OPL4GPT：硬件设计最佳编程语言的应用空间探索.pdf

1、OPL4GPT:An Application Space Exploration of Optimal Programming Language for Hardware Design by LLMKimia Tasnia,Sazadur Rahman Department of Electrical and Computer Engineering,University of Central FloridaOutline Motivation Designing an SoC with LLMo Conversational flow to explore the optimal langu

2、age for hardware designo OPL4GPT Framework Modular Prompting and Profiling o One case study-RSA Cryptography Experimental Resultso Experimental Setup o Comparison between C/C+and Verilog code Conclusion2Intro:Emerging Design Automation55%55%5%70%5%70%Hardware0.3%Scripting0.8%HDL Generation55%Verific

3、ation30%Others10%Scripting5%General Programming 78%Others20.9%Survey of 120 Research PapersGitHub Code DistributionPass Rate in HDL codingFunc.SyntaxOpen Source Products Involved Models Llama,Claude,Mistral Synthesis ABC,Yosys Simulation iVerilog Benchmarks OpenCores,RTLLM Evaluation VerilogEval,Hum

4、anEvalEvaluation Metrics Syntactical Correctness Functional Correctness Synthesizability Power,Performance,Area Security and Trustability ScalabilityOSE Activities Open-source benchmark management External IP developer feedback Community user engagement Sustainable human-on-the-loop Hardware design

5、by non-STEM users Bridge gap to enter electronics industryOpenHDLOpen BenchmarkIP ContentDeveloperCommunity DrivenOrganizational GovernanceEvaluation MetricsDevelopment ToolsDistributed EffortBridge Gap to Indsutry3 LLM is showing promising results to advance automated HDL generation Existing works

6、remain applicable mostly to Verilog code generation However,LLM generated codes are falling short to meet syntax/functionality.Requires multiple iterations to debug and converge to the optimal code55%55%5%70%5%70%Hardware0.3%Scripting0.8%HDL Generation55%Verification30%Others10%Scripting5%General Pr