Can spintronics help make cognitive computing more efficient?

Abstract: In this talk, I discuss why computer designers look to the brain for inspiration on how to develop novel ways of computing for cognitive tasks, why spintronic devices, particularly magnetic tunnel junctions, might make valuable contributions to this process, and why success requires considering all levels of the computational stack from devices through the architecture. I illustrate these points in terms of three recent computational platforms my colleagues and I have worked on using magnetic tunnel junctions in different ways. The first example shows experimental results on binary neural networks based on crossbar arrays of magnetic tunnel junctions used as programmable resistors. The second example is a complex-valued Hopfield network that we model using magnetic tunnel junctions as spin-torque oscillators. Finally, I describe a calculation scheme using thermally unstable magnetic tunnel junctions to generate random signals.

There will be light refreshments at the conclusion of the colloquium in the faculty lounge (Wean Hall 7425).