Accelerating Probabilistic Computing Hardware using Topology and Symmetry-breaking
Abstract
Emergent computing paradigms, such as probabilistic computing, are envisioned to solve complex problems efficiently through reducedoperation time and reduced power-consumption as compared to the existing classical computing methods. The working principle of stochastic computing is based on probabilistic bits, commonly referred to as p-bit, whose states randomly fluctuate between up and down, and the output state probabilities can be continuously tuned by an input parameter. This is distinct from binary #classical bits# that toggles between two stable high and low voltage levels assigned as up and down states respectively. The interconnected network of correlated and ultrafast-fluctuating p-bits has been envisioned to be a well-suited hardware platform for solving optimization and inference tasks. Since energy efficient optimization problem solving using conventional computing methods is challenging, the use of p-bit can open new avenues for computing tasks such as intrinsic optimization and stochastic inference. A thermally unstable classical nanomagnet has been put forth as an optimal solution to physically realize a p-bit node because the magnetic fluctuations can be used to implement random number generation, which is a key requirement for building networks of p-circuits for probabilistic computing machines. Here, we propose a research program aimed at accelerating probabilistic computing hardware by exploiting topology and symmetry-breaking in a new class of material systems, namely Weyl semimetals (WSM). Specifically, we propose to exploit large and unconventional spin current in WSM to demonstrate an energy efficient magnetization control for p-bit operation. Through the proposed research activities, we will provide the experimental realization of external magnetic-field free and an energy-efficient operation of a magnetic p-bit device driven by topological tilted spin current generated in WSM and demonstrate the working principle of a two-terminal magneticp-bit device enabled by the discovery of a novel magnetoresistance phenomena in WSM by the PIs. Our proposed research will establish the utility of topology and other phenomena in emergent quantum materials for providing new functionalities to realize modular probabilistic computing hardware. If successful, our proposed research will have far-reaching implications to develop hardware for probabilistic computing based on spin-based modular p-bits and is of utmost importance to the ONR to maintain superiority in future computing technologies.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Aug 11, 2023
- Source ID
- N000142312751
Entities
People
- Simranjeet Singh
Organizations
- Carnegie Mellon University
- Office of Naval Research
- United States Navy