Single Neuromorphic Memristor Closely Emulates Multiple Synaptic Mechanisms For Energy Efficient Neural Networks

This article discusses the use of a modified STPN unit to estimate energy consumption in multi-functional hardware synapses for deep neural networks. The unit is integrated into a network simulator and tested on the Atari Pong game, showing faster and more stable training compared to the original version. The study also compares energy consumption between GPU and memristive synapses, demonstrating a significant gain in efficiency with the latter.