Interfacial and Bulk Switching MoS ₂ Memristors for an All-2D Reservoir Computing Framework

In this study, we report non-filamentary resistance switching with short- and long-term memory dynamics in Au/Ti/MoS 2 /Au memristive devices. The temporal dynamics is engineered by controlling the thickness of the chemical-vapor-deposited (CVD) MoS 2 films. Monolayer (1L)-MoS 2 devices exhibit forming-free and unipolar resistance switching with short-term memory (STM). Bipolar and non-volatile resistance switching with outstanding cycle-to-cycle uniformity is observed for the multilayer (ML) MoS 2 memristive devices. Detailed transport analysis and high-resolution scanning tunneling electron microscopy (HRSTEM) studies correlate the non-volatile switching in ML-MoS 2 with a trap-assisted space charge limited conduction (SCLC) mechanism, leading to a bulk-limited resistance switching behavior. We fabricate a 16 × 16 array of ML-MoS 2 memristors and demonstrate long-term potentiation/depression behavior. The 1L-MoS 2 memristive devices with STM are used to generate four-bit reservoir states. Finally, we simulate a small reservoir computing (RC) framework by using experimental data from 5 reservoir neurons and the 16 × 16 synaptic array. This small RC network achieves 89.56% accuracy in a spoken-digit recognition task, owing to the excellent cycle-to-cycle uniformity in both the reservoir neurons and the synaptic array. Our work demonstrates the possibility of creating neural networks with reliable performance using non-filamentary memristive devices on a single 2D-material platform.