by Max Maxfield
Things continue to gain traction in “chiplet space” (where no one can hear you scream). In fact, before I forget, Chiplet Summit 2026 takes place in just a couple of weeks as I pen these words. The organizers say this is the place to “see and be seen” if you are in any way dabbling your toes in the chiplet waters (I’m paraphrasing, of course).
Not surprisingly, one of the keynote presentations at the Chiplet Summit will be given by Cadence, which is doing some very interesting things on the chiplet front. This presentation will be given by David Glasco, VP Compute Solutions Group at Cadence. David oversees hardware/software design and implementation of soft IP for AI, and his presentation will cover modular multi-die designs for AI, edge, and physical AI applications.
Hmmm, “physical AI applications.” Reading this made me think, “What’s that when it’s at home?” as people say in England. Fortunately, I was just chatting with Michael Posner, Sr Product Marketing Group Director, Chiplets & IP solutions at Cadence. Michael explained that physical AI is any system that performs AI processing at the edge, including autonomous vehicles, drones, and robots.
Physical AI systems are the next evolution in the AI journey (Source: Cadence)
Michael also noted that if you look at what the folks in Aerospace and Defense are doing with their edge AI, it’s fundamentally targeted at their autonomous vehicles, autonomous drones, and autonomous robotic systems, but they have some unique requirements for their physical AI, which is why Aerospace and Defense are typically assigned their own niche in the scheme of things.
As an aside, the phrase “physical AI systems” is relatively new and still emerging, rather than long-established or universally standardized. In fact, it wasn’t widely used in mainstream engineering or AI literature prior to the early-to-mid 2020s. Usage has grown quickly in the last few years, especially in discussions about robotics and autonomous machines, embodied AI and edge AI, and AI that senses, decides, and acts in the real world, not just in software.
Since Michael and I have both been around for a long time, we discussed how the current excitement over the term “multi-die” can be confusing for older engineers, since they’ve “been there, done that, and seen it all before.” On the off chance you’re interested:
Multiple silicon dies mounted on a single ceramic substrate interconnected with thin-film metallization and wire bonds began to appear in the mid-1960s.
Hybrid microelectronics matured (thick-film and thin-film on ceramic) in the 1970s, leading military, aerospace, and high-reliability computing to widely adopt multi-die ceramic hybrids.
... continue reading