US-Japan Workshop: Probabilistic computing hardware for tomorrow’s energy efficient AI accelerators
Wednesday, September 9
Good afternoon, everyone, and welcome to Purdue and the Birck Nanotechnology Center.
Some of you were with us over the past two days for the 10th Annual U.S.–Japan Workshop. For those who were, welcome. And for those joining us today, we’re very pleased to have you here.
The theme of the larger workshop was the importance of the U.S.–Japan partnership in science and technology and the critical, transformative role that AI will play in research and technology development.
This workshop gives us a chance to move from talking about that partnership in broad terms to seeing what it can actually produce.
There is a wonderful history of scientific collaboration between Japan and the United States. And today, with technology increasingly intertwined with economic competitiveness and national security, those collaborations are becoming even more important.
But the best collaborations are not simply agreements between governments or institutions. They begin when individual researchers discover that they can do something together that neither could do as well alone.
We have a particularly good example right here.
Several years ago, Hideo Ohno and his colleagues at Tohoku University joined forces with Supriyo Datta and his colleagues here at Purdue. Tohoku brought extraordinary capabilities in spintronics and magnetic tunnel junctions. Purdue brought new ideas about probabilistic computing and the p-bit.
That collaboration helped demonstrate something quite remarkable — the natural randomness of nanoscale devices — something engineers traditionally worked very hard to eliminate — could instead become a computational resource.
Their work led to a Nature paper in 2019 demonstrating integer factorization using stochastic magnetic tunnel junctions.
And, more importantly, it helped open a much larger question that brings us together today:
We are learning that extraordinary progress in AI comes with an extraordinary appetite for computing — and therefore for energy. Simply scaling today's computing architectures will not carry us very far into the future. And that is exactly what you will be discussing this afternoon.
Can probabilistic computing provide a fundamentally more energy-efficient way to perform some of the computations that will be increasingly important in the AI era?
This workshop brings together people from Tohoku and Purdue, but also Hitachi, Northwestern, NYU, and NIST. Putting the best minds together is how an emerging technology moves forward — from an interesting scientific to a useful technology.
There is much discussion these days about the need to accelerate the pace of innovation — to more quicky move advances in research labs into manufacturing and deployment in real systems. I hope you’ll find time in your panel discussion to discuss that too.
So thank you for being here. I hope today produces some good science and some spirited discussions.
Best wishes for a productive workshop.