Spiking Neural Network Based on Memory Capacitors and Metal-Oxide Thin-Film Transistors

Combining its ultra-low leakage current and coupling of the input and weight signals through its threshold voltage modulation mechanism, dual-gate thin-film transistors have been deployed as artificial synapses in an artificial neural network (ANN) employing an array of capacitors as memory elements. Spiking neural network (SNN), a more biomimetic flavor of an ANN, works with information coded in spike signals and demands the implementation of synaptic spike-timing-dependent plasticity and post-synaptic lossy charge-accumulation mechanisms. Incorporating both, a 4×4 SNN is implemented and its application to the classification of a set of 2×2 patterns is demonstrated.