Coin Flipping PUF: A Novel PUF with Improved Resistance against Machine Learning Attacks

Abstract

We propose a novel coin-flipping physically unclonable function (CF-PUF) that significantly improves the resistance against machine-learning attacks. The proposed PUF utilizes the strong nonlinearity of the convergence time of bistable rings (BRs) with respect to variations in the threshold voltage. The response is generated based on the instantaneous value of a ring oscillator at the convergence time of the corresponding BR, which is running in parallel. SPICE simulations show that the prediction accuracy of support-vector machine (SVM) on the responses of CF-PUF is around 50 percent, which means that SVM cannot predict better than random guesses.