Oxygen Plasma–Assisted Bipolar Doping of WSe ₂ for Reinforcement Learning Synaptic Devices

Controllable bipolar doping in two-dimensional (2D) transition metal dichalcogenides (TMDs) remains a major challenge that limits their application in semiconductor information technologies. Realizing both controllable doping and nonvolatile carrier modulation of TMDs via a single fabrication process is crucial for implementing weight and activation regulation in neuromorphic computing. Here, we propose a facile strategy for WSe₂ based on oxygen plasma-induced oxide layer formation and the generation of surface oxygen vacancies within the layer, in which precise regulation of the plasma flow rate enables tunable p-type and n-type doping. High flow rates induce the formation of WO_<i>x</i>, while low flow rates generate oxygen vacancy-rich WO_<i>x</i>, as confirmed by X-ray photoelectron spectroscopy. Kelvin probe force microscopy measurements further reveal surface potential shifts of -99 mV and +66 mV for WSe₂ modified by stoichiometric WO_<i>x</i> and vacancy-rich WO_<i>x</i>, respectively, validating the bipolar doping mechanism. By exploiting the nonvolatile characteristics of oxygen vacancies, this doping strategy is integrated into a neuromorphic platform to fabricate high-performance transistor-type artificial synaptic devices with WO_<i>x</i> serving as the surface-tunable oxide layer. Regulation of the ionization and migration of oxygen vacancies in the WO_<i>x</i> layer via an external electric field endows the device with the capability to mimic key synaptic behaviors, including short-term plasticity/long-term plasticity, long-term potentiation, and long-term depression, and to exhibit 64 stable, programmable conductance states. When implemented in a reinforcement learning framework, the synaptic devices demonstrate adaptive decision-making capabilities. Importantly, a 6-bit in-memory computing model based on the proposed artificial synapse achieves an average dynamic-interaction reward of 5759.06, comparable to 5894.38 for a conventional 32-bit von Neumann system. This work provides an approach to the design of multifunctional hardware that achieves bipolar doping and neuromorphic computing in 2D semiconductors by introducing the same oxygen vacancy defects under different electrical modulation conditions.