To say that cars are turning into ‘computers on wheels’ is already a bit of a cliché, it’s also quickly becoming an understatement, says Imec, standard compute power inside the electronic control unit (ECU) will not be able to process the enormous workloads that come with the ADAS, communication, and entertainment functions of tomorrow’s vehicles. Only high-performance compute can rise to that challenge.
Such supercomputing can no longer be achieved in one package using monolithic IC design – as the
size and complexity would become unmanageable. Chiplet design – a modular approach based on heterogeneous integration – allows to scale up the number of transistors and other components without hitting the physical limits of a single chip. It’s being implemented in all kinds of supercomputing applications, and cars cannot afford to trail behind.
“Standard compute power inside the ECU will not be able to process the workloads that come with the ADAS, communication, and entertainment functions of tomorrow’s vehicles,” says Imec.
Does driving a supercomputer sound like an expensive idea to you? Then the good news is that using chiplets leads to a reduction of cost. That’s because:
Different processes can be used for different chiplets, because not all functions need the most advanced (and expensive) semiconductor technology nodes.
The opportunity to use optimized processes for specific tasks, combined with the sharing of resources such as wiring and cooling infrastructure leads to overall energy savings.
Chiplets are less complex to manufacture than large monolithic chips. That leads to fewer defects, thus higher yields.
The lower production costs are partly offset by higher packaging costs. Nevertheless, the savings realized by using chiplets vs. monolithic design is estimated to be 40%.
Digital innovation has become a key differentiator for car companies. Automotive OEMs that want to distinguish themselves from their competitors have plenty of options: more safety features, extraordinary entertainment options, a fluid communication interface, … But no matter the choices they make, they most likely involve changes to the soft- and/or hardware. And they need to be implemented quickly to keep the time to market as short as possible.
While upgrading a monolithic design is a long process, replacing or adding a chiplet should be as simple as swapping a yellow LEGO block for a blue one. It can happen during the lifetime of a vehicle line. This gives OEMs the opportunity to build a dependable yet flexible electronic architecture with a base function chiplet augmented by workload-specific chiplets in the same package.
And that brings us to the final point …
“Replacing or adding a chiplet should be as simple as swapping a yellow LEGO block for a blue one.”
Ever since COVID, a stable semiconductor supply chain has been a top priority within the automotive sector. And while the lockdowns have passed, there’s still reason for concern.
For chip suppliers, automotive is a challenging market (with stringent reliability requirements) that takes relatively low volumes. And development costs in advanced nodes are increasing exponentially.
For these reasons, there’s a dwindling number of automotive processor vendors. And they have to stick to a one-size-fits-all solution across multiple market segments to reach sufficient volume per design.
When the number of suppliers is down to three or even less, the supply chain becomes vulnerable. The move to chiplet-based processors, that are less taxing to develop, could lead to the rise of new (niche) players that make the market more diverse and resilient. It could even allow OEMs to make their own chiplets.
For this mix-and-match strategy to work, there’s a need for standardization across the industry when it comes to packaging and interconnects.
