Synopsys brings autonomous chip-design workflows to Microsoft Discovery

Synopsys on July 27 announced that its autonomous debugging and physical implementation workflows for chip design are available for evaluation through Microsoft Discovery, with AMD testing their use on future products. The workflows, built around Synopsys's AgentEngineer framework, aim to reduce debug cycle time by 25% to 40% and improve implementation results.

Ravi Subramanian https://www.synopsys.com/content/dam/synopsys/company/about/bio-pdfs/ravi-subramanian-bio-2025.pdf?ref=runtimewire , a two-time semiconductor and chip-design founder, now serves as Synopsys https://www.synopsys.com/?ref=runtimewire 's chief product management officer. Synopsys said in a July 27 announcement https://www.prnewswire.com/news-releases/synopsys-advances-agentic-ai-chip-design-with-amd-and-microsoft-302834852.html?ref=runtimewire that its autonomous debugging and physical implementation workflows are available for evaluation through Microsoft Discovery https://azure.microsoft.com/en-us/solutions/discovery/?ref=runtimewire , with AMD https://www.amd.com/en/blogs/2026/how-agentic-ai-is-reshaping-chip-design.html?ref=runtimewire testing their use on future products. According to his Synopsys biography https://www.synopsys.com/content/dam/synopsys/company/about/bio-pdfs/ravi-subramanian-bio-2025.pdf?ref=runtimewire , Subramanian joined Synopsys in 2022 after serving as senior vice president and general manager of IC verification at Siemens EDA. He founded MorphICs Technologies, a fabless wireless semiconductor startup acquired by Infineon, and Berkeley Design Automation, an analog verification software developer acquired by Mentor Graphics, now part of Siemens. That path helps explain the focus of Synopsys's latest product move. Subramanian has spent much of his career around verification, where engineering teams can burn weeks sorting failures, tracing root causes and rerunning tools before a design is ready for manufacturing. Synopsys is trying to turn that iterative work into a job that specialized agents can execute with less human intervention. From assistant to operator The first workflow targets debug closure. Synopsys says domain-specific and task-level agents can identify design failures, conduct root-cause analysis and automate parts of validation. Early evaluations produced a claimed 25% to 40% reduction in debug cycle time, according to Synopsys. The second workflow applies agents to implementation and closure, the process of converting a chip design into a physical layout that meets requirements for timing, power, area and manufacturability. It uses Synopsys implementation agents and Fusion Compiler https://www.synopsys.com/implementation-and-signoff/physical-implementation/fusion-compiler.html?ref=runtimewire on Azure to tune quality of results and iterate toward closure. Synopsys says initial tests improved those results, though it did not publish a numerical comparison. Both workflows are built around AgentEngineer https://www.synopsys.com/blogs/chip-design/agentengineer-technology-transforming-engineering-workflows.html?ref=runtimewire , Synopsys's framework for agents that can plan work, invoke engineering tools, evaluate intermediate output and continue running over extended periods. Closure requires repeated decisions across many tool runs, with each result determining the next action. Microsoft made Discovery generally available on June 2 https://azure.microsoft.com/en-us/blog/announcing-microsoft-discovery-general-availability-and-microsoft-discovery-app-preview/?ref=runtimewire and positions it as an enterprise platform for agentic AI workflows in scientific and engineering work. Synopsys and Microsoft say the two applications are the first EDA workflows offered for evaluation on Discovery. Synopsys says customers can request evaluation access through Synopsys https://investor.synopsys.com/news/news-details/2026/Synopsys-Advances-Agentic-AI-Chip-Design-with-AMD-and-Microsoft/default.aspx?ref=runtimewire . Neither pricing nor a date for broad commercial availability was disclosed. AMD supplies the real-world test AMD gives Synopsys a semiconductor engineering partner for testing the workflows against commercial chip-development complexity. Synopsys reports that AMD is actively evaluating the applications; it does not claim production deployment across design programs. Alex Starr https://www.prnewswire.com/news-releases/synopsys-advances-agentic-ai-chip-design-with-amd-and-microsoft-302834852.html?ref=runtimewire , an AMD Corporate Fellow who has worked at AMD for more than 20 years and leads Applied AI and Shift Left AI at AMD, gives the evaluation a real chip-design setting. In Synopsys's July 27 announcement, Starr said AI-driven workflows powered by Discovery and RCA with deep EDA domain knowledge are a paradigm AMD sees "accelerating scaling and deployment, helping improve both design velocity and overall silicon quality." That wording keeps human engineers in the loop. Agents still need access controls, workflow design, trusted engineering tools and clear points for human review. A fast wrong answer can carry manufacturing-scale consequences when the output becomes silicon. Synopsys connects AgentEngineer to Microsoft and Nvidia platforms The Microsoft collaboration is one part of a wider partner strategy. On July 26, Synopsys separately introduced autonomous engineering workflows built with Nvidia technology https://investor.synopsys.com/news/news-details/2026/Synopsys-Showcases-Comprehensive-Autonomous-Engineering-Workflows-from-Silicon-to-Systems-Developed-with-NVIDIA-Technology/default.aspx?ref=runtimewire , using Nvidia Nemotron models and the Nvidia OpenShell runtime. The two announcements place AgentEngineer above infrastructure supplied by different partners. Microsoft supplies Discovery and Azure infrastructure for the workflows announced July 27. Synopsys's July 26 release says its Nvidia-based workflows use Nvidia Nemotron on Nvidia accelerated computing and are secured by the Nvidia OpenShell runtime. In its July 27 announcement, Synopsys said it is advancing an "open, interoperable agentic AI stack." The Microsoft announcement does not identify supported third-party EDA tools, outside models or customer-controlled orchestration systems. Cadence and Siemens are also building autonomous EDA agents Cadence said on June 1 https://www.cadence.com/en US/home/company/newsroom/press-releases/pr/2026/cadence-unveils-industrys-first-fully-autonomous-virtual.html?ref=runtimewire that it had developed a fully autonomous version of its ChipStack AI Super Agent. Cadence says the system iterates across specification understanding, RTL generation, verification planning, formal analysis, simulation, debugging and design convergence. It expects ChipStack's autonomous capabilities and its AgentStack orchestration framework to reach early-access customers in the second half of 2026. Siemens announced Fuse EDA AI Agent on March 16 https://news.siemens.com/en-gb/siemens-fuse-eda-ai-agent/?ref=runtimewire for semiconductor, 3D IC and circuit-board workflows spanning design, verification and manufacturing sign-off. On July 26, Siemens expanded Fuse https://news.siemens.com/en-us/siemens-nvidia-dac-2026/?ref=runtimewire with agents that continuously check their decisions against deterministic, physics-based EDA engines. Cadence and Siemens are building parallel autonomous EDA-agent products. That narrows Synopsys's claim here to Microsoft Discovery availability and AMD's evaluation, rather than autonomous EDA broadly. The 40% claim needs a production baseline Synopsys's 25% to 40% debug cycle-time reduction is the announcement's central performance figure, and the least explained. Synopsys describes the results as early evaluations without naming the designs tested, the number of trials, the previous workflow, the engineering hours involved or the share of the process that remained under human supervision. The results have not been presented as an independent benchmark. Even a smaller sustained reduction could carry weight. Verification and closure consume engineering time late in a chip program, when unresolved failures can threaten tapeout schedules and delay revenue. The commercial question is whether AgentEngineer can reproduce its early results across different designs, process nodes and customer tool chains while protecting sensitive chip data. Subramanian's startup career was built around specialized semiconductor problems that general-purpose software could not solve. At Synopsys, he can apply the same thesis across a much larger engineering portfolio: pair specialized chip-design tools and data with models and compute supplied by infrastructure partners. For founders and operators building vertical agents, access to domain tools and proprietary data may form the most defensible part of the product. For now, Synopsys has an evaluation program, AMD's participation and an early performance range. Production-scale evidence across multiple customers remains the missing proof.