Towards a high bandwidth waveguide photodetector in an InP membrane on silicon

An InP membrane based waveguide photodetector is designed to reach a 3 dB bandwidth beyond 100 GHz. A uni-traveling-carrier (UTC) configuration is used for minimizing the carrier transit time. A novel double-side process technology is developed to enhance the bandwidth. In addition, a low resistance n-type contact technology is optimized to control the metal spiking into semiconductors. Introduction and design: InP Membrane on Silicon (IMOS) technology provides a new platform for integration of passive and active photonic devices on top of CMOS chips. These devices are fabricated in an InP-based membrane which is bonded to a silicon wafer by using benzocyclobutene (BCB). Recently, a uni-traveling-carrier photodetector (UTC-PD) in the IMOS platform is being developed for high speed applications. The carrier transport in UTC-PDs is dominated by electrons due to the utilization of a p-type doped absorption layer. The higher velocity of electrons compared to holes results in a higher bandwidth than in conventional PIN-PDs. The designed structure is shown in Fig. 1. The width of the PD mesa is designed as 3 μm. A 300 nm thick intrinsic InP layer is used both as the passive waveguide and as the electron collector (depleted) in the UTC region. A 150 nm thick ptype doped InGaAs layer is used both as the absorber and the p-contact. This layer is doped with a gradient concentration from 10 cm at the collector-absorber interface to 10 cm at the contact surface. Optical simulation gives a modal absorption coefficient of 3600 cm at 1.55 μm. This high absorption is due to the strong confinement of the optical mode in such a high index contrast membrane. The PD is designed to be 10 μm long, which is sufficient for an absorption efficiency of 97%. The small device area results in a junction capacitance of less than 10 fF. The carrier transit time of a UTC-PD consists of two components: the electron diffusion time in the ptype InGaAs absorber, and the electron drift time in the depleted InP collector. The first term can be reduced below 1 ps with the doping gradient in the p-type InGaAs layer. In terms of the electron drift time, UTC-PDs can be set with an optimal bias voltage so that electrons travel at the velocity overshoot (above 2×10 cm/s). At this bias, the electron drift time through the 300 nm collector is only 1.5 ps. The bandwidth limited by a total transit time of 2.5 ps is calculated to be 150 GHz. Double-side process technology: While the transit time and the capacitance are minimized, the high sheet resistance of a thin membrane device remains the major limiting factor for the bandwidth. In order to reduce the series resistance of the p-contact layer, a double-side process scheme (Fig. 2) is used to evaporate the p-metal (Ti/Pt/Au) at the back side of the device before bonding to Si. In this way the spacing between p-metal can be made sufficiently narrow to minimize the series resistance. An optimized spacing of 3 μm (equal to the mesa width) is chosen as a trade-off between the metal loss and the resistance (Fig. 3). At this point, the metal loss (340 cm) is very limited compared to the modal absorption (3600 cm). The resistance is simulated by considering a conductivity of 9000 S/m and a specific contact resistance of 3×10 Ω cm of the InGaAs layer. In case of p-metal on top of the membrane (n-side, same side as the mesa), a practical limit from lithography determines the smallest Fig. 1: Cross-section of the UTC-PD. Th 4a R12