Memristor Devices and Systems for Analog and Neuromorphic Computing

Recent breakthroughs in memristive devices have paved the way for unprecedented opportunities in computing, including analog and neuromorphic computing. Analog computing is known for its high energy efficiency and throughput but often suffers from poor precision. To address this, we have investigated the sources of reading noise, successfully mitigating them to achieve an unprecedented 2048 conductance levels in individual memristors. This equates to 11 bits per cell, setting a record precision across various memory types. Recognizing the ongoing demand for single or double precision in numerous applications, we have proposed and developed a novel circuit architecture and programming protocol. This innovation allows analog memories to achieve arbitrarily high precision with minimal circuit overhead. Our experimental validation includes a memristor System-on-Chip fabricated in a standard foundry, demonstrating significant improvements in precision and power efficiency compared to traditional digital systems.<br/>Neuromorphic computing, a paradigm with more bio-inspirations than analog computing, holds the potential to approximate natural intelligence while maintaining efficiency. The nonlinear dynamics of ion migration are crucial for biological intelligent systems, making them highly desirable for artificial neuromorphic computing systems. Certain volatile memristive devices function based on ion dynamics, similar to biological systems, and can emulate biological components like neurons efficiently. We have demonstrated comprehensive neurons exhibiting six key neuronal functions within a footprint no larger than a single transistor and with very low energy consumption.<br/>These demonstrations highlight the transformative potential of memristive devices in computing, transcending historical limitations and ushering in a new era of precision and efficiency.