The path beyond EDA copilots: building a purpose-built, customizable, secure, and open AI solution with Fuse EDA AI System

Over the past few years, AI has been pervading the semiconductor industry by being integrated into chip and board design workflows and Electronic Design Automation (EDA) tools, which are used to design new SoCs. Early machine learning (ML) and reinforcement learning (RL) solutions have already delivered significant speed improvements in EDA workflows. More recently, generative AI has introduced “copilot-like” functionalities, while agentic AI solutions, such as Model Context Protocol (MCP)-based flows, are still in early stages of implementation. Despite this progress, key challenges persist that hinder AI from realizing its full potential in EDA. Current solutions are often fragmented across siloed tools, locked into closed ecosystems, and built on.

Limitations of Current AI Approaches in EDA

Unlocking AI’s full potential in EDA hinges on addressing fundamental limitations in both established and emerging approaches (see figure 1). Traditional ML/RL approaches follow a paradigm where task-specific models developed by vendors are deployed at customer sites. While some finetuning of customer data (e.g. for telemetry predictions) is possible, these models are typically managed as separate point tools, leading to version control issues, cumbersome management overhead, and a steep learning curve for users.

Generative AI approaches face similar constraints. The current approach primarily involves utilizing an off-the-shelf large language model (LLM) and fine-tuning it on a company’s internal, text-only documents, such as PDFs and manuals. These systems, sometimes augmented by a simple retrieval augmented generation (RAG) framework, are often general-purpose AI models not optimized for the unique demands of chip and board design. These limitations hinder seamless integration and restrict customization.

A New Foundation for EDA AI Systems

This white paper examines these challenges and presents a comprehensive solution: an EDA AI system built on four foundational pillars - purpose-built for EDA, customizable, secure, and open. We introduce Siemens’ Fuse EDA AI System, which has been developed precisely to tackle the above challenges, and examine how such a system can fundamentally transform chip and board design, equipping engineers with generative and agentic AI capabilities that drive productivity and innovation across every stage of electronic design.