[22a-E102-1~8] FS.1 Focused Session "AI Electronics"

[22a-E102-1] [Highlight]Floating Gate Technology-based Stochastic Neurons for Spiking Neural Network Hardware

〇Akira Goda¹, Chihiro Matsui¹, Ken Takeuchi¹ (1.Univ. Tokyo)

Keywords:Spiking Neural Network, Floating gate memory, Stochastic neuron

Non-volatile memory technologies can be utilized for leaky integrate fire (LIF) circuits in spiking neural network hardware to realize energy and circuit area savings. An analytical model for floating gate technology-based neurons has been developed by considering the Fowler Nordheim tunneling and direct tunneling mechanisms as well as the electron injection stochasticity originated from the quantum effects of the tunneling process. The model successfully demonstrated LIF operations and stochasticity of inter-spike-intervals . The tunability of LIF and stochastic characteristics in a wide range has been shown by engineering the device parameters and operational conditions.