Digit Recognition Through Unsupervised Learning by Lithium Silicate Synapses

The spike timing-dependent plasticity (STDP) of synapses has been shown as a fundamental rule of brain learning. In this work, we demonstrate that lithium silicate (LiSiOx) can be used to achieve STDP by exploiting its resistive switching properties. We illustrate that when performing MNIST database recognition the benefit by HRS(high resistance state) variance is utilized. Considering the widely variant HRS is a common phenomenon that many resistors will meet used as synapses, it should be a general applicability in the future development of neuromorphic computing.