Driving With Chiplets

The first examples of the upper class of vehicles that can drive autonomously on the highway already have arrived on the market or will be introduced to the market in the coming years.

Travel on the highway was selected as the first application because the number of objects that have to be taken into account in front of, next to, and behind the vehicle is manageable. This means the required processing power can be estimated and is realizable with current high-performance computing platforms. If the processing power of such a platform is not sufficient, it can be cascaded.

In contrast to autonomous driving on the highway, in complex environments such as in the city, significantly more objects must be detected. Current estimates expect up to 400 objects, which must be registered simultaneously. However, there is no upper limit, and in some situations this number may be higher. That makes it very difficult to estimate the processing power, because it correlates strongly with the environmental situation. It is certain, however, that existing high-performance platforms for autonomous driving are not able to supply the necessary processing power.

A new system architecture approach is required, and various options are available. Those range from complex systems-on-chip (SoCs) to new approaches from the packaging area, such as chiplets.

Chiplets allow various circuits to be mounted on a package substrate, as with a system-on-package (SiP). Unlike SiP, however, with chiplets the interface between the circuits is also specified and standardized. This allows prefabricated circuits to be assembled into new types of systems more or less arbitrarily, from the perspective of the application.

The chiplet approach has a number of advantages for developers of autonomous vehicles. Currently, autonomous driving is being introduced by means of various intermediate steps, and mid-class vehicles will only permit partially autonomous driving in the near future. This poses the challenge that different levels of processing power and system interfaces must be made available for the various vehicle classes. It is significantly easier to meet this requirement with a modular approach, such as chiplets, than with a monolithic approach, such as SoCs. Discussions are therefore underway in the automotive industry to hasten the development of chiplets.

However, the implementation of chiplets in the area of autonomous driving will likely place different requirements on the interface between the circuits than in the currently discussed standards. It is therefore important for this application area to advance its own interface definition, or to bring both activities into alignment. For this reason, some promoters of chiplet technology, such as Fraunhofer IIS/EAS, are involved in various standardization activities.

In addition, the ever-increasing expansion of autonomous driving will also place significantly higher demands on integration technology. The data transfer rate between the circuits will be very high because extensive image and radar data is processed, leading to large data quantities per unit of time. Then this data must be processed in the circuits of the chiplets, and therefore regularly exchanged between the circuits. This requires high data rates that can only be realized with fast and massively parallel interfaces, so the corresponding package technology also has to be prepared. Only approaches such as 2.5D integration (interposers) or fan-out technologies can satisfy these requirements.

New approaches also are required with regard to verification. Particularly when it comes to the signal integrity of the lines between the circuits and the power/ground supply, further simulations and analysis are indispensable.