Memristor Synaptic Array with Spiking Neurons for Neuromorphic Audio Signal Recognition

After successful realization of memristor by HP, it has gained momentum for realization of neuromorphic circuits and systems, since it demonstrates a similar characteristic of biological neural systems. In this work, audio signal processing with a spiking neural network architecture using the memristor as the memory element has been realized. This paper implements an integrated computer-in-memory architecture that uses a memristor crossbar array to reduce latency bandwidth using vector-matrix multiplication. With minimum latency the architecture identifies the new feed audio signal from the mixture of stored signals. Further, a TTFS (Time to First Spike) encoding scheme is realized using an LIF neuron for generating a spike train for enhancing the digitization of input audio signals, and in the output stage, it classifies the voltage levels for audio enhancement. The proposed design achieves 98% accuracy, consuming 0.676mW of power and 22pJ per synapse during audio signal processing.