Pith. sign in

REVIEW 66 cited by

ChipNeMo: Domain-Adapted LLMs for Chip Design

Not yet reviewed by Pith; the record is open.

Re-run · record.json Download PDF Read on arXiv ↗

This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.

SPECIMEN: schema-true, not a live event

T0 review · schema-true

One-sentence machine reading of the paper's core claim.

pith:XXXXXXXX · record.json · timestamp

arxiv 2311.00176 v5 pith:QTQUXNNZ submitted 2023-10-31 cs.CL

Mingjie Liu , Teodor-Dumitru Ene , Robert Kirby , Chris Cheng , Nathaniel Pinckney , Rongjian Liang , Jonah Alben , Himyanshu Anand
show 34 more authors
Sanmitra Banerjee Ismet Bayraktaroglu Bonita Bhaskaran Bryan Catanzaro Arjun Chaudhuri Sharon Clay Bill Dally Laura Dang Parikshit Deshpande Siddhanth Dhodhi Sameer Halepete Eric Hill Jiashang Hu Sumit Jain Ankit Jindal Brucek Khailany George Kokai Kishor Kunal Xiaowei Li Charley Lind Hao Liu Stuart Oberman Sujeet Omar Ghasem Pasandi Sreedhar Pratty Jonathan Raiman Ambar Sarkar Zhengjiang Shao Hanfei Sun Pratik P Suthar Varun Tej Walker Turner Kaizhe Xu Haoxing Ren
This is my paper · ORCID
classification cs.CL
keywords modelsapplicationschipdesigndomain-adaptivelanguagellmsanalysis
verification ladder T0 review T1 audit T2 compute T3 formal

Signed reviews

No signed human review yet.

0 comments
read the original abstract

ChipNeMo aims to explore the applications of large language models (LLMs) for industrial chip design. Instead of directly deploying off-the-shelf commercial or open-source LLMs, we instead adopt the following domain adaptation techniques: domain-adaptive tokenization, domain-adaptive continued pretraining, model alignment with domain-specific instructions, and domain-adapted retrieval models. We evaluate these methods on three selected LLM applications for chip design: an engineering assistant chatbot, EDA script generation, and bug summarization and analysis. Our evaluations demonstrate that domain-adaptive pretraining of language models, can lead to superior performance in domain related downstream tasks compared to their base LLaMA2 counterparts, without degradations in generic capabilities. In particular, our largest model, ChipNeMo-70B, outperforms the highly capable GPT-4 on two of our use cases, namely engineering assistant chatbot and EDA scripts generation, while exhibiting competitive performance on bug summarization and analysis. These results underscore the potential of domain-specific customization for enhancing the effectiveness of large language models in specialized applications.

Discussion (0). Continue with ORCID to comment.

Forward citations

Showing 60 of 66 Pith papers that cite this

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. See all 66 Pith citations

  1. Why Do Prefetchers Fail? Let Agents Answer

    cs.AR 2026-08 reject novelty 7.0 of 10

    An agent-driven loop that turns a prefetcher's residual misses into new sub-prefetchers yields a 17-engine design claimed to beat human designs on held-out workloads.

  2. SynAct: A Reasoning-Acting Large Language Model Agent for Adaptive Synthesis Optimization

    cs.AR 2026-08 conditional novelty 7.0 of 10

    SynAct, a closed-loop LLM agent with retrieval and latent-space Bayesian optimization, reduces average WNS to 27.03% of bootstrap synthesis across 14 designs.

  3. NetlistBench: Evaluating LLM Reliability in SPICE Netlist Recognition and Manipulation

    eess.SY 2026-08 accept novelty 7.0 of 10

    Using a deterministic canonical-circuit oracle, NetlistBench finds that LLM accuracy on SPICE netlist tasks drops sharply as structural complexity and edit horizon increase.

  4. PCBWorld: A Benchmark Environment for Engine-Grounded PCB Design Automation

    cs.AI 2026-07 conditional novelty 7.0 of 10

    An open-source KiCad-grounded PCB routing environment and benchmark where agents interactively route boards via engine-native operations, with a PPO policy trained on synthetic boards achieving zero-shot transfer to r...

  5. SABLE: An NDA-Safe Closed-Loop LLM Framework for Analog Circuit Optimization in Industrial EDA Flows

    cs.AR 2026-07 conditional novelty 7.0 of 10

    An NDA-safe scrubbed boundary lets cloud LLMs optimize analog circuits in real Cadence flows; multi-model PVT benchmarks show successful closure on LC-VCO (7/11) and two-stage op-amp (4/11) tasks.

  6. Bridging the Last Mile of Circuit Design: PostEDA-Bench, a Hierarchical Benchmark for PPA Convergence and DRC Fixing

    cs.AR 2026-05 unverdicted novelty 7.0 of 10

    PostEDA-Bench shows LLM agents succeed reasonably on basic DRC and single-objective PPA tasks but struggle on practical DRC reasoning (best 36.66% success) and multi-objective PPA (best 20% success).

  7. FVRuleLearner: Operator-Level Reasoning Tree (Op-Tree)-Based Rules Learning for Formal Verification

    cs.AR 2026-03 unverdicted novelty 7.0 of 10

    FVRuleLearner retrieves learned operator-level reasoning rules to boost the functional correctness of LLM-generated SystemVerilog assertions by roughly 30 percentage points over simple prompting baselines.

  8. Quantum Circuit Vision: Cost-Aware Evaluation of Visual AI Agents for Quantum Code Generation

    quant-ph 2026-07 conditional novelty 6.5 of 10

    On a 132-circuit visual quantum-to-code benchmark, Claude Sonnet matches Opus accuracy at ~18% cost, depth predicts failure better than qubit count, CoT does not help, and cascade routing reaches 84% accuracy at 38% cost.

  9. AgenticECO: An Agentic Framework for ECO on 3D Integrated Circuits

    cs.AI 2026-08 conditional novelty 6.0 of 10

    An agentic framework with a minimal-disturbance router and independent verifier clears post-route hybrid-bond spacing defects in 3D-IC designs (7/9 on one backbone, 9/9 on another) with low disturbance and zero clock-...

  10. VeriTrace: Human-Like Temporal Exploration Completes Agentic Action Space

    cs.AI 2026-08 conditional novelty 6.0 of 10

    VeriTrace, a multi-agent LLM system whose Inspector agent controls signal and time-window selection during debugging, reports 100% functional correctness on VerilogEval-V2, the first reported perfect score on that benchmark.

  11. PICopilot: An LLM-based Agentic Framework for Assisting Photonic Integrated Circuit Design via Script Generation

    cs.ET 2026-08 conditional novelty 6.0 of 10

    A training-free multi-agent LLM framework with summary-based hybrid retrieval generates correct PIC design tool scripts from natural language, passing all 48 self-built benchmark tasks where the best baseline passes 27.

  12. FinHardBench: Can LLMs Generate Latency-Aware Hardware for Financial Computing?

    cs.CL 2026-08 conditional novelty 6.0 of 10

    FinHardBench, 33 financial FPGA tasks, finds LLMs pass functional tests 19-61% of the time and produce many routed-but-wrong designs, while top models tune a 6-stage pipeline to optimal latency more reliably than thre...

  13. ARES: Adaptive Reasoning-Effort Steering for PPA- and Cost-Aware RTL Optimization with LLM Agents

    cs.AR 2026-07 conditional novelty 6.0 of 10

    Escalating an LLM agent's reasoning effort only when cheap edits stall improves RTL power-area-delay by 23–27% versus 16–23% for the best fixed effort, at equal dollar cost.

  14. ContractHIL-HLS: Contract-Aligned Multi-Agent Workflow with Hardware-in-the-Loop Feedback for HLS Design

    cs.AI 2026-07 conditional novelty 6.0 of 10

    A contract-aligned multi-agent workflow with hardware-in-the-loop feedback raises HLS-Eval testbench pass@1 from 64.0% to 70.4% and enables a dual-bitstream PQC accelerator with 4x faster runtime.

  15. SCALE: Self-Supervised Constraint-Aware Layout GEneration for Local P&R DRV Fixing at Advanced Nodes

    cs.CV 2026-07 conditional novelty 6.0 of 10

    Self-supervised generation of DRC-annotated layouts plus VLM fine-tuning lifts agent solve rates for local DRV fixing from 63–85% to 88–97% on 100 real sub-2nm cases.

  16. DRC-Aid: Design-Rule Correction via Agentic Framework utilizing Inference-Time Large Language Models

    cs.AR 2026-07 conditional novelty 6.0 of 10

    A closed-loop agentic framework using LLM action selection over a rule-engine menu repairs ~92.5% of DRC-violating FreePDK45 layouts to DRC-clean, LVS-equivalent states.

  17. Symbol and Footprint Database for Electronic Components by Agentic Recognition and Generation

    cs.AI 2026-07 conditional novelty 6.0 of 10

    An MLLM-driven agentic pipeline generates PCB component symbols and footprints from datasheets with reported 86%/80% accuracy and builds a 1,000-component library.

  18. RF-Agent: A Practical Framework for Building Language Agents for RFIC Design

    cs.CL 2026-07 conditional novelty 6.0 of 10

    A multi-agent LLM pipeline distills seven RF textbooks into an 11k-sample QTSA dataset and benchmark, with SFT and RAG experiments showing accuracy gains on that benchmark.

  19. MAGE: Human-Like Macro Placement via Agentic Multimodal Reasoning

    cs.AI 2026-07 conditional novelty 6.0 of 10

    A multi-agent vision-language framework for chip macro placement improves post-route timing (WNS/TNS) over commercial, human, and Hier-RTLMP baselines while trading higher wirelength, and introduces four human-likenes...

  20. Retrieve, Schedule, Reflect: LLM Agents for Chip QoR Optimization

    cs.AR 2026-03 conditional novelty 6.0 of 10

    An LLM scheduler using tree-structured retrieval and Pareto-based reflection automates chip QoR optimization, matching human experts and beating an RL baseline on eight designs.

  21. MACO: A Multi-Agent LLM Framework for Automated CGRA Hardware/Software Co-Design

    cs.AR 2025-09 reject novelty 6.0 of 10

    A multi-agent LLM framework that iteratively co-designs CGRA hardware and software parameters, reporting power and performance improvements over LLM and manual baselines.

  22. GENIE-ASI: Generative Instruction and Executable Code for Analog Subcircuit Identification

    cs.AR 2025-08 conditional novelty 6.0 of 10

    A training-free LLM pipeline can generate executable Python code that identifies analog subcircuits in flattened SPICE netlists, matching rule-based labels on simple blocks and partially on complex ones.

  23. VERIRL: Boosting the LLM-based Verilog Code Generation via Reinforcement Learning

    cs.LG 2025-08 unverdicted novelty 6.0 of 10

    VERIRL applies reinforcement learning with a trace-back rescore mechanism and sample-balanced weighting to achieve state-of-the-art Verilog code generation.

  24. AI Agents for Photonic Integrated Circuit Design Automation

    cs.AR 2025-08 conditional novelty 6.0 of 10

    PhIDO, a multi-agent LLM framework, generates GDSII photonic-chip layouts from natural language and, across seven models, the best achieve about 57% end-to-end pass@5 success for designs with 15 or fewer components.

  25. MCP4EDA: LLM-Powered Model Context Protocol RTL-to-GDSII Automation with Backend Aware Synthesis Optimization

    cs.AR 2025-07 conditional novelty 6.0 of 10

    MCP4EDA is an MCP server that lets LLMs orchestrate the open-source RTL-to-GDSII flow and iteratively refine synthesis scripts from post-layout metrics.

  26. RealBench: Benchmarking Verilog Generation Models with Real-World IP Designs

    cs.LG 2025-07 conditional novelty 6.0 of 10

    RealBench measures LLM Verilog generation on complex open-source IP cores with formal verification, and all tested models score near zero on full system designs.

  27. AnalogFed: Privacy-Preserving Discovery of Analog Circuits at Scale with Federated Generative AI

    cs.LG 2025-07 reject novelty 6.0 of 10

    AnalogFed combines federated learning with a generative analog-topology model, adding dummy-token input perturbation and partial homomorphic encryption to resist membership inference and model inversion attacks.

  28. TD-Interpreter: Enhancing the Understanding of Timing Diagrams with Visual-Language Learning

    cs.LG 2025-07 reject novelty 6.0 of 10

    A fine-tuned LLaVA model, trained on synthetic timing-diagram Q&A pairs generated from Verilog simulations, answers design and verification questions about timing diagrams.

  29. MMCircuitEval: A Comprehensive Multimodal Circuit-Focused Benchmark for Evaluating LLMs

    cs.LG 2025-07 conditional novelty 6.0 of 10

    A new multimodal benchmark of 3,614 circuit QA pairs shows that large language models perform worst on back-end layout and computation tasks, and that current models generally underperform on circuit design questions.

  30. CoVE: Compressed Vocabulary Expansion Makes Better LLM-based Recommender Systems

    cs.IR 2025-06 conditional novelty 6.0 of 10

    CoVE assigns each item a unique token ID, tunes item embeddings and the LM head, and predicts the next item from logits, beating finetune-and-retrieval baselines by up to 62 percent with a 16x compressed embedding table.

  31. Controlling Context: Generative AI at Work in Integrated Circuit Design and Other High-Precision Domains

    cs.HC 2025-06 conditional novelty 6.0 of 10

    Qualitative interviews with 17 IC-design engineers show that the main difficulty with generative AI tools is not output accuracy but context control, supporting a shift toward interactive context-steering features.

  32. Spec2RTL-Agent: Automated Hardware Code Generation from Complex Specifications Using LLM Agent Systems

    cs.AR 2025-06 conditional novelty 6.0 of 10

    On three NIST crypto standards (AES, DSS, HMAC), Spec2RTL-Agent generates RTL via a multi-agent pipeline from pseudocode to Python to synthesizable C++, reporting 3/3 correct designs with about 4.3 human interventions...

  33. JARVIS: A Multi-Agent Code Assistant for High-Quality EDA Script Generation

    cs.SE 2025-05 conditional novelty 6.0 of 10

    A multi-agent LLM framework with rule enforcement, compiler feedback, and retrieval achieves 92/93/81% pass@1 on three self-built EDA benchmarks, up from 67/62/43% for the best single model.

  34. VeriReason: Reinforcement Learning with Testbench Feedback for Reasoning-Enhanced Verilog Generation

    cs.AI 2025-05 conditional novelty 6.0 of 10

    VeriReason shows that fine-tuning a code model with GRPO and reward signals from simulated testbenches raises first-attempt Verilog generation accuracy by up to 2.8x and lifts a 7B model to 83.1% pass@5 on VerilogEval...

  35. RTL++: Graph-enhanced LLM for RTL Code Generation

    cs.PL 2025-05 conditional novelty 6.0 of 10

    RTL++ encodes RTL code as textualized CFG/DFG graphs to generate richer instructions for fine-tuning CodeLlama-7B, claiming improved VerilogEval pass@k over comparable open-source models.

  36. SysVCoder: An LLM-Driven Framework for Systematic Generation of System-Level Design

    cs.SE 2025-04 conditional novelty 6.0 of 10

    A two-stage LLM pipeline converts natural-language hardware specs into a structured intermediate representation and then into system-level Verilog, beating open-source Verilog generators and matching a GPT-4 agent wit...

  37. VeriCoder: Enhancing LLM-Based RTL Code Generation through Functional Correctness Validation

    cs.AR 2025-04 conditional novelty 6.0 of 10

    An LLM fine-tuned on 125,777 Verilog examples that each passed a generated unit test beats prior open-source models on functional correctness benchmarks for RTL code generation.

  38. VRank: Enhancing Verilog Code Generation from Large Language Models via Self-Consistency

    cs.AR 2025-01 conditional novelty 6.0 of 10

    VRank selects better LLM-generated Verilog code by clustering candidates that produce identical simulation outputs and ranking the clusters by consistency, yielding an average 10.5% pass@1 gain on VerilogEval-Human.

  39. Paradigm-Based Automatic HDL Code Generation Using LLMs

    cs.PL 2025-01 conditional novelty 6.0 of 10

    A paradigm-based workflow with information-list reuse and a two-phase loop improves LLM-generated Verilog pass rates on VerilogEval, with the full-dataset result built from a hybrid of baseline and proposed-method outputs.

  40. HaVen: Hallucination-Mitigated LLM for Verilog Code Generation Aligned with HDL Engineers

    cs.PL 2025-01 conditional novelty 6.0 of 10

    HaVen combines chain-of-thought translation of diagrams with knowledge and logic augmented training data to improve LLM-generated Verilog code correctness.

  41. AGON: Automated Design Framework for Customizing Processors from ISA Documents

    cs.AR 2024-12 conditional novelty 6.0 of 10

    LLM-generated nano-operator functions allow automated synthesis of customized out-of-order RISC-V processors from ISA documents, with simulated 2.35x average speedups over BOOM.

  42. MAGE: A Multi-Agent Engine for Automated RTL Code Generation

    cs.AR 2024-12 reject novelty 6.0 of 10

    MAGE, a multi-agent LLM pipeline with high-temperature sampling and state-checkpoint debugging, reports 95.7% Pass@1 on VerilogEval-V2, but the evaluation includes the golden testbench as generation input.

  43. RTL-Breaker: Assessing the Security of LLMs against Backdoor Attacks on HDL Code Generation

    cs.CR 2024-11 conditional novelty 6.0 of 10

    RTL-Breaker shows that fine-tuning an LLM on a small fraction of poisoned Verilog samples can make it generate malicious RTL code when a specific trigger appears, while still passing VerilogEval functional checks.

  44. UVLLM: An Automated Universal RTL Verification Framework using LLMs

    cs.AR 2024-11 conditional novelty 6.0 of 10

    UVLLM uses LLM-generated reference models inside UVM testbenches to localize and repair RTL errors, reporting 86.99% syntax and 71.92% functional fix rates on a new 331-instance benchmark.

  45. LLMPirate: LLMs for Black-box Hardware IP Piracy

    cs.CR 2024-11 conditional novelty 6.0 of 10

    LLMPirate uses eight LLMs to rewrite Verilog netlists into functionally equivalent circuits that evade GNN4IP, MOSS, Jplag, and SIM on most tested designs.

  46. Masala-CHAI: A Large-Scale SPICE Netlist Dataset for Analog Circuits by Harnessing AI

    cs.AR 2024-11 conditional novelty 6.0 of 10

    Masala-CHAI automatically converts schematic images into SPICE netlists using object detection, line detection, and LLMs, producing a 7,500-example open dataset that boosts analog netlist generation Pass@1 by up to 46...

  47. Towards Reliable AI-Assisted Analog Design: Template-Constrained LLM Agents for SAR ADC Generation

    cs.SE 2026-07 conditional novelty 5.0 of 10

    ATLAS combines template-constrained LLM agents with Bayesian optimization to produce SAR ADC netlists that meet user specs in simulation.

  48. VeriOpt: PPA-Aware High-Quality Verilog Generation via Multi-Role LLMs

    cs.SE 2025-07 conditional novelty 5.0 of 10

    A multi-role LLM prompting framework with PPA-aware in-context learning reports 25/29 functional correctness on RTLLM and up to 88% power, 76% area, and 73% timing gains over its own baseline.

  49. EvoVerilog: Large Langugage Model Assisted Evolution of Verilog Code

    cs.AR 2025-06 unverdicted novelty 5.0 of 10

    EvoVerilog uses multiobjective evolutionary search with LLMs to generate Verilog code, reporting higher pass@10 than prior methods on VerilogEval-Machine and VerilogEval-Human.

  50. Improving LLM-Powered EDA Assistants with RAFT

    cs.CL 2025-06 conditional novelty 5.0 of 10

    Synthetic Q/A pairs generated by DeepSeek-V3 from EDA documents, combined with RAFT fine-tuning, improve retrieval-augmented EDA assistant accuracy, with real-user few-shot examples further helping on human-style questions.

  51. Large Processor Chip Model

    cs.AR 2025-06 reject novelty 5.0 of 10

    The paper proposes a three-level LLM-based framework for end-to-end computer architecture design, with a claimed 3DGS case study whose quantitative results are not fully supported.

  52. TelcoLM: collecting data, adapting, and benchmarking language models for the telecommunication domain

    cs.CL 2024-12 conditional novelty 5.0 of 10

    For Llama-2-7B on telecommunications tasks, instruction tuning on a telco-generated dataset suffices; continuing pretraining on raw telco text adds little (max +0.03 accuracy).

  53. AiEDA: Agentic AI Design Framework for Digital ASIC System Design

    cs.AR 2024-12 reject novelty 5.0 of 10

    AiEDA combines LLM agents with open-source EDA tools in a four-stage concept-to-GDSII flow, but it has not yet demonstrated the full flow on its KWS case study.

  54. SuperCode: Sustainability PER AI-driven CO-DEsign

    astro-ph.IM 2024-12 unverdicted novelty 5.0 of 10

    The paper proposes an AI-driven hardware-software-science co-design methodology for radio astronomy, using sustainability as the key performance indicator, with no empirical results yet.

  55. On the effective transfer of knowledge from English to Hindi Wikipedia

    cs.CL 2024-12 conditional novelty 5.0 of 10

    A retrieval, neutralization, and machine-translation pipeline can add relevant factual text to Hindi Wikipedia biography sections, but the claimed 65% and 62% gains are not fully supported by the reported evaluation.

  56. EDA-Aware RTL Generation with Large Language Models

    cs.AR 2024-11 conditional novelty 5.0 of 10

    AIvril2, a multi-agent LLM framework with EDA tool feedback, raises RTL functional pass rates to 77% for Verilog and 66% for VHDL on VerilogEval-Human.

  57. Chain-of-Descriptions: Improving Code LLMs for VHDL Code Generation and Summarization

    cs.CL 2025-07 conditional novelty 4.0 of 10

    Chain-of-Descriptions, a plan-then-generate prompt strategy, improves VHDL code generation and summarization scores for eight LLMs on the VHDL-Eval and VHDL-Xform benchmarks.

  58. QiMeng: Fully Automated Hardware and Software Design for Processor Chip

    cs.AR 2025-06 conditional novelty 4.0 of 10

    QiMeng is a proposed three-layer architecture for automating processor hardware and software design, with several published components but no integrated implementation yet.

  59. Customizing a Large Language Model for VHDL Design of High-Performance Microprocessors

    cs.AR 2025-05 conditional novelty 4.0 of 10

    Extended pretraining on proprietary VHDL code and documents improved a Granite LLM's expert-rated explanations of microprocessor VHDL from 43% to 69%, and instruction tuning pushed the expected rating to 71%.

  60. ChipAlign: Instruction Alignment in Large Language Models for Chip Design via Geodesic Interpolation

    cs.AR 2024-12 conditional novelty 4.0 of 10

    Merging a chip-specialized LLM with a general chat LLM via spherical geodesic interpolation improves instruction following while preserving chip knowledge.

See all 66 Pith citations

Pith tools