A binarized spiking neural network based on auto-reset LIF neurons and large signal synapses using STT-MTJs

Abstract

A binarized spiking neural network using auto-reset leaky integrate-and-fire (LIF) neurons with a two-transistor and three-magnetic tunnel junction (2T3MTJ) core and large signal synapses with two-transistor and two-magnetic tunnel junction (2T2MTJ) is designed. The network is applied to a classifier of the MNIST handwritten digit dataset with a 784×400 synapse crossbar array. The weights are trained offline by using the spike-timing-dependent plasticity (STDP) learning algorithm and deployed to the spin-transfer torque magnetic tunnel junction (STT-MTJ) resistances in the synapses after being binarized. Its performance is evaluated by HSPICE using the STT-MTJ device model which takes stochastic change in the angle between the two magnetic moments in the free and the pinned layers into consideration. 75% test accuracy is achieved for 1200 patterns with 1ns read and 1ns write operations and 0.23pJ/SOP energy consumption.

A training method for deep neural network inference accelerators with high tolerance for their hardware imperfection

“…4) The concept of IMC, first mentioned by Kautz in 1969, 5) has since inspired the exploration of many engineers in this direction to mitigate the bottleneck. They implemented emerging memory technologies such as ReRAM, 6) FeRAM, 7) MRAM, 8) PCM, 9) and so on to achieve the IMC architecture.…”