A 4096-Neuron 1M-Synapse 3.8PJ/SOP Spiking Neural Network with On-Chip STDP Learning and Sparse Weights in 10NM FinFET CMOS

A 4096-neuron, 1M-synapse SNN in 10nm FinFET CMOS achieves a peak throughput of 25.2GSOP/s at 0.9V, peak energy efficiency of 3.8pJ/SOP at 525mV, and $2.3\mu \text{W}$ /neuron operation at 450mV. The SNN skips zero-valued activations for up to $9.4\times$ lower energy. Fine-grained sparse weights reduce memory by up to $16\times$. On-chip STDP trains RBMs to de-noise MNIST digits and to reconstruct natural scene images with RMSE of 0.036. A 50% sparse weight MLP classifies MNIST digits with 97.9% accuracy at $1.7\mu \text{J}$ /classification.