Neuromorphic hardware for rapid sparse coding

Abstract

Sparse coding is an important optimization problem in many signal processing applications. A neuromorphic system based on the Locally Competitive Algorithm (LCA) is proposed to solve an overcomplete l1 sparse coding problem. The system includes integrate and fire (IF) neurons and current-based synapses. A network of 18 neurons with 12 inputs is implemented on the RASP 2.9v chip, a Field Programmable Analog Array (FPAA) with directly programmable floating gate elements. The circuit successfully solves the optimization problem, converging to within 4.8% RMS of a digital solver, with an objective cost only 1.7% higher on average. The active circuit consumes 559 μA of current at 2.4V, and converges on solutions in 25 μs.