Molecular Neuromorphic Hardware
Brain, inspired AI chips: 14, bit precision, 200× less energy consumption than current standard
Contact:
sreetosh@iisc.ac.in
Vision
To build the substrate for AI that learns continuously, locally, and at orders, of, magnitude lower energy than digital hardware.
Problem
AI compute is bottlenecked by energy: today's GPUs spend most of their power moving data between memory and processor, not computing, making AI expensive, slow, and impossible to run on, device.
Solution
Molecular memristor crossbars that fuse memory and computation in one physical medium, delivering 14, bit analog precision with orders, of, magnitude lower energy than digital AI hardware.
Potential
Edge AI silicon is a $50B+ market by 2030. Our platform targets every device that needs to learn locally: phones, drones, satellites, wearables, defence sensors; potentially billions of endpoints.
Target Beneficiaries
AI, hardware OEMs, defence and space agencies, edge, device manufacturers, healthcare wearables firms, and any nation seeking sovereign, low, power AI infrastructure beyond the cloud.
Current Users
Chip, level prototypes running CNN, ANN, LSTM and language model inference and training; validated on James Webb Space Telescope data; in active translation with DRDO and MeitY for 200, mm wafer, scale integration.
