Dynamic Modulation of Space Charge Resistances in ECRAM Devices

With the rapid rise in the prevalence of artificial intelligence, the energy consumed by training neural networks has also skyrocketed. Electrochemical random-access memory (ECRAM) is a promising technology for brain-inspired, energy efficient training of neural networks. Resistance modulation in ECRAM devices is achieved by electrochemically controlled dynamic intercalation of small cations like H⁺ and Mg²⁺ into the channel material, accompanied by electron doping. In this work we aim to understand the role that channel microstructure and space charge regions play in resistance modulation. Using electrochemical impedance spectroscopy (EIS), we find that for polycrystalline WO₃, a very common channel material for ECRAM, the resistance of undoped films is dominated by space charge resistances originating from grain boundaries. Upon intercalation of small concentrations of cations, the space charge resistance rapidly decreases compared to the bulk resistance, with the bulk resistance dominating the total resistance for higher cation concentrations. The experimental results are supported by an equilibrium electrostatics model of space charge regions in M_xWO₃. This work provides new insights into ECRAM working mechanisms, which could enable the informed design of devices and help them meet the demanding requirements for application in energy-efficient training of neural networks.