The semiconductor industry has so far managed to increase chip processing power at regular intervals while also reducing product size and relative cost—this is known colloquially as Moore’s law. Systems on a chip (SoCs), in which multiple functions are consolidated on a single device, represent one aspect of this evolution, and OEMs are increasingly invested in them. The global automotive SoC market’s value is expected to reach US$70bn by 2030, up from US$37bn in 2022, according to Zion Market Research.
However, Big Tech players like IBM already foresee a deceleration and possible end to Moore’s law. This is bad news for software-defined vehicle (SDV) developers working on advanced features and autonomous capabilities, which will require greater processing power than current generations of hardware can provide. This is where chiplets can make a difference.
As opposed to monolithic architectures, where all functions are consolidated on a single die, chiplets break down SoCs into small, interconnected, and modular blocks that can be assembled according to specific use cases. This innovation yields more powerful, customisable, and energy and cost efficient computing. David Glasco, Vice President of R&D for Compute Solutions Group at Cadence, tells Automotive World that buzz around the technology started building in the data centre sector around five years ago. Now, it’s coming to vehicles.
Chiplets enter automotive
Glasco observes that initial chiplet development focused on proprietary solutions with little applicability beyond data centres. However, when players like Arm—a significant semiconductor IP supplier to the global automotive industry—began to expound and expand on the idea, a “turning point” was reached. In March 2024, Cadence and Arm announced their collaboration on a chiplet-based reference design and software development platform specifically to accelerate SDV innovation.
By November, they revealed their “ground-breaking” results: the first tape-out of a system chiplet using both companies’ IP combined with Cadence’s electronic design automation solutions. The chiplet integrates multiple processors, controllers, and memory IP to manage SoC resources and functionality. It is compliant with Arm’s Chiplet System Architecture for interoperability and fast time-to-market, while the complementary software stack is provided with a digital twin of SOAFEE-compliant hardware. This means software can be prototyped virtually instead of using physical silicon.
“There’s a growing number of automakers that want to follow the Teslas of the world, but they don’t have any silicon design teams,” says Glasco. “For an industry that’s still primarily based in mechanics, it needs help reducing the element of risk—both in terms of investment and safety.” As such, he believes partnering with companies like Cadence is essential, as OEMs can specify the exact chiplet that allows their software engineers to incorporate desirable vehicle functionalities.
A ’Lego-like’ solution
New advanced driver-assistance systems and autonomous driving systems are of particular interest at the moment, and this is an area Glasco helped develop first-hand during his tenure as a Full-Self Driving SoC architect at Tesla. Going forward, specialised chiplets will facilitate enhanced data fusion from sensor hardware and handle the processing necessary for machine learning in these functions. Cadence estimates that in-vehicle processing and networking power could eventually be equivalent to a data centre processing node.
Scalability is a crucial ingredient for introducing autonomous functions, and chiplets are well suited for the iterative journey from SAE Level 2 to 5. “Big automakers have a broad product portfolio,” states Glasco. “They want a common architecture across low-, mid-, and high-range vehicles that can be made more complex according to the price point targeted.” A low-end solution, for example, might feature a CPU, GPU and some AI functionalities. Mid-range could add multimedia capabilities and digital signal processing, with high-end then incorporating multiple CPU and AI chiplets on top.
“The idea is that there’s a lot of reuse within those offerings,” he continues. “As automakers iterate their SDVs, they can keep the same basic architecture and just swap a couple of chiplets.” Besides saving R&D costs, this could also help address prominent challenges concerning software verification—by retaining the same core hardware, developers do not need to worry about the performance and safety of their software for each new vehicle generation. “It’s a Lego-like solution: just replace the piece that doesn’t do what you want.”
Although a chiplet approach as described by Glasco would likely entail a degree of hardware standardisation across automotive, he does not believe this will be problematic for brand differentiation. “The real value proposition is the look and feel of a vehicle as shaped by the software; that’s where new experiences will come from.”
Inflection point oncoming?
With the automotive industry on the cusp of what Cadence considers to be a “technological renaissance”, the company is seeking to guide the adoption of chiplets. In October 2024, Cadence joined the Automotive Chiplet Alliance (ACA) led by the Interuniversity Microelectronics Centre (IMEC), a nanoelectronics and digital tech hub based in Belgium. ACA aims to form an ecosystem of companies from across automotive to prototype, test, and standardise chiplets in Europe.
Incumbents will need to decide whether they become the next Oldsmobile or the next Tesla
“Cadence became involved to understand the European market’s direction and concerns with building out silicon,” says Glasco. Most prominently, he relates that automakers are struggling with the cultural shift required for SDVs: instead of matching products to specific release timeframes, silicon innovations generally come ‘as and when’. This is contrary to established practice among automakers, so ACA will try to smooth the runway. “We don’t advocate for our products or a particular architecture; it’s more about sharing our knowledge and capabilities.”
Glasco states that the value of IMEC’s project is that it can “share the risk” of chiplet integration among automakers evenly, which could provide the push necessary to accelerate the technology’s adoption. At CES 2025, he observed that interest in the possibilities and capabilities of chiplets is growing, yet the automotive industry is “waiting for someone to go first.” Software-shaped mobility was ultimately a key theme of the event, and Glasco predicts that an inflection point could be reached within the next 12 months.
However, the danger for hesitant legacy OEMs is that they never catch up to new contenders moving more aggressively in the SDV space, particularly Chinese brands. “The market initially thought China would produce unsafe and low-quality vehicles, but the exact opposite has turned out to be true,” Glasco concludes. “Non-traditional automakers are more likely to adopt chiplets quicker, and incumbents will need to decide whether they become the next Oldsmobile or the next Tesla.”
