Wearable EEG inference is constrained not only by model size, but also by the channel-dependent sensing front end that must process high-dimensional inputs before compact downstream computation. We present EEG2MCU, a sensing-to-execution co-design framework that treats electrode count as a hardware design variable through task-aware montage selection. EEG2MCU retains a compact time-frequency representation and generates only the INT8 regions required by each first-layer tile, avoiding full input materialization. Halo reuse and access-aware placement across Flash, SRAM, and tightly coupled memory further reduce redundant generation and data movement. On a Cortex-M7 MCU, EEG2MCU reduces peak SRAM by 11.38x relative to the baseline, including a 1.71x reduction from the 62-to-19 electrode montage, while reducing latency by 2.57x and energy by 2.48x. It retains 94-96% of full-montage accuracy, demonstrating practical high-dimensional EEG inference under tight embedded-resource constraints.
Figures & tables
Fig. 1: EEG2MCU system overview . Offline co-design selects a hardware-feasible 19-electrode montage, while the wearable MCU streams the compact STFT through tiled first-layer execution.
Fig. 2: Per-layer peak memory (top) and MACs (bottom, zero-MAC layers omitted) for SE-ResNet-8 with 62 electrodes, log scale. The first layer dominates both: 1374.4 KiB against the 1024 KiB AXI-SRAM budget, and 80.5% of all MACs, about 21 times the next-largest layer.
Fig. 3: EEG2MCU preprocessing pipeline. Raw multichannel EEG is segmented into 4-s windows, filtered and normalized per channel, and aligned to the 19-electrode deployment montage to form the 19×128×128 model input.
Fig. 4: Evaluated model families. All models share a common first layer that consumes the full multi-electrode input and constitutes the primary MCU feasibility bottleneck, followed by a CNN/Transformer/Mamba backbone and a classification head.
Fig. 5: Standard (left) versus streamed tiled (right) first-layer execution. Per-tile interpolation and quantization replace full FP32/INT8 input materialization with a bounded INT8 tile buffer.
Fig. 6: EEG2MCU first-layer memory hierarchy and row-wise tiling. Weights are copied once from Flash to AXI-SRAM, the im2col workspace is placed in DTCM, and the minimum feasible tile count is selected under the SRAM budget; halo rows are reused across adjacent tiles.
Ch
Tiles
**Hierarchy-Aware Execution
Streamed
Layer 0
Latency (ms)
Energy (mJ)
Peak
Weight Res.
DTCM
Halo
Gen. (ms)
(ms)
RN 8
SE
Eff. *
Mamba *
RN 8
SE
Eff. *
Mamba *
(KiB)
62
1
✗
✗
✗
—
—
—
—
—
—
—
—
—
—
1,374.4
62
2
✗
✗
✗
30.2
4913.6
5492.7
5554.8
5492.0
5629.4
791.9
801.6
788.5
806.1
850.0
62
2
✓
✗
✗
—
—
—
—
—
—
—
—
—
—
1,077.4
62
4
✗
✗
✗
30.2
4924.3
5503.4
5565.5
5502.7
5640.1
795.3
804.9
791.8
809.5
617.5
62
4
✓
✗
✗
30.2
4622.3
5201.4
5263.5
5200.7
5338.1
700.2
709.9
696.8
714.5
826.7
TABLE I: MCU Execution Breakdown of Streamed Generation and Hierarchy-Aware Execution
SE-ResNet8
EfficientFormer [ 32 ]
Mamba [ 33 ]
No. Electrodes
Acc.
F1
Params
Acc.
F1
Params
Acc.
F1
Params
62
0.781
0.782
1.5M
0.769
0.767
3M
0.785
0.772
2.5M
19 (Std 10–20)
0.619
0.618
1.5M
0.625
0.614
3M
0.685
0.668
2.5M
19 (Saliency-guided)
0.734
0.730
1.5M
0.736
0.736
3M
0.737
0.735
2.5M
TABLE II: Model Performance on SEED Across Montage Configurations
Framework
Target
No ext. mem.
On-demand input gen.
Sensing in scope [a]
Arb. kern. geometry [b]
Cache/TCM placement [c]
DORY [ 24 ]
GAP-8 + ext. L3
✗
✗
✗
✓
✗
TinyOps [ 25 ]
M4/M7 + SDRAM
✗
✗
✗
✓
✗
TinyEngine [ 12 ]
Cortex-M
✓
✗
✗
✗
✗
MLonMCU [ 26 ]
Retargetable
✓
✗
✗
✓
✗
EEG2MCU (ours)
Cortex-M7
✓
✓
✓
✓
✓
TABLE III: Comparison with MCU deployment frameworks and prior EEG-on-MCU deployment
Fig. 7: Peak SRAM across cumulative front-end optimizations relative to the 1024 KiB budget, decreasing from 5289.8 KiB for the unfused 62-channel baseline to 464.9 KiB (11.38 × reduction) for the optimized 19-channel design. Callouts show the peak-memory breakdown at each stage.
Fig. 8: MACs/cycle under cumulative SRAM weight residency, DTCM placement, and halo reuse for 62- and 19-channel inputs: (a) layer-0 convolution and (b) end-to-end SE-ResNet-8.
Fig. 9: Montage-aware ResNet-8 scaling at N=4 with all memory optimizations enabled: (a) latency, (b) energy, and (c) peak AXI-SRAM. Versus 62 electrodes, the selected 19-electrode montage reduces these metrics by 2.57× , 2.48× , and 1.71× , respectively (Table I ), while meeting the 4 s window and 1024 KiB AXI-SRAM limits.
The growing use of electroencephalography (EEG) motivates automated analysis that is accurate, transferable, and deployable on constrained hardware. Recent EEG foundation models learn general representations from large-scale pretraining, but their size and computational cost limit edge and wearable deployment. We introduce BRIDGE-EEG, an efficient multi-task EEG classification pipeline that preserves the benefits of pretraining while reducing model size. A unified preprocessing scheme maps heterogeneous recordings with different channel counts, montages, and sampling rates to a device-agnostic 62-channel time--frequency representation. We pretrain an SE-ResNet18 teacher (11.84 M parameters) with SimCLR on unlabeled EEG from five heterogeneous datasets, then compress it into SE-ResNet8 (1.56 M) and SE-ResNet4 (0.48 M) students using task-agnostic and task-specific distillation. We evaluate six benchmarks spanning abnormality detection, motor imagery, and emotion recognition. For abnormality detection and emotion recognition, the students achieve accuracy comparable to or better than several recent EEG foundation models with 10--1,000× more parameters. Motor imagery shows a remaining representation gap, highlighting the importance of pretraining diversity. Inference profiling on a server GPU, desktop CPU, and NVIDIA Jetson Orin Nano shows up to 3.0× lower edge energy per inference (15.64 mJ vs. 46.67 mJ). The compact models further support future deployment on MCU-class wearables.
Meghna Roy Chowdhury, Chengwei Zhou, Haotian Yu +2
Purdue University, Indiana, USA · Case Western University, Ohio, USA
Recent electroencephalography (EEG) foundation models increasingly emphasize model scale, potentially limiting their deployment on resource-constrained devices and in everyday environments. We argue that transferable EEG features can instead be learned with a compact architecture and introduce BrainCodes, a lightweight foundation model pretrained through self-supervised learning on more than 800 hours of recordings from over 1,000 participants. Our architecture combines a quantization-based neural tokenizer with a spatiotemporal encoder that supports arbitrary electrode configurations. The pretraining objective explicitly encourages representations that capture spatial connectivity through channel positional encoding, and sensitivity to narrowband neural oscillations through an auxiliary training objective. Our model matches or surpasses the state of the art across diverse neurophysiological monitoring tasks, including sleep staging, attention assessment, and cognitive load estimation. BrainCodes transfers effectively to low-resource wearable devices while using up to 97% fewer trainable parameters than comparable architectures. Beyond predictive accuracy, our evaluation framework characterizes the geometry and invariances of learned EEG representations across measurable spatial, temporal, and spectral signal properties.
Kleanthis Avramidis, Ivan Tashev, Dimitra Emmanouilidou
Signal Analysis and Interpretation Lab, University of Southern California, Los Angeles, CA 90007, USA · Microsoft Research, Redmond, WA, USA
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 μNPUs 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 (∼370KB with 8-bit quantization). We also profile its component-wise latency and resource consumption on both the MAX78000 and HX6538 WE2 μNPUs, 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.
Imperial College London London, United Kingdom · Nokia Bell Labs London, United Kingdom · Brave Software & University of Cambridge London, United Kingdom