NVIDIA, Cadence collaborating seems like a natural progression in this AI-first world. But can AI truly parent its next generation of hardware? Seems questionable.
AI for design or design for AI- this is the question as NVIDIA enters into a partnership with Cadence Design Systems (CDS). The overview is that NVIDIA aims to create a virtuous cycle of AI design by breaking physical and computational bottlenecks.
Moore’s theory is a significant observation, a trend that the entire manufacturing industry operates on. And since 1965, the industry has been finding loopholes to shrink as many transistors as possible. But forcefully fitting several transistors together creates heat, and one cannot remove a single transistor without melting the chip.
And that was merely one of the many challenges that threaten to stall Moore’s law.
The NVIDIA-Cadence alliance is a strategic workaround to this dilemma.
Training inside simulations is obviously much easier than training robots in the real world. There are physical limitations (Moore’s law is one), and the training data is also readily available. Now Cadence is generating them through its physics engines- to train robots inside simulations.
But even that faces a conundrum. There’s little understanding of how real-world materials interact. However, this partnership might truly change that.
Cadence has designed a head agent, called the AgentStack, that’s fuelled by NVIDIA’s Nemotron models. This AI sifts through thousands of design possibilities to find the best one- it’s basically AI designing another AI.
It is the future of AI design.
Meanwhile, NVIDIA is using these head agents to design their own chips- it’s a loop: NVIDIA’s chips are being designed by AI running on NVIDIA’s chips.
It’s a dual-track strategy.
Cadence’s agents are basically expert copilots who can observe a design and suggest changes accordingly. AI is leveraging AI to build the next generation of AI hardware
– a feedback loop like this:
NVIDIA designs and builds a quicker GPU ⇒ Cadence leverages it to make their software more effective and speed up output ⇒ Software engineers use this to build faster GPUs.
Rinse and repeat.
The goal is to decrease the time needed to complete significant tasks- the focus is on building AI for robotic systems. And we’re beginning by actively zeroing in on the designs.
