Programmable neuronal-synaptic transistors based on 2D MXene for a high-efficiency neuromorphic hardware network

Developing a high-efficiency neuromorphic hardware network is essential to achieve complex artificial intelligence. Here, for the first time, we propose a multi-neuromorphic functional device based on the 2D material Mxene—a switchable neuronal-synaptic transistor (SNST) that can be programmed to realize synaptic or neuronal function—and break the boundaries between neuronal and synaptic modules for a high-efficiency neuromorphic network, including fabrication efficiency, resource utilization efficiency, and operational efficiency. A neural network composed of multiple SNSTs is designed for authenticity data recognition that can equitably redistribute the neuromorphic hardware sources and adjust the topological structure of the hardware network, improving operational speed and reducing the number of devices in the network. Finally, an SNST-based hardware system is developed for facial recognition with a recognition accuracy of about 80%. This work demonstrates that programmable SNST-based neuromorphic chips with a simple fabrication process, equitable resource utilization, and high operational speed have great potential for highly efficient and accurate neuromorphic networks.