Accelerated neuromorphic computing with site specific joule heating of ECRAM arrays

Electrochemical random access memory (ECRAM) devices are attractive for their linear, symmetric, and area-scalable switching behavior, useful for artificial synapses in neuromorphic computing. Large scale, network-level reconfigurability of ECRAM has been limited by a lack of knowledge of the underlying thermophysical properties of their constituent materials. Here we use finite-element and numeric simulations with constraints matching typical ECRAM materials to elucidate the electrothermal interactions of ECRAM devices coupled with microheaters.

Writing information in ECRAM occurs at elevated temperatures. Microheaters address two key issues in densely networked memory systems: write speed and write-disturb. With nearly a million times less thermal mass than convectional full-chip ambient temperature control, microheaters enable highly increased write speeds. Microheaters also minimize write disturb as targeted heating of individual devices eliminates the excess exposure of neighboring cells to elevated temperatures. We fabricated microheaters guided by our simulation framework and confirmed localization and rapid temperature changes. The results offer a simulation platform to design microheaters for ECRAM and other devices that use thermal stimulation for storing information.