Phase Change Memory: A Review on Electrical Behavior and Use in Analog In-Memory-Computing (A-IMC) Applications

Abstract

Recent development and progress of Artificial Intelligence (AI) algorithms made clear that this topic is a paradigm shift with respect to the past. High throughput and ability to do complex tasks makes AI a great field of opportunity. This advancement is somehow limited by the physical implementation of the chips that are still bound to the historical von-Neumann Architecture with processing units and memory hardware spatially separated. The way data is bussed and processed needs disruptive innovation, rather than an evolutionary approach, too. In Analog In-Memory Computing (A-IMC) the typical properties of resistance-based memory technologies are used to both store and compute information. This allows for incredibly high parallelism and removes the problems related to the known von-Neumann bottleneck. In the present work, A-IMC networks based on resistive memories and on the Phase Change Memory (PCM) technology, in particular, are extensively discussed. After a first review of the general features of PCM devices, their application to A-IMC is described, aiming at a full description of the current technological scenario.