LiNbO3 dynamic memristors for reservoir computing

In conventional digital computers, data and information are represented in binary form and encoded in the steady states of transistors. They are then processed in a quasi-static way. However, with transistors approaching their physical limits and the von Neumann bottleneck, the rate of improvement in computing efficiency has slowed down. Therefore, drawing inspiration from the dynamic and adaptive properties of biological systems, research in neural morphology computing has great potential for development. Memristors, a class of nonlinear resistors with memory function, naturally embody dynamics through their internal electrical processes. This enables nonconventional computing paradigms with enhanced capability and energy efficiency, such as reservoir computing. In this paper, we propose a volatile memristor made of LiNbO3 with nonlinear I-V characteristics and short-term memory function. This memristor is well-suited to be used as a nonlinear node in the storage layer of reservoir computing. With this system, we can achieve the same functionality as traditional reservoir computing with a single device, instead of a large number of interconnected nodes. The collective states of memristors after the application of trains of pulses to the respective memristors are unique for each combination of pulse patterns. This provides a more reliable and efficient way for subsequent output layer classification processing. The system was successfully used to recognize images of Arabic numerals 0-9. This work not only broadens the application scope of materials such as lithium niobate in neural morphology computing but also provides new ideas for developing more efficient neural morphology devices and systems.