Stochastic p-bits for invertible logic
Seminar | November 27 | 4-5 p.m. | 540 Cory Hall
Kerem Camsari, Purdue University
The field of electronics is at a crossroads. For many decades, the field has been driven by the process of continuous miniaturization, as embodied in the celebrated Moore's Law, but it is not clear how long this can continue, in view of the problems of increasing heat dissipation and feature sizes reaching atomic dimensions. Meanwhile, the field of magnetics has made enormous improvements in the past two decades or so, culminating in the commercial development of the spin-transfer-torque MRAM as a memory device. In this talk, I will show that starting from fundamental theories and available experiments, a physics-based, modular circuit approach can be used to make sense of the growing list of emerging materials and phenomena based on spins and magnets. These essential building blocks can be creatively used to understand and evaluate novel device concepts.
Next, I will talk about a new neuromorphic hardware framework that we call Probabilistic Spin Logic (PSL), based on probabilistic bits (p-bits) and p-circuits. Among natural applications for non-Boolean problems and deep learning algorithms, I will show intriguing examples of an enhanced type of (invertible) Boolean logic, where a given p-circuit can be used to compute the inverse of a Boolean relation it is designed to realize, for example a multiplier circuit operates invertibly to factorize a given number. This feature has practically significant implications for modern cryptography algorithms like RSA, that rely on the directional difficulty of a given problem: multiplication is easy, factorization is hard.
I will conclude with a vision for a beyond-Moore electronics and emphasize the need for efficient neuromorphic hardware that can intelligently exploit the emerging materials and phenomena in light of the ongoing revolution in the fields of machine learning and artificial intelligence.
Kerem Y. Camsari is a post-doctoral research associate at the School of Electrical and Computer Engineering at Purdue working with the Supriyo Datta group. His PhD thesis focused on establishing the "Modular Approach to Spintronics", bringing a wide range of physical methods such as the Non-Equilibrium Green's Function (NEGF) method, spin diffusion (Valet-Fert) equations for transport, and LLG for magnet dynamics into a unified circuit framework. His recent work has been on a neuromorphic hardware framework based on probabilistic bits (p-bit) and p-circuits that can efficiently implement the emerging machine learning algorithms in hardware and can also potentially be useful as a hardware solution for many hard problems of computer science. He has published 15 papers in journals and or conferences and has delivered more than 10 invited talks in international conferences and workshops on his work.