NeuroMatrix

A cross-scale validation of an analog spiking neural network for associative memory.

year 2026 status paper with Sabra, Baker, Talarico, Morehead touched this month

Analog spiking neural network in IHP SG13G2 130nm BiCMOS, paired with a C++/Rust/Python simulation framework that validates the architecture across scales — from a handful of neurons to ~25,000.

The same architecture works for sparse associative memory: in simulation, a 993-concept memory operates at 96% of the Hopfield capacity limit with 82% cued recall accuracy. Paper currently under review at MWSCAS 2026.

enter read case study o open paper pdf r open repo