"An engineer can tell an artificial intelligence (AI) system, 'I'm going to bed now. Finish this job.' It's like saying, 'Verify it, debug it, and if there are problems, fix them. Don't ask me questions—solve it yourself.' It's like assigning an overnight task to a subordinate."
Anchor Gupta, senior vice president of Siemens Electronic Design Automation (EDA) integrated circuit (IC) products, said this at the Siemens EDA Forum Seoul 2026 held at Lotte Hotel Jamsil in Seoul on the 11th. EDA is software that designs semiconductor circuits and verifies errors and performance.
Gupta, the senior vice president, picked "the ability to work on its own" as the biggest change from combining AI and EDA. He said AI is evolving from an assistive tool into an "agent" that actually performs tasks. In Siemens EDA, when you give it a goal, multiple AI agents divide roles to carry out design and verification, and the results are checked again by an engine that reflects the physical laws of semiconductors.
◇ From Conversational AI to overnight verification agents
Gupta, the senior vice president, said, "Customers now expect new features every few months, not every few years." Siemens released Fuse in Jun. last year, which lets users converse in natural language with AI to use EDA tools. Fuse is a platform that links AI models with Siemens EDA software. In Mar. this year, it applied a single AI agent, and last month it added a feature in which multiple agents collaborate, work for long hours, and even verify results on their own.
With these features added, the "cell characterization" task of calculating the power and speed of standard cells by temperature, voltage, and process conditions has also been automated. An AI agent finds the necessary files and tools to compute and verify, and if problems arise, it repeats debugging (analyzing error causes) and fixes. Gupta, the senior vice president, said, "People don't intervene in the middle," adding, "You can run the work with a single instruction: characterize, verify, debug, and if there's an issue, fix it."
Results produced by AI are reverified by an EDA engine that reflects mathematical algorithms and the physical laws of semiconductors. Gupta, the senior vice president, said, "In semiconductors, even a small error can lead to chip failure or lower Production yield," adding, "The most important task is to provide reliable results."
The more complex semiconductors become, the heavier the verification burden grows. Siemens assessed that with designs integrating more than 200 billion transistors and thermal densities exceeding 100W/㎠, it has become difficult to keep pace using only conventional methods. The expansion of "heterogeneous integration," which packs chips made with different types and processes into a single package, also increases the complexity of design and verification. Siemens projected that semiconductors with AI capabilities will surge, with 70% of all chips equipped with AI compute acceleration by 2028.
According to Siemens, using the Questa for functional verification together with Tessent for semiconductor testing can increase automatic test pattern generation (ATPG—technology that automatically generates test patterns to find semiconductor defects) simulation speeds by three to five times over conventional methods.
◇ Collaborating with Nvidia, while keeping AI models open
Siemens chose Nvidia as a key partner for advancing AI EDA. Gupta, the senior vice president, said, "Nvidia is a GPU manufacturer and has long built CUDA (a software platform for GPU computing), and it has the foundation for EDA vendors to port code and link it to AI agent systems."
Siemens uses Nvidia's open AI model Nemotron while also enabling Fuse to connect AI models from OpenAI, Google, and Anthropic. Gupta, the senior vice president, said, "In a rapidly changing AI ecosystem, relying on a single model will slow innovation," adding, "From the start, we designed it so customers can bring and use the models they want."
Siemens put forward this open architecture and its verification EDA engine as its competitive edge. On differences with rivals such as Synopsys and Cadence, Gupta, the senior vice president, said, "In the agent era, multiple agents can converse with different tools, reducing dependence on a single platform," adding, "Our top priority is to provide a high-quality EDA engine."
German company Siemens has steadily built up its software capabilities by investing more than 25 billion euros (about 4.081 trillion won) since 2007. In 2017, it significantly expanded its business by acquiring U.S. EDA firm Mentor Graphics.
Siemens' fiscal year 2026 third-quarter (Apr.–Jun.) revenue was 20.794 billion euros (about 3.394 trillion won), up 7% from a year earlier. Industrial business profit—an indicator Siemens uses to show the profitability of its main industrial divisions—rose 25% to 3.524 billion euros (about 575 billion won). Digital industries revenue, which includes EDA, increased 12% to 4.932 billion euros (about 805 billion won), and business profit rose 44% to 923 million euros (about 151 billion won).
Gupta, the senior vice president, said the EDA business grew more than 30% in the recent quarter, adding, "Many customers are Korean corporations." He added, "Korea is a leading market not only in memory but also in AI chip and accelerator design," and said, "We will work with Korean customers to help develop the products needed in the AI era more quickly."