March 11th, 2026 NYC Quantum Computing In Person Meetup (w/hybrid option)

Title:
Scaling "optical GPUs" using arbitrarily programmable wave propagation

Abstract:
Classical optical computing has recently made waves by promising to efficiently perform linear algebra, including the most computationally expensive operations of advanced artificial intelligence models. I will explain where an "optical GPU" would have fundamental advantages over its electronic counterparts, as well as the persistent pain points that hinder their development. It turns out there is a break-even point in the size of linear operations beyond which optics can be more energy-efficient.

However, one of the main problems is plainly that no one can build optical processors large enough to reach the break-even point. The central question is: How to scale up optical computers in the face of analog error accumulation, optical losses, and their large spatial footprint?

We demonstrated a solution that piggybacks on the massive scale of off-the-shelf electronic chips to bypass these scaling limits. Instead of designing optical circuitry, we have developed a chip that utilizes the millions of pixels of electronic displays to arbitrarily control complex wave propagation. The controlled propagation of light enables larger and more robust optical computations than previously possible.

I will conclude by discussing what new capabilities these advances in programmable optics open up for quantum computing.

Bio:
Martin Stein is a postdoctoral researcher at NTT Research, Inc., based in Logan Wright’s lab at Yale University. He earned his PhD from Cornell University under Peter McMahon, focusing on "Physical Neural Networks" and the development of highly programmable photonic chips.