TL;DR
This paper introduces a game-theoretic model for hardware Trojan detection, treating the IC manufacturer as an attacker and the testing facility as a defender. It uniquely incorporates Prospect Theory (PT) to account for the subjective, potentially irrational decision-making of both parties under uncertainty, offering a more realistic analysis than traditional Expected Utility Theory (EUT).
What Problem Does This Paper Address?
The increasing reliance on offshore manufacturing for Integrated Circuits (ICs) introduces significant security risks, primarily through the insertion of "hardware Trojans"—maliciously designed components that can cause detrimental system failures when activated. Detecting these Trojans is challenging due to resource limitations, meaning not all potential Trojan types can be tested for. This paper addresses the strategic interaction between the party inserting the Trojan (the manufacturer/attacker) and the party detecting it (the testing facility/defender). Traditional game theory assumes rational actors, but real-world decision-making, especially under risk and uncertainty, often deviates from this. The paper aims to bridge this gap by modeling this interaction using Prospect Theory, which better captures human behavior under risk and uncertainty.
Methods
The core of the paper is a noncooperative, zero-sum game formulation between an attacker and a defender.
- Game Formulation: The attacker chooses a Trojan type to insert, aiming to maximize damage. The defender chooses a subset of Trojan types to test for, aiming to minimize damage and impose penalties.
- Prospect Theory (PT) Integration: Unlike standard Expected Utility Theory (EUT), which assumes rational decision-making, this work incorporates Prospect Theory. PT accounts for how individuals subjectively perceive probabilities and values, especially under risk and uncertainty. This is modeled using a "weighting effect" where players distort their perception of the opponent's strategies based on their own rationality parameter (α).
- Fictitious Play Algorithm: To solve the game under both EUT and PT, a novel algorithm based on fictitious play is proposed. This algorithm simulates players iteratively adjusting their strategies based on observed opponent behavior, converging to a mixed-strategy Nash equilibrium (MSNE). The paper proves that this algorithm converges to an MSNE for both EUT and PT.
Datasets
Not discussed in the paper. The paper uses an illustrative case study with defined Trojan types and damage values rather than real-world datasets.
Key Insights and Ideas
- Beyond Rationality: The most significant insight is that assuming purely rational actors (EUT) can lead to misleading conclusions in hardware Trojan detection. Prospect Theory reveals how subjective risk perception and uncertainty can drastically alter optimal strategies and game outcomes.
- Risk Aversion's Double-Edged Sword: Both attackers and defenders can exhibit risk aversion. A risk-averse defender might focus testing on the most damaging Trojans, potentially leaving them vulnerable to other types. A risk-averse attacker might opt for less damaging Trojans, which could be less likely to be detected.
- PT's Impact on Equilibrium: The paper demonstrates that PT-based equilibria can differ significantly from EUT equilibria. For instance, under PT, the attacker might be more inclined to insert lower-value Trojans, and the defender might be more aggressive in testing for the highest-damage Trojans.
- Fine Tuning the Fine: The value of the penalty (fine) imposed on the attacker plays a crucial role. The paper shows that under PT, the defender might need to set higher fines to achieve a favorable outcome compared to the EUT scenario, especially when both parties are irrational.
Key Results
- Algorithm Convergence: The proposed fictitious play algorithm is proven to converge to a mixed-strategy Nash equilibrium for both EUT and PT frameworks.
- EUT vs. PT Outcomes: Simulations show that PT can lead to different equilibrium strategies and expected utilities compared to EUT. For example, in one scenario, the overall value of the game (attacker's utility) decreased by 30% under PT compared to EUT.
- Impact of Rationality Parameters: The paper analyzes how varying rationality parameters (α) for the attacker and defender affects game outcomes. It shows that when both players are significantly irrational (low α), the defender might emerge as the winner by adopting conservative strategies. Conversely, when one player is rational and the other irrational, the irrational player can exploit the rational one's predictable behavior.
- Fine Value Analysis: The study demonstrates that the optimal fine value for the defender to achieve a positive utility differs between EUT and PT. Under PT, a higher fine might be necessary.
Limitations and Future Work
- Simplified Model: The paper assumes a single attacker and a single defender, and a uniform fine for all Trojan types. Real-world scenarios might involve multiple attackers, multiple defenders, and varying fine structures.
- Illustrative Case Study: The numerical analysis uses a simplified scenario with four Trojan types and specific damage values. Extending this to more complex scenarios with a larger number of Trojans and more nuanced damage models would be valuable.
- Weighting Function: While the Prelec function is used for its established properties, the paper notes that other weighting functions could be explored, potentially requiring new experimental data.
- Future Work: The authors suggest extending the framework to scenarios with multiple attackers or defenders, and incorporating more detailed system models to quantify damage more precisely. They also highlight the potential for future work in applying PT to other security games.
Summary updated