mlsanityData to 7 Oct 2026

BEHAVE: Functional Behavior Modeling Enables Self-Improving Agents for Hardware Design and Verification

arXiv 2609.34785 ↗First posted 28 Sep 2026cs.AI cs.AR cs.LGcited by 0

On the map: Synthetic data and self-improvement for LLMs → Self-evolving LLM agents

Abstract (arXiv)

Developing agents for hardware design and verification requires reliable correctness feedback. As a hardware specification may permit correct implementations with different latencies, matching design and reference outputs cycle by cycle can reject valid designs. To address this, we introduce BEHAVE, an agentic framework for multi-turn joint hardware design and verification through functional behavior modeling. We define Behavior IR to express task functionality as executable behavior models without prescribing implementation timing beyond the specification. The agent iteratively develops a register-transfer-level (RTL) design and a behavior model as the design's verification reference. Our evaluator, BEHAVE-Sim, checks both artifacts separately against a hidden golden behavior model using input stimuli generated by random sampling and solver-guided search. BEHAVE thus supports power, performance, and area (PPA) exploration across task-permitted latencies and microarchitectures. During training, the same evaluator provides verifiable reinforcement learning (RL) rewards from specification-behavior pairs without reference RTL. For self-improvement, the agent continually searches for high-level implementations relevant to its capability gaps, constructs and checks specification-behavior pairs, and trains on the expanded task pool. We release BEHAVE-Train and BEHAVE-Eval with 600 human-reviewed specification-behavior pairs for realistic hardware workloads. Starting from 60 seed tasks and acquiring 100 new tasks, self-improvement raises Qwen3.8-27B's RTL pass@1 on BEHAVE-Eval from 55.0% to 75.0%, reaching performance comparable to RL using a 540-task pool.

Key facts. No citing papers in the map yet (data through 7 Oct 2026). Placed in Self-evolving LLM agents, one of 13 directions in Synthetic data and self-improvement for LLMs. Builds on 51 of its 131 references that are in the map. First posted on arXiv 28 Sep 2026.

Builds on

51 of 131 references.

Closest ancestors · 1

Same direction, newest first

  1. 2026Dr. RTL: Autonomous Agentic RTL Optimization through Tool-Grounded Self-Improvement16 Apr

Other references · 50

  1. 2024DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Modelscited by 8,851
  2. 2023ChipNeMo: Domain-Adapted LLMs for Chip Designcited by 168
  3. 2025Comprehensive Verilog Design Problems: A Next-Generation Benchmark Dataset for Evaluating Large Language Models and Agents on RTL Design and Verificationcited by 41
  4. 2026CUDA Agent: Large-Scale Agentic RL for High-Performance CUDA Kernel Generationcited by 41
  5. 2026Agent-World: Scaling Real-World Environment Synthesis for Evolving General Agent Intelligencecited by 35
  6. 2025Socratic-Zero : Bootstrapping Reasoning via Data-Free Agent Co-evolutioncited by 24
  7. 2023Using LLMs to Facilitate Formal Verification of RTLcited by 17
  8. 2024LLM-Aided Testbench Generation and Bug Detection for Finite-State Machinescited by 15
  9. 2026TTCS: Test-Time Curriculum Synthesis for Self-Evolvingcited by 15
  10. 2025Insights from Verification: Training a Verilog Generation LLM with Reinforcement Learning with Testbench Feedbackcited by 14
  11. 2025RealBench: Benchmarking Verilog Generation Models with Real-World IP Designscited by 14
  12. 2026ACE-RTL: When Agentic Context Evolution Meets RTL-Specialized LLMscited by 12
38 more
  1. 2026POET: Power-Oriented Evolutionary Tuning for LLM-Based RTL PPA Optimizationcited by 11
  2. 2025MCP4EDA: LLM-Powered Model Context Protocol RTL-to-GDSII Automation with Backend Aware Synthesis Optimizationcited by 9
  3. 2026OpenSkill: Open-World Self-Evolution for LLM Agentscited by 9
  4. 2025VerilogDB: The Largest, Highest-Quality Dataset with a Preprocessing Framework for LLM-based RTL Generationcited by 8
  5. 2026EvolVE: Evolutionary Search for LLM-based Verilog Generation and Optimizationcited by 7
  6. 2026ChipBench: A Next-Step Benchmark for Evaluating LLM Performance in AI-Aided Chip Designcited by 7
  7. 2025PRO-V-R1: Reasoning Enhanced Programming Agent for RTL Verificationcited by 6
  8. 2025ChatModel: Automating Reference Model Design and Verification with LLMscited by 6
  9. 2026Agentic Hardware Design as Repository-Level Code Evolutioncited by 6
  10. 2026A New Benchmark for the Appropriate Evaluation of RTL Code Optimizationcited by 5
  11. 2026Evolutionary System Prompt Learning for Reinforcement Learning in LLMscited by 5
  12. 2026ChatSVA: Bridging SVA Generation for Hardware Verification via Task-Specific LLMscited by 5
  13. 2026SiliconMind-V1: Multi-Agent Distillation and Debug-Reasoning Workflows for Verilog Code Generationcited by 4
  14. 2026VeriAgent: A Tool-Integrated Multi-Agent System with Evolving Memory for PPA-Aware RTL Code Generationcited by 4
  15. 2026RTLSeek: Boosting the LLM-Based RTL Generation with Multi-Stage Diversity-Oriented Reinforcement Learningcited by 4
  16. 2025Pluto: A Benchmark for Evaluating Efficiency of LLM-generated Hardware Codecited by 3
  17. 2025CorrectHDL: Agentic HDL Design with LLMs Leveraging High-Level Synthesis as Referencecited by 3
  18. 2025NotSoTiny: A Large, Living Benchmark for RTL Code Generationcited by 3
  19. 2026CktEvo: Repository-Level RTL Code Benchmark for Design Evolutioncited by 3
  20. 2026UVMarvel: an Automated LLM-aided UVM Machine for Subsystem-level RTL Verificationcited by 3
  21. 2025EARL: Entropy-Aware RL Alignment of LLMs for Reliable RTL Code Generationcited by 2
  22. 2026LLM-FSM: Scaling Large Language Models for Finite-State Reasoning in RTL Code Generationcited by 2
  23. 2026SpecLoop: An Agentic RTL-to-Specification Framework with Formal Verification Feedback Loopcited by 2
  24. 2026QiMeng-CodeV-SVA: Training Specialized LLMs for Hardware Assertion Generation via RTL-Grounded Bidirectional Data Synthesiscited by 2
  25. 2026CHIA: An open-source framework for principled, agentic AI-driven hardware/software co-design researchcited by 2
  26. 2025Faver: Boosting LLM-based RTL Generation with Function Abstracted Verifiable Middlewarecited by 1
  27. 2026FormalRTL: Verified RTL Synthesis at Scalecited by 1
  28. 2026AutoVeriFix+: High-Correctness RTL Generation via Trace-Aware Causal Fix and Semantic Redundancy Pruningcited by 1
  29. 2026Exploring the Agentic Frontier of Verilog Code Generationcited by 1
  30. 2026DRTriton: Large-Scale Synthetic Data Driven Reinforcement Learning for Triton Kernel Generationcited by 1
  31. 2026VeriGraphi: A Multi-Agent Framework of Hierarchical RTL Generation for Large Hardware Designscited by 1
  32. 2026RefEvo: Agentic Design with Co-Evolutionary Verification for Agile Reference Model Generationcited by 1
  33. 2026ChipMATE: Multi-Agent Training via Reinforcement Learning for Enhanced RTL Generationcited by 1
  34. 2026Evidence-Driven LLM Agent for C-to-Synthesizable-C Conversion and Verificationcited by 1
  35. 2026GoGoTB: Agentic RTL Verification with Specification-Grounded Coverage Closurecited by 1
  36. 2026ARES: Adaptive Reasoning-Effort Steering for PPA- and Cost-Aware RTL Optimization with LLM Agentscited by 1
  37. 2026HINT: Toward an Executable Hardware-Intent Representation Layer for LLM-Driven RTL Generationcited by 1
  38. 2026LLM-based Hardware Development with Hierarchical IRs and End-to-End Multi-Agent Workflowcited by 1

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2025DUET: Agentic Design Understanding via Experimentation and Testing
  2. 2026CovR: Coverage-Aware Hardware Verification via Reasoning-Guided Reinforcement Learning
  3. 2026From Pragmas to Partners: A Symbiotic Evolution of Agentic High-Level Synthesis
  4. 2026CLOSER-Bench: Evaluating Budgeted Cross-Stage Design Closure for Hardware Agents
  5. 2026AgentDV: Closed-Loop Agentic AI for Hardware Design Verification
  6. 2026A Survey on LLM-Integrated Hardware Design Verification
  7. 2026SeqFeed: Improving Agentic RTL Code Generation with Sequential Behavior Feedback
  8. 2026Beyond Scaling: Self-Evolving LLM Agents for Hardware Kernel Optimization via an Experience-Driven Workflow and Experience Graph Memory