Cadence launches AuraStack AI agent for PCB and packaging workflows

Cadence Design Systems announced AuraStack on Wednesday, an agentic AI system for electrical engineers designing printed circuit boards and advanced packages. Cadence did not disclose pricing, contract value, customer count or revenue targets, but said Nvidia and other large electronics companies have signed up.

The product is aimed at a practical gap in engineering software: many board and packaging workflows already rely on automation, yet engineers still spend substantial time setting up, sequencing and interpreting many small tasks around simulation and verification. Cadence says AuraStack uses AI as a control layer for its existing tools, rather than as a replacement for the high-precision simulation engines used in reliability and physics-heavy design work.

Michael Jackson, corporate vice president of Cadence’s system design and analysis division, told The Register that “AI is amplifying the value of our engineering products and technologies.” According to Cadence, AuraStack can work with a range of open and proprietary AI models and can take natural-language instructions that it turns into multi-step workflows across Cadence’s test and simulation suites.

AI as workflow controller

The distinction matters for EDA buyers. PCB design and advanced packaging are areas where wrong answers can turn into expensive respins, and the relevant workloads often depend on single- or double-precision numerical methods. AuraStack’s role, as described by Cadence, is to plan and coordinate the steps that feed those simulations, which then run on CPUs, GPUs and other accelerators.

Jackson gave an example involving IR reliability checks. In that workflow, an engineer has to identify power-management components, build a simulation-ready power tree, run the simulation and return feedback to the designer. Cadence says parts of those processes are already automated in its software, but the volume of work across a board or package project remains high.

Jackson said engineers spend 65% of their day dealing with such tasks, and claimed AuraStack can improve productivity by 15 times by taking over coordination work. Cadence did not provide a detailed benchmark basis for that productivity figure in the details reported, so buyers will have to evaluate whether the gains hold across their own design flows.

Part of a broader agent push in engineering software

AuraStack is not Cadence’s first agent effort. The company has also built AI agents for digital and analog chip design, according to Cadence. The new system extends that pattern into PCB and advanced-packaging work, where the design process depends on many linked tools rather than a single model-generated output.

The approach also reflects a broader industry bet: low-precision AI models can be useful when they steer higher-precision computing instead of replacing it. Nvidia has been one of the more visible backers of that model, helped by the fact that its GPUs are used for both AI workloads and scientific or engineering simulations.

Related work has appeared outside commercial EDA. Researchers at the Department of Energy’s Sandia National Laboratories have used AI agents to generate and test hypotheses in what they described as a self-driving lab. Those systems used more established model types, including variational auto-encoders, rather than large language models.

For Cadence, the immediate commercial question is narrower: whether an AI interface can reduce time spent managing toolchains without weakening trust in the underlying simulation process. AuraStack’s early customer list gives the launch credibility, but the undisclosed pricing and unverified productivity claims leave the economics for engineering teams unresolved.

This story draws on original reporting from The Register.