Automotive chiplets are small, separately made dies that are packaged together to act as one vehicle computer, so a carmaker can scale from basic driver assistance to high-end autonomy by changing the mix of dies rather than designing a new monolithic chip each time. The hard part is making chiplets from different suppliers work together and meet automotive safety and reliability rules. In October 2024 the Belgian research centre imec launched its Automotive Chiplet Program (ACP) to tackle exactly that, with carmakers, suppliers and chip companies working together. This post, first published on 25 October 2024, covers that launch and why it matters.
Key takeaways
- Chiplets let designers mix and match functional dies (compute, AI, I/O, memory interfaces) in one package.
- For cars, the main benefits are scalability across vehicle models, reuse of proven dies to save development cost, and data-centre class performance where needed.
- imec's ACP, announced in October 2024, is pre-competitive research on chiplet architectures and packaging for automotive use.
- The first companies to commit included BMW Group, Bosch, Valeo, Siemens, SiliconAuto, Cadence, Synopsys, Arm, ASE and Tenstorrent.
- The open problems are reliability qualification, cost of integration, and common standards so dies from different suppliers interoperate.
What are automotive chiplets?
Chiplets are small functional blocks, each a die of silicon, that can be combined in different ways to build a complex integrated circuit (IC). Think of them as building bricks for semiconductors. Manufacturers can mix and match dies to suit a specific need, which matters a great deal in the automotive sector, where requirements vary widely from one vehicle model to another.
A chiplet-based vehicle computer might combine a general-purpose CPU die, an AI accelerator die, an I/O die for cameras and networks, and memory, all joined by short, high-bandwidth die-to-die links inside an advanced package. Each die can be made on the process node that suits it best: leading-edge for compute, a mature and cheaper node for analogue and I/O.
Monolithic SoC versus chiplets
| Aspect | Monolithic SoC | Chiplet-based system |
|---|---|---|
| Scaling across models | New chip, or one large chip with unused parts in low-end cars | Add or remove dies for each tier |
| Process choice | One node for everything | Best node for each function |
| Yield on very large designs | Falls quickly with die area | Smaller dies yield better; only known-good dies are assembled |
| Interconnect | On-die wires, lowest latency and power | Die-to-die links add some power, latency and design effort |
| Supply chain | One vendor, one design | Several vendors, needs agreed interfaces and test methods |
| Qualification | Well-established automotive flows | Package-level reliability still being worked out |
Worked example: why smaller dies help yield
Illustrative numbers using the simple Poisson yield model, yield = e-D x A, with a defect density D of 0.1 per cm2:
- One 600 mm2 (6 cm2) monolithic die: yield = e-0.6, about 55%.
- One 150 mm2 (1.5 cm2) chiplet: yield = e-0.15, about 86%.
Because each chiplet is tested before assembly, four good 150 mm2 dies can be picked from a wafer with far less scrap than one large die. The saving is partly offset by packaging cost, die-to-die interface area and the risk of losing good dies in a failed assembly. Real defect densities and costs vary by node and foundry; these figures only show the trend.
Why are they important?
- Scalability: chiplets let performance scale from basic advanced driver assistance systems (ADAS) to fully autonomous driving.
- Cost efficiency: by reusing existing chiplet designs, manufacturers can cut research and development costs significantly.
- Enhanced functionality: specialised chiplets can process data from sensors such as cameras and radar, improving vehicle intelligence.
- High performance: they can bring data-centre class processing power to next-generation automotive applications.
imec's Automotive Chiplet Program
imec, the Belgian research institute, leads this research through its Automotive Chiplet Program (ACP), announced on 10 October 2024 at its Automotive Chiplet Forum in Ann Arbor, Michigan. The programme aims to address the challenges specific to the automotive industry, such as strict safety standards and high reliability requirements, and to identify which chiplet architectures and packaging technologies suit high-performance automotive computing.
Key players involved
imec has partnered with a range of influential companies to speed up this work. The companies named in imec's October 2024 announcement as the first to commit were:
- BMW Group
- Bosch
- Siemens
- SiliconAuto
- Cadence
- Arm
- Valeo
- ASE
- Synopsys
- Tenstorrent
The earlier version of this post listed the first seven; ASE, Synopsys and Tenstorrent were also named in the announcement and have been added. These partnerships matter because they bring together carmakers, Tier 1 suppliers, EDA tool vendors, IP and processor companies and packaging houses: the whole chain that has to agree before chiplets from different suppliers can be combined safely.
The Automotive Chiplet Conference
imec hosted its second automotive chiplet event, where industry leaders discussed progress and next steps. As reported at the time, the number of participating companies rose from 30 to 50, reflecting the growing interest in chiplet technology. Attendees included carmakers (OEMs), Tier 1 and Tier 2 suppliers and semiconductor design houses, which shows the kind of collaboration this field needs.
Highlights from the conference
- Standardisation: a strong push for common industry standards so that chiplet designs from different suppliers can interoperate.
- Pre-competitive research: the ACP focuses on shared research that benefits all partners without touching their competitive products.
- Future-proofing: flexible architectures that can keep up with fast-moving vehicle computing needs.
Challenges ahead
Despite the promising outlook, several challenges remain:
- Reliability standards: meeting strict automotive qualification such as AEC-Q100, which today is written for packaged ICs rather than multi-die assemblies, along with functional safety under ISO 26262. Cars expect operation from about -40 °C and service lives of 15 years or more, so solder bumps, interposers and underfill must survive long thermal cycling and vibration.
- Cost-effective integration: fitting chiplets into existing manufacturing and supply chains without packaging and test costs eating the savings.
- A standardised approach: agreed die-to-die interfaces, such as UCIe (Universal Chiplet Interconnect Express, first released in 2022), plus common test, security and safety methods so dies from different suppliers can be integrated easily.
- Known-good die and traceability: every die must be fully tested before assembly, and a failure in the field must be traceable to the right die and supplier.
How reviewers probe an automotive chiplet design
- Which safety mechanisms cover the die-to-die links, and how faults there are detected and reported.
- Who owns qualification of the complete package when dies come from several companies.
- How thermal hot spots on the AI die affect neighbouring dies over a vehicle's life.
- Whether the interface is a published standard or a proprietary link that locks in one supplier.
Update
The programme has continued to add members since this post was first written; for example, GlobalFoundries was reported to have joined as a foundry partner in October 2025. Check imec's own pages for the current list.
FAQ
What is the difference between a chiplet and an SoC?
An SoC puts all functions on one die. A chiplet system splits them across several dies that are joined inside one package and behave like a single chip.
Why do carmakers want chiplets?
To use one family of dies across many vehicle models, from entry level to premium, and to upgrade compute without redesigning everything.
What does imec's ACP actually do?
It runs pre-competitive research with its partners on which chiplet architectures, interfaces and packaging technologies suit automotive computing, and on how to make them reliable and standard.
Conclusion
The future of automotive computing is closely tied to chiplets. With organisations like imec bringing carmakers, suppliers and chip companies together, the industry can work towards a more connected and efficient ecosystem. As these modular components become common, they will extend vehicle capabilities and open the way to new kinds of in-car computing. We will keep following how these developments unfold over the coming years.
Your comments will be moderated before it appears here.