A novel compound synapse using probabilistic spin-orbit-torque switching for MTJ based deep neural networks

Vaibhav Ostwal,Ramtin Zand,Ronald DeMara,Joerg Appenzeller

DOI: https://doi.org/10.48550/arXiv.1910.00171

2019-10-01

Abstract:Analog electronic non-volatile memories mimicking synaptic operations are being explored for the implementation of neuromorphic computing systems. Compound synapses consisting of ensembles of stochastic binary elements are alternatives to analog memory synapses to achieve multilevel memory operation. Among existing binary memory technologies, magnetic tunneling junction (MTJ) based Magnetic Random Access Memory (MRAM) technology has matured to the point of commercialization. More importantly for this work, stochasticity is natural to the MTJ switching physics e.g devices referred as p-bits which mimic binary stochastic neurons. In this article, we experimentally demonstrate a novel compound synapse that uses stochastic spin-orbit torque (SOT) switching of an ensemble of nano-magnets that are located on one shared spin Hall effect (SHE) material channel, i.e. tantalum. By using a properly chosen pulse scheme, we are able to demonstrate linear potentiation and depression in the synapse, as required for many neuromorphic architectures. In addition to this experimental effort, we also performed circuit simulations on an SOT-MRAM based 784*200*10 deep belief network (DBN) consisting of p-bit based neurons and compound synapses. MNIST pattern recognition was used to evaluate the system performance, and our findings indicate that a significant reduction in recognition error rates can be achieved when using our incremental pulse scheme rather than a non-linear potentiation and depression as obtained when employing identical pulses.

Applied Physics,Disordered Systems and Neural Networks

What problem does this paper attempt to address?

This paper aims to solve the problem of achieving efficient and reliable multi - level synaptic operations in deep neural networks based on magnetic tunnel junctions (MTJ). Specifically, the authors propose a new composite synaptic structure based on spin - orbit torque (SOT) switching to overcome the nonlinear problems in existing technologies and improve the performance of neuromorphic computing systems. Through experiments and circuit simulations, the ability of this new composite synapse to achieve linear enhancement and suppression, as well as its application effect in pattern recognition tasks, are studied. The paper pays special attention to how to improve the linearity of synaptic behavior through an improved pulse scheme, thereby significantly reducing the recognition error rate.