Energy efficient iontronic neuromorphic computing

In our most recent publication in Faraday Discussions we simulate a simple fluidic reservoir computing network and show that its power efficiency could potentially offer an improvement of multiple orders of magnitude in power efficiency when analysing time series inputs in real-time. The dynamics of iontronic devices align well with reservoir computing frameworks and intrinsically feature natural timescales, making them interesting candidates for real-time analysis of temporal signals found in natural systems. Crucially, these features scale down to exceedingly small device scales, which therefore require extremely little power to operate.

Over the past few years, a surge of interest has emerged in iontronic neuromorphic computing, yet identifying how to exploit this fundamentally different substrate remains largely an open challenge. While our research focused on a minimal model and a fundamental task, these results represent an important initial suggestion for domains where iontronics could potentially be of relevance. Due to the Faraday Discussions format, I had the great opportunity to then discuss these results, and iontronic's wider challenges in general, with an expert audience. The resulting questions and my responses will be published alongside the paper.