Sub-Project 5

High-Precision Molecular Neuromorphic Accelerator

OVERVIEW

Real-time neural decoding requires extraordinary computational speed at minimal power. Traditional silicon platforms struggle to meet this bar for implantable systems. This sub-project harnesses a breakthrough in molecular electronics — memristor crossbars made from molecular materials — to build a neuromorphic computing platform that performs AI inference with analog precision, ultra-low latency, and power consumption orders of magnitude below conventional approaches.

THE TECHNOLOGY

Molecular memristors are nanoscale electronic devices that can store and process information simultaneously — like the synapses in a biological brain. Organized into crossbar arrays, they can perform vector-matrix multiplication (the core operation of neural networks) in a single step, reducing computational complexity from O(N²) to O(1). Our platform achieves a signal-to-noise ratio of 78 dB — four orders of magnitude better than existing analog platforms.

RESEARCH GOALS

Design and implement a PCB-based neuromorphic accelerator using 64x64 molecular crossbar structures

Demonstrate AI/ML operations including deep learning inference, signal denoising, Fourier and wavelet transforms

Fabricate a compact System-on-Chip (SoC) integrating molecular crossbars with silicon for real-time EEG decoding

Benchmark against state-of-the-art platforms on energy efficiency, SNR, and processing speed

KEY TASKS

Patients perform carefully designed cognitive tasks during their monitoring: visual search and attention tasks, working memory tasks with distractors, economic decision-making, and self-initiated versus externally cued motor tasks. Resting state recordings (30 min) and task-based recordings (up to 60 min per day) are collected across sessions, respecting patient readiness and comfort.

INFRASTRUCTURE

  • Molecular device fabrication facilities at IISc Centre for Nanoscience and Engineering (CeNSE)
  • Lock-in amplifier and precision measurement setup for device characterization
  • TSMC tape-out access for SoC fabrication
  • EEG signal processing and AI inference validation platform

CLINICAL PARTNERS

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

EXPECTED OUTPUTS

  • World-leading analog AI computation platform with 78 dB SNR
  • Integrated SoC capable of real-time EEG and brain-signal decoding
  • Publication in high-impact journals demonstrating molecular neuromorphic computing for neurofeedback