EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Ultra-Low-Power Signal Processing and Edge-AI Hardware for Brain–Computer Interfaces: A Survey from Analog Front-Ends to On-Chip Inference

Tuan-Anh Tran, Thithu-Huyen Tran, Cong-Kha Pham

Electronics · 23.10.2026 · 15 (21) · 4914

DOI: 10.3390/electronics15214914 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Electronics, online erschienen am 23.10.2026. Verfasst von Tuan-Anh Tran und 2 weiteren Autor:innen. Die Arbeit nennt 123 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

The evolution of Brain–Computer Interfaces (BCIs) toward unconstrained, daily life deployment requires a fundamental shift from cloud-tethered processing to low-power, wearable System-on-Chip (SoC) architectures. This survey presents a narrative, full-stack taxonomy of the acquisition-to-inference pipeline, examining the co-design space spanning neural signal acquisition, hardware-friendly feature extraction, and on-chip AI acceleration for edge BCIs. We first evaluate high-fidelity acquisition technologies, detailing low-noise Analog Front-Ends (AFEs) and non-invasive dry-electrode arrays optimized for signal integrity under motion artifacts. Next, we categorize lightweight neural network topologies—including compact Convolutional Neural Networks (CNNs), Spiking Neural Networks (SNNs), and Hyperdimensional Computing (HDC)—tailored for real-time decoding under limited memory constraints. To bridge algorithmic complexity with battery-constrained edge execution, we analyze state-of-the-art model compression methodologies and dedicated hardware acceleration paradigms across FPGA and ASIC platforms. Finally, we highlight the monolithic integration of mixed-signal analog conditioning and high-efficiency inference engines into specialized “EEG-aware” SoCs, offering a quantitative comparison of power efficiency (power per channel), decoding latency, and silicon area. We conclude by identifying hardware–algorithm co-design as the critical path toward autonomous, ultra-low-power neural prosthetics and closed-loop bio-interactive systems.

Zitieren

APA

Tran, T., Tran, T., & Pham, C. (2026). Ultra-Low-Power Signal Processing and Edge-AI Hardware for Brain–Computer Interfaces: A Survey from Analog Front-Ends to On-Chip Inference. Electronics, 15(21), 4914. https://doi.org/10.3390/electronics15214914

Vancouver

Tran T, Tran T, Pham C. Ultra-Low-Power Signal Processing and Edge-AI Hardware for Brain–Computer Interfaces: A Survey from Analog Front-Ends to On-Chip Inference. Electronics. 2026;15(21):4914. doi:10.3390/electronics15214914

BibTeX
@article{Tran2026,
  author = {Tran, Tuan-Anh and Tran, Thithu-Huyen and Pham, Cong-Kha},
  title = {Ultra-Low-Power Signal Processing and Edge-AI Hardware for Brain–Computer Interfaces: A Survey from Analog Front-Ends to On-Chip Inference},
  journal = {Electronics},
  year = {2026},
  volume = {15},
  number = {21},
  pages = {4914},
  doi = {10.3390/electronics15214914}
}
Kostenlos starten →