Physics-aware analog circuits for edge artificial intelligence
Artificial intelligence (AI) models are a widespread tool employed across a broad range of applications, made possible by advanced architectures, learning algorithms and hardware. A growing consensus in the field holds that the development of highly efficient models can no longer be decoupled from the underlying hardware, making the algorithm-hardware co-development a central research topic and leading to various application-specific integrated circuits (ASICs) tailored to execute specific models efficiently. Neuromorphic computing aims to further improve AI efficiency by drawing inspiration from biological neural systems. This work presents two novel computing primitives, specifically neuron circuits, to advance the state of the art in low-power analog neuromorphic computing for edge applications.