Thomas Defferriere1,Yiyang Li1
University of Michigan1
Thomas Defferriere1,Yiyang Li1
University of Michigan1
Modern machine learning and artificial intelligence workflows consume vast amounts of energy due to the need to move massive quantities of data between memory and processor. Analog, in-memory computing can be much more energy efficient by simultaneously acting as both memory and processor. In-memory computing has been limited by the lack of an analog nonvolatile memory with reliable switching and low-current and voltage operations.<br/>In this work, we show how electrochemical random access memory (ECRAM) can fulfill this needed. Inspired from a battery, ECRAM electrochemically shuttles oxygen vacancy point defects between two transition metal oxides with a solid electrolyte sandwich. The oxygen vacancy concentration is able to deterministically control the analog resistance states, just as how the state of charge of a battery is a reproducible analog value. Whereas earlier ECRAM devices based on lithium ions and protons would rapidly lose state (self-discharge) when scaled to nanosized dimensions, our solid-state, oxygen-based ECRAM is not only nonvolatile at room temperature, but can also switch and retain state at 200C, well above the operating temperature of any solid-state memory. ECRAM’s ability to electrochemically move point defects within solids provide new approaches for both analog and high-temperature memory.