This project proposes to investigate the vulnerability of analog neural network integrated circuits, to side-channel attacks seeking to exfiltrate the model implemented by these networks. Considering the proprietary nature of datasets that are used to train these models, it is imperative to obtain an understanding of their vulnerability to model reverse engineering attacks so that appropriate countermeasures can be eventually developed. In this work we will leverage an existing platform for analog neural network experimentation [1] which the co-PI designed and implemented in the past, to develop and assess the effectiveness of various side channel-based model exfiltration attacks. Building on the co-PI’s experience and infrastructure for testing analog/mixed-signal integrated circuits, as well as the PI’s expertise and infrastructure for side channel attacks using mixed signal as well as optical/laser measurements [2], we will develop an array of such attacks specifically geared toward obtaining the weights that are stored in the synapses and the neurons of a trained analog neural network IC in the form of charge trapped in analog floating gate transistors.