A 2.86-TOPS/W Current Mirror Cross-Bar-Based Machine-Learning and Physical Unclonable Function Engine For Internet-of-Things Applications

Abstract

Energy-efficient machine-learning and physical unclonable function (PUF) has drawn significant attention for Internet-of-Things (IoT) application in wake-up detection for bandwidth/computation reduction and privacy protection at sensor node or autonomous device. A machine-learning and PUF engine for IoT applications is presented in this paper with a current mirror cross-bar (CMCB) being a shared core circuit for both functions, leading to reduction in overhead area by 48.5×. A novel dimension expansion technique is proposed to increase weight matrix dimension beyond the physically implemented array with small hardware and energy overhead. A signed multiply-accumulation is realized in CMCB with differential current path and 2-phase conversion. The proposed engine achieves an error rate of 6.34% on MNIST digit recognition task with an energy efficiency of 2.86 TOPS/W. The PUF achieves a native bit error rate of 2.3% across corners and extremely low area per challenge response pair (CRP) of 4.17×10 <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">-59</sup> μm <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> /CRP due to exponentially more CRP enabled by ternary input mode.