AI-Guided Stimulus and Debug Triage: A Methodology for Accelerating Pre-Silicon Emulation of Flagship SoC Programs
Abstract
Raw emulation capacity for flagship system-on-chip (SoC) programs keeps growing, and engineer-hour supply keeps falling further behind it. That gap, not machine availability, is the actual bottleneck in pre-silicon verification today, driven by workload diversity across artificial intelligence (AI), extended reality (XR), and mobile platforms converging on a single die. This article argues that closing the gap requires a bounded, governed layer of AI-guided assistance rather than either extreme - full automation or continued reliance on engineer-only throughput. A three-layer adoption framework is proposed: scoping, which restricts AI involvement to stimulus generation for known-workload classes and anomaly-flagged debug triage while keeping root-cause sign-off and tape-out risk calls fully engineer-owned; validation, which requires every AI-generated output to be checked against a trusted reference - simulation, formal methods, or historical regression baselines - before an engineer may act on it; and governance, an operating-review cadence with rollback criteria defined before deployment, not improvised afterward. Pilot deployment of emulation-friendly RTL on a premium-tier mobile SoC program produced a shift of approximately 60 percent of RTL earlier in the program schedule, with no coverage-model exception granted to reach that outcome. The finding that emerges is not that AI accelerates verification - that much is already argued elsewhere - but that the acceleration only survives contact with a tape-out schedule when it is deliberately bounded. Emulation productivity under AI assistance is a governance problem before it is a technical one, and the three-layer framework is offered as a transferable operating model for programs where the cost of an unvalidated miss is unacceptable.
Keywords
Downloads
References
- S. Fine, A. Freund, I. Jaeger, Y. Naveh, A. Ziv, Y. Mansour, "Harnessing Machine Learning to Improve the Success Rate of Stimuli Generation," IEEE HLDVT, 2005. https://ieeexplore.ieee.org/document/1568823
- L. Pierre, F. Pancher, R. Suescun, J. Quévremont, "On the Effectiveness of Assertion-Based Verification in an Industrial Context," FMICS, LNCS vol. 8187, pp. 78-93, 2013. https://link.springer.com/chapter/10.1007/978-3-642-41010-9_6
- Accellera Systems Initiative, Portable Test and Stimulus Standard, Version 1.0a, 2019. https://accellera.org/images/downloads/standards/pss/Portable_Test_Stimulus_Standard_v2.1.pdf
- IEEE, IEEE Standard for SystemVerilog, IEEE Std 1800-2017, 2018. https://ieeexplore.ieee.org/document/8299595
- Accellera Systems Initiative, Universal Verification Methodology (UVM) 1.2 User's Guide, 2015. https://www.accellera.org/downloads/standards/uvm
- G. Huang et al., "Machine Learning for Electronic Design Automation: A Survey," ACM TODAES, vol. 26, no. 5, art. 40, 2021. https://arxiv.org/abs/2102.03357
- M. Liu, N. Pinckney, B. Khailany, H. Ren, "VerilogEval: Evaluating Large Language Models for Verilog Code Generation," IEEE/ACM ICCAD, 2023. https://arxiv.org/abs/2309.07544
- N. Pinckney, C. Batten, M. Liu, H. Ren, B. Khailany, "Revisiting VerilogEval," ACM TODAES, vol. 30, no. 6, art. 91, 2025. https://arxiv.org/abs/2408.11053
- H. Liu, Y. Lu, M. Wang, X. Yao, B. Yu, "LLM-Assisted Circuit Verification: A Comprehensive Survey," ASP-DAC, 2026. https://ieeexplore.ieee.org/document/11420692
- D. N. Gadde, T. Nalapat, A. Kumar, S. Simon, D. Lettnin, W. Kunz, "Efficient Stimuli Generation using Reinforcement Learning in Design Verification," SMACD, 2024. https://arxiv.org/abs/2405.19815
- Z. Yan, W. Fang, M. Li, M. Li, S. Liu, Z. Xie, H. Zhang, "AssertLLM," ASP-DAC, 2025. https://zhiyaoxie.com/files/ASPDAC25_AssertLLM.pdf
- R. Qiu, G. L. Zhang, R. Drechsler, U. Schlichtmann, B. Li, "AutoBench," ACM/IEEE MLCAD, 2024. https://portal.fis.tum.de/en/publications/autobench-automatic-testbench-generation-and-evaluation-using-llm/
- J. Ye et al., "From Concept to Practice: an Automated LLM-aided UVM Machine for RTL Verification," IEEE/ACM ICCAD, 2025. https://arxiv.org/abs/2504.19959
- K. Xu, J. Sun, Y. Hu, X. Fang, W. Shan, X. Wang, Z. Jiang, "MEIC: Re-thinking RTL Debug Automation using LLMs," IEEE/ACM ICCAD, 2024. https://arxiv.org/abs/2405.06840
- S. Liu, Y. Lu, W. Fang, M. Li, Z. Xie, "OpenLLM-RTL," IEEE/ACM ICCAD, 2024. https://arxiv.org/abs/2503.15112
- R. Gal, E. Haber, A. Ziv, "Using DNNs and Smart Sampling for Coverage Closure Acceleration," ACM/IEEE MLCAD, 2020. https://ieeexplore.ieee.org/document/9394648
- A. Jayasena, P. Mishra, "Directed Test Generation for Hardware Validation: A Survey," ACM Computing Surveys, 2023. https://dl.acm.org/doi/10.1145/3638046
- L. Zulberti, S. Di Matteo, P. Nannipieri, S. Saponara, L. Fanucci, "A Script-Based Cycle-True Verification Framework to Speed-Up Hardware and Software Co-Design," Electronics, vol. 11, no. 22, art. 3704, 2022. https://www.mdpi.com/2079-9292/11/22/3704
- M. Abdollahi, S. F. Yeganli, M. A. Baharloo, A. Baniasadi, "Hardware Design and Verification with Large Language Models: A Scoping Review," Electronics, vol. 14, no. 1, art. 120, 2025. https://www.mdpi.com/2079-9292/14/1/120
- A. Rammal, K. Ezukwoke, A. Hoayek, M. Batton-Hubert, "Root Cause Prediction for Failures in Semiconductor Industry," Scientific Reports, vol. 13, art. 4934, 2023. https://www.nature.com/articles/s41598-023-30769-8
- S. Alsaedi, A. A. A. Gad-Elrab, A. Noaman, F. Eassa, "Two-Level Information-Retrieval-Based Model for Bug Localization," Electronics, vol. 13, no. 2, art. 321, 2024. https://www.mdpi.com/2079-9292/13/2/321
- A. Newcomb, O. Ochoa, "Formal Methods for Safety-Critical Machine Learning: A Systematic Literature Review," Frontiers in AI, vol. 9, art. 1749956, 2026. https://www.frontiersin.org/journals/artificial-intelligence/articles/10.3389/frai.2026.1749956/full