength edge of the L-band (1611 nm). The device also shows very low polarization dependent responsivity. The inset in Fig. 3 shows the measured polarization dependent responsivity (PDR) of the device over the measured wavelength range where a very low PDR of 0.3 dB is demonstrated. The roll-off in responsivity after 1560 nm is expected due to the decreased absorption of Ge. The responsivity at longer wavelengths can be improved simply by increasing the length of Ge without sacrificing the speed of the device, as the device is not RC limited.

The frequency response of the above device was measured by a vector network component analyzer. A high-speed rf signal was applied to an external high-speed modulator with a bandwidth of about 40 GHz. A reverse voltage bias was applied to the device through a bias-tee. The modulated light at 1550 nm was then coupled to the device and the electrical output was measured through a high speed rf probe. Figure 4 shows the normalized optical response for the detector. The measured optical bandwidths are 17.5, 32.6, and 36.8 GHz at bias of 0, −1, and −3 V, respectively. The inset in Fig. 4 shows the dependence of optical bandwidth with reverse bias. As the drift velocity already reaches the saturation velocity in Ge at −1 V bias, high optical bandwidth is achieved at this relative small voltage. The device at −1 V bias is fast enough to detect a 40 Gbs/s optical signal. Figure 5 shows the measured optical bandwidth for the devices with different i-Ge width and the transit limited bandwidth according to Eq. (1). The measured results match the transit limit very well. This proves the speed of the devices is transit limited.

In conclusion, we have demonstrated a high-speed Ge photodetector integrated with a large cross-section SOI waveguide. A butt-coupled Ge horizontal p-i-n junction was used to enable very high-speed operation. The demonstrated device has a very compact active area of only $8μm2$⁠, a 3 dB bandwidth of over 32 GHz and responsivity as high as 1.1 A/W.

The authors acknowledge funding of this work by DARPA MTO office under UNIC program (Contract agreement No. HR0011–08–9-0001 with SUN Microsystems) managed by Dr. J. Shah. The authors greatly acknowledge Dr. Wei Qian and Dr. Jonathan Luff from Kotura Inc., Dr. Guoliang Li, Dr. Xuezhe Zheng and Dr. Kannan Raj from SUN Microsystems for helpful discussions. The views, opinions, and/or findings contained in this article/presentation are those of the author/presenter and should not be interpreted as representing the official views or policies, either expressed or implied, of the Defense Advanced Research Projects Agency or the Department of Defense.