[18p-G203-1~18] 13.5 Semiconductor devices and related technologies

▼ [18p-G203-13] Impact of solid-state memristor variability on perceptron supervised learning via STDP

〇Radu M Berdan¹, Takao Marukame¹, Yoshifumi Nishi¹ (1.Toshiba RDC)

Keywords:memristor, neuromorphic engineering, unsupervised learning

Neuromorphic engineering offers great promise in creating compact and power efficient non-von-Neumann computing systems capable of brain-like computation. Memristors are non-volatile memory elements which aid in achieving this goal, due to their nano-scale size and dynamics akin to biological synapses. In this paper we explore the impact of practical device variability so often reported in literature on the performance of large scale artificial neural networks trained via Spike-Timing Dependent Plasticity (STDP) on the task of hand-written digit recognition.