Enhanced IGZO Neuromorphic Vision via Interface and Film Quality Optimization
摘要
The significance of artificial vision lies in its deep physical-level integration of visual information “acquisition” and “understanding,” representing an advanced frontier and a crucial technology for overcoming current artificial intelligence (AI) computational and energy efficiency bottlenecks as well as taking a solid step in achieving artificial general intelligence and enabling machines to actually “see” and intelligently interact with the world. Substantial efforts are being devoted to the development of materials and structures to advance artificial vision. Among these materials, IGZO has garnered significant attention, leveraging its unique advantage of seamless integration with existing industries. However, limitations in reliability and modulation characteristics remain unresolved because of interface issues, the need for device structure innovations (e.g., three-terminal transistor structures being superior to two-terminal memristors), and the formation of heterojunctions with two-dimensional (2D) materials. In this study, we optimize the interface structure and film-forming quality of IGZO through a carefully designed annealing process. Beyond achieving excellent electrical performance (threshold voltage: 1.05 V, carrier mobility: ~14.9 cm2 V−1 s−1, subthreshold swing: 70.4 mV/decade, paired-pulse facilitation: 142%), we reveal a defect-engineering mechanism linking deep-level oxygen vacancy dynamics to optical synaptic plasticity. Leveraging this mechanism, we demonstrate a system-level optical reservoir computing architecture that reduces training parameters by 98.93%. This work contributes to unlocking the potential of IGZO for highly efficient, low-cost neuromorphic computing systems.