[20p-B01-1~9] Total bio-mimetic material science for harmonized functions among sensing, processing and action

[20p-B01-5] Capacitor-free leaky-integration for an artificial neuron

〇Isao Inoue¹ (1.AIST)

Keywords:artificial neuron, leaky integrate, field effect transistor

The artificial neuron of spiking neural networks (SNN), in general, is composed of a large capacitor to integrate trains of input pulses from other neurons. The large size of the capacitor prevents the miniaturisation of the artificial neuron for an integrated circuit. Our SrTiO₃ FET with a 2um channel can do the signal integration by itself as long as above 10 sec without an external capacitor. The mechanism will be discussed in this talk. Moreover, we will introduce our research target at present that is to develop an SNN consisting of these kinds of devices, which will realise a platform that can perform real-time learning and inference at the edge without the power-consuming logical operation such as a back-propagation.