SCANN: Side Channel Analysis of Spiking Neural Networks

Spiking neural networks (SNNs) are quickly gaining traction as a viable alternative to deep (DNNs). Compared DNNs, SNNs computationally more powerful and energy efficient. The design metrics (synaptic weights, membrane threshold, etc.) chosen for such SNN architectures often proprietary constitute confidential intellectual property (IP). Our study indicates that implemented using conventional analog neurons susceptible side channel attack (SCA). Unlike the SCAs aimed leak private keys from cryptographic implementations, SCANN (SCA̲ of spiking n̲eural n̲etworks) can reveal sensitive IP within through power channel. We demonstrate eight unique attacks by taking common neuron (axon hillock neuron) test case. chose this particular model since it is biologically plausible hence good fit SNNs. Simulation results indicate different synaptic neurons/layer, thresholds, capacitor sizes (which building blocks SNN) yield distinct spike timing signatures, making them vulnerable SCA. show an adversary use templates (using foundry-calibrated simulations or fabricating known parameters in chips) analysis identify specifications SNN.