Abstract
Traditional security chip architecture design often relies heavily on subjective expert experience, which can lack scientific rigor. Conversely, purely mathematical multi-attribute decision-making (MADM) models frequently ignore the value of subjective intuition or struggle with heterogeneous data types. This paper proposes a decision-making method based on "satisfaction degree" and "consistency degree" weighting. By integrating expert preference information with objective attribute distinguishability, the framework provides a robust and effective approach for evaluating and selecting optimal security chip designs.
Introduction
Designing a modern security chip requires balancing conflicting performance metrics, such as high clock frequency and robust security features, against constraints like silicon area and power consumption. Historically, these trade-offs were managed through qualitative "gut feelings." While standard MADM tools like the Analytic Hierarchy Process (AHP) exist, they often fail to handle the "fuzzy" nature of qualitative data (e.g., "High Security") or datasets where candidate values are too similar to distinguish effectively.
Methodology
The proposed framework bridges the gap between objective data and subjective expertise through a five-step process:
- Fuzzy-to-Interval Conversion: Qualitative attributes are first mapped to Triangular Fuzzy Numbers (TFNs) and subsequently converted into interval numbers. This allows the model to mathematically process linguistic variables like "Very High" alongside precise numerical data.
- Satisfaction Degree (SD) Filtering: To streamline the decision process, a "satisfaction degree" function acts as a preliminary filter. Candidates that fail to meet a designer's baseline expectations for any given attribute are removed from consideration.
- Consistency Degree Weighting: This is the core innovation. The model calculates the "consistency degree" of each attribute—essentially measuring how much an attribute varies across candidates. If all chips have nearly identical power consumption, that attribute is assigned a lower weight because it provides little information for distinguishing between options.
- Preference Factor Integration: To ensure the model aligns with human goals, designers provide a "preference factor" ($a_j$). The final attribute weights are a hybrid of objective distinguishability and subjective priority.
- Final Ranking: The remaining candidates are ranked using an algorithm that compares interval numbers to determine the overall most "satisfactory" architecture.
Case Study and Dataset
The methodology was validated using a simulated design project involving eight architecture schemes ($A_1$ through $A_8$). Each scheme was evaluated against four criteria: Main Frequency ($C_1$), Core Area ($C_2$), Security Level ($C_3$), and Power Consumption ($C_4$).
The dataset reflects the complexity of real-world design, featuring a mix of precise integers (frequency), linguistic labels (security), and interval ranges (power consumption).
Key Results and Discussion
The application of the Satisfaction Degree filter immediately narrowed the field from eight candidates to five: $A_1, A_2, A_3, A_5,$ and $A_8$. Candidates $A_4, A_6,$ and $A_7$ were excluded for failing to meet minimum performance or security thresholds.
The final evaluation identified architecture $A_5$ as the optimal choice, achieving the highest satisfaction score.
A comparison between this hybrid approach and purely objective ranking methods revealed that while both might identify the same top candidate, the internal rankings of other schemes differ. The satisfaction-based approach more accurately captures how closely a design aligns with a designer's specific "ideal" range, rather than just its relative performance against other candidates.
Conclusion
The proposed method successfully integrates objective data analysis with subjective expert preference. By using consistency degree to determine attribute importance and satisfaction degree to filter candidates, the framework offers a scientifically grounded yet flexible tool for security chip architecture selection. Future work will focus on refining the satisfaction functions—moving from linear to more complex non-linear models—and developing methods to reconcile conflicting preferences from multiple experts.
Summary updated