金莲直播 Cloud
Cloud native EDA tools & pre-optimized hardware platforms
Request a Free Trial
Table of Contents
Note: The following article contains statements made during a panel discussion at 金莲直播 Executive Forum, held March 19, 2025, in Santa Clara, California.
A tool is only as good as its operator. To harness the full potential of any instrument — from a hammer to a Fusion Compiler — one must apply the right knowledge and skill to achieve the desired outcome.
Artificial intelligence is no different. With assistive AI already augmenting longstanding chip design workflows and entirely new workflows on the horizon with agentic AI, industry leaders are assessing how they can best use these fast-evolving technologies. Their highly trained workforces are a key part of that equation.
As AI capabilities and use cases take shape, leaders need to carefully consider the roles their engineering teams play today, how those roles will change in the near future, and what it will take for them to adapt.
Productivity boost
AI is already delivering productivity gains. But to date, large language models (LLMs) have mostly played the role of supercharged researcher, processing and interpreting massive amounts of structured data. At NVIDIA, for example, engineers ask chatbots free-form questions to extract information from their intranet databases.
The industry knows it has only scratched the surface of AI’s transformative potential.
“We’ve become more productive, we’re doing things faster,” said Manoj Selva, VP and GM of product enablement at Intel, during the inaugural 金莲直播 Executive Forum. “[But] we haven’t cracked how to do things in a different way yet. I think that’s the next piece of AI.”
Design verification is one example, he noted. While AI can now shrink one million clock cycles down to 100,000 or less, the next step is auto debugging.
“Can I temporally, spatially locate the bug so an engineer doesn't have to spend two, three weeks trying to figure it out?” asked Selva. “Can I then take [that intelligence] back to my architecture where I correct the construction? Those are the paradigms that are still in front of us.”
Left to right: Junko Yoshida (moderator), Brian Amick (AMD), Manoj Selva (Intel), Reynold D'Sa (Microsoft), Sameer Halepete (NVIDIA), and Shankar Krishnamoorthy (金莲直播) at 金莲直播 Executive Forum
Agents among us
That paradigm shift may be just around the corner. Agentic AI, for instance, will increasingly act as an independent and self-directed design partner — proactively identifying improvements, foreseeing potential bottlenecks, and adjusting workflows.
“Complexity is growing at such a rapid pace that the execution cycles are getting so short,” said Shankar Krishnamoorthy, chief product development officer at 金莲直播. “If you don’t introduce these types of disruptions in the workflow, there’s no way you can meet these kinds of schedules [that chip companies face].”
Even with AgentEngineer? technology working at progressively higher levels of autonomy, Krishnamoorthy said humans still need to watch over it.
“Engineers have to evolve from being chefs to being good food critics,” he said. “A lot of stuff is not going to be made by them, it’s going to be made by the agent. But engineers have to evaluate it and give feedback so the quality of the capability improves.”
That will require engineers to shift their mindset.
“Our current workforce is used to doing things from scratch,” said Brian Amick, SVP of technology and engineering at AMD. “Now, with AI solutions, we can often create solutions way faster. But you have to be able to judge if it’s actually a good solution.”
A shift in flow
AI will change the workforce in other ways too, reducing the demand for specialist engineers in favor of broader expertise.
“There will still be specialists, that does not go away. But more generalists can really leverage AI and leverage the knowledge base that comes from AI tools,” said Reynold D’Sa, CVP of silicon, cloud hardware, and infrastructure engineering at Microsoft. “That also gives a better design capability, because they’re looking at it from a bigger picture perspective versus the specialists.”
Humans are often resistant to change, and engineers are no exception. Industry leaders acknowledge the difficulty of transforming well-established roles and workflows.
“It’s such a disruptive change in the way we work, primarily around the feeling of loss of control and accountability,” said D’Sa.
“It prompts questions of, ‘So what am I doing here? What is my part in this?’” Selva said. “We need to make sure people understand that there’s higher value, higher productivity tasks for them to do.”
To ease the transition and help their workforce embrace change, chip companies need to put AI tools in more engineers’ hands — and then give them the space to experiment.
“It goes against very tight schedules and shifting deadlines, but you can’t just do these types of transitions while also doing your day job,” said Sameer Halepete, VP of VLSI engineering at NVIDIA. “You need to give a little slack for people to retool their workflow.”
Lowering the cost of change
Despite the challenges that come with change, industry leaders are overwhelmingly optimistic about the transformative potential of agentic AI. But the pursuit of innovation and acceleration cannot come at the expense of product quality, workforce morale, or business performance.
“We really need to understand our customers’ workflows and introduce these technologies in a way that the cost of change is low,” said Krishnamoorthy.
There is little doubt that the future of the industry lies in engineers and AI agents working together.
“The chip design process has become so complex, and smart, well-paid engineers spend a lot of cycles in ‘high-toil’ steps that should be automated or even eliminated from the workflow,” said Krishnamoorthy. “The promise of agentic AI in EDA is to re-engineer workflows, address these ‘high-toil’ steps, and move them to progressively higher levels of autonomy.”
Continue Reading
Blog
4 min read
/
Apr 30, 2025
Ethernet Evolution: Trends, Challenges, and the Future of Interoperability
