Detection of Hardware Trojan using CNN with Residual Network

Due to very high demand of low cost IoTs and smart devices, a lot of company outsources their design for fabrication to third party. This raises a deep security concern of malicious modification of integrated circuits (IC) which is known as Hardware Trojan (HT). By lowering the effective entropy of a random-number generator or revealing information on the internal operation of the processor to an attacker, very simple but difficult-to-detect trojans could be used to weaken the security of a crypto processor. The purpose of side-channel measurements for identifying fraudulent IC tampering has been intensively investigated during the previous decade. This paper proposes a deep convolutional neural network (DCNN) model with residual network for hardware trojan detection, with the goal of high detection accuracy. The accuracy of the model in detecting malicious hardware was then assessed against cutting-edge algorithms. A convolutional neural network model was trained using power traces from an FPGA-based implementation (HT-free) of a symmetric Cryptographic algorithm (AES). In this phase, the convolutional neural network with residual unit has been implemented, and evaluation of the trojan detection indicates a high accuracy, with an average percentage accuracy of 90.48%.