Strained MTJs with latch-based sensing for stochastic computing
Sudipta Bhuin, Ayan K. Biswas, Larry Pileggi · 2017
This paper presents the design and modeling of a strained magneto-tunnel-junction (S-MTJ) for stochastic computing. Due to the inherent half-way deterministic switching offered by an S-MTJ when operated with voltage generated strain, a true random number generator can be implemented due to the random thermal noise that acts on the free layer at room temperature. Using a compact model for the S-MTJ, we demonstrate its applicability for stochastic computing. Since the read operation of a S-MTJ dominates the total power dissipation, we propose a novel dynamic latch-based sense circuit that consumes two orders of magnitude less power than a bias-based sensing scheme.