System-on-Chip Message Flow Mining with Masked-Language Models
Md Rubel Ahmed, Bardia Nadimi, Hao Zheng · 2023
High-quality system-level specifications are necessary to validate system-on-chip (SoC) designs comprehensively. However, manual development and maintenance of such specifications across the entire SoC design flow are daunting tasks. We propose a novel method that utilizes a masked-language model to infer comprehensive and accurate specifications from SoC communication traces. The proposed method can overcome the inherent complexity of SoC traces induced by the concurrent executions of system transactions that existing mining tools often find highly challenging. Furthermore, the mined specifications provide a high-level and abstract view of SoC communication behaviors, which can significantly facilitate design understanding and validation. Experiments on traces from a simple SoC model and a more detailed, close-to-reality model developed in gem5 show promising results. The proposed method outperforms several existing state-of-the-art trace mining tools.