The resurrection of analogue computing

Abstract: Brain-like information processing needs brain-like computing hardware. In recent decade, a range of brain-inspired analogue computing technologies has emerged. These technologies, such as in-memory computing with memristors, are expected to run machine learning tasks with high energy efficiency. It is widely recognized that these brain-inspired hardware gain efficiency through their non-von Neumann architecture, namely collocated processing and memory units. However, digital computing can also adopt non-von Neumann architecture to improve efficiency. Commercial analogue processors that can share the workloads of their digital counterparts are still beyond the horizon. Whether or not analogue computing can meet expectations remains uncertain. Here, we review the emergent analogue computing technologies, and interpret the rationale behind the advancements. Rather than emphasizing only the non-von Neumann architecture, we also stress the significance of space-time continuity in brain-like analogue systems. From a balanced perspective, we discuss the challenges and future directions for analogue computing.

Key words: Artificial intelligence; neuromorphic computing; energy-based models; analogue computing

Cite as: JOSarXiv.202404.0001

Version History