New-Tech Europe | Q3 2026 | Digital Edition

EDA AI Agents: Intelligent Automation in Semiconductor & PCB Design by Niranjan Sitapure, Central AI Product Manager, Siemens EDA

The next era of semiconductor and PCB design will be defined by two parallel imperatives: making core engines faster and making engineers more productive.

need more than a chatbot; they need autonomous systems capable of intelligent reasoning, multi-step execution, and real-time adaptation across diverse EDA tools. This is the promise of agentic automation: a unified orchestration layer that delivers expert-level decision-making across the complete design lifecycle. Realizing it, however, requires overcoming domain-specific hurdles that generic AI frameworks are simply not equipped to handle. Five Core Challenges of EDA Complexity Generic, off-the-shelf AI models struggle with chip design because the industry relies on a highly specialized foundation. Any solution must seamlessly span the entire end-to-end workflow - from initial concept to manufacturing sign-off - automating critical tasks across front-end design, verification, physical implementation, PCB sign-off, and manufacturing readiness to serve as a true unified intelligence layer. To effectively deploy agentic AI in EDA, developers must address five distinct challenges: 1. Proprietary Chip Design Expertise: Chip design relies on physics-based methodologies absent from public training data. Generic agents lack the specific expertise needed to configure specialized tools, orchestrate sequences, or generate precise production code. 2. Rigid EDA Environments and Data Flows: EDA relies on secure, on-premise clusters rather than fast cloud

On the engine side, the industry is embedding machine learning and reinforcement learning directly into EDA tools - enabling, for example, local models built from a small set of SPICE simulations to dramatically accelerate verification while maintaining near-SPICE accuracy. Simultaneously, leading EDA vendors are partnering with hardware companies such as NVIDIA to GPU-accelerate core algorithms, unlocking vastly higher throughput across simulation, design exploration, coverage analysis, and OPC. Addressing the second imperative of engineering productivity demands a fundamentally different kind of AI solution. For faster engineers, generative EDA AI copilots were the industry’s first answer - but they are no longer sufficient. As design complexity and tool fragmentation accelerate, manual scripts and isolated point solutions fail to scale. Engineers

28 l New-Tech Magazine Europe