Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
The paper introduces PhysioLite, a lightweight neural network architecture tailored for real-time ECG/EMG signal analysis on microcontroller-scale NPUs. PhysioLite achieves comparable performance to state-of-the-art Transformer models while being significantly smaller and compatible with resource-constrained hardware by using learnable wavelet filter banks, CPU-offloaded positional encoding, and hardware-aware layer design. Profiling on MAX78000 and HX6538 WE2 $μ$ NPUs demonstrates the model's feasibility for battery-powered wearable devices.
You can now run near state-of-the-art ECG/EMG analysis on a microcontroller, thanks to a model that's 10x smaller than transformers.
The miniaturisation of neural processing units (NPUs) and other low-power accelerators has enabled their integration into microcontroller-scale wearable hardware, supporting near-real-time, offline, and privacy-preserving inference. Yet physiological signal analysis has remained infeasible on such hardware; recent Transformer-based models show state-of-the-art performance but are prohibitively large for resource- and power-constrained hardware and incompatible with $μ$ NPUs due to their dynamic attention operations. We introduce PhysioLite, a lightweight, NPU-compatible model architecture and training framework for ECG/EMG signal analysis. Using learnable wavelet filter banks, CPU-offloaded positional encoding, and hardware-aware layer design, PhysioLite reaches performance comparable to state-of-the-art Transformer-based foundation models on ECG and EMG benchmarks, while being <10% of the size ($\sim$370KB with 8-bit quantization). We also profile its component-wise latency and resource consumption on both the MAX78000 and HX6538 WE2 $μ$ NPUs, demonstrating its viability for signal analysis on constrained, battery-powered hardware. We release our model(s) and training framework at: https://github.com/j0shmillar/physiolite.