|
|
![]() |

Automotive Chiplet Ecosystem Heralds Agility and Flexibility
To learn more about the technical challenges and potential use cases behind chiplet adoption, EE Times Europe spoke to Kurt Herremans, director of imec's automotive chip program.
www.eetimes.eu, Jan. 28, 2025 –
In late 2024, imec hosted a gathering of automotive semiconductor vendors, EDA organizations, and tier 1 suppliers to debate the evolution of chiplets for automotive applications. At the event, imec launched its Automotive Chiplet Program (ACP) and announced that Arm, ASE, BMW Group, Bosch, Cadence Design Systems, Siemens, SiliconAuto, Synopsys, Tenstorrent, and Valeo were its initial stakeholders. According to imec, the program's goals are "to evaluate which chiplet architectures and packaging technologies are best suited to support car manufacturers' specific high-performance computing and strict safety requirements, while striving to extend the benefits of chiplet technology–such as increased flexibility, improved performance, and cost savings–to the entire automotive industry."
The foundations behind establishing this ecosystem of leading automotive partners are rooted in the belief that existing monolithic chip architectures have struggled to keep pace with the increasingly demanding needs of applications such as advanced driver assistance systems (ADAS) and in-vehicle infotainment (IVI) services. The program aims to investigate a more efficient and seamless approach to architecting compute-intensive chips for such specialized services. Chiplets offer opportunities for quick customization with lower develo
EE Times Europe: Does this initiative signal a resurgence of interest in chiplets?
Herremans: Interconnect is a crucial aspect of the chiplet approach. It's a different interconnect from what was traditionally built for monolithic chips with high-density buses running between the various IP blocks. Now, the focus is on making those functions available as discrete chiplets with high-density buses connecting them. Also, the packaging technology has evolved in the direction of 2.5D and 3D to enable these interconnects to be made. That may have led to a resurgence of using chiplets, but it's also down to the pure growth of compute. As you try to shoehorn more and more compute onto the die, it becomes bigger. Even with the advancements of Moore's law, you can print more transistors per square millimeter, but even so, the chip becomes bigger. In many designs, like 400- and 500-millimeter die, the yield becomes problematic, and this is an aspect where chiplets deliver benefits. You can chop your monolithic design into multiple individual chips for better yields. There is a cost associated to switching to advanced packaging, and you obviously need to reserve a bit of area on each chip for interconnect, but fundamentally, the chips get smaller, offer improved yields, and the total cost reduces.
EE Times Europe: Are chiplets part of the move to more centralized compute in vehicles?
Herremans: The amount of compute required in a car continues to grow. That growth comes from the evolution of ADAS, but even if you say that truly advanced self-driving is still many years out, any driver-assist functions will benefit from more CPU resources to deliver better fidelity. Many vehicle systems today would perform better with more compute horsepower. It's the same with infotainment systems. Most vehicles today have an infotainment system, but the interaction with many of these systems could be better. This is where, for instance, the introductions of things like large language models (LLMs) inside the vehicle become a reality; the interaction with the car can become much more natural in that sense. That leads to the requirement to pool even more processing horsepower inside the vehicle. Re-architecting the vehicle's compute infrastructure by moving what is typically in separate ECUs into a centralized computer turns many systems into software functions.