“Chiplet architectures can solve real performance, yield and cost challenges”

The shift towards software-defined vehicles is increasingly exposing a structural dependency that the automotive industry has long been able to abstract away: semiconductor architecture is no longer just a supplier topic, but a core element of vehicle platform strategy. Decisions about compute, integration and scalability are moving closer to the OEM.

Jim Foresi, Director of the Automotive Semiconductor R&D Center at Imec, examines this shift from a system-level perspective. With a background spanning semiconductor development and automotive applications, he focuses on how chiplet architectures and open interfaces can redefine how compute platforms are designed, sourced and evolved over time.

At the Automotive Computing Conference in Detroit Foresi will discuss how open, standardised chiplet approaches can accelerate innovation while reducing dependency on monolithic system designs. In the following interview, he outlines which architectural decisions will shape automotive platforms for years to come.

ADT: Why is now the right time for the automotive industry to move beyond monolithic SoCs?

Foresi: Advanced vehicles require AI-class compute, high-speed networking and rapid platform updates. That puts automotive directly in competition with data center and mobile markets for advanced semiconductor process nodes. Unlike those markets, automotive has lower volumes and longer lifecycles which makes monolithic SoCs at leading-edge nodes both expensive and prone to supply chain risk. Not everything in a vehicle needs 3 or 5 nm silicon, but monolithic designs force that. Chiplet architectures allow automakers to separate what truly requires advanced nodes from what doesn‘t, reducing exposure to supply constraints as other industries consume more high-end wafer supply. The time to plan for that transition is now, before compute demand grows further and automotive loses priority in access to global semiconductor capacity.

How can open and standardised chiplet architectures accelerate innovation across the automotive value chain?

“AI makes diligence affordable”

AI-driven tools are reshaping embedded automotive development – not by replacing engineers, but by accelerating testing, validation and review workflows. Felix Martin, Research Engineer at Tasking, explains where AI already adds value and where safety-critical limits remain.

Benjamin Müller

Open and standardised chiplet architectures change who can participate and where innovation happens. Instead of requiring a single supplier to deliver an entire SoC, they allow OEMs and Tier 1s to combine compute, networking, power and safety functions from different sources around a common interface. That naturally supports scalability. The same architectural foundation can support different compute and accelerator configurations across low, mid and high-end ADAS. It also lowers the barrier for new silicon suppliers and start-ups to contribute differentiated functionality without having to build a full automotive platform. The result is faster iteration, more supplier diversity and options for differentiation, while keeping software and safety architectures stable. Standardised chiplets let automotive evolve compute incrementally instead of in large, slow and risky jumps.

What lessons can automotive learn from other industries already using chiplet approaches?

The first lesson is that chiplets already exist and work. High-performance compute markets have shown that chiplet architectures can solve real performance, yield and cost challenges that are increasingly difficult for large monolithic dies at advanced semiconductor process nodes. A second lesson is that chiplets only scale when the supporting ecosystem is in place. Other industries now benefit from a growing EDA, advanced packaging and test infrastructure specifically designed for multi-die systems. Finally, those markets treat chiplets as a system-level strategy, not a packaging exercise. The value comes from deciding what belongs on advanced nodes, what does not and how systems evolve over time. Automotive doesn’t need to start from scratch.

What needs to happen for a sustainable and interoperable automotive chiplet ecosystem to emerge?

The industry needs standards, not just for die-to-die interfaces, but for qualification, test, safety integration and lifecycle management. Automotive chiplets have to be interoperable at the system level, not just electrically compatible. The ecosystem also has to mature beyond individual products. That means shared reference architectures, validated packaging flows and co-design across silicon, software and system teams. This is where neutral, pre-competitive collaboration matters. Organisations like imec play a critical role by bringing OEMs, Tier 1s, EDA vendors, packaging houses and silicon suppliers together around common roadmaps and demonstrators. The hard questions need to be worked out early. OEMs also need to stay architecturally engaged. A sustainable chiplet ecosystem only works if automakers define system intent and scalability requirements up front rather than inheriting the choices made in a single supplier’s SoC product. That architectural ownership is what turns chiplets from a component strategy into a vehicle platform strategy.

Sponsored

Automotive Software Strategies Europe

19-20 May, 2026 SZ Tower, Munich

Which silicon decisions being made today will shape vehicle platforms for the next decade?

The most consequential decisions are architectural: whether vehicle platforms are built around large monolithic SoCs or modular multi-die systems, and how tightly they are coupled to specific semiconductor process nodes. Choices about where advanced nodes are truly required versus where mature nodes are sufficient will determine long-term cost, scalability and supply resilience. Once those boundaries are set, they’re hard to walk back. Equally important is whether OEMs retain architectural control or delegate those decisions entirely to silicon suppliers. That choice will shape platform flexibility, upgrade paths and dependency for years to come.

In open ecosystems such as chiplets and RISC-V, who carries the integration and certification risk, OEMs, Tier 1s or semiconductor partners?

Even if OEMs don’t want to build the system themselves, the risk doesn‘t disappear; it just shifts. Tier 1s can integrate and certify complete platforms, and semiconductor partners can deliver qualified chiplets and IP, but ultimately OEMs own the end-to-end platform. In practice, Tier 1s will absorb much of the execution burden, especially for integration and validation. However, OEMs still define requirements, system boundaries and platform lifecycles, which means they implicitly carry architectural risk. The upside is control and flexibility; the trade-off is that OEMs need to be involved at a deeper level, even if they source complete solutions from Tier 1s. Open systems don’t remove risk; they make ownership of that risk visible and manageable.

Partnerships are becoming essential across the automotive computing stack. Which type of partnerships will matter most in the next three years, strategic, technological or regulatory, and what should OEMs prioritise first?

Technological partnerships matter most right now. Automotive is entering a phase where architecture choices, packaging, software integration and silicon partitioning are all moving at once. OEMs need partners who can help co-develop platforms, not just supply components. This is where pre-competitive environments like imec are valuable. They give OEMs, Tier 1s and silicon providers a space to align on architectures, reference flows and roadmaps before anyone has committed to a product. There is a clear need for partnerships that build something together. OEMs who treat this as a traditional procurement decision will find themselves reacting to semiconductor constraints rather than shaping them.

What challenges will the wiring harness industry face over the next five years?

Which challenges will shape the wiring harness and EDS industry most strongly over the next five years? Four experts – all speakers at the 2026 Bordnetzkongress in Germany – provide surprisingly different answers.

Benjamin Müller