Electrical Engineering and Computer Science, Seoul National University, Seoul , Republic of Korea; ABSTRACT 1. INTRODUCTION

Transcription

1 Design optimization of an optically drivable heterogeneous MOSFET with silicon compatibility Seongjae Cho *a, Hyungjin Kim b, S. J. Ben Yoo c, Byung-Gook Park b, and James S. Harris, Jr. a a Department of Electrical Engineering, Stanford University, Stanford, CA 94305, USA; b Department of Electrical Engineering and Computer Science, Seoul National University, Seoul , Republic of Korea; c Department of Electrical and Computer Engineering, University of California, Davis, CA 95616, USA *felixcho@stanford.edu; tel ; fax , bgpark@snu.ac.kr, harris@snow.stanford.edu ABSTRACT Optical and electronic devices for optoelectronic integrated circuits have been extensively studied, and now, more efforts for the conversion between optical and electrical signals are accordingly required. In this work, a silicon (Si)-compatible optically drivable III-V-on-Si metal-oxide-semiconductor field-effect transistor (MOSFET) is studied by simulation. The proposed optoelectronic device provides a strong interface between the optical and the electronic platforms as a key component of the optical interconnect. The optically driven MOSFET device is analogously analyzed into a photodetector and its complementary device, getting rid of receiver circuitry, which improves the integration density and simplifies the fabrication processes. To realize the optical switching with maximized photo-sensing region, a bottom gate is formed to modulate the channel, where germanium (Ge) and gallium arsenide (GaAs) are the active materials on Si platform. Both direct-current (DC) and alternating-current (AC) performances of an optimized device are evaluated. Keywords: Optoelectronics, integrated circuit, silicon compatibility, metal-oxide-semiconductor field-effect transistor, optical interconnect, photodetector, optical switching 1. INTRODUCTION Optical interconnect is considered for the next-generation interconnect technology as one of the promising copper (Cu) replacements due to its merits of high bandwidth, low power, low latency, and noise immunity [1]. From a system view, optical interconnect includes not only optical waveguide but also the components processing light signals in connection with a waveguide in the optoelectronic integrated circuit, especially with the function of optical and electrical signal conversion. In this work, an optically drivable MOSFET for III-V-on-silicon (Si) integrated optoelectronics is proposed and designed. The device is operated by an optical signal when the device is electrically turned on, and vice versa. The base materials of interest in this work are germanium (Ge) and gallium arsenide (GaAs) on the Si substrate. Although the properties of individual materials are already well known and have been conventionally used for highspeed complementary metal-oxide-semiconductor (CMOS) devices [2 4], they are re-highlighted for various applications via the heterogeneous integration [5 6]: higher Si-compatibility of III-V and an efficient optical source on Si are made more expectable by the intermediate lattice constant of Ge between those of Si and GaAs, and its unique energy-band structure. 2. DEVICE STRUCTURE AND OPERATION Figure 1 shows the schematic cross-sectional view of the simulated optical MOSFET device in the channel direction. Although the detailed process architecture is not dealt in this work, a possible fabrication method can be schemed: Ge and GaAs are epitaxially grown on Si substrate in sequence. SiGe can be optionally used in growing Ge on Si. The Ge layer is degenerately p-type doped for the bottom gate formation. Ge is selectively removed by anisotropic etch to construct the photo-sensitive area of GaAs. Before growing GaAs, thin gate oxide is deposited and patterned for defining the active. GaAs is epitaxially grown with in situ p-type doping and sits on the gate oxide by lateral overgrowth [7]. For more efficient lateral growth process, nanoscale holes can be formed over the dielectric layer [8 9]. The grown GaAs is chemically and mechanically planarized (CMP). Degenerately n-type doped GaAs is grown and patterned to construct raised source and drain (S/D) junctions. Electrodes are formed by proper metal depositions and the fabrication is completed by annealing. Physics and Simulation of Optoelectronic Devices XXI, edited by Bernd Witzigmann, Marek Osinski, Fritz Henneberger, Yasuhiko Arakawa, Proc. of SPIE Vol. 8619, 86191M 2013 SPIE CCC code: X/13/$18 doi: / Proc. of SPIE Vol M-1

2 The optical MOSFET in Fig. 1 is normally in the off-state no matter the gate voltage (V GS ) is high or low when there is no light input. However, when an optical signal is incident over the device, electrons in the GaAs region are excited by the photons and the potential barrier seen by source electrons is slightly lowered. The source electrons and optically generated electrons are swept into the drain when V GS is high. Either high optical signal or high electrical signal turns on the device only when the other is activated. Source hi Drain Gate Ge Si Substrate Figure 1. Schematic view of the simulated optical MOSFET device. The device structure and fabrication process are not limited to those suggested in this study to realize the proposed heterogeneous optoelectronic device. In simulations, the thickness of GaAs active and gate oxide were 490 nm and 5 nm, respectively, as shown in Fig. 1. The inversion layer is formed in the GaAs region between source and gate when the device is illuminated. For this reason, it can be termed as the optical channel and its length (L OC ) was 5 μm. The GaAs photo-sensing region between gate and drain ends is defined as the electrical channel controlled by the gate and its length (L EC ) was 250 nm. Also, the workfunction of the p + Ge bottom gate was set to 4.70 ev. We used an optical input with a wavelength of 800 nm in the near-infrared (NIR) regime that can be detected by GaAs of which bandgap energy (E G ) is 1.42 ev at 300 K. An 800-nm optical source can be achieved by AlGaAs [10 11], by which its integration with the optical MOSFET is more viable. 3. OPTIMIZATION AND DC PERFORMANCE 3.1 Operation without light signal Several parameters can be considered for the design optimization, including critical dimensions and process conditions. Among the parameters, the optimization is focused on doping concentration of the GaAs photo-sensing region in this work. The optical MOSFET should be turned off in the absence of light signal even though the device in the operation mode, V GS = drain voltage (V DS ) = High, which is the most fundamental requirement that doping condition in GaAs channel should meet above all. Figure 2(a) shows the simulated I D -V GS curves of the optical MOSFET in Fig. 1. GaAs channel doping concentration (N ch ) was varied from cm -3 to cm -3 to identify the permissible range of N ch. V GS was swept from -0.5 V to 2 V without light at V DS = 2 V. As V ch increases, the electrical on-state current (I on,dark ), I D at V GS = 2 V with no light input, monotonically decreases as confirmed by Fig. 2(a). Above N ch = cm -3, I on,dark drastically decreased and V GS had no controllability over the channel with N ch higher than cm -3. The drastic decrease of I on,dark as a function of N ch is more explicitly redrawn in the upper part of Fig. 2(b). In the MOSFET operation, the energy barrier seen by the source electrons is elevated as the channel doping concentration gets higher. Optically generated electrons make the barrier lowering more efficient in a lightly doped channel. The number of conduction electrons optically excited from the valence band is much smaller than that of source electrons. In the operation of the optical MOSFET, the majority of current conduction is composed of the source electrons by drift-diffusion. Thus, it is reasonable that the optical electrons Proc. of SPIE Vol M-2

3 make a contribution to the increased current conduction by lowering the energy barrier through the instantaneous elevating of Fermi-level rather than by supplying a number of electrons comparable to the amount of source electrons. Fig. 2(b) (upper) proves that the permissible range of N ch is above cm -3. The electrical off-state current (I off,dair ) is defined as I D at V GS = 0 V without light. Here, the subscript dark in I on,dark and I off,dark specified the on- and off-state currents determined by whether V GS is high or low (only electrically) in the absence of the light signal ' 10' =. 10' - Q100 10` 10',. 10-e ç 10s i 10i 10-n " U 10" 10'4 I- 10's Channel Doping t 1x1014 cm' -o- 1x1010 cm-' -A- 1x1016 cm' -0-5x1016 cm4-4- 6x1016 cm-' 1x10" cm' t 3x10" cm-' 6x10" cm' -0-1 x1016 cm',- Light Off, VDS = 2.0 V Higher Channel Doping M Gate Volgate (VGs) [V] 2.0 (a) 10. Q Q15 Q15 Ó-1 Ê 2.0 w_ Drive mode with light c VGS = VDS = 2.O V, ff Permissible I Channel Doping Concentration [cm'] Standby mode with light off (V0S, VDS = 0 V, 2.0 V) All permissible 0.5 D 10" 10" ' Channel Doping Concentration [cm.'] D 10" Figure 2. Direct-current (DC) characteristics with no light at different N ch values. (a) I D -V GS curves. (b) I on,dark and I off,dark. (b) There is no noticeable issue in I off,dark as confirmed by both Fig. 1 and Fig. 2(b) (lower). The fluctuations in Fig. 2(b) (lower) are not from a certain tendency depending on N ch but can be understood as noises. The design optimization has begun from finding out the permissible range of N ch where the device is turned off regardless of the value of V GS, particularly when V GS = 2 V (high), only if the light signal is not supplied. Proc. of SPIE Vol M-3

4 3.2 Operation with light signal The device operation is investigated in the existence of light signal with a power of 1 W/cm 2 for further optimization. Here, N ch is more specifically limited by the permissible range obtained in the previous subsection. As shown in Fig. 3(a), I on,light reaches 10-4 A/μm while I on,dark was in the order of femto-ampere (fa). I on,light is defined as I D at V GS = 2 V with the illumination on a device with a high enough N ch. I on,light is comparable to the maximum I on,dark that can be reached by a normally operating device even without light signal owing to the low enough N ch. I off,light of a few tens of nano-amperes (na) mainly originates from the electrons optically generated in the channel and swept into the drain by high V DS. However, the optically excited electrons are not increased by higher V GS. In comparing the I on,light and I off,light of a fully operating optical MOSFET, it is reassured that the optically generated electrons contributes to more efficient formation of the inversion channel rather than to a remarkable increase in drift electrons by participating in the drift. As N ch gets higher, both I on,light and I off,light monotonically decreases and the former shows more prominent degradation Channel Doping 6x1016 cm' 1x1017 cm 3x1017 cm-3 t 6x1017 cm' 1x1078 cm 3 Light On, Vps = 2.0 V Higher Channel Doping Gate Voltage (VGs) [V] (a) 2.0, d 3 r 10 - Light On (1 W /cm2), Vps = 2.0 V 0 ó10' d 350 > Channel Doping Concentration [cm] Light On (1 Wlcm2), Ups =2.0 V '1018 U Channel Doping Concentration [cm'] (b) Figure 3. DC characteristics with light signal at different N ch values. (a) I D -V GS curves. (b) Current ratio and subthreshold swing. Proc. of SPIE Vol M-4

5 Fig. 3(b) depicts the I on /I off current ratio (more properly, I on,light /I off,light ) (upper) and subthreshold swing (S) (lower) as a function of N ch. Since higher N ch degrades both the current ratio and swing characteristics, it would be desirable to lightly dope the channel as long as the N ch falls into the permissible range. 4. HIGH-FREQUENCY PERFORMANCE For an optimized optical MOSFET with N ch = cm -3, the high-frequency transfer characteristics between optical and electrical signals are evaluated. More specifically, I D as the result of available photo-current (I opt ) is investigated in the high-frequency regime. Since an I opt as low as the order of a few tens of na regardless of V GS draws a 10 3 times higher I on,light, the optical MOSFET plays a role of optical amplification. Fig. 4 demonstrates the frequency response of the optical MOSFET as an optical amplifier. 90 d 45 fmax, opt = 1.5 GHz ' " Frequency [Hz] ' Frequency [Hz] Figure 4. Frequency response of the optical MOSFET to the light signal. Magnitude (left) and phase (right) of I D /I opt. A positive decibel (db) value of the I D /I opt magnitude (left in Fig. 4) implies that the optically generated electrons help in constructing the inversion channel and successfully draw a certain amount of I D. The maximum optical cut-off frequency (f max,opt ) is 1.5 GHz. Below this frequency, conduction current by drift-diffusion can promptly respond to the optical signal change. However, at frequencies higher than f max,opt, the I D is even lower than I opt, which infers that the speed of the input signal begins to exceed that of the optical process. In consequence, the optical generation time of a photo-sensing region would be one of the factors limiting f max,opt of an optical MOSFET. It is confirmed from the results that a GHz-bandwidth is obtained from a long-channel device with 5-μm L OC. As plotted in Fig. 4 (right), a significant phase mismatch between I opt and I D at high frequencies above 10 MHz occurs, which shows that I D severely lags behind the high-frequency optical signal. Although the main focus in designing has been made on the channel doping concentration, there is much room for improving the high-frequency performances by geometric optimization. Reflecting that the bandwidth of a p-i-n photodetector has a strongly dependence on the distance between anode and cathode (the thickness of intrinsic region) [12], designing a short-channel optical MOSFET would be definitely strategic for achieving improved performances and realizing the optoelectronic circuits with higher integration density. Proc. of SPIE Vol M-5

6 5. CONCLUSIONS In this work, an optically drivable III-V-on-Si MOSFET was proposed and its performances were evaluated. A design optimization was carried out in terms of the channel doping concentration which is the most crucial parameter in device performance. DC and high-frequency performances were investigated at the optimum doping, where a current ratio (I on,light /I off,light ) higher than 10 3 and a maximum optical cut-off frequency (f max,opt ) of 1.5 GHz were obtained. The proposed optical MOSFET has a strong potential as a core component in the optical interconnect for high-speed and high-density optoelectronic integrated circuits through the convergences between materials and between functions. ACKNOWLEDGEMENTS This work was supported by the Center for Integrated Smart Sensors funded by the Korean Ministry of Education, Science and Technology as Global Frontier Project (CISS-2012M3A6A ). REFERENCES [1] Interconnect, International Technology Roadmap for Semiconductors, (2011). [2] Saraswat, K. C, Chui, C. O., Krishnamohan, T., Nayfeh, A. and McIntyre, P., Ge based high performance nanoscale MOSFETs, Microelectron. Eng. 80(1), (2005). [3] Kamata, Y., High-k/Ge MOSFETs for future nanoelectronics, Mater. Today 11(1 2), (2008). [4] Adachi, S., [Physical Properties of III-V Semiconductor Compounds: InP, InAs, GaAs, GaP, InGaAs, and InGaAsP], John Wiley & Sons, (1992). [5] Cho, S., Kang, I. M., Kamins, T. I., Park, B.-G. and Harris, Jr., J. S., Silicon-compatible compound semiconductor tunneling field-effect transistor for high performance and low standby power operation, Appl. Phys. Lett. 99(24), (2011). [6] Cho, S., Park, B.-G., Yang, C., Cheung, S., Yoon, E., Kamins, T. I., Yoo, S. J. B. and Harris, Jr., J. S., Roomtemperature electroluminescence from germanium in an Al 0.3 Ga 0.7 As/Ge heterojunction light-emitting diode by Γ- valley transport, Opt. Express 20(4), (2012). [7] Ujiie, Y. and Nishinaga, T., Epitaxial Lateral Overgrowth of GaAs on a Si Substrate, Jpn. J. Appl. Phys. 28(3), L337 L339 (1989). [8] Langdo, T. A., Leitz, C. W., Currie, M. T., Fitzgerald, E. A., Lochtefeld, A. and Antoniadis, D. A., High quality Ge on Si by epitaxial necking, Appl. Phys. Lett. 76(25), (2000). [9] Wang, L. S., Tripathy, S., Wang, B. Z., Teng, J. H., Chow, S. Y. and Chua, S. J., Nanoscale epitaxial overgrowth process and properties of GaN layers on Si (111) substrates, Appl. Phys. Lett. 89(1), (2006). [10] Sakamoto, M., Endriz, J. G. and Scifres, D. R., 120 W CW output power from monolithic AlGaAs (800 nm) laser diode array mounted on diamond heatsink, Electron. Lett. 28(2), (1992). [11] Knauer, A., Bugge, F., Wenzel, E. H., Vogel, K., Zeimer, U. and Weyers, M., Optimization of GaAsP/AlGaAs- Based QW Laser Structures for High Power 800 nm Operation, J. Electron. Mater. 29(1), (2000). [12] Cho, S., Kim, H., Sun, M.-C., Park, B.-G. and Harris, Jr., J. S., Process Considerations for 80-GHz High- Performance p-i-n Silicon Photodetector for Optical Interconnect, J. Semicond. Technol. Sci. 12(3), (2012). Proc. of SPIE Vol M-6