Neuromorphic computing builds chips that mimic the architecture of biological neural networks — spiking, sparse, adaptive. At Lexcore, we research how these principles can replace brute-force deep learning with something more fundamental: hardware that learns the way life does.
Unlike standard deep learning which processes dense matrices in synchronized batches, neuromorphic systems use sparse, event-driven spikes — exactly how biological neurons communicate. The result: 10–100x less energy, real-time adaptation, and on-device learning without cloud dependency.
Cortina Zero is our proof-of-concept in software. The architecture — k-WTA activation, Hebbian LTP, sparse connectivity — was designed from Phase 1 to be neuromorphically implementable. Our research now moves toward specification and simulation of a dedicated neuromorphic substrate for Cortina intelligence.
"The brain uses 20 watts. A data centre uses 20 megawatts for comparable tasks. The difference is not intelligence — it is architecture."
— Lexcore Neuromorphic Research, 2026k-WTA and Hebbian LTP implemented in software — neuromorphic-ready architecture
212M parameter model with RoPE — software maturity before silicon
Port Cortina Zero core to Intel Loihi 2 neuromorphic SDK, profile efficiency
Publish India's first neuromorphic AI chip specification
Partnership with ISRO/DRDO for sovereign neuromorphic edge deployment
We are seeking partnerships with hardware labs, ISRO affiliates, and semiconductor researchers. India needs sovereign neuromorphic capability.
Read Whitepapers Collaborate