Implementation of Probabilistic Bits (Pbits) using Low Barrier Magnets: Investigation and Analysis

Abstract

Probabilistic spin logic (PSL) is a recently proposed computational paradigm targeting a specific class of problems. Earlier works show the better performance of PSL compared to digital computation in terms of speed and energy efficiency for a few optimization problems such as traveling salesman problem (TSP), max-cut problem, etc. PSL is implemented using probabilistic bits or pbits that fluctuate randomly between two metastable states. One of the popular methods of realizing a pbit is a 3T-1MTJ implementation, with a low barrier magnet (LBM) acting as the free layer of the magnetic tunnel junction (MTJ). In this paper, we perform extensive simulations to examine the suitability of LBMs for 3T-1MTJ pbit implementation. We have analyzed the impact of various material parameters on the attributes of a pbit. Further, we propose a method of selecting material parameters for an LBM-based pbit implementation that improves flips per second (fps), a crucial system-level figure of merit (FOM), without sacrificing the ideal sigmoid-type output response.