Turning to Information Theory to Bring In-Memory Computing Into Practice

This paper explores the emerging field of in-memory computing, which has the potential to significantly improve energy efficiency in many signal processing and machine learning applications. In-memory computing systems are impacted by various sources of noise and perturbations that can affect their computation accuracy. Therefore, this paper aims to identify the key challenges to be addressed from an information-theoretic point of view in this area. It first identifies relevant computation structures and noise models from the literature of hardware implementation, and then reviews existing works in information theory and error-correction for in-memory computing. Finally, it identifies key open avenues for establishing information-theoretic foundations of in-memory computing systems, and for providing insightful design rules leading to highly energy-efficient computing systems.