Using Programmable Graphene Channels as Weights in Spin-Diffusive Neuromorphic Computing

Abstract

A graphene-based spin-diffusive neural network is presented in this paper that takes advantage of the locally tunable spin transport of graphene and the non-volatility of nanomagnets. By using electrostatically gated graphene as spintronic synapses, a weighted summation operation can be performed in the spin domain while the weights can be programmed using circuits in the charge domain. Four-component spin/charge circuit simulations coupled to magnetic dynamics are used to show the feasibility of the neuron-synapse functionality and quantify the analog weighting capability of the graphene under different spin-relaxation mechanisms. This spin-diffusive neural network using a graphene-based synapse design achieves total energy consumption of 0.55–0.97 fJ per cell <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$\cdot $ </tex-math></inline-formula> synapse and attains significantly better scalability compared to its digital counterparts, particularly as the number and bit accuracy of the synapses increases.