TinyML Hits the Memory Wall: A Benchmark of Ten Sub-Watt Edge Devices

Neural network inference at the edge must fit tight energy budgets. Existing benchmarks compare few devices, do not report energy on a common basis, and omit the newest microcontrollers with an NPU or the 1 GHz Cortex-M85. Hardware-aware design also ranks networks by MAC count, yet no study has tested whether this proxy holds across accelerators. We benchmark sub-watt edge AI hardware by evaluating four CNNs (MobileNetV3, MCUNetV1, EfficientNet, SqueezeNet) and two tiny transformers (MobileViT-XXS, DeiT-tiny) on ten commercial devices: an in-sensor DSP (Sony IMX500), four MCUs with an NPU (STM32N6, Alif Ensemble E7/E8, GreenWaves GAP9), and five MCUs without an NPU (STM32U5, Apollo510B, ESP32-S3 Xtensa LX7, RA8M2, STM32V8). Four devices are novel (Alif E8, Apollo510B, RA8M2, STM32V8). Measured as energy above idle, the IMX500 (0.13-0.35 mJ) uses 1.3-3x less energy than the STM32N6 (0.31-1.05 mJ) on every model; the GAP9 uses the least on MobileNetV3 and the Alif E7 the least on MCUNetV1 and SqueezeNet. All four use 6-34x less energy than the STM32V8 (1.85-3.98 mJ), and the Alif E7 and IMX500 form the power-performance Pareto front. Latency on all four accelerators scales with memory traffic (bytes of weights and activations, r >= 0.98) and not with MACs, so they are memory-bound, while no MCU without an NPU is. The NPU advantage grows with intensity from 1.4x over a 1 GHz Cortex-M85 on MCUNetV1 to 8.2x on SqueezeNet. MAC count is therefore a misleading cost proxy for sub-watt accelerators.