# FEMBA on the Edge: Physiologically-Aware Pre-Training, Quantization, and Deployment of a Bidirectional Mamba EEG Foundation Model on an Ultra-low Power Microcontroller

arXiv: https://arxiv.org/abs/2603.26716 · HTML: https://mlsanity.com/paper/2603.26716/

## Key facts

- Cited by 1 papers in the map (data through 14 Oct 2026).
- Placed in Physiological signal modeling, one of 10 directions in State-space sequence models (S4, Mamba).
- Builds on 32 of its 65 references that are in the map.
- First posted on arXiv 18 Mar 2026.

## Placed in

- State-space sequence models (S4, Mamba) / [Physiological signal modeling](https://mlsanity.com/ssm/physiological-signal-modeling/)

## Builds on (most cited first, 30 of 32)

- [Representation Learning with Contrastive Predictive Coding](https://mlsanity.com/paper/1807.03748/) (2018)
- [Mamba: Linear-Time Sequence Modeling with Selective State Spaces](https://mlsanity.com/paper/2312.00752/) (2023)
- [EEGNet: A Compact Convolutional Network for EEG-based Brain-Computer Interfaces](https://mlsanity.com/paper/1611.08024/) (2016)
- [Jamba: A Hybrid Transformer-Mamba Language Model](https://mlsanity.com/paper/2403.19887/) (2024)
- [Large Brain Model for Learning Generic Representations with Tremendous EEG Data in BCI](https://mlsanity.com/paper/2405.18765/) (2024)
- [BENDR: using transformers and a contrastive self-supervised learning task to learn from massive amounts of EEG data](https://mlsanity.com/paper/2101.12037/) (2021)
- [CBraMod: A Criss-Cross Brain Foundation Model for EEG Decoding](https://mlsanity.com/paper/2412.07236/) (2024)
- [Transformer-based Spatial-Temporal Feature Learning for EEG Decoding](https://mlsanity.com/paper/2106.11170/) (2021)
- [Unsupervised Time-Series Representation Learning with Iterative Bilinear Temporal-Spectral Fusion](https://mlsanity.com/paper/2202.04770/) (2022)
- [BrainBERT: Self-supervised representation learning for intracranial recordings](https://mlsanity.com/paper/2302.14367/) (2023)
- [A Comprehensive Survey on Model Quantization for Deep Neural Networks in Image Classification](https://mlsanity.com/paper/2205.07877/) (2022)
- [EEG-TCNet: An Accurate Temporal Convolutional Network for Embedded Motor-Imagery Brain-Machine Interfaces](https://mlsanity.com/paper/2006.00622/) (2020)
- [Transformer Convolutional Neural Networks for Automated Artifact Detection in Scalp EEG](https://mlsanity.com/paper/2208.02405/) (2022)
- [EEGFormer: Towards Transferable and Interpretable Large-Scale EEG Foundation Model](https://mlsanity.com/paper/2401.10278/) (2024)
- [EEG2Rep: Enhancing Self-supervised EEG Representation Through Informative Masked Inputs](https://mlsanity.com/paper/2402.17772/) (2024)
- [PULP-NN: Accelerating Quantized Neural Networks on Parallel Ultra-Low-Power RISC-V Processors](https://mlsanity.com/paper/1908.11263/) (2019)
- [Data augmentation for learning predictive models on EEG: a systematic comparison](https://mlsanity.com/paper/2206.14483/) (2022)
- [Self-supervised EEG Representation Learning for Automatic Sleep Staging](https://mlsanity.com/paper/2110.15278/) (2021)
- [Modeling Multivariate Biosignals With Graph Neural Networks and Structured State Space Models](https://mlsanity.com/paper/2211.11176/) (2022)
- [MambaQuant: Quantizing the Mamba Family with Variance Aligned Rotation Methods](https://mlsanity.com/paper/2501.13484/) (2025)
- [EEG-GNN: Graph Neural Networks for Classification of Electroencephalogram (EEG) Signals](https://mlsanity.com/paper/2106.09135/) (2021)
- [LUNA: Efficient and Topology-Agnostic Foundation Model for EEG Signal Analysis](https://mlsanity.com/paper/2510.22257/) (2025)
- [CADDA: Class-wise Automatic Differentiable Data Augmentation for EEG Signals](https://mlsanity.com/paper/2106.13695/) (2021)
- [EEGMamba: Bidirectional State Space Model with Mixture of Experts for EEG Multi-task Classification](https://mlsanity.com/paper/2407.20254/) (2024)
- [FEMBA: Efficient and Scalable EEG Analysis with a Bidirectional Mamba Foundation Model](https://mlsanity.com/paper/2502.06438/) (2025)
- [ECG-TCN: Wearable Cardiac Arrhythmia Detection with a Temporal Convolutional Network](https://mlsanity.com/paper/2103.13740/) (2021)
- [CEReBrO: Compact Encoder for Representations of Brain Oscillations Using Efficient Alternating Attention](https://mlsanity.com/paper/2501.10885/) (2025)
- [Addressing Class Imbalance in Classification Problems of Noisy Signals by using Fourier Transform Surrogates](https://mlsanity.com/paper/1806.08675/) (2018)
- [xTern: Energy-Efficient Ternary Neural Network Inference on RISC-V-Based Edge Systems](https://mlsanity.com/paper/2405.19065/) (2024)
- [GAPses: Versatile smart glasses for comfortable and fully-dry acquisition and parallel ultra-low-power processing of EEG and EOG](https://mlsanity.com/paper/2406.07903/) (2024)

## Built on by (most cited first, 1 of 1)

- [TinyCast: Probabilistic Zero-Shot Forecasting with Computed Periodicity](https://mlsanity.com/paper/2608.15767/) (2026)

## Abstract (arXiv)

Objective: To enable continuous, long-term neuro-monitoring on wearable devices by overcoming the computational bottlenecks of Transformer-based Electroencephalography (EEG) foundation models and the quantization challenges inherent to State-Space Models (SSMs). Methods: We present FEMBA, a bidirectional Mamba architecture pre-trained on over 21,000 hours of EEG. We introduce a novel Physiologically-Aware pre-training objective, consisting of a reconstruction with low-pass filtering, to prioritize neural oscillations over high-frequency artifacts. To address the activation outliers common in SSMs, we employ Quantization-Aware Training (QAT) to compress the model to 2-bit weights. The framework is deployed on a parallel ultra-low-power RISC-V microcontroller (GAP9) using a custom double-buffered memory streaming scheme. Results: The proposed low-pass pre-training improves downstream AUROC on TUAB from 0.863 to 0.893 and AUPR from 0.862 to 0.898 compared to the best contrastive baseline. QAT successfully compresses weights with negligible performance loss, whereas standard post-training quantization degrades accuracy by approximately 30%. The embedded implementation achieves deterministic real-time inference (1.70~s per 5~s window) and reduces the memory footprint by 74% (to ≈2~MB), achieving competitive accuracy with up to 27× fewer FLOPs than Transformer benchmarks. Conclusion: FEMBA demonstrates that Mamba-based foundation models can be effectively quantized and deployed on extreme-edge hardware without sacrificing the representation quality required for robust clinical analysis. Significance: This work establishes the first full-stack framework for deploying large-scale EEG foundation models on ultra-low-power wearables, facilitating continuous, SSM based monitoring for epilepsy and sleep disorders.