Scaled non-volatile bulk analogue memory for neuromorphic computing

The opportunity to harness the full potential of artificial intelligence rests upon the vast quantities of data and the enormous computing power for data processing. Such computing power come at the cost of ever-increasing demand for electricity, often supplied by fossil sources. With silicon transistors nearing its physical limits, these vast data and computing needs for future artificial intelligence applications may necessitate novel, unconventional computing approaches.

Neuromorphic computing aims to replace the centralized digital computing architecture with a high parallel, distributed, and analogue computing architecture inspired by the human brain. Its core advantage for artificial intelligence comes from the ability to co-locate information processing and information storage on the same element. This results in many orders of magnitude improvements in energy efficiency compared to conventional digital computers, which must constantly fetch data between memory and processor.

This research aims to develop and improve a novel electrochemical resistive memory cell that serves this memory-processor hybrid. This cell is inspired by a battery; however, instead of storing energy, it stores information. This project also trains undergraduate and graduate students in electrochemical and electronic materials through both research and teaching.

It also aims to develop curriculum for secondary school teachers in functional materials through an ASM teacher’s camp.

This research aims to develop an electrochemical cell called bulk resistive random access memory (bulk-RRAM). Conventional resistive random access memory, or RRAM, stores information in a nanosized filament; this nanosized filament is extremely difficult to control owing to the stochastic behavior of a discrete number of defects that control the resistance state.

In contrast, bulk-RRAM utilizes an ion-conducting solid electrolyte to block filament formation and is instead sensitive to the collective behavior of all oxygen vacancy defects in the material’s bulk. This results in predictable and deterministic switching even with greater than 100 separable analogue information states. This research aims to solve the outstanding challenges with bulk-RRAM, including the information retention time, scaling to nanosized dimensions, device variability, and integration into crossbar arrays.

The goal is to determine whether the bulk-RRAM can provide the elusive hybrid memory-logic cell to realize the promises of neuromorphic computing that yields substantial improvements in energy efficiency.

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.