Scalable transition metal dichalcogenide memtransistor arrays with Schottky-barrier control for energy-efficient artificial neural networks.

Abstract

Memtransistors that integrate memristor and transistor functionalities are promising candidates for scalable, energy-efficient neuromorphic computing. However, achieving high performance in memtransistor arrays-particularly in terms of resistive switching ratio, uniformity, and scalability-remains a significant challenge for practical deployment in artificial neural networks. Here, we present scalable memtransistor arrays based on transition metal dichalcogenides (TMDCs), where the Schottky barrier is precisely controlled by modulating vacancy distribution and migration behavior. This approach enables a substantial improvement in the resistive switching ratio, reaching 10⁵ through gate modulation. The device-to-device variation is maintained below 6.8%, and the power consumption is as low as 1 pJ per operation. These devices demonstrate high performance in artificial neural network applications, achieving greater than 98% accuracy in image recognition tasks. Furthermore, the devices exhibit remarkable scalability, with a cell size as small as 4.65 F², and can be further miniaturized by adjusting the channel size without affecting the switching performance. This work highlights the potential of TMDC-based memtransistor arrays for energy-efficient, high-performance artificial neural networks, offering a scalable solution for next-generation neuromorphic computing hardware.