Always-On 674μ W@4GOP/s Error Resilient Binary Neural Networks With Aggressive SRAM Voltage Scaling on a 22-nm IoT End-Node

• Published in: IEEE transactions on circuits and systems. I, Regular papers Vol. 67; no. 11; pp. 3905 - 3918
• Main Authors: Mauro, Alfio Di, Conti, Francesco, Schiavone, Pasquale Davide, Rossi, Davide, Benini, Luca
• Format: Journal Article
• Language: English
• Published: New York IEEE 01-11-2020; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: binary neural networks; Bit error rate; Datasets; Hardware; Inference; Insects; IoT; Low voltage; Memory management; near-threshold computing; Neural networks; Prototypes; Random access memory; Resilience; Scaling; Smart sensors; SRAM voltage scaling; Static random access memory; System on chip; ultra-low power
• ISSN: 1549-8328; 1558-0806
• DOI: 10.1109/TCSI.2020.3012576

Binary Neural Networks (BNNs) have been shown to be robust to random bit-level noise, making aggressive voltage scaling attractive as a power-saving technique for both logic and SRAMs. In this work, we introduce the first fully programmable IoT end-node system-on-chip (SoC) capable of executing software-defined, hardware-accelerated BNNs at ultra-low voltage. Our SoC exploits a hybrid memory scheme where error-vulnerable SRAMs are complemented by reliable standard-cell memories to safely store critical data under aggressive voltage scaling. On a prototype in 22nm FDX technology, we demonstrate that both the logic and SRAM voltage can be dropped to 0.5V without any accuracy penalty on a BNN trained for the CIFAR-10 dataset, improving energy efficiency by 2.2X w.r.t. nominal conditions. Furthermore, we show that the supply voltage can be dropped to 0.42V (50% of nominal) while keeping more than 99% of the nominal accuracy (with a bit error rate ~1/1000). In this operating point, our prototype performs 4Gop/s (15.4 Inference/s on the CIFAR-10 dataset) by computing up to 13 binary ops per pJ, achieving 22.8 Inference/s/mW while keeping within a peak power envelope of 674uW - low enough to enable always-on operation in ultra-low power smart cameras, long-lifetime environmental sensors, and insect-sized pico-drones.