Structural Transformation of 2D InSe Toward Ultrafast and Energy-Efficient Non-Volatile Memristive Switching

2D materials provide a versatile platform for developing memristor-based in-memory computing systems, with potential to address some limitations of conventional von Neumann architectures. However, meeting the stringent requirements for precision, stability, and energy efficiency in neural network hardware remains a challenge due to the intrinsic properties of many 2D materials. In this work, a controllable ultraviolet ozone (UVO) treatment is introduced to engineer the properties of 2D indium selenide (InSe) by introducing a tailored combination of defects, amorphous regions, and oxidized phases. This modification improves the structural stability and vertical conductivity of InSe, and promotes ion migration, enabling a transition from non-switching to stable nonvolatile resistive switching (RS) behavior. The resulting memristors exhibit uniform RS characteristics, with low variability in switching voltage (5.8%) and a tunable on/off ratio ranging from 10² to 10⁵. In addition, the devices demonstrate sub-20 ns switching speeds and can emulate artificial neural networks (ANNs) with recognition accuracy comparable to software-based implementations. Hardware-based convolutional image processing with improved power efficiency is further demonstrated, underscoring the potential of UVO-InSe memristors for energy-efficient neuromorphic computing applications.