Analog Performance Locking through Neural Network-Based Biasing.

We introduce a method for protecting analog Integrated Circuits (ICs) against unauthorized use by obfuscating their operating point using an analog neural network. With the model of the trained analog neural network acting as a lock and its inputs as the key, only the correct key combination will unlock the analog IC, by providing it with the required bias conditions to operate within its specification limits. By defining the key combinations in the continuous analog space and by using floating gate transistors to realize the neural network, the proposed method defends itself against efforts to guess the correct key through model approximation attacks. Moreover, by inhibiting retraining of the analog neural network, the proposed solution enables customization of the lock and key combination to each IC. The proposed solution has been implemented in silicon through a proof-of-concept experimental setup comprising a Low-Noise Amplifier (LNA) and a programmable analog neural network. Experimental results demonstrate the effectiveness of the method in preventing unauthorized use of an analog IC.