Agentic AI reshapes PCB and advanced packaging design

Phil Bishop, VP of Corporate Marketing, Global Customer Success Team at Cadence, talks to Phil Alsop, Contributing Editor of Advanced Packaging Magazine, about how agentic AI orchestration, combined with trusted EDA and system design and analysis tools, can help engineering teams manage the growing complexity of PCB and advanced package development. As hyperscale data centres deploy massive AI clusters and industries such as automotive and power electronics push towards increasingly intelligent, high-performance systems, Bishop discusses why connected design workflows, multiphysics analysis and intelligent automation are becoming critical to faster, higher-quality design realisation.

Phil Alsop: AI infrastructure, automotive electronics and high-performance computing systems are becoming increasingly complex. What are the biggest design challenges facing PCB and advanced packaging teams today?

Phil Bishop: One of the biggest challenges is that the design process is complex and the teams are often disparate. PCB layout, system architecture, packaging and analysis may all sit within different groups.

Advanced packaging is now critical to high-performance electronics, but the handoffs between those different teams can slow the overall workflow. A major opportunity is therefore to bring those teams together and allow data and design intent to move more effectively across the entire system design process.

For PCB and advanced packaging teams, improving that end-to-end workflow is becoming fundamental to increasing throughput.

Phil Alsop: Cadence describes AuraStack as an agentic AI platform for PCB and advanced packaging design. What distinguishes agentic AI from the AI-assisted tools engineers are already familiar with?

Phil Bishop: AI-assisted tools can help optimise an individual step of the design process. Cadence Agentic AI goes further by looking across the entire workflow.

That means working from the schematic and design intent through architecture, PCB development and packaging, while coordinating the different tools and datasets involved.

Multiphysics simulation is another important part of that approach. The AuraStack AI Super Agent acts as an assistant to the lead engineer throughout the PCB and package design process, allowing electrical, thermal, mechanical and other analyses to be performed as the design evolves.

The main difference is therefore not simply applying AI to an individual task, but optimising across the complete workflow.

Phil Alsop: Cadence also offers AI super agents spanning chip, package and PCB development. Why is it important to connect those traditionally separate design domains?

Phil Bishop: The AuraStack AI Super Agent focuses on PCB and packaging, but Cadence also has super agents addressing other parts of the design flow.

The ChipStack AI Super Agent addresses areas such as verification and RTL creation, InnoStack focuses on digital and physical design, while the ViraStack AI Super Agent addresses analogue design, including layout migration and analogue performance optimisation.

Together they provide a broader suite of agentic capabilities.

The AuraStack AI Super Agent can, in some cases, sit above these different domains because it operates at the system level. For example, it can call other technologies where 3D-IC integration or analysis is required.

The important point is that system design increasingly cannot be treated as a series of isolated domains. Chip, package and PCB behaviour are interconnected, so the design workflow needs to reflect that.

Phil Alsop: The AuraStack AI Super Agent places a strong emphasis on integrated electrical, thermal, mechanical, signal integrity and power integrity analysis. Why is multiphysics co-optimisation becoming so important?

Phil Bishop: Power integrity, signal integrity, thermal analysis, EM-IR and other forms of multiphysics analysis have always been necessary. What changes with an agentic workflow is the ability to bring those analyses together and apply them continuously throughout development.

The system also builds what we call a mental model around the design intent and the environmental constraints in which the PCB and package will operate.

A 5G system, for example, will have very different requirements from an automotive electronic system. Thermal behaviour, vibration, power integrity and other factors will therefore be assessed against the intended operating environment.

The value comes from linking agentic workflow optimisation with principled simulation technologies and the underlying Cadence tools. That allows the system to consider multiple physical effects together rather than treating them as separate checks at the end of the process.

Phil Alsop: In power-dense applications such as data centres, EVs and energy infrastructure, how does earlier visibility into thermal and power integrity issues improve system performance and reliability?

Phil Bishop: The earlier you can analyse thermal performance, power integrity and signal integrity, the better you can optimise the design.

Historically, these analyses were sometimes performed relatively late. A design team might complete much of the work before handing the system to a separate analysis group.

With an integrated multiphysics workflow, analysis can happen incrementally throughout development. At each stage, engineers can ask whether the design is delivering the performance required by the end application.

That is particularly important in environments such as data centres, where power consumption and thermal behaviour are critical.

Instead of discovering an issue near the end of the design cycle, engineers can evaluate performance continuously and adjust the architecture much earlier.

Phil Alsop: Cadence has cited significant productivity improvements from agentic AI. Beyond reducing manual work, how do you see AI changing the role of engineers?

Phil Bishop: AI is not a replacement for good engineers. What it does is allow engineers to move up a level and focus more attention on architectural and system-level decisions.

Agentic solutions are only as good as the design intent, constraints and hierarchy provided to them, and that knowledge ultimately comes from experienced engineers.

The opportunity is to augment engineering expertise and allow teams to explore far more alternatives than would have been practical with traditional workflows.

A strong system engineer can evaluate different architectures earlier instead of discovering late in the process that another approach may have been better.

There is a useful analogy with software development. Engineers gradually moved away from working directly with machine code and assembly languages as higher-level abstractions became available. AI represents another step in that direction.

Engineers still need deep technical knowledge, but natural-language interaction and agentic automation can allow them to spend less time on lower-level tasks and more time defining what they want the system to achieve.

Phil Alsop: Cadence has worked with companies including NVIDIA, TSMC, Socionext, FORVIA HELLA and Schneider Electric. What early lessons are emerging from these collaborations?

Phil Bishop: One of the clearest benefits is workflow optimisation.

With the AuraStack AI Super Agent, we are seeing productivity improvements by connecting different elements of PCB and packaging design, sharing data more efficiently and applying multiphysics analysis across the workflow.

The emphasis varies between collaborations. With Socionext and NVIDIA, for example, much of the value has been around process and workflow optimisation. With TSMC, the focus has been more on packaging, including substrate design, routing and placement.

Across these examples, the common theme is improving end-to-end throughput.

Previously, different people or teams might have had to invoke separate tools for individual parts of the design process. AI-assisted tools could improve those local tasks, but agentic AI enables optimisation across system design, PCB design, chip design and packaging together.

That broader orchestration is where we see an important difference.

Phil Alsop: Does identifying problems earlier also help shorten the design cycle and improve time to market?

Phil Bishop: Absolutely. There are two major benefits.

The first is getting to market sooner.

If the workflow becomes significantly more productive, customers can shorten development cycles and ultimately reach revenue earlier.

The second benefit is design quality.

When tools are connected and optimisation happens continuously, engineers can identify errors earlier and improve the quality and reliability of the final product.

So, this is not just about time to market. It is also about improving what is ultimately produced.

A workflow that can look at multiple issues simultaneously, run multiphysics simulation throughout the process and optimise continuously can improve both turnaround time and the quality of the end system.

Phil Alsop: How do you expect agentic AI to change PCB and advanced packaging development over the next few years?

Phil Bishop: We are still at a relatively early stage.

At the Design Automation Conference, there were many companies exploring different forms of agentic EDA. A lot of the early activity has focused on verification, while other developments are beginning to address IC physical design.

The important point is that agentic systems still need strong underlying tools and principled simulation. If you move too far away from those core technologies and rely only on natural-language techniques or large language models, you are unlikely to produce a reliable engineering result.

The agents need to orchestrate trusted design and simulation technologies.

Over time, I expect agentic AI to become increasingly common across EDA. These systems should improve as they learn from more design activity, while experienced lead engineers and system architects will become even more important because they will define the constraints and prompt the systems at a deeper level.

We will see improvements not only in turnaround time but also in design quality across more areas of the EDA flow.

Phil Alsop: Is there one area of electronic system design where you think agentic AI will create the greatest value?

Phil Bishop: The overall system workflow is probably one of the biggest opportunities.

Bringing PCB and package design together and then being able to simulate the complete system is potentially transformative because, for many years, those processes have largely depended on handoffs between different teams.

Now we have the opportunity to optimise across the entire flow.

I also think analogue design will be an important area. Analogue layout migration and optimisation are inherently complex processes, and complexity is exactly where agentic technologies can have significant impact.

For decades, there has been an ambition to build a single tool that does everything. A better model is to allow multiple specialised tools to exchange data intelligently.

That intelligence and orchestration layer is where technologies such as the AuraStack and ViraStack AI Super Agents can make a major difference.