Siemens Deploys Agentic AI for Chip Design

The Siemens Questa One Agentic Toolkit represents a strategic move to address the growing productivity gap in integrated circuit verification, a challenge exacerbated by rising chip complexity and a shortage of skilled engineers. The toolkit employs AI agents to automate and accelerate workflows, integrating with Siemens' broader Fuse EDA AI system to provide a context-aware, natural language interface across the entire electronic design automation (EDA) workflow. This shift from assistive automation to more autonomous, goal-driven agents is a key evolution in the EDA industry. Under the hood, the toolkit leverages technologies from a partnership with Nvidia, specifically the Llama Nemotron and NIM models, to power its AI-driven workflows. It features five distinct AI agents: an RTL Code Agent for generating synthesizable code, a Lint Agent for error checking, a CDC Agent for clock domain crossing verification, a Verification Planning Agent, and a Debug Agent for failure analysis. Early adopters like MediaTek have reported that engineers can become proficient with the toolkit within hours, indicating a focus on developer experience. Siemens' strategy extends beyond verification, with generative and agentic AI being integrated across its portfolio, including the Solido custom IC platform and the Aprisa digital implementation workflow. This creates a centralized multimodal EDA data lake, designed to break down data silos between design teams and create a data flywheel effect for continuous AI learning. The system is designed for both on-premise and cloud deployment, featuring enterprise-grade security and custom access controls to protect intellectual property. The move comes as the EDA market, an oligopoly dominated by Synopsys and Cadence with roughly 30% market share each, sees all major players investing heavily in AI. Siemens holds about 13% of the market and is actively forming alliances, such as with Samsung and Nvidia, to increase its share. The industry is at a critical inflection point, with agentic AI seen as essential to continuing the productivity trends at the heart of Moore's Law. A significant challenge for agentic AI in chip design is that the underlying AI models are often trained on open-domain data, not the highly specific data of the semiconductor industry. Furthermore, much of the data produced by current EDA tools is trapped in human-readable logs and tool-specific databases, which are not structured for machine parsing. Effective AI governance is therefore critical, establishing clear frameworks for model behavior, data handling, and human oversight to manage risks associated with automated decision-making in such a precise and high-stakes environment.