Deep RNN-Oriented Paradigm Shift through BOCANet
Fatemeh Tehranipoor, Nima Karimian, Mehran Mozaffari Kermani, Hamid Mahmoodi · 2019
Logic encryption obfuscation has been used for thwarting counterfeiting, overproduction, and reverse engineering but vulnerable to attacks. However, it was recently shown that satisfiability - checking (SAT) can potentially compromise hardware obfuscation circuits. In this paper, we develop a novel attack called BOCANet that can be beneficial from deep learning architecture to compromise hardware obfuscation circuits's key. Our approach involves exploiting deep recurrent neural network (D-RNN) model, and developing attack model to compromise the obfuscated hardware at least an order-of magnitude more efficiently and under resource-constrained scenarios. In our experiments, the BOCANet approach achieves an average success rate of 100% for 32 bit key size, 93.4% for 64 bit key size, 92.2% and 91.7% for 128 and 256 bit key size, respectively.