Memristor Crossbar Scaling Limits and the Implementation of a Large
Neural Network Using 3D Stacked Crossbars
Abstract
Memristor crossbar-based neural networks perform parallel operations in
the analog domain. Ex-situ training approach needs to program the
predetermined resistance values to the memristor crossbar. Because of
the stochasticity of the memristor devices, programming a memristor
needs to read the device resistance value iteratively. Reading a single
memristor in a crossbar (without an isolation transistor) is challenging
due to the sneak path current. Programming a memristor in a crossbar to
either the RON or ROFF
state is relatively straight-forward. A neural network implemented using
higher precision weights provides higher classification accuracy
compared to a Ternary Neural Network (TNN). This paper demonstrates an
implementation of memristor-based neural networks using only the two
resistance values ( RON, ROFF).
We have examined the impact of the device RON/
ROFF ratio, driver size on the scaling of the
memristor-based neural network circuit. We have implemented a large
neural network using multiple smaller 3D stacked crossbar arrays. We
also have proposed novel neuron circuits to achieve higher weight
precision. Our experimental result shows that the proposed higher
precision synapses are easy to program and provide better classification
accuracy compared to a TNN.