Silicon Photonics Computing: Remedy for Energy-draining AI

Dharanidhar Dang, PhD

Assistant Professor in the Department of Electrical & Computer Engineering

University of Texas, San Antonio (UTSA)

9/1/2023

11AM- 12PM CST

Location: SAN PEDRO 1 YOTTA ROOM 430

506 Dolorosa St, San Antonio, TX 78204

Zoom: https://utsa.zoom.us/j/91049981814

Abstract: AI and in particular deep learning compute demand is growing exponentially. In fact, 10 times a year. What makes it worse is, the AI growth is 5 times more than that of Moore’s Law. To give a perspective, training the latest GPT model with the best of GPUs costs 4.6M $ of electricity and it is going to get worse further. Now, to think of fast-decision making with limited power budget, traditional computing systems cannot offer the required energy efficiency. Gladly, silicon photonics with its ultra-low power characteristics, high-speed nature, and large-scale parallelism gives us immense hope to address these challenges.

In this talk, I will demonstrate how I utilize a variety of silicon photonic components along with emerging memory such as resistive memory to design next-generation end-to-end deep learning accelerator which can demonstrate two order or more improvement in energy efficiency and speedup compared to the state-of-the-art. The talk also includes the showcase of a novel photonic backpropagation accelerator which is a first of its kind. Finally, I will end the talk with several future directions to tackle critical problems in the real world.