Wan Xiang, Yuan Qiujie, Sun Lianze, Chen Kunfang, Khim Dongyoon, Luo Zhongzhong
College of Integrated Circuit Science and Engineering, Nanjing University of Posts and Telecommunications, Nanjing 210023, China.
College of Electronic and Optical Engineering, Nanjing University of Posts and Telecommunications, Nanjing 210023, China.
Polymers (Basel). 2025 Apr 25;17(9):1178. doi: 10.3390/polym17091178.
This study presented a novel reservoir computing (RC) system based on polymer electrolyte-gated MoS transistors. The proposed transistors operate through lithium ion (Li) intercalation, which induces reversible phase transitions between semiconducting 2H and metallic 1T' phases in MoS films. This mechanism enables dynamic conductance modulation with inherent nonlinearity and fading memory effects, rendering these transistors particularly suitable as reservoir nodes. Our RC implementation leverages time-multiplexed virtual nodes to reduce physical component requirements while maintaining rich temporal dynamics. Testing on a spoken digit recognition task using the NIST TI-46 dataset demonstrated 95.1% accuracy, while chaotic time-series prediction of the Lorenz system achieved a normalized root mean square error as low as 0.04. This work established polymer electrolyte-gated MoS transistors as promising building blocks for efficient RC systems capable of processing complex temporal patterns, offering enhanced scalability, and practical applicability in neuromorphic computation.
本研究提出了一种基于聚合物电解质门控二硫化钼晶体管的新型储层计算（RC）系统。所提出的晶体管通过锂离子嵌入来运行，这会在二硫化钼薄膜中诱导半导体2H相和金属1T'相之间的可逆相变。这种机制能够实现具有固有非线性和衰退记忆效应的动态电导调制，使得这些晶体管特别适合作为储层节点。我们的储层计算实现利用时分复用虚拟节点来减少物理组件需求，同时保持丰富的时间动态特性。使用美国国家标准与技术研究院（NIST）TI-46数据集进行的语音数字识别任务测试显示准确率达到95.1%，而对洛伦兹系统的混沌时间序列预测实现了低至0.04的归一化均方根误差。这项工作确立了聚合物电解质门控二硫化钼晶体管作为高效储层计算系统的有前景的构建模块，该系统能够处理复杂的时间模式，具有增强的可扩展性以及在神经形态计算中的实际适用性。
