Temporal versatility from intercalation-based neuromorphic devices exhibiting 150 mV non-volatile operation

Abstract

Memristors are a promising technology to surpass the limitations of the current silicon complementary metal-oxide-semiconductor architecture via the realization of neuromorphic computing. Here, we demonstrate intercalation-based non-volatile lithium niobite (Li1 – xNbO2) memristors for highly scalable, efficient, and dense neuromorphic circuitry. Volatile, semi-volatile, and non-volatile operation is achieved using a single material, where each operational mode provides a timescale that enables short-term, medium-term, and long-term memory in conjunction with computation-in-memory. The two-terminal non-volatile devices exhibit conductance changes of up to ∼2000% and have inherent non-binary operations proportional to flux linkage, allowing for analog neuromorphic functions mimicking synaptic weight updates. It is shown that Li1 – xNbO2 devices are highly scalable due to the intercalation-based mechanism, with non-volatile operation requiring a mere 150 mV for a 4 μm2 device, the lowest reported operating voltage for an inorganic non-volatile memristor. The programming voltage scales linearly with device size, projecting millivolt operation and attojoule energy consumption for nanoscale devices.