Reconfigurable Ambipolar Non-Volatile Transistors for Fuzzy Logic‐Enabled Edge Computing

Abstract

Edge computing can reduce bandwidth requirements and enable superior response speeds. Fuzzy logic offers significant potential for complex system control and modeling at the edge, owing to its proficiency in handling uncertain information. However, CMOS-based fuzzy analog hardware faces bottlenecks such as structural complexity, high power consumption, and limited reconfigurability. Here, we propose a non-volatile field-effect transistor based on CuInP₂S₆/hBN/Al₂O₃-encapsulated BP/HfO₂ heterostructure, which effectively mitigates the poor polarization performance inherent to CuInP₂S₆ materials. By modulating the channel carrier distribution via the polarization gate voltage, the device realizes linear tunability of the threshold voltage and reconfigurability of ambipolar electrical characteristics. The transistor is integrated into the fuzzy layer of a convolutional neural network (CNN). Experimental results demonstrate that in edge detection and processing tasks for pneumonia CT images, the proposed scheme improves the recognition accuracy by 10% compared with traditional CNNs. Symbol recognition in space-ground optical communication scenarios further validates the effectiveness of the proposed reconfigurable hardware fuzzy neural network. This work replaces complex traditional circuits with a single device, reducing the component count to less than one-tenth of that required by conventional CMOS circuits, thus significantly lowering circuit complexity and enhancing system energy efficiency.