Scoping the Landscape of (Extreme) Edge Machine Learning Processors

At the (extreme) edge, machine learning applications for IoT have gained immense popularity, with academic and commercial players providing various hardware solutions. This chapter outlines and organizes this vast and general machine learning hardware solutions field. To achieve this objective, the survey utilizes the three building blocks of ML at the (extreme) edge: application, algorithm, and hardware (architecture and circuit), to categorize different solutions and to understand the interplay between the blocks. The chapter presents qualitative and quantitative survey of the SotA hardware platforms designed for ML accelerations. Several design trends and prospects of embedded ML learning processors are extracted through a systematic comparison of the hardware platforms based on critical performance metrics.