Resistance-Restorable Nanofluidic Memristor and Neuromorphic Chip
Ke Liu, Yong Wang, Miao Sun, Jiajia Lu, Deli Shi, Yanbo Xie
Abstract
Resistance drift due to residual ions limits the accuracy of memristor-based neuromorphic computing. Here, we demonstrate nanofluidic memristors based on voltage-driven ion filling within Ångström channels, immersed in asymmetrically concentrated electrolyte solutions. Inspired by the brain's waste clearance, we restore conductance after 20,000 cycles by removing trapped ions, paving the way for endurance enhancement. The devices exhibit hour-long retention and ultralow energy consumption (∼0.2 fJ per spike per channel). By tuning the voltage, frequency, and pH, we emulate short-term synaptic plasticity. Finally, we demonstrated the first 4 × 4 nanofluidic memristor array capable of recognizing mathematical operators. Our work demonstrated that fluidic memristors are promising for energy-efficient, long-retention, and endurance neuromorphic chips.