Phosphosilicate Glass-Based Nanosecond Protonic Programmable Resistors for Analog Deep Learning

Abstract

Programmable resistors are widely explored devices as building blocks for analog deep learning accelerators. In this work, we demonstrate Si-compatible nanoscale protonic devices based on phosphosilicate glass (PSG) electrolyte. The conductance modulation characteristics of our devices are ideal in terms of fast operation (5 ns/pulse), high energy efficiency (~fJ/pulse), large dynamic conductance range, linearity, symmetry, reversibility, retention, and high endurance. This paper summarizes the fabrication and characteristics of these devices with particular attention to the proton dynamics during fast operation.