Gate insulator stack engineering for fully CMOS-compatible reservoir computing

The need for processing complex and temporal datasets has increased with the rise of artificial intelligence. In this context, reservoir computing, which utilizes the short-term memory of the reservoir to map input data into a high-dimensional space, has gathered significant interest. In this study, for the first time, fully CMOS-compatible reservoir computing is demonstrated through gate insulator stack engineering. Integrated on a single wafer, CMOS circuits, Al₂O₃/Si₃N₄ (A/N) devices for both reservoir and leaky integrate-and-fire neuron applications, and Al₂O₃/Si₃N₄/SiO₂ (A/N/O) devices as synaptic devices are verified. Furthermore, the influence of various bias conditions on reservoir performance is analyzed. The proposed co-integrated reservoir computing system efficiently handles temporal data, reducing ~ 53% of network resources with only ~ 0.17%p accuracy drop while being robust to device variations.