The Future of AI and HPC: A Synergy in Technological Excellence?
The impact of AI on scientific computing is still uncertain. One relies on extremely accurate double-precision calculations, whereas the other is quite content with 4 bits. At first glance, these two seem completely opposed, representing opposite ends of the spectrum known as high-performance computing (HPC) — and yes, whether it’s accepted or not, AI falls within the realm of HPC. Nevertheless, the latest AI development from Cadence Design Systems, a leading figure in industrial HPC, provides insight into how high- and low-precision computing can not only coexist but also collaborate to tackle larger and more intricate challenges more efficiently with reduced resources.
Introducing AuraStack: Where AI Meets PCB Design
Unveiled on Wednesday, Cadence’s AuraStack is an autonomous AI system designed to assist electrical engineers in the design and testing of printed circuit boards (PCBs), as well as in advanced packaging design and testing — two activities that have traditionally depended on highly accurate simulations. AI undoubtedly plays a significant role in Cadence’s offering; however, the company is not substituting these tools with AI models prone to errors. Rather, AuraStack functions similarly to Anthropic’s Claude Code or OpenAI’s Codex, but instead of writing, compiling, debugging, and executing C or Rust in a controlled environment, Cadence’s newest agent is crafted to manage its existing testing and simulation frameworks.
Enhancing Engineering with AI
“AI is enhancing the value of our engineering products and technologies,” stated Michael Jackson, CVP of Cadence’s system design and analysis division, to GadgetLad. To put it differently, the AI model — reportedly integrating a diverse array of open and proprietary models — acts as a natural language interface capable of planning and executing complex multi-step workflows for circuit design and testing that operate at higher precision using CPUs, GPUs, and other accelerators.
Real-World Implications for Engineering Tasks
“For instance, if I’m going to verify and address the IR reliability, I must pinpoint the power management components. I need to construct a simulation-ready power tree, then execute the simulation, and afterward provide feedback to the designer,” Jackson explained. Cadence’s current product suite already automates several of these functions. The challenge, as Jackson points out, is that creating a PCB or package design often entails completing thousands of tasks throughout its lifecycle.
Increasing Efficiency: The AuraStack Edge
“Sixty-five percent of an engineer’s day is consumed by managing and tackling numerous tasks,” he added. By managing that mundane work, Jackson asserts that AuraStack can enhance productivity by 15 times, allowing the designer to focus on design and engineering decisions instead of individual tasks. These improvements are substantial enough that numerous major players in the electronics industry, including Nvidia, have already subscribed to the service.
Beyond PCBs: AI’s Wider Influence
Cadence isn’t merely integrating AI with HPC for chip design or advanced packaging. The engineering software provider has developed similar agents for both digital and analog chip design. The concept of employing low precision computing to operate AI models that direct more precise single- and double-precision physics simulations is not novel. Nvidia stands as one of the primary proponents of this strategy, which is logical considering its GPUs extend beyond training and executing AI models, even if that’s primarily what consumers are purchasing them for lately.
AI in Action: Beyond Cadence’s Realm
This year, we looked into how researchers at the Department of Energy’s Sandia National Laboratories utilized AI agents to formulate and evaluate new hypotheses. They described their system as a self-operating laboratory. Yet, those tests, while conceptually akin to what Cadence is achieving with AuraStack, did not employ LLMs, choosing instead to utilize more established architectures like variational auto-encoders. However, with the success of code assistants, it’s easy to envision agents akin to AuraStack being employed to automate laboratory equipment, conduct simulations, and then iterate on the findings, allowing scientists to pursue their research even after they’ve drifted off to sleep for the night.
Conclusion: AI and HPC – Harmony in Innovation
Will AI and HPC coexist harmoniously, or may they clash like a chaotic night out in Geordie? Cadence’s AuraStack could very well serve as the mediator we’ve needed, fusing the accuracy of conventional computing with the flexibility of AI. Indeed, it’s like blending the finest of Newcastle with a dash of pioneering technology. Stay alert, this story is just beginning!
