Stochastic Spiking Neural Networks based on nanoelectronic spin devices canbe a possible pathway at achieving "brain-like" compact and energy-efficientcognitive intelligence. The computational model attempt to exploit theintrinsic device stochasticity of nanoelectronic synaptic or neural componentsto perform learning or inference. However, there has been limited analysis onthe scaling effect of stochastic spin devices and its impact on the operationof such stochastic networks at the system level. This work attempts to explorethe design space and analyze the performance of nano-magnet based stochasticneuromorphic computing architectures for magnets with different barrierheights. We illustrate how the underlying network architecture must be modifiedto account for the random telegraphic switching behavior displayed by magnetswith low barrier heights as they are scaled into the superparamagnetic regime.We perform a device to system level analysis on a deep neural networkarchitecture for a digit recognition problem on the MNIST dataset.
展开▼