A 345mW heterogeneous many-core processor with an intelligent inference engine for robust object recognition

Abstract

A 228GOPS 345 mW heterogeneous many-core processor combines bottom-up and top-down attention for high accuracy. It uses analog-digital mixed-mode neuro-fuzzy inference circuits to realize robust object recognition. Weight perturbation learning and workload-aware voltage/frequency control are adopted. The 5 × 10 mm <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> chip with 96% recognition accuracy contains 4 SIMD vector processing elements and 32 MIMD processors.