Personal tools

Hardware/software codesign neural decoding algorithm for “neural dust”

From iis-projects

Revision as of 15:51, 7 June 2021 by Liaoj (talk | contribs) (Created page with "thumb|600px === Description === A brain-machine interface (BMI) acquires brain activity and translates the information into actions to control software and...")
(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)
Jump to: navigation, search
ReMote.png

Description

A brain-machine interface (BMI) acquires brain activity and translates the information into actions to control software and hardware such as computers and prostheses. As a potential treatment for many neurological diseases, it has won great attention in academia and industry.

Non-invasive BMIs, mostly based on EEG signals, do not require surgery to implant sensing nodes in the brain. However, they strongly suffer from low spatial resolution and low signal-to-noise ratio, making them often not accurate. Implantable BMIs, on the other hand, offer high resolution and high signal quality, making them necessary for applications where decoding accuracy is crucial.

Minimizing the damage to the brain is one of the primary goals of implantable BMI systems. Most of the existing systems are bulky and wired for communication and power transfer. Wireless, miniaturized, and implantable BMI systems (sometimes called “neural recording dust”) hold the promise of restoring motor function while reducing the damage caused by the implantation [1][2]. However, such system also poses stringent constraints on the power consumption and area.

Researchers have developed mm-scale neural probe [1][2] and efficient algorithms [3][4] to tackle the problem. Our goal is to hardware/software codesign novel deep learning algorithm for neural decoding based on the spiking band power (SBP) information from the mm-scale neural probe.

In this project, the student will: 1. Study prior art 2. Get familiar with the dataset and the system 3. Explore ML algorithms (CNNs, RNNs, SNNs, …) 4. H/S codesign efficient algorithms

Status: Available

Looking for master or semester thesis students
Supervisor: Jiawei Liao, Xiaying Wang

Prerequisites

  • Machine Learning
  • Deep Learning
  • Python
  • VLSI is a plus

Character

  • 20% Literature review
  • 20% Theory
  • 60% Programming

Professor

Prof. Taekwang Jang <tjang@ethz.ch>

Reference

[1] J. Lim et al., “26.9 A 0.19×0.17mm 2 Wireless Neural Recording IC for Motor Prediction with Near-Infrared-Based Power and Data Telemetry,” in 2020 IEEE International Solid- State Circuits Conference - (ISSCC), San Francisco, CA, USA, Feb. 2020, pp. 416–418. doi: 10.1109/ISSCC19947.2020.9063005.

[2] E. Moon et al., “Bridging the ‘Last Millimeter’ Gap of Brain-Machine Interfaces via Near-Infrared Wireless Power Transfer and Data Communications,” ACS Photonics, vol. 8, no. 5, pp. 1430–1438, May 2021, doi: 10.1021/acsphotonics.1c00160.

[3] S. R. Nason et al., “A low-power band of neuronal spiking activity dominated by local single units improves the performance of brain–machine interfaces,” Nat Biomed Eng, vol. 4, no. 10, pp. 973–983, Oct. 2020, doi: 10.1038/s41551-020-0591-0.

[4] A. K. Vaskov et al., “Cortical Decoding of Individual Finger Group Motions Using ReFIT Kalman Filter,” Front. Neurosci., vol. 12, 2018, doi: 10.3389/fnins.2018.00751.

↑ top

Practical Details