Accelerating Hardware Verification with Graph Models

Raghul Saravanan, Sreenitha Kasarapu, Sai Manoj Pudukotai Dinakarrao · 2024

Read Accelerating Hardware Verification with Graph Models by Raghul Saravanan, Sreenitha Kasarapu, Sai Manoj Pudukotai Dinakarrao (2024) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2412-13374.txt.

Introduction

Raghul Saravanan, Sreenitha Kasarapu, and Sai Manoj Pudukotai Dinakarrao fuzzed hardware designs with graph learning in a 2024 paper. Chip bugs planted during synthesis hide at the gate level, where simulation runs slowly and classic fuzzing scales poorly. Their GraphFuzz tool turns a gate-level design into a graph, gates as nodes with behavior features, so learning algorithms can study its patterns directly.

Graph models then flag suspicious node patterns that suggest vulnerabilities, steering the search for flaws early in the design cycle. On benchmark circuits and open-source processors, predictions averaged well above chance for both general accuracy and bug detection specifically. The authors present graph-based checking as a scalable complement to simulation-heavy verification.

Catalog record

CREATORRaghul Saravanan, Sreenitha Kasarapu, Sai Manoj Pudukotai Dinakarrao
YEAR2024
COLLECTIONPAPERS
FORMATTXT
FILE SIZE68 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2412-13374

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Accelerating Hardware Verification with Graph Models Raghul Saravanan, Sreenitha Kasarapu, Sai Manoj Pudukotai Dinakarrao Abstract The ever-expanding intricacies within the architectural design spectrum of modern processors and intellectual property (IP) designs foster a challenging environment for identifying and mitigating potential flaws in the early stage of the IC design cycle. Ensuring system integrity and security amidst this complexity demands continuous innovation in design verification methodologies. In the era of hardware design verification, Hardware Fuzzing – a robust technique inspired by software testing has recently gained traction for its effective bug detection in complex hardware designs. Lately, hardware fuzzing frameworks are emerging and successfully identifying bugs in cutting-edge processors. Detecting the bugs and vulnerabilities demands a complex network of instruction seeds to ensure thorough coverage of intricate design elements of the CPU. However, hardware bugs can manifest at different abstraction levels, including RTL (pre-synthesis), gate-level netlist (post-synthesis), and physical layout. Comprehensive verification across all levels is crucial to ensure security and functionality. Gate-level verification is particularly important, as it can uncover bugs introduced during synthesis that are missed at higher abstraction levels. However, the complexity of gate-level netlists in modern SoCs poses significant challenges, leading to longer simulation times and resource-intensive verification. In this paper, we propose a novel hardware fuzzer, GraphFuzz , a graph-based hardware fuzzer to overcome the limitations and enhance the fuzzing process at the gate-level netlist. In this approach, hardware designs are modeled as graph nodes, where intrinsic hardware behaviors of the gates are encoded as features. By applying graph learning algorithms, we analyze node patterns and detect potential vulnerabilities effectively. We test our fuzzer on varied hardware designs, including benchmark circuits, peripherals, and intricate design elements of popular open-source processors. Our graph model representing the netlist during fuzzing achieves an average accuracy of 80% and detects bugs with an average accuracy of 70% demonstrating the effectiveness of using graph-based techniques to enhance hardware verification and identify potential vulnerabilities at the gate-level netlist. [email protected], [email protected], [email protected] I Introduction The ever-increasing threat landscape of hardware designs in the modern IC design flow, exacerbated by global collaboration between vendors, necessitates sophisticated hardware verification mechanisms to ensure system integrity [ 1 ] . As the complexity of integrated circuits (ICs) and the reliance on heterogeneous third-party intellectual property (IP) blocks—especially in system-on-chips (SoCs)—grows, so do the vulnerabilities that can be exploited [ 2 ] . These security vulnerabilities arise not only from the involvement of third-party vendors but also throughout the design, implementation, and integration phases of the modern IC design flow, posing significant challenges to SoC security [ 3 , 4 , 5 , 6 ] . Such vulnerabilities can manifest in various forms, including information leakage, access control violations, side-channel attacks, and the insertion of hardware Trojans [ 7 , 8 , 9 , 10 , 11 , 12 , 13 , 14 , 15 , 16 , 17 ] . Furthermore, existing countermeasures such as logic locking, camouflaging, and watermarking are proving inadequate in addressing these threats due to the increasing complexity of large designs [ 18 , 19 , 20 , 21 ] . Unlike software, hardware patches are costly and not practical, making it crucial to address vulnerabilities before fabrication. For instance, Intel’s machine check bug, for instance, enabled denial-of-service attacks, costing nearly half a billion dollars in re-engineering. Security vulnerabilities stem from the underlying hardware, such as RTL and gate-level netlists, necessitating a concentrated focus on hardware verification at the logic level. To nullify the effects of the vulnerabilities and bugs in the SoCs and ICs, hardware verification community has developed various pre-silicon verification techniques [ 22 , 23 , 24 , 25 , 26 , 27 , 28 , 29 ] and Electronic Design Automation (EDA) tools [ 30 , 31 , 32 ] . Formal methods and information flow tracking (IFT) are the most popular verification techniques. However, recent advancements in formal methods, such as satisfiability checking (SAT) [ 33 , 34 ] and satisfiability modulo theories (SMT) [ 35 ] , have encountered significant challenges due to scalability limitations and the requirement for expert knowledge. These constraints diminish their effectiveness in verifying increasingly complex modern hardware designs. Furthermore, while Information Flow Tracking (IFT) techniques [ 36 , 37 , 38 ] are effective in tracking sensitive data flows, they require design instrumentation, which in turn demands additional computational time and resources, further complicating their application to large-scale systems. There is a growing need for hardware verification methodologies and frameworks for modern SoCs that can scale to large, complex designs [ 39 , 40 , 41 , 42 , 43 , 44 , 45 ] . To address the shortcomings of the existing pre-silicon verification, recently hardware community researchers introduced hardware fuzzing inspired from software testing paradigm [ 46 , 47 , 48 , 49 , 50 , 51 , 52 , 53 , 54 , 55 , 56 ] . Fuzzing is a widely used software testing technique for bug detection in software applications. Industry-based fuzzing frameworks such as Google’s OSS-Fuzz [ 57 ] and Microsoft’s Security Risk Detection [ 58 ] have proved their efficacy and effectively identified a plethora of security vulnerabilities. The popular software fuzzer, American Fuzzy Lop (AFL) [ 59 ] , is predominantly used in software fuzzing and is predominantly used in majority of the software fuzzing techniques. The initial work on hardware fuzzing, RFuzz [ 46 ] adopted software fuzzing methodologies directly onto hardware, while the latter works translated the hardware to an equivalent software and the fuzzing the resultant software code [ 49 ] . Unfortunately, these fuzzing methods inherently fail to capture the intrinsic hardware behaviors including logic transitions, and register computations. To circumvent the aforementioned challenges, recent works have proposed fuzzing the hardware at its native hardware abstraction level, making it compatible with traditional hardware verification techniques [ 52 , 60 , 48 , 61 ] . These existing works have proven efficient in identifying bugs and vulnerabilities in state-of-the-art open-source CPU RTL designs [ 62 , 63 , 64 , 65 ] , highlighting the prominence of fuzzing at hardware level abstraction [ 66 , 52 , 60 , 61 ] . To this end, there have been several works on hardware fuzzing [ 46 , 52 , 60 , 49 , 48 , 67 , 68 ] that aim to fuzz hardware at the RTL abstraction level, focusing on verifying the design’s adherence to its specification. However, it is important to note that hardware vulnerabilities can manifest at different levels of hardware abstraction [ 69 ] , including the RTL level (pre-synthesis), gate-level netlist (post-synthesis) and the physical level (layout) . To ensure comprehensive security and functionality, hardware verification must be performed at every abstraction level of the design flow. As we elaborate in Section II-C , verification at the gate-level is critical as it can capture bugs and vulnerabilities introduced during the synthesis that go unnoticed in higher level of abstraction, ensuring that the final implementation adheres to the intended behavior specified at the RTL level before entering physical design phase. Conversely, verification at the gate-level netlist, which encompasses thousands or even millions of gate components in complex SoC

Read the complete text