NORP: A Compact Neuromorphic Olfactory Recognition Processor With On-Chip Hybrid Learning

This work proposes a low-cost neuromorphic olfactory recognition processor for electronic noses. The processor targets a two-layer spiking neural network, enabling on-chip unsupervised spike-timing dependent plasticity (STDP) learning for the feature extraction layer and supervised spike-error dependent plasticity (SEDP) learning for the feature classification layer. An FPGA prototype of the SNN processor was implemented. It attained a comparably high accuracy of 96.25% at a 188 sample/s inference speed on the Twin Gas Sensor Arrays odor dataset.