《電子技術(shù)應(yīng)用》
您所在的位置:首頁(yè) > EDA與制造 > 設(shè)計(jì)應(yīng)用 > 使用Cadence AI技術(shù)加速驗(yàn)證效率提升
使用Cadence AI技術(shù)加速驗(yàn)證效率提升
電子技術(shù)應(yīng)用
徐加山1,,姚舒雨1,徐志磊2
1.深圳市中興微電子技術(shù)有限公司,;2.上??请娮涌萍加邢薰?/div>
摘要: 隨著硬件設(shè)計(jì)規(guī)模和復(fù)雜程度的不斷增加,驗(yàn)證收斂的挑戰(zhàn)難度不斷增大,,單純依靠增加 CPU 核數(shù)量并行測(cè)試的方法治標(biāo)不治本,。如何在投片前做到驗(yàn)證關(guān)鍵指標(biāo)收斂,是驗(yàn)證工程師面對(duì)的難題,。為解決這一難題,,提出了采用人工智能驅(qū)動(dòng)的驗(yàn)證EDA工具和生成式大模型兩種提效方案,其中EDA工具有Cadence利用人工智能驅(qū)動(dòng)的Verisium apps和采用機(jī)器學(xué)習(xí)技術(shù)Xcelium ML,,前者用來(lái)提升驗(yàn)證故障定位效率,,包括Verisium AutoTriage、Verisium SemanticDiff,、Verisium WaveMiner等,,后者可用來(lái)提升驗(yàn)證覆蓋率收斂效率。生成式大模型可輔助智能debug和自動(dòng)生成驗(yàn)證用例,,主要介紹各實(shí)現(xiàn)方案,,并給出了項(xiàng)目實(shí)驗(yàn)提升結(jié)果。
中圖分類號(hào):TN402 文獻(xiàn)標(biāo)志碼:A DOI: 10.16157/j.issn.0258-7998.240803
中文引用格式: 徐加山,,姚舒雨,,徐志磊. 使用Cadence AI技術(shù)加速驗(yàn)證效率提升[J]. 電子技術(shù)應(yīng)用,2024,,50(8):32-36.
英文引用格式: Xu Jiashan,,Yao Shuyu,Xu Zhilei. Accelerating verification efficiency with Cadence AI technology[J]. Application of Electronic Technique,,2024,,50(8):32-36.
Accelerating verification efficiency with Cadence AI technology
Xu Jiashan1,Yao Shuyu1,,Xu Zhilei2
1.Shenzhen Sanechips Technology Co.,, Ltd., China,;2.Cadence Design Systems,, Inc.
Abstract: With the increasing scale and complexity of hardware design, the verification convergence challenge is becoming more difficult. Simply increasing the number of CPU cores to increase parallel testing cannot solve this problem fundamentally. How to achieve verification convergence before tape-out is a difficult problem that verification engineers have to face. To solve this problem, this article proposes two efficiency improvement solutions: AI-driven verification EDA tools and large-scale model generation. The EDA tools include Cadence's AI-driven Verisium apps and Xcelium ML using machine learning technology. The former is used to improve the fault location efficiency of verification, including Verisium AutoTriage, Verisium SemanticDiff, and Verisium WaveMiner. The latter can be used to improve the verification coverage convergence efficiency. Large-scale model generation can assist intelligent debugging and automatically generate verification cases. This article mainly introduces each implementation solution and gives the project experimental improvement results.
Key words : IC verification;AI,;Verisium apps,;efficiency improvement

引言

隨著高性能和低延時(shí)等需求的增加,芯片設(shè)計(jì)復(fù)雜度越來(lái)越高,,漏洞也越來(lái)越多,,修復(fù)也變得愈加困難,,IC驗(yàn)證工程師要在規(guī)定時(shí)間內(nèi)完成所有功能驗(yàn)證工作變得日益艱巨。因此對(duì)能加速驗(yàn)證工作的工具需求也日益迫切,。引入新的工具和基于人工智能的方法,,是可以提高驗(yàn)證效率的手段。一個(gè)完整的IC驗(yàn)證流程包含驗(yàn)證需求/策略制定,、驗(yàn)證平臺(tái)搭建,、驗(yàn)證用例編寫(xiě)、仿真執(zhí)行,、debug調(diào)試,、覆蓋率收斂、驗(yàn)證報(bào)告生成等階段,。其中EDA工具在debug調(diào)試和覆蓋率收斂提效方面推出了解決方案,。Verisium(人工智能驅(qū)動(dòng)的驗(yàn)證平臺(tái))就是利用大數(shù)據(jù)提高驗(yàn)證效率,主要體現(xiàn)在debug調(diào)試方面,,它能實(shí)現(xiàn)以下功能:對(duì)存在相同錯(cuò)誤而導(dǎo)致失敗的測(cè)試進(jìn)行自動(dòng)分組,,幫助驗(yàn)證工程師在正確與錯(cuò)誤測(cè)試中更方便地比較并找到錯(cuò)誤點(diǎn),以及在編輯工具上固定某個(gè)標(biāo)簽并分析仿真日志和代碼簽入之間的關(guān)系,。Xcelium ML(Machine Learning)采用機(jī)器學(xué)習(xí)技術(shù)可實(shí)現(xiàn)功能覆蓋率快速收斂,,大大提高驗(yàn)證仿真效率,同時(shí)有效節(jié)省服務(wù)器計(jì)算資源,。另外隨著人工智能的發(fā)展,,本文也探索了生成式大模型對(duì)debug調(diào)試及用例編寫(xiě)等方面解決方案。


本文詳細(xì)內(nèi)容請(qǐng)下載:

http://forexkbc.com/resource/share/2000006118


作者信息:

徐加山1,,姚舒雨1,,徐志磊2

(1.深圳市中興微電子技術(shù)有限公司,江蘇 南京 210012,;

2.上??请娮涌萍加邢薰荆?上海200120)


Magazine.Subscription.jpg

此內(nèi)容為AET網(wǎng)站原創(chuàng),,未經(jīng)授權(quán)禁止轉(zhuǎn)載,。