Unsupervised Learning in a Ternary SNN Using STDP

This paper proposes a novel implementation of a ternary Spiking Neural Network (SNN) and investigates it using a hierarchical simulation framework. The proposed ternary SNN is trained in an unsupervised manner using the Spike Timing Dependent Plasticity (STDP) learning rule. A ternary neuron is implemented using a Dual-Pocket Tunnel Field effect transistor (DP-TFET). The synapse consists of a Magnetic Tunnel Junction (MTJ) with a Heavy Metal (HM) underlayer, allowing for the adjustment of its conductance by directing a current through the HM layer. Further, we show that a pair of dual-pocket Fully-Depleted Silicon-on-Insulator (FD-SOI) MOSFETs can be utilized to generate a current, which reduces exponentially with increasing duration of firing events between pre- and post-synaptic neurons. This current modulates the synapse’s conductance according to STDP. Furthermore, it is demonstrated that the proposed ternary SNN can be trained to classify digits in the MNIST dataset with an accuracy of 82%, which is better (75%) than that obtained using a binary SNN. Moreover, the runtime required to train the proposed ternary SNN is $8\times $ less than that required for a binary SNN.