时间：2026年7月26日
Date: July 26, 2026
西门子携手英伟达推出自我验证AI Agent设计流程，芯片与PCB签核周期从”天”压缩至”小时”
Siemens and NVIDIA Unleash Self-Verifying AI Agents for Chip and PCB Design, Compressing Signoff from Days to Hours
摘要 :
2026年7月26日，西门子宣布与英伟达深化合作，共同推出用于电子设计自动化（EDA）的”自我验证型”Agentic AI工作流，并将Fuse EDA AI Agent系统接入Intelligence Center X，为芯片和PCB工程团队提供可持续验证的设计结果。
联合方案通过英伟达NeMo Gym库对AI Agent进行调优，提高应答质量并降低Token成本；借助OpenShell运行时提供企业级安全边界；并结合Nemotron推理模型与CUDA-X加速，将传统上耗时数天的签核流程压缩至数小时。西门子EDA首席AI战略官Amit Gupta指出，能长时间自主运行的EDA Agent必须持续用成熟的工程工具校验自身决策，才能被产线接受。除芯片版图外，能力矩阵进一步延伸至用于库特性建模的Solido Characterization Suite和用于验证的Questa One Agentic Toolkit，让设计、制造与供应链团队围绕机器可校验的共享状态协同工作。西门子将本次发布定位为Agentic AI在”零容忍幻觉”的高风险工程领域的重要拐点。
Summary :
Siemens on July 26, 2026 announced an expanded partnership with NVIDIA to deliver self-verifying agentic AI workflows for electronic design automation, plugging its Fuse EDA AI Agent system into the broader Intelligence Center X so that semiconductor and PCB engineering teams get continuously validated results.
The joint stack uses NVIDIA’s NeMo Gym library to fine-tune agents for higher answer quality and lower token cost, adds enterprise-grade guardrails through an OpenShell runtime, and taps Nemotron reasoning models plus CUDA-X acceleration to shorten what were traditionally multi-day signoff loops. Siemens’ Amit Gupta highlighted that autonomous, long-running EDA agents must continuously re-check their own decisions against proven engineering tools to be trusted in production. Beyond chip layout, the fabric extends into the Solido Characterization Suite for library modeling and the Questa One Agentic Toolkit for verification, letting design, manufacturing and supply-chain teams coordinate on shared, machine-verifiable state. Siemens positioned the announcement as an inflection point for agentic AI in high-consequence engineering domains where hallucinations cannot be tolerated.
入选理由 :
“自我验证”是关键词——这是主流厂商首批将AI Agent与强制性真值校验循环打包交付的案例之一，直击”Agent能否被工程/受监管领域采用”的最大障碍。
Why Selected:
“Self-verifying” is the key word—this is one of the first mainstream releases where an AI agent is packaged with mandatory ground-truth validation loops, addressing the single biggest blocker to agent adoption in EDA and other regulated engineering fields.
浏览量：104,200
Views：104,200
独立IP：31,800
Unique Clicks ：31,800
