- 學位論文
植入組合硬體木馬與使用節點特徵之機器學習檢測木馬方法
Implanting combinational hardware Trojan and using node features for machine learning Trojan detection
摘要
在迅速發展的半導體領域，積體電路（IC）的製造已經變得高度專業分工，使得不受信任的第三方智慧財產權（3PIP）供應商有機會在IC中植入硬體木馬（Hardware Trojan, HT），這對IC的安全性帶來了嚴厲挑戰。因此，確保IC安全性和品質控制成為半導體產業極為關切的問題。硬體木馬的植入，通常會影響電路節點的結構。因此我們使用電路節點的邏輯機率和電路的測試圖樣建立硬體木馬組合，對組合電路進行植入形成硬體木馬組合電路，並提出基於電路節點結構資訊的21個特徵以及節點分類分群方式，接著，利用參數調校找出最佳參數之XGBoost分類器，最後，使用訓練完成的XGBoost分類器，辨識測試電路的可疑節點，實驗結果顯示所提出的特徵可以有效地幫助識別植入硬體木馬的電路， 分群方法A和方法B在測試電路的真陽性(TPR)與真陰性(TNR)都達到99.9%，方法C則是能有效辨識HT，並且在測試電路的準確率(ACR)達到99.92%。
並列摘要
In the rapidly developing semiconductor industry, manufacturing of integrated circuits (ICs) has become highly specialized, which provides untrustworthy third-party intellectual property (3PIP) vendors the opportunity to implant Hardware Trojans (HT) in ICs. This presents a significant challenge to the security of ICs. Therefore, ensuring the safety and quality control of ICs has become a matter of great concern to the semiconductor industry. Implantation of Hardware Trojans usually affects the structure of circuit nodes. Therefore, we utilize logic probabilities of circuit nodes and test patterns of combinational circuits to implant combinational Hardware Trojan into the combinational circuits. We propose a set of 21 features based on structural information of circuit nodes and three kinds of node classification methods. Subsequently, we use parameter tuning to find the optimal parameters for an XGBoost classifier. Finally, we identify suspicious nodes in test circuits using the trained XGBoost classifier. Experimental results indicate that the proposed features can effectively help recognize circuits with HT implants. For the testing circuits, node classification methods A and B achieve 99.9% for True Positive Rate (TPR) and True Negative Rate (TNR). At the same time, method C effectively identifies Hardware Trojans (HT) and achieves an Accuracy Rate (ACR) of 99.92% for the testing circuits.