Experimental study demonstrates environmentally stable analog weight modulation in electrochemical random-access memory, highlighting a practical pathway for neuromorphic computing hardware.
Analog electrochemical memories provide ionic control of conductance for neuromorphic computing, but reliable retention of programmed states remains a critical challenge. Here we report a three-terminal Li-based electrochemical random-access memory (Li-ECRAM) consisting of a LiCoO2 channel, a mechanically buffered and moisture protective SiOx ion reservoir and a LiPON interlayer. The device supports 64 distinguishable conductance states and maintains representative programmed states over 103 s at temperatures up to 80 °C and 80% relative humidity. First-principles-informed thermodynamic analysis indicates that LiPON adjusts the interface band alignment and the interfacial electrostatic potential. This interface potential redistribution suppresses lithium migration and parasitic electronic transport, stabilizing the lithium–vacancy profile in LiCoO2. The Li-ECRAM exhibits paired pulse facilitation, excitatory and inhibitory postsynaptic currents, and long-term potentiation and depression with nonlinearity factors of αP = 0.501and αD = 0.826, while keeping cycle-to-cycle variation below 0.4% and device-to-device variation at 0.52%. When the measured conductance updates are mapped into a YOLOv5 fire detection network, the system reaches a mean average precision of 76.7%, or 91.1% of the software performance. These results identify LiPON/SiOx synergistic channel–reservoir interface regulation as a route to environmentally stable Li-ECRAM synapses.
No takes yet. Share an insight, caveat, or question.
Zhang et al. (2026) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: