At GTC, Nvidia’s GPU Technology Conference, MediaTek announced the first Dimensity Auto Cockpit SoCs that are a result of its previously announced partnership with Nvidia. According to MediaTek, the new Dimensity CV-1, CM-1, CY-1, and CX-1 will target entry-level in-vehicle experiences with the CV-1 to premium in-vehicle experiences with CX-1. The new products will feature a combination of Arm V9-A CPU cores, an HDR Image Signal Processor (ISP), a multi-audio DSP, support for multiple cameras, AI processing capabilities, and a Nvidia RTX GPU. Additionally, the platform will support Nvidia’s Drive OS software platform. According to the company, the new SoCs will provide a safe and entertaining experience with support for multiple cameras, displays, audio formats, plus the latest graphics capabilities and generative AI applications. While the technical details of the products are limited at this time, it is the mere announcement of the products that presage significant milestones for the industry.
First, the announcement signals positive progress of the partnership to produce a family of products using a chiplet strategy in less than a year from when the partnership was announced. While the use of chiplets is becoming the standard for high-end processors and accelerators, this is noteworthy for more mainstream embedded/IoT-centric applications. However, Tirias Research believes that the use of chiplets will become more common across these types of applications toward the end of the decade as the UCIe chiplet standards are adopted, testing standards for chiplets are established between semiconductor foundries, and volume production reduces the cost of the packaging.
Chiplets will allow for more rapid development of chips that are highly differentiated, such as with the Dimensity Auto Cockpit product family. The Dimensity Auto Cockpit products combined the licensed intellectual property (IP) from Arm, IP and the design expertise of MediaTek, and the graphics and software expertise of Nvidia without the need to have everything on a single die. Tirias Research predicts that this could initially reduce the time from concept to product testing by upwards of 50% from today’s timelines, and it could be even more as the use of chiplets from multiple vendors and foundries becomes more common. In this case, all the chiplets are produced by one foundry, TSMC, eliminating the challenge of having multiple foundries.
Another significance of this announcement is what this represents for Nvidia’s ever changing business model. While Nvidia has offered GPU IP for licensing in the past, there were no takers. This may open the doors to leveraging Nvidia semiconductor IP, especially when paired with the company’s leading software platforms and AI foundation models, for other market segments and applications.
The final takeaway is what this means for not only MediaTek, but the rest of the semiconductor industry. With the use of licensed IP combined with chiplets, the barriers to entering new markets or segments are lowering while the ability to differentiate between products is increasing. There is little doubt that OEMs across industry segments will seek to leverage the ability to customize chips for specific applications. This allows for greater product differentiation, as well as the ability to pack more of the semiconductor content into a single package for higher performance efficiency while reducing the system complexity and cost.
While we await additional details about the new Dimensity Auto Cockpit SoCs and other Dimensity Auto Drive, Connect, and Component products, the announcement does mark a milestone for the MediaTek-Nvidia partnership and the industry as it evolves around the changing dynamics of semiconductor design and manufacturing.