IoT & Embedded

last updated 2026-05-04

Physics / mechanism

IoT and embedded systems span microcontrollers (MCUs), sensors, edge processors, and the connectivity stacks linking them to cloud or fog compute. Core compute is 32-bit ARM Cortex-M or RISC-V cores running at 10–500 MHz, with power envelopes from sub-µW (energy harvesting nodes) to ~100 mW (edge ML inference). Flash/SRAM integration, analog front-ends, and mixed-signal IP define differentiation. Connectivity layers—BLE 5.x, 802.15.4/Thread, LoRaWAN, NB-IoT—add protocol overhead. State-of-the-art: TSMC 22nm/GF 22FDX MCUs achieving <10 µA deep-sleep, on-device ML inference via NPU tiles (e.g., Ambiq Apollo, Nordic nRF54), and MEMS sensor fusion at <1 mW. Security (TrustZone, PSA Certified) is now table-stakes.

Competitive landscape

Key axis of competition is power vs. compute density. Nordic, STM, NXP, and Renesas dominate general-purpose MCUs. Ambiq and Syntiant attack ultra-low-power ML inference. RISC-V cores (SiFive, Andes) threaten ARM licensing economics at the low end.

ApproachPower floorML capability
ARM Cortex-M (STM32, NXP)~1 µA sleepSoft, CMSIS-NN
Ultra-LP custom (Ambiq)~10 nA sleepDedicated NPU
RISC-V MCU (GigaDevice, WCH)~0.5 µA sleepMinimal

Companies using

Connected ideas

Sources

Frontier (open questions)

Frontier questions