Programmable electrical coupling between stochastic magnetic tunnel junctions

Abstract

Recent probabilistic computing frameworks have used a microprocessors-in-the-loop approach to sensing output signals and applying control voltages. Here the authors study the dynamics of an analog, two-spin Ising computer built from superparamagnetic tunnel junctions (SMTJs). The op-amp-based circuit features polarity selection and programmable gain, allowing both positive and negative coupling and the performance of simulated annealing. While scaling of this design requires only trivial circuit modifications, large networks of spins would require SMTJs with enhanced properties, suggesting that a co-design approach between devices, architectures, and algorithms is necessary.