Stochastic Spiking Neural Networks Enabled by Magnetic Tunnel Junctions: From Nontelegraphic to Telegraphic Switching Regimes

Artificial neural networks built around nanoelectronic components are a means to realizing compact, energy-efficient cognitive intelligence. The authors use the inherent device physics of nanomagnets to emulate the computational primitives of a neural network, reducing the energy and area requirements of the underlying hardware. They analyze the performance of stochastic neuromorphic computing platforms with magnets of different barrier heights, and show how the core network architecture must be modified as the magnets scale down to the superparamagnetic regime.