Introduction
Modern System-on-Chip (SoC) designs face a critical challenge: the increasing complexity of integrated circuits (ICs) makes exhaustive verification nearly impossible, leading to hardware bugs discovered only after manufacturing. Unlike software or firmware, hardware is traditionally immutable once deployed. To address this, designers incorporate "patching logic"—additional hardware such as multiplexers (MUXes) and control registers—that allows for the observation or modification of internal signals at run-time. However, the lack of a formal metric for "patchability" often results in inefficient designs that either waste silicon area or fail to cover critical vulnerabilities.
Quantifying Patchability
This paper introduces a formal framework to quantify Patching Controllability (PC) and Patching Observability (PO). PC measures the extent to which a signal can be forced to a specific value via patching logic, while PO assesses how effectively a signal's state can be monitored.
The core of the methodology lies in mathematical propagation rules for these scores through standard Register Transfer Level (RTL) operators. For instance, the PC score for logic gates like OR or AND is derived from the average of their input scores, while concatenation operations sum the scores to reflect the combined bit-width controllability.
A key innovation of this framework is its handling of conditional logic. The method recognizes that controlling a multiplexer's selector signal effectively grants controllability over the output, even if the data inputs are hardcoded constants. This insight is crucial for security, as many vulnerabilities involve fixed values like hardcoded permission bits. By controlling the logic that selects these constants, designers can effectively patch bugs without needing to replace every constant with a register, significantly reducing resource overhead.
Evaluation and Results
The methodology was evaluated using the reglk_wrapper IP block from the Hack@DAC-21 SoC. The researchers compared five strategic configurations (V1-V5) against a "Greedy" baseline that attempts to make almost every signal patchable.
The results demonstrate significant efficiency gains. The "V3" configuration achieved a normalized patchability score of 0.9—representing 90% of the theoretical maximum—while requiring 65% fewer resources (110 bits of investment) than the greedy approach (319 bits). This highlights that strategic placement of patching logic on high-leverage control signals is far more effective than a high-volume, unoptimized approach.
Furthermore, the framework successfully identified configurations capable of mitigating specific Common Weakness Enumerations (CWEs), such as CWE-1262 (Improper Access Control) and CWE-1231 (Lock Bit Modification). The precision of the framework was further validated by configuration "V4," which appeared promising to human intuition but yielded a significantly lower score of 0.4 compared to the more strategically designed V3.
Conclusion and Future Work
By treating patchability as a quantifiable design constraint similar to power or area, this framework enables designers to perform rigorous design space exploration. Future work will focus on refining the model to include directional controllability (distinguishing between the ability to force a signal to '0' versus '1') and enhancing the security of the patching infrastructure itself to ensure the "fix" does not become a secondary attack vector.
Summary updated