TY - GEN
T1 - A Hybrid Optical-Electrical Analog Deep Learning Accelerator Using Incoherent Optical Signals
AU - Yang, Mingdai
AU - Jokar, Mohammad Reza
AU - Qiu, Junyi
AU - Lou, Qiuwen
AU - Liu, Yuming
AU - Udupa, Aditi
AU - Chong, Frederic T.
AU - Dallesasse, John M.
AU - Feng, Milton
AU - Goddard, Lynford L.
AU - Hu, X. Sharon
AU - Li, Yanjing
N1 - We thank all anonymous reviewers for their comments. This work is sponsored in part by NSF grants 1640192 and 1405959, and also E2CDA-NRI, a funded center of NRI, a Semiconductor Research Corporation (SRC) program sponsored by NERC and NIST.
PY - 2021/6/22
Y1 - 2021/6/22
N2 - We present a hybrid optical-electrical analog deep learning (DL) accelerator, the first work to use incoherent optical signals for DL workloads. Incoherent optical designs are more attractive than coherent ones as the former can be more easily realized in practice. However, a significant challenge in analog DL accelerators, where multiply-Accumulate operations are dominant, is that there is no known solution to perform accumulation using incoherent optical signals. We overcome this challenge by devising a hybrid approach: Accumulation is done in the electrical domain, while multiplication is performed in the optical domain. The key technology enabler of our design is the transistor laser, which performs electrical-To-optical and optical-To-electrical conversions efficiently to tightly integrate electrical and optical devices into compact circuits. As such, our design fully realizes the ultra high-speed and high-energy-efficiency advantages of analog and optical computing. Our evaluation results using the MNIST benchmark show that our design achieves 2214× and 65× improvements in latency and energy, respectively, compared to a state-of-The-Art memristor-based analog design.
AB - We present a hybrid optical-electrical analog deep learning (DL) accelerator, the first work to use incoherent optical signals for DL workloads. Incoherent optical designs are more attractive than coherent ones as the former can be more easily realized in practice. However, a significant challenge in analog DL accelerators, where multiply-Accumulate operations are dominant, is that there is no known solution to perform accumulation using incoherent optical signals. We overcome this challenge by devising a hybrid approach: Accumulation is done in the electrical domain, while multiplication is performed in the optical domain. The key technology enabler of our design is the transistor laser, which performs electrical-To-optical and optical-To-electrical conversions efficiently to tightly integrate electrical and optical devices into compact circuits. As such, our design fully realizes the ultra high-speed and high-energy-efficiency advantages of analog and optical computing. Our evaluation results using the MNIST benchmark show that our design achieves 2214× and 65× improvements in latency and energy, respectively, compared to a state-of-The-Art memristor-based analog design.
KW - deep learning accelerator
KW - optical computing
UR - http://www.scopus.com/inward/record.url?scp=85109219866&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85109219866&partnerID=8YFLogxK
U2 - 10.1145/3453688.3461531
DO - 10.1145/3453688.3461531
M3 - Conference contribution
AN - SCOPUS:85109219866
T3 - Proceedings of the ACM Great Lakes Symposium on VLSI, GLSVLSI
SP - 271
EP - 276
BT - GLSVLSI 2021 - Proceedings of the 2021 Great Lakes Symposium on VLSI
PB - Association for Computing Machinery
T2 - 31st Great Lakes Symposium on VLSI, GLSVLSI 2021
Y2 - 22 June 2021 through 25 June 2021
ER -