Palimpsest Memories Stored in Memristive Synapses

Abstract

Biological synapses store multiple memories on top of each other in a palimpsest fashion and at different timescales. Palimpsest consolidation is facilitated by the interaction of hidden biochemical processes that govern synaptic efficacy during varying lifetimes. This arrangement allows idle memories to be temporarily overwritten without being forgotten, in favour of new memories utilised in the short-term. While embedded artificial intelligence can greatly benefit from such functionality, a practical demonstration in hardware is still missing. Here, we show how the intrinsic properties of metal-oxide volatile memristors emulate the hidden processes that support biological palimpsest consolidation. Our memristive synapses exhibit an expanded doubled capacity which can protect a consolidated long-term memory while up to hundreds of uncorrelated short-term memories temporarily overwrite it. The synapses can also implement familiarity detection of previously forgotten memories. Crucially, palimpsest operation is achieved automatically and without the need for specialised instructions. We further demonstrate a practical adaptation of this technology in the context of image vision. This showcases the use of emerging memory technologies to efficiently expand the capacity of artificial intelligence hardware towards more generalised learning memories.