Sub-Project 6

AI-Based Neuromorphic Co-Processor for Cognitive Rehabilitation

OVERVIEW

This sub-project integrates the scientific insights from Sub-Project 1 with the hardware developed in Sub-Projects 4 and 5 to build the Brain Co-Processor itself: a closed-loop, AI-powered system that reads neural signals, decodes cognitive states in real time, and delivers targeted neurofeedback or stimulation to enhance attention, working memory, and motor recovery. The system is designed to be deployable at the patient’s bedside or at home — not just in a laboratory.

PRELIMINARY RESULTS

We have already demonstrated the core concept. Using an SSVEP-based cognitive brain-machine interface (cBMI), participants successfully learned to regulate their own attentional brain states through real-time neurofeedback. A single training session produced significant and generalized improvements in attentional accuracy across both trained and untrained hemifields — and machine learning classifiers trained on post-neurofeedback neural responses showed significantly improved decoding of attended locations.

RESEARCH GOALS

Develop and scale AI/ML-based real-time decoding of attention and working memory from EEG

Build a compact, portable ML-enabled BCI system using off-the-shelf components for clinical deployment

Design and fabricate a high-density neuromorphic System-on-Chip (SoC) supporting 256+ channels at low power

Develop a modular chiplet architecture enabling ultra-high-density (1,024+ channel) neural recording with on-chip decoding

THE FOUR DEVELOPMENT PHASES

  • Phase 1 (COTS-based): A portable, 32-channel ML-enabled BCI board for real-time cognitive neurofeedback
  • Phase 2 (Neuromorphic SoC): A 256-channel sparsity-driven neuromorphic co-processor with on-chip AI inference
  • Phase 3 (Chiplet Architecture): Modular tiling of front-end chiplets for 1,024+ simultaneous recording channels
  • Phase 4 (Clinical Integration): Full closed-loop system validated in stroke and healthy-participant studies

INFRASTRUCTURE

  • EEG acquisition and real-time neurofeedback platform
  • High-precision eye-tracking for covert attention monitoring
  • FPGA-based ML co-processor prototyping environment
  • TSMC tape-out facilities for SoC fabrication
  • Chiplet integration and characterization hardware

CLINICAL PARTNERS

  • Amrita Advanced Centre for Epilepsy (Kochi)
  • MS Ramaiah Memorial Hospitals (Bengaluru)
  • Sree Chitra Tirunal Institute (Thiruvananthapuram)
  • Deenanath Mangeshkar Hospital (Pune)

EXPECTED OUTPUTS

  • Low-density COTS-based BCI board validated in healthy and stroke participants
  • High-density neuromorphic frontend chip and ML co-processor SoC
  • Integrated chiplet architecture for ultra-high-density neural recording
  • Comprehensive cognitive rehabilitation platform validated across EEG, ECoG, and intracranial modalities