1School of Electrical Engineering & Computer Sciences, National University of ... and Computer Engineering, Tennessee Tech University, Cookeville, TN, USA.
Power Profiling of Microcontroller’s Instruction Set for Runtime Hardware Trojans Detection without Golden Circuit Models Faiq Khalid Lodhi1, Syed Rafay Hasan2, Osman Hasan1, Falah Awwad3 1School of Electrical Engineering & Computer Sciences, National University of Sciences & Technology, Islamabad, Pakistan 2Department of Electrical and Computer Engineering, Tennessee Tech University, Cookeville, TN, USA 3College of Engineering, United Arab Emirates University, Al-Ain, UAE
Introduction
Problem Formulation Malicious insertions (Hardware Trojans) in the integrated circuits (IC) poses significant threat to the security of IC [1]. • • • • •
Disables or Destructs the circuits Leakage of Confidential Information Changes the characteristics of ICs Denial of service (DoS) Degradation of performance
• The possibility of detecting these Trojans during test phase is very low, and they may get activated once the chip is in use.
One of the prominent techniques to detect the changes in characteristics of integrated circuits (ICs) is: Electrical Signatures based Trojan Detection: It extracts the values of important characteristics of integrated circuits i.e. timing, frequency, current and power etc., form golden circuits[1].
Proposed Framework
S_n
S_(n-1)
S_2
Trained Model
S_1
“Our proposed framework extract and implement the trained machine learning model, based on Instruction’s power profile, for Runtime Trojan Detection in Microprocessors”
The Proposed Framework Consists of two phases
Micro-Controller/Processor Pipeline Stages
Test Program (Instruction Set)
• In Conventional Electrical signature based solutions • Intruder can overshadow signatures • Trojans depending on the aging of the chip
Instruction’s Power Profile Machine Learning Algorithms
Training and Validation
Power Profiling of instruction set Identify the modules for each pipeline stage of a particular instruction Insert the power ports on each pipeline stage Extract the power profile for a particular Instruction Set Extraction of Trained Machine Learning Model Obtain the power profile of each Instruction Train and validate the model by using different ML tools Integrate the extracted trained model in Microprocessor/controller at SoC Integration level
Case Study: MC8051 MC8051 Benchmark Trojans (Trust-Hub)[2] Six Instructions ADD, MOV, SUBB, INC, JNC and JNMP Xpower Analyzer Power Profiling RapidMiner Machine Learning Training and Validation
Performance Analysis True Trojan True No Trojan K-Nearest Neighbor (k-NN) (Accuracy = 99.02%) Predicted Trojan 1229 (TP) 43 (FP) Predicted No Trojan 31 (FN) 6257 (TN) Decision Tree (DT) (Accuracy = 94.84%) Predicted Trojan 870 (TP) 0 (FP) Predicted No Trojan 390 (FN) 6300 (TN) Deep Learning (DL) (Accuracy = 87.09%) Predicted Trojan 526 (TP) 242 (FP) Predicted No Trojan 734 (FN) 6058 (TN)
Precision
Instructions
Fetch
Decode
Execute
Access
Write Back
MOV
3.15 mW
8.15 mW
11.97 mW
50.55 mW
19.05 mW
96.62% 99.51%
ADD
3.02 mW
8.2 mW
22.98 mW
73.65 mW
31.15 mW
SUBB
3.09 mW 8.745 mW
23.64 mW
75.84 mW
34.56 mW
INC
3.19 mW
8.21 mW
20.65 mW
61.25 mW
23.91 mW
JNC
2.98 mW
8.15 mW
16.08 mW
22.15 mW
18.95 mW
JMP
2.94 mW
8.19 mW
15.47 mW
22.74 mW
20.15 mW
100.00% 94.17% 68.49% 89.19%
Naïve Bayesian Classification (NBC) (Accuracy = 86.46%) Predicted Trojan
312 (TP)
76 (FP)
80.41%
Predicted No Trojan
948 (FN)
6224 (TN)
86.78%
References 1. M. Tehranipoor and F. Koushanfar, “A survey of hardware trojan taxonomy and detection,” IEEE Des. Test Comput., vol. 27, no. 1, pp. 10–25, 2010. 2. M. Tehranipoor and H. Salamani, “trust-HUB,” 2016. [Online]. Available: https://www.trust-hub.org/