Enhanced GaSe-Based Vertical Memristors with HfO₂ Layer for High-Performance Neuromorphic Computing

In neuromorphic computing, gallium selenide (GaSe)-based memristors have emerged as promising candidates for energy-efficient and high-performance synaptic devices due to their inherent ability to emulate synaptic plasticity. However, the switching ratio (SR) of vertically structured GaSe-based memristors is often limited, restricting their effectiveness in neural network weight mapping. In this study, we introduce an improved GaSe vertical memristor featuring a copper (Cu) top electrode and a gold (Au) bottom electrode. By incorporating a thin hafnium oxide (HfO2) layer between the GaSe material and the bottom electrode, we significantly enhance the SR and expand the number of available conductive states and the conductance modulation range. The ION/OFF ratio rises with increasing HfO2 layer thickness, achieving values of 30, 3.9 × 104, and 5.9 × 105 for thicknesses of 1, 2, and 3 nm, respectively. However, as the HfO2 layer thickness increases from 1 to 2 or 3 nm, the GaSe vertical memristor shifts from analog to digital behavior, resulting in heightened conductance nonlinearity. Comparing the performance of GaSe vertical memristors with varying HfO2 layer thicknesses, the device with a 1 nm HfO2 layer stands out, offering both an improved ION/OFF ratio and an optimized synaptic linearity factor. Furthermore, simulations conducted on the “NeuroSim+” platform demonstrate that a three-layer neural network leveraging this optimized GaSe vertical memristor achieves an impressive handwritten digit recognition accuracy of 93%. This study establishes a solid foundation for the development of high-performance GaSe memristors with improved switching characteristics for future neuromorphic applications.