by Joe Abajian | 05/14/2025

To the dismay of skeptics, Moore’s law has held true for 60 years. The number of transistors per microchip has roughly doubled every two years, but the cost of adding transistors has increased exponentially—not linearly. Additionally, more transistors in a smaller area generate new manufacturing and engineering challenges. As transistor counts exploded, silicon wafer quality became more critical. Crystal lattice defects, dopant concentration variations, and surface contamination can disable circuit pathways and create entire regions of non-functional transistors. Put simply, silicon defects are extremely expensive problems that are much more likely to occur on larger dies for monolithic system-on-chips (SoCs). To combat the risks associated with packing transistors onto a large die, semiconductor companies have focused research dollars on chiplets, which feature interlocking sets of discrete components on separate pieces of silicon, enabling modular customization that addresses the unique performance and I/O needs of industries such as automotive and medical.

Chiplet architectures feature several benefits. Most notably, they can reduce long-term manufacturing costs and optimize resource consumption. Traditional, monolithic chip development is akin to building an entire house from scratch. Chiplets, however, are similar to prefabricated home construction. SoC makers or device manufacturers can independently assemble specialized, high-quality components, thus reducing overall complexity and cost. Chiplets also enable developers to prioritize high performance for certain functions based on their needs. Rather than relying on prebuilt configurations from chip makers that feature high performance components in areas chosen by ODMs, engineers can access highly tuned or tailored chiplets.

Despite their potential benefits, few chipmakers have chiplet offerings due to design and implementation challenges. Nevertheless, chiplet R&D is at an all-time high. Baya Systems, a RISC-V chiplet maker, has raised $36M this month. Arm (NASDAQ: ARM) released the first public specification of the Arm Chiplet System Architecture (CSA) and reported that over 60 companies are already engaging with the architecture. The majority of chiplet innovation over the past decade, however, has come from AMD (NASDAQ: AMD) and Intel (NASDAQ: INTC).

So far, AMD and Intel have largely limited chiplet designs to high-performance computing applications, making the future of chiplets for embedded systems uncertain. Chiplets have the potential to optimize resource management and outperform monolithic SoCs, but heat management is still a critical issue for chiplets due to the sheer number of cores in a small area. Additionally, the interconnects between chiplets can delay data transfer, making chiplets unsuitable for real-time safety-critical applications in the near term. Intel’s recent advancements in silicon interposer technology, however, have reduced interconnect communication time significantly.

Currently delayed by technical limitations, chiplets will make their way to the broader embedded market over the next several years. In the status quo, however, scalability, design flexibility, and cost efficiency will catalyze chiplet use in datacenter and HPC use cases. As hardware providers continue to innovate on chiplet designs and architectures, technical barriers will erode, making chiplets a key technology for both embedded and enterprise use cases.

To learn more about VDC’s coverage of processor technology, read the Executive Brief from our most recent processors report and our VDC View covering processor architectures.