A 28nm AI Microcontroller with Embedded Flash Memory for Edge Computing

Explore a groundbreaking 15-minute conference talk that unveils a revolutionary AI microcontroller designed to overcome the memory bottlenecks that truly limit Edge AI performance. Discover how Anaflash has engineered an innovative solution featuring zero-standby, power-efficient memory with 4-bit per-cell embedded flash technology seamlessly integrated with computation resources on a single chip. Learn about the Near Memory Computing Unit (NMCU) architecture that eliminates external memory dependencies through wide I/O interfaces, enabling immediate data access after booting or waking from deep sleep - a critical advancement for battery-powered devices. Examine the sophisticated three-part NMCU design including control logic for weight management, 16 processing elements with high-bandwidth weight sharing, and quantization blocks for efficient result conversion. Understand how this technology, fabricated using Samsung Foundry's 28nm standard logic process in a compact 4 by 4.5 mm² die, achieves over 95% accuracy on MNIST datasets and 0.878 AUC on Deep Auto Encoder models while matching software baseline performance. Gain insights into the overstress-free waterline driver circuit that extends flash cell margins for enhanced reliability, and discover how this embedded flash microcontroller represents the future of intelligent edge computing by uniting memory and computation for unprecedented efficiency, performance, and cost-effectiveness in smart device applications.