pith. sign in

arxiv: 2604.15359 · v1 · submitted 2026-04-12 · 💻 cs.AR · cs.LG· cs.SE

AutoFlows++: Hierarchical Message Flow Mining for System on Chip Designs

Bardia Nadimi , Hao Zheng This is my paper

Pith reviewed 2026-05-10 15:31 UTC · model grok-4.3

classification 💻 cs.AR cs.LGcs.SE
keywords message flow miningSoC verificationcommunication traceshierarchical miningtrace analysisfunctional verificationGEM5 simulation
0
0 comments X

The pith

AutoFlows++ mines message flows hierarchically from SoC communication traces by first extracting local patterns then composing global flows.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper presents AutoFlows++ as a two-stage framework for extracting communication specifications from traces in complex SoC designs. Local mining pulls simple patterns from traces at individual component interfaces. Global mining then combines these into higher-level flows that span multiple components. This hierarchical approach aims to handle interleaving and ambiguity better than existing methods, leading to more accurate results on both synthetic and simulated SoC traces from GEM5.

Core claim

AutoFlows++ is a design-architecture-guided hierarchical framework that performs local mining of simple communication patterns from individual interfaces followed by global mining to compose higher-level flows across multiple components, achieving significantly improved flow extraction accuracy compared to prior approaches on synthetic and GEM5-generated traces.

What carries the argument

The two-stage hierarchical mining process guided by design architecture, with local pattern extraction followed by global composition of flows.

If this is right

  • Provides concise communication specifications useful for functional verification of SoC designs.
  • Supports performance analysis by revealing actual communication behaviors across components.
  • Facilitates post-silicon debugging with more accurate extraction from real traces.
  • Avoids the explosion of candidate flows that occurs in prior methods on interleaved data.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The method could apply to traces from physical hardware if architecture details remain accessible.
  • Similar local-to-global composition might help in mining patterns from other interleaved event logs such as network traffic or software execution.
  • Without the architecture guidance, accuracy may drop in fully black-box systems lacking component interface definitions.

Load-bearing premise

That local patterns extracted from individual interfaces can be reliably composed into accurate global flows without missing causal relationships or generating an unmanageable number of candidates.

What would settle it

A comparison on a new complex SoC trace with known ground-truth flows where AutoFlows++ produces incorrect or incomplete global flows compared to the ground truth.

Figures

Figures reproduced from arXiv: 2604.15359 by Bardia Nadimi, Hao Zheng.

Figure 1
Figure 1. Figure 1: (a) Simplified SoC architecture example, (b) Message [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: (a) Graph representation of CPU downstream flows [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: illustrates the overall workflow of the proposed AutoFlows++ framework. The method takes as input a set of system communication traces and begins with a local mining stage, which extracts binary patterns (BPs) from interface￾sliced traces to capture localized interaction dependencies among system components. It then proceeds to a global mining stage, where a global causality graph is constructed and a path… view at source ↗
Figure 4
Figure 4. Figure 4: Causality graph from trace (1): edges marked with [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Sub-causality graphs for the example trace 5. [PITH_FULL_IMAGE:figures/full_fig_p009_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: (a) GEM5 Full-System (FS) design. (b) GEM5 System [PITH_FULL_IMAGE:figures/full_fig_p010_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Counts of instances of the mined message sequences of different lengths using the AutoFlows++, AutoFlows, and AutoModel method [PITH_FULL_IMAGE:figures/full_fig_p012_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: (a) A flow mined from the GEM5 threads trace. (b) GEM5 documentation’s memory write miss sequence. (c) Two flows mined from [PITH_FULL_IMAGE:figures/full_fig_p012_8.png] view at source ↗
read the original abstract

Understanding communication behavior in modern system-on-chip (SoC) designs is critical for functional verification, performance analysis, and post-silicon debugging. Communication traces capture message exchanges among system components and provide valuable insights into system behavior. However, deriving concise communication specifications from such traces remains challenging due to interleaved instances of communication flows, and ambiguous causal relationships among messages. Existing mining approaches often struggle with scalability and ambiguity when traces contain complex interleaving of message patterns across multiple components. These conditions often lead to an explosion in the number of candidate flows and inaccurate extraction of communication behaviors. This paper presents AutoFlows++, a design-architecture-guided hierarchical framework for mining message flows from communication traces of complex SoC designs. AutoFlows++ operates in two stages: local mining followed by global mining. In the local mining stage, simple communication patterns are extracted from traces observed at individual communication interfaces between components. In the global mining stage, these local patterns are composed to identify higher-level message flows that characterize communication behavior across multiple components. Experimental results on both synthetic traces and traces generated from SoC models in GEM5 demonstrate that AutoFlows++ significantly improves flow extraction accuracy compared with prior approaches, highlighting its effectiveness for practical SoC validation tasks.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The paper introduces AutoFlows++, a design-architecture-guided hierarchical framework for mining message flows from SoC communication traces. It operates in two stages: local mining extracts simple patterns from individual component interfaces, while global mining composes these into higher-level flows across multiple components. The central claim is that this approach significantly improves flow extraction accuracy over prior methods, as demonstrated on synthetic traces and GEM5-generated SoC traces.

Significance. If the empirical claims hold with proper validation, the work could meaningfully advance automated analysis tools for SoC functional verification, performance evaluation, and post-silicon debugging by addressing interleaving and causality challenges in communication traces. The hierarchical local-to-global structure offers a plausible path to scalability, and the use of architecture guidance is a reasonable design choice. However, the current lack of detailed quantitative evidence and unresolved questions around global composition limit its assessed impact.

major comments (2)
  1. [Experimental evaluation] Experimental evaluation section: The abstract asserts significant accuracy improvements on synthetic and GEM5 traces, yet no quantitative metrics (e.g., precision/recall/F1), baseline algorithm descriptions, trace counts, statistical tests, or analysis of post-hoc parameter choices are supplied. This absence directly undermines evaluation of the central empirical claim.
  2. [Global mining stage (Section 4)] Global mining stage (Section 4): The composition of local interface patterns into global flows is described at a high level using architecture guidance, but provides no explicit mechanisms (e.g., timing windows, message IDs, or causality rules) for resolving ambiguous ordering and interleaving across components. This is load-bearing for the reported gains, as the skeptic concern about missed cross-component links or candidate explosion remains unaddressed.
minor comments (2)
  1. [Abstract] The abstract would be strengthened by including at least one concrete accuracy metric or baseline name to ground the superiority claim.
  2. [Method description] Notation for local vs. global patterns could be clarified with a small example diagram or pseudocode snippet early in the method description.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive and detailed feedback on our manuscript. We appreciate the identification of areas where additional clarity and evidence are needed to strengthen the presentation of AutoFlows++. We address each major comment below and commit to revisions that directly respond to the concerns.

read point-by-point responses

  1. Referee: [Experimental evaluation] Experimental evaluation section: The abstract asserts significant accuracy improvements on synthetic and GEM5 traces, yet no quantitative metrics (e.g., precision/recall/F1), baseline algorithm descriptions, trace counts, statistical tests, or analysis of post-hoc parameter choices are supplied. This absence directly undermines evaluation of the central empirical claim.

    Authors: We agree that the experimental evaluation section requires substantial expansion to properly support the accuracy claims. In the revised manuscript, we will add quantitative results including precision, recall, and F1 scores for flow extraction on the synthetic traces and GEM5-generated SoC traces. We will describe the baseline algorithms in detail, specify the number and characteristics of traces evaluated, report statistical tests (such as paired t-tests with p-values) to confirm significance of improvements, and include sensitivity analysis for key parameters. These changes will enable a rigorous assessment of the empirical contributions. revision: yes

  2. Referee: [Global mining stage (Section 4)] Global mining stage (Section 4): The composition of local interface patterns into global flows is described at a high level using architecture guidance, but provides no explicit mechanisms (e.g., timing windows, message IDs, or causality rules) for resolving ambiguous ordering and interleaving across components. This is load-bearing for the reported gains, as the skeptic concern about missed cross-component links or candidate explosion remains unaddressed.

    Authors: We acknowledge that the current description of the global mining stage is high-level and does not fully specify the resolution mechanisms. In the revision, we will expand Section 4 with explicit details on how architecture guidance is operationalized: timing windows derived from component latency specifications in the design, message ID matching to link local patterns across interfaces, and causality rules extracted from the SoC architecture to enforce valid ordering and prune invalid compositions. This will directly address potential issues of missed cross-component links and candidate explosion, clarifying the load-bearing aspects of the hierarchical approach. revision: yes

Circularity Check

0 steps flagged

No circularity in algorithmic framework or empirical claims

full rationale

The paper presents a two-stage algorithmic framework (local pattern mining at interfaces followed by global composition of flows) without any equations, derivations, fitted parameters, or 'predictions' that reduce to inputs by construction. Claims of improved accuracy rest on experimental results using synthetic traces and GEM5-generated SoC traces, which constitute independent validation rather than self-referential fitting. No self-citations are invoked as load-bearing uniqueness theorems or ansatzes. The contribution is a new hierarchical mining procedure whose correctness is assessed externally via trace experiments, making the derivation self-contained.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

No free parameters, axioms, or invented entities are referenced in the abstract.

pith-pipeline@v0.9.0 · 5512 in / 1055 out tokens · 55911 ms · 2026-05-10T15:31:28.354722+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

47 extracted references · 47 canonical work pages

  1. [1]

    Detecting api documentation errors

    Hao Zhong and Zhendong Su. Detecting api documentation errors. OOPSLA’13, pages 803–816, New York, NY , USA, 2013. Association for Computing Machinery

    work page 2013

  2. [2]

    Harris, Goerschwin Fey, and Mathias Soeken

    Sandip Ray, Ian G. Harris, Goerschwin Fey, and Mathias Soeken. Mul- tilevel design understanding: From specification to logic invited paper. InProceedings of the 35th International Conference on Computer-Aided Design, ICCAD ’16, pages 133:1–133:6, 2016

    work page 2016

  3. [3]

    Ray, J Bhadra, M Abadir, and Li-C Wang

    W Chen, S. Ray, J Bhadra, M Abadir, and Li-C Wang. Challenges and trends in modern soc design verification.IEEE Design Test, 34(5):7–22, Oct 2017

    work page 2017

  4. [4]

    Perracotta: Mining temporal api rules from imperfect traces

    Jinlin Yang, David Evans, Deepali Bhardwaj, Thirumalesh Bhat, and Manuvir Das. Perracotta: Mining temporal api rules from imperfect traces. InProceedings of the 28th International Conference on Software Engineering, ICSE ’06, pages 282–291, 2006

    work page 2006

  5. [5]

    Liu and S

    L. Liu and S. Vasudevan. Automatic generation of system level assertions from transaction level models.Journal of Electronic Testing, 2013

    work page 2013

  6. [6]

    Learning temporal specifications from imperfect traces using bayesian inference

    Artur Mrowca, Martin Nocker, Sebastian Steinhorst, and Stephan G¨unnemann. Learning temporal specifications from imperfect traces using bayesian inference. InProceedings of the 56th Annual Design Automation Conference 2019, DAC ’19, pages 96:1–96:6, 2019

    work page 2019

  7. [7]

    Learning concise models from long execution traces

    Natasha Jeppu, Tom Melham, Daniel Kroening, and John O’Leary. Learning concise models from long execution traces. InDAC ’20, June 2020

    work page 2020

  8. [8]

    Binkert et al

    N. Binkert et al. The gem5 simulator.SIGARCH Comput. Archit. News, 39(2):1–7, aug 2011

    work page 2011

  9. [9]

    Ivan Beschastnikh, Yuriy Brun, Sigurd Schneider, Michael Sloan, and Michael D. Ernst. Leveraging existing instrumentation to automatically infer invariant-constrained models. InProceedings of the 19th ACM SIG- SOFT Symposium and the 13th European Conference on Foundations of Software Engineering, ESEC/FSE ’11, pages 267–277, 2011

    work page 2011

  10. [10]

    J. Yang, D. Evans, D. Bhardwaj, T. Bhat, and M. Das. Perracotta: Mining temporal api rules from imperfect traces. In28th International Conference on Software Engineering, pages 282–291, 2006

    work page 2006

  11. [11]

    Mrowca, M

    A. Mrowca, M. Nocker, S. Steinhorst, and S. G ¨unnemann. Learning temporal specifications from imperfect traces using bayesian inference. InProceedings of the 56th Annual Design Automation Conference 2019, 2019

    work page 2019

  12. [12]

    Wil M. P. van der Aalst.Process Mining: Data Science in Action. Springer, 2nd edition, 2016

    work page 2016

  13. [13]

    G ¨unther and Wil M

    Christian W. G ¨unther and Wil M. P. van der Aalst. Fuzzy mining – adaptive process simplification based on multi-perspective metrics. In Gustavo Alonso, Peter Dadam, and Michael Rosemann, editors, Business Process Management, pages 328–343, Berlin, Heidelberg,

    work page

  14. [14]

    Springer Berlin Heidelberg

    work page

  15. [15]

    Specification mining for smart contracts with trace slicing and predicate abstraction

    Ye Liu, Yixuan Liu, Yi Li, and Cyrille Artho. Specification mining for smart contracts with trace slicing and predicate abstraction. In 2025 IEEE International Conference on Software Analysis, Evolution and Reengineering (SANER), pages 147–158, 2025

    work page 2025

  16. [16]

    Graph-based trace analysis for microservice architecture understanding and problem diagnosis

    Xiaofeng Guo, Xin Peng, Hanzhang Wang, Wanxue Li, Huai Jiang, Dan Ding, Tao Xie, and Liangfei Su. Graph-based trace analysis for microservice architecture understanding and problem diagnosis. In Proceedings of the 28th ACM Joint Meeting on European Software Engineering Conference and Symposium on the Foundations of Software Engineering, ESEC/FSE 2020, pag...

    work page 2020

  17. [17]

    Association for Computing Machinery

    work page

  18. [18]

    Wenchao Li, Alessandro Forin, and Sanjit A. Seshia. Scalable specifi- cation mining for verification and diagnosis. InProceedings of the 47th Design Automation Conference, DAC ’10, pages 755–760, New York, NY , USA, 2010

    work page 2010

  19. [19]

    Hertz, D

    S. Hertz, D. Sheridan, and S. Vasudevan. Mining hardware assertions with guidance from static analysis.Transactions on Computer-Aided Design of Integrated Circuits and Systems, 2013

    work page 2013

  20. [20]

    Danese, F

    A. Danese, F. Filini, and G. Pravadelli. A time-window based approach for dynamic assertions mining on control signals. InIFIP/IEEE International Conference on Very Large Scale Integration, 2015

    work page 2015

  21. [21]

    Danese, T

    A. Danese, T. Ghasempouri, and G. Pravadelli. Automatic extraction of assertions from execution traces of behavioural models. InProceedings of the Design, Automation, & Test in Europe Conference, 2015

    work page 2015

  22. [22]

    Danese, N

    A. Danese, N. D. Riva, and G. Pravadelli. A-team: Automatic template- based assertion miner. InProceedings of the 54th Annual Design Automation Conference, 2017

    work page 2017

  23. [23]

    Chang and L

    P. Chang and L. Wang. Automatic assertion extraction via sequential data mining of simulation traces. InASPDAC, 2010

    work page 2010

  24. [24]

    Artmine: Automatic association rule mining with temporal behavior for hardware verification

    Mohammad Reza Heidari Iman, Gert Jervan, and Tara Ghasempouri. Artmine: Automatic association rule mining with temporal behavior for hardware verification. In2024 Design, Automation & Test in Europe Conference & Exhibition (DATE), pages 1–6, 2024

    work page 2024

  25. [25]

    M. J. Heule and S. Verwer. Software model synthesis using satisfiability solvers.Empirical Software Engineering, 18(4), 2013

    work page 2013

  26. [26]

    K. J. Lang, B. A. Pearlmutter, and R. A. Price. Results of the abbadingo one dfa learning competition and a new evidence-driven state merging algorithm. InICGI, 1998

    work page 1998

  27. [27]

    Ulyantsev and F

    V . Ulyantsev and F. Tsarev. Extended finite-state machine induction using sat-solver. InICMLAW, volume 2, 2011

    work page 2011

  28. [28]

    Inferring extended finite state machine models from software executions

    Neil Walkinshaw, Ramsay Taylor, and John Derrick. Inferring extended finite state machine models from software executions. In2013 20th Working Conference on Reverse Engineering (WCRE), pages 301–310, 2013

    work page 2013

  29. [29]

    Zheng, M

    H. Zheng, M. R. Ahmed, P. Mukherjee, M. C. Ketkar, and J. Yang. Model synthesis for communication traces of system designs. In2021 IEEE 39th International Conference on Computer Design (ICCD), pages 492–499, CA, USA, Oct 2021

    work page 2021

  30. [30]

    Automodel: Auto- matic synthesis of models from communication traces of soc designs

    Md Rubel Ahmed, Bardia Nadimi, and Hao Zheng. Automodel: Auto- matic synthesis of models from communication traces of soc designs. IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems, pages 1–1, 2024

    work page 2024

  31. [31]

    Autoflows: Inferring message flows from system communication traces

    Bardia Nadimi and Hao Zheng. Autoflows: Inferring message flows from system communication traces. In2025 26th International Symposium on Quality Electronic Design (ISQED), pages 1–8, 2025

    work page 2025

  32. [32]

    M. R. Ahmed, H. Zheng, P. Mukherjee, M. C. Ketkar, and J. Yang. Mining message flows from system-on-chip execution traces. In22nd ISQED. IEEE, 2021

    work page 2021

  33. [33]

    X. Meng, K. Raj, S. Ray, and K. Basu. Sevnoc: Security validation of system-on-chip designs with noc fabrics.IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems, 2022

    work page 2022

  34. [34]

    Time, clocks, and the ordering of events in a distributed system.Commun

    Leslie Lamport. Time, clocks, and the ordering of events in a distributed system.Commun. ACM, 21(7):558–565, July 1978

    work page 1978

  35. [35]

    Virtual time and global states of distributed systems

    Friedemann Mattern. Virtual time and global states of distributed systems. 2002

    work page 2002

  36. [36]

    Trace-based multi-dimensional root cause localization of performance issues in microservice systems

    Chenxi Zhang, Zhen Dong, Xin Peng, Bicheng Zhang, and Miao Chen. Trace-based multi-dimensional root cause localization of performance issues in microservice systems. In2024 IEEE/ACM 46th International Conference on Software Engineering (ICSE), pages 1347–1358, 2024

    work page 2024

  37. [37]

    Sigelman, Luiz Andr ´e Barroso, Mike Burrows, Pat Stephenson, Manoj Plakal, Donald Beaver, Saul Jaspan, and Chandan Shanbhag

    Benjamin H. Sigelman, Luiz Andr ´e Barroso, Mike Burrows, Pat Stephenson, Manoj Plakal, Donald Beaver, Saul Jaspan, and Chandan Shanbhag. Dapper, a large-scale distributed systems tracing infrastruc- ture. InGoogle Technical Report, 2010

    work page 2010

  38. [38]

    A multi-expert large language model architecture for verilog code generation

    Bardia Nadimi and Hao Zheng. A multi-expert large language model architecture for verilog code generation. In2024 IEEE LLM Aided Design Workshop (LAD), pages 1–5, 2024

    work page 2024

  39. [39]

    Pyranet: A multi- layered hierarchical dataset for verilog

    Bardia Nadimi, Ghali Omar Boutaib, and Hao Zheng. Pyranet: A multi- layered hierarchical dataset for verilog. In2025 62nd ACM/IEEE Design Automation Conference (DAC), pages 1–7, 2025

    work page 2025

  40. [40]

    Codev: Empowering llms with hdl generation through multi- level summarization.IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems, 45(4):1893–1906, 2026

    Yang Zhao, Di Huang, Chongxiao Li, Pengwei Jin, Muxin Song, Yinan Xu, Ziyuan Nan, Mingju Gao, Tianyun Ma, Lei Qi, Yansong Pan, Zhenxing Zhang, Rui Zhang, Xishan Zhang, Zidong Du, Qi Guo, and Xing Hu. Codev: Empowering llms with hdl generation through multi- level summarization.IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems, ...

    work page 1906

  41. [41]

    Verimind: Agentic llm for automated verilog generation with a novel evaluation metric, 2025

    Bardia Nadimi, Ghali Omar Boutaib, and Hao Zheng. Verimind: Agentic llm for automated verilog generation with a novel evaluation metric, 2025

    work page 2025

  42. [42]

    Specgen: Au- tomated generation of formal program specifications via large language models

    Lezhi Ma, Shangqing Liu, Yi Li, Xiaofei Xie, and Lei Bu. Specgen: Au- tomated generation of formal program specifications via large language models. InProceedings of the IEEE/ACM 47th International Conference on Software Engineering, ICSE ’25, page 16–28. IEEE Press, 2025

    work page 2025

  43. [43]

    Nl2ctl: Automatic generation of formal requirements specifications via large language models

    Mengyan Zhao, Ran Tao, Yanhong Huang, Jianqi Shi, Shengchao Qin, and Yang Yang. Nl2ctl: Automatic generation of formal requirements specifications via large language models. In Kazuhiro Ogata, Dominique Mery, Meng Sun, and Shaoying Liu, editors,Formal Methods and Software Engineering, pages 1–17, Singapore, 2024. Springer Nature Singapore

    work page 2024

  44. [44]

    System-on-chip message flow mining with masked-language models

    Md Rubel Ahmed, Bardia Nadimi, and Hao Zheng. System-on-chip message flow mining with masked-language models. In2023 IEEE 66th International Midwest Symposium on Circuits and Systems (MWSCAS), pages 496–500, 2023

    work page 2023

  45. [45]

    Tb or not tb: Coverage-driven direct preference optimization for verilog stimulus generation, 2025

    Bardia Nadimi, Khashayar Filom, Deming Chen, and Hao Zheng. Tb or not tb: Coverage-driven direct preference optimization for verilog stimulus generation, 2025

    work page 2025

  46. [46]

    Llms for hardware verification: Frameworks, techniques, and future directions

    Khushboo Qayyum, Sallar Ahmadi-Pour, Chandan Kumar Jha, Muham- mad Hassan, and Rolf Drechsler. Llms for hardware verification: Frameworks, techniques, and future directions. In2024 IEEE 33rd Asian Test Symposium (ATS), pages 1–6, 2024

    work page 2024

  47. [47]

    Blumer, A

    A. Blumer, A. Ehrenfeucht, D. Haussler, and M. K. Warmuth. Occam’s razor.Information Processing Letters, 24(6), 1987. Bardia Nadimireceived the B.S. degree in computer-hardware engineering and the M.S. degree in computer system architecture both from Shahid Beheshti University, Tehran, Iran, in 2017, and 2020 respectively. He is currently a Ph.D. candidat...

    work page 1987