Lego-Style Software For Automotive And Industrial Chiplet Systems?

Chiplets are a key topic in the semiconductor industry today, as they offer the potential to greatly increase the performance and flexibility of chips. The current focus is primarily on implementation, in particular on the architecture and the development of die-to-die interfaces that enable efficient communication between the individual chips. These technologies hold out the promise of meeting growing demands for computing power and energy efficiency. At the same time, a distinction is increasingly being made between different application areas. While mature solutions already exist for data centers, the search is heating up for solutions that can serve the industrial and automotive sectors.

Another crucial aspect that is increasingly coming into focus is the topic of boot-up and software deployment. Although this topic is already being discussed, standards have yet to become thoroughly established. Boot-up and software deployment requirements vary widely depending on the application. For data centers, solutions are being developed for central, high-performance systems. In the automotive and industrial sectors, meanwhile, development work is increasingly centered on distributed, synchronized multi-processor systems that also integrate additional accelerators such as GPUs or FPGAs. These systems have special requirements for boot-up and software deployment that are very different from those in the data center sector.

The automotive sector, in particular, is driving the development of open chiplet ecosystems, which require a high level of interoperability and compatibility. The topic of boot-up and software deployment is particularly important here, as different chiplet modules, which may come from different manufacturers, need to communicate seamlessly with each other—not only at the electrical level, but also at the software level. This calls for standardized processes that ensure efficient integration and long-term scalability.

However, one key consideration for the future development of chiplet systems is software. This area can be divided into two main categories: real-time operating systems and application software. At present, the developers of chiplet systems for automotive applications don’t give much, if any, thought to either of these categories, although they are becoming far more important. Adopting a modular structure here, similar to the Lego building block principle of chiplets, could help create flexible and scalable software architecture.

In the area of real-time operating systems, there are two viable options. One is to develop distributed solutions for each individual chiplet and have them communicate with each other via standardized interfaces—similar to the die-to-die interfaces in hardware. The other is to develop a holistic solution and dynamically adapt it to the various chiplets. In this case, the interface would be located directly between the chiplet and the operating system, with the operating system providing uniform access to the application software. This would enable more efficient, flexible, and scalable integration of chiplet technologies into different applications, driving the development of reliable high-performance systems.