The physics and promise of phase change memory in non-von Neumann computing and machine learning

Abstract

As artificial intelligence (AI) and machine learning (ML) push the limits of conventional computing architectures, the need for high-speed, energy-efficient, and scalable memory technologies has become paramount. Phase change memory (PCM) is a transformative non-volatile memory solution, offering nanosecond-speed switching, multilevel storage, and inherent compatibility with neuromorphic and non-von Neumann computing paradigms. This review comprehensively examines PCM's underlying physics, material innovations, and operational mechanisms from crystallization kinetics and thermal dynamics to resistance drift and threshold switching enabling its unique capabilities. We further explore the interplay between nanoscale material properties and macroscopic device performance, emphasizing advances in deposition techniques and characterization methods. Critical challenges in reliability (endurance and data retention) and scalability are addressed alongside emerging material discoveries that could redefine PCM's role in AI hardware. By bridging fundamental science with cutting-edge applications, this work underscores PCM's potential to revolutionize next-generation computing, from edge AI to brain-inspired systems, paving the way for a post-von Neumann era.