By Ron Wilson
Chiplets and automotive. Can the two ever live happily together?
Chiplets: born in the crucible of high-performance computing and AI, dependent on advanced packaging and interconnect standards that barely exist. Automotive: the paragon of conservative, regulation-bound, and cost-averse late adopters. It doesn’t sound promising.
But Bart Plackle, vice president for automotive at global semiconductor technology consortium Imec, sees chiplets as the only hope for the automotive industry to get the massive computing performance it needs into cars.
Plackle says that twin forces are driving a fundamental change in what a car is — from a vehicle to a mobile advanced computing center. One of those forces is increasing driving automation, from advanced ADAS to some degree of autonomy. Only massive computing power, Plackle says, can provide an excellent automated driving experience. The other force is inside the cab — the user’s interaction with the sounds, displays, and controls. Once again, the user experience designers envision, including augmented reality and enveloping experiences, will require massive computing power.
And, Plackle says, chiplets are the only way to get there. Continued process scaling will not provide enough transistors on one die, and even if it could, yields could not be made commercially viable. So future automotive computing systems must be assembled from chiplets.
Presented with this scenario, auto vendors ask two questions: Will chiplets even work for automotive? And can we afford them?
The first question has no sure answer today. The advanced packaging used in datacenter chiplets cannot withstand the environmental demands of in-vehicle use. And no one knows if known good dies can be produced to automotive quality standards. Pulling together contributors from across the industry, IMEC has convened a series of working groups that are formulating the right questions and working toward firm recommendations.
The second question may hinge on a key attribute of chiplet-based systems: composability. The automotive industry is not a high-volume market. But the ability for system designers to compose an architecture from chiplets, and to create derivatives by mixing and matching, could be incredibly valuable. Ideally, Plackle says, designers could mix and match chiplets with different computing power, function, reliability levels, vendors, and even product lifetimes to meet a wide range of needs without new chip designs.
That, however, would require a plug-and-play bus standard. Today’s UCIe is not there, Plackle warns. It offers basic connectivity, but allows variations in protocols, reliability measures, security, and other layers vital to the automotive market.
Can the industry work together, with IMEC’s assistance, to build these additional layers around the existing groundwork — and the high-volume chiplets — that the datacenter industry is providing? Can solutions to the mechanical and cost issues be found? And if so, when will we see chiplet-based computing systems in cars? To see Bart Plackle’s answers, watch Why Carmakers Want Chiplets.
The webinar is now available in the Ojo-Yoshida report YouTube channel:
What’s a Chiplet? Why Now?
The innaugural episode of our Dig Deeper on Chiplet series, aired last month, can be viewed on this link.
Ron Wilson is a contributing editor for The Ojo-Yoshida Report. He contracts with Intel, but the views are his, not Intel’s.
Copyright permission/reprint service of a full Ojo-Yoshida Report story is available for promotional use on your website, marketing materials and social media promotions. Please send us an email at firstname.lastname@example.org for details.