SNN on ARM Edge

Spiking neural networks with continuous on-device learning.

What we're building

A lightweight SNN runtime for ARM edge devices. Models learn continuously after deployment — no separate training and inference phases. Weights adapt in response to real-world input without ever going back to a server.

Why SNN on ARM

Sparse spike events don't map well to GPU SIMD — CPUs are a better fit
ARM dominates edge and mobile, where SNN power efficiency matters most
Continuous learning means the model improves the longer it runs

Pre-training

ARM cloud instances (GCP C4A Axion) for initial weight calibration. Same ISA as target devices, consistent clock without Android scheduler interference.

Edge runtime

Snapdragon 870 and similar. Single-thread optimized. Event-driven spike processing with continuous weight adaptation.