Boolean Computation in Single‐Transistor Neuron

• DOI: 10.1002/adma.202409040
• Publication Date: 2024-10-16T05:35:31Z
• Authors: Hanxi Li; Jiayang Hu; Yishu Zhang; Anzhe Chen; Li Lin; Ge Chen; Yance Chen; Jian Chai; Qian He; Hailiang Wang; Shiman Huang; Jiachao Zhou; Yang Xu; Bin Yu
• Keywords: Neuromorphic engineering; XOR gate

Brain neurons exhibit far more sophisticated and powerful information-processing capabilities than the simple integrators commonly modeled in neuromorphic computing. A biological neuron can fact efficiently perform Boolean algebra, including linear nonseparable operations. Traditional logic circuits require a dozen transistors combined as NOT, AND, OR gates to implement XOR. Lacking competency, artificial neural networks multilayered solutions exercise XOR operation. Here, it is shown that single-transistor neuron, harnessing intrinsic ambipolarity of graphene ionic filamentary dynamics, enable situ reconfigurable multiple operations from separable an ultra-compact design. By leveraging spatiotemporal integration inputs, bio-realistic spiking-dependent computation fully realized, rivaling efficiency human brain. Furthermore, soft-XOR-based network via algorithm-hardware co-design, showcasing substantial performance improvement, demonstrated. These results demonstrate how form single transistor, may function platform for findings are anticipated be starting point implementing computations at individual transistor level, leading super-scalable resource-efficient brain-inspired information processing.