Chipping Away at the Software-Defined Vehicle: How the Chiplet Trend Is Enabling Collaborative Innovation

Last week at NVIDIA’s GPU Technology Conference (GTC), MediaTek showcased the four first Dimensity Auto Cockpit Systems-on-Chip (SoCs) designed in partnership with NVIDIA, less than 10 months after the initial announcement of their collaboration. The combination of NVIDIA’s Artificial Intelligence (AI) and graphics Intellectual Property (IP), through a Graphics Processing Unit (GPU) chiplet, with MediaTek’s high-speed connectivity, mobile computing, and general automotive SoC expertise have allowed them to produce a family of SoCs that addresses a range of in-vehicle experiences, from entry-level to premium lines. The Dimensity Auto Cockpit CX-1, CY-1, CM-1, and CV-1 ship with integrated NVIDIA DRIVE Operating system (OS) support and several Deep Learning (DL) functionalities, such as Large Language Models (LLMs), AI-based gaze correction, and audio enhancement for video conferencing, and driver alertness detection, among other AI-based safety applications.The Benefits of Chiplet Architectures

The chiplet trend has seen repeated mentions at automotive shows over the last few years, with various silicon companies announcing chiplet initiatives, such as Intel’s collaboration with imec to support integration of third-party chiplets into automotive SoCs. The benefits of a chiplet architecture approach to automotive SoC design has been apparent to Original Equipment Manufacturers (OEMs) and other industry players alike, but MediaTek’s announcement marks the first time that some of these benefits have been realized in practice. It proves that the automotive industry is capable of accelerating innovation cycles through chiplet design, and different vendors can effectively combine their respective expertise to produce a solution that addresses multiple problems for OEMs. Some other benefits of a chiplet-based architecture include:

• Flexibility: Enabling OEMs to make decisions later in the production cycle regarding their silicon hardware platform; for example, substituting a GPU to enable higher resolution screens to match a new product update roadmap that includes triple-A gaming potential 3 years after shipping. This enables flexibility in adapting to trends and demand that is closer to the standard of consumer electronics.
• Scalability: This hardware model is simpler and more modular for substituting into other vehicle lines within the same OEM by allowing the change of GPU or Central Processing Unit (CPU) specs; for example, while retaining software efforts due to the same architecture across all solutions. This allows OEMs to use the architecture for their mass market to premium vehicle models and make significant cost and time savings as a result.
• Domain Consolidation: Against the backdrop of consolidation of domains in mixed-criticality SoCs, as analyzed in ABI Research’s Accommodating Mixed-Criticality Compute in Digital Cockpit Domain Controllers report, OEMs are becoming more limited in what silicon vendor they can go to for their hardware. With the traditional automotive chip model, they need their silicon partner to be capable of consolidating domains such as connectivity, Advanced Driver-Assistance Systems (ADAS), and infotainment into a single SoC; therefore, the silicon vendor must have this expertise already. With a chiplet design, silicon companies can share their expertise to offer OEMs a consolidated domain and mixed-criticality SoCs that address all of the OEMs’ specifications.

If more silicon vendors can gain ground in the chiplet space, the benefits will only become more significant for OEMs and silicon companies alike. A more complete ecosystem with expertise shared across several vendors will allows OEMs to avoid vendor lock-in to the maximum degree, and allows smaller vendors to compete with more established players by giving them a more focused potential point of entry to an OEM or market through a specific chiplet specialty, such as a focus on cutting-edge neural network processing through Neural Processing Unit (NPU) chiplets. This motivates innovation and diversifies the supply chain, creating a key resilience that OEMs have been looking for.

Additionally, the OEM’s desire for ownership of their technology stack, which the industry has seen partly materialize through in-house software divisions, such as Woven by Toyota, can be achieved through the growth of this trend. An OEM will find it much simpler to design its own in-house CPUs for its vehicle goals than attempting to build entire stacks that includes the entire SoC, OS, Real-Time Operating System (RTOS), and hypervisor stack. If an OEM does want to own even more of its stack, it can do so by taking “baby steps”—slowly introducing more of its own solutions in its vehicles over time, instead of undertaking large investments to produce entire systems that may or may not meet their specifications upon completion.

However, to reap the benefits of a chiplet ecosystem, the industry needs to address some key obstacles. For chiplets from any vendor in the ecosystem to complement any other chiplet, there must be standardization around the interconnect technology, data security standards, and low-cost, high-reliability packaging technology. imec’s Automotive Chiplet Alliance is an initiative that addresses this exact issue—there are several working groups that focus on architecture, standardization, and quality & reliability. Increased participation in these initiatives is crucial, from all agents of the ecosystem, from the entrenched Tier Ones with extensive experience in automotive safety and security standards to the new entrant silicon specialists with expertise in chiplet interconnect standards in the data center domain that they can transfer to the automotive industry.

As these standards are reached, more and more vendors will be able to engage in complementary partnerships like NVIDIA and MediaTek, combining expertise across the range of software technologies that OEMs need to create the in-vehicle experiences that consumers are beginning to expect from their cars. This acceleration of innovation will be a primary enabler of the software-defined vehicle transition.