Collaborating with Black Sesame Technologies, Intel Aims at Next-Generation Intelligent Cockpit Chips | Exclusive Interview by 36Kr

Text | Li Anqi

Editor | Li Qin

The chip giant Intel is still advancing into the intelligent vehicle industry.

"The future of the global automotive industry is being shaped by China today. We hope to be a part of it." On April 23 this year, Intel participated in the Shanghai Auto Show for the first time. At the media communication meeting, Jack Weast, vice president of Intel Corporation and general manager of the automotive business unit, once again emphasized the importance of the Chinese market.

However, with Qualcomm and NVIDIA firmly holding the market share of chips for intelligent cockpits and intelligent driving, how can Intel open up the market?

In fact, at the beginning of 2024, Intel had already released the first - generation SDV (AI - enhanced software - defined) cockpit SOC, which is basically of the same origin as its Core chips for the PC market.

However, the domestic cockpit market competition is already quite fierce. Besides Qualcomm, the dominant player, there are also domestic players such as Xinqing and Xinchi. Intel's cockpit chips haven't made much of a splash.

This time, Intel decided to adopt a different strategy.

At the Shanghai Auto Show, Intel reached a cooperation with Black Sesame Technologies and jointly launched a cockpit - driving integrated platform. The platform will integrate Intel's cockpit SoC, as well as Black Sesame Technologies' Huashan A2000 and Wudang C1200 family of chips.

Black Sesame Technologies' Huashan A2000 is mainly targeted at full - scenario assisted driving, while the C1200 family is for the demand of cross - domain integrated computing.

The two parties said that the solution can meet the driving needs of automakers from L2+ to L4. Currently, the two parties have established a joint working group and are expected to release the platform reference design in the second quarter of this year and make preparations for mass production.

In addition, Intel also launched an in - vehicle pure edge - side GUI intelligent agent with the AI company Mianbi Intelligence, providing users with functions such as offline voice command understanding, context memory, personalized service recommendation, and screen operation, and understanding users' complex conversations and instructions.

In other words, by pairing with different - level chips from Black Sesame Technologies, Intel hopes to more flexibly meet the intelligent driving and cockpit needs of automakers.

However, Jack Weast said that although they have launched the cockpit - driving integration with Black Sesame Technologies, the cooperation is not limited to Black Sesame. "Intel is willing to cooperate with any ADAS manufacturer."

In addition to the progress of cooperation, Intel's own automotive product capabilities are also evolving.

At the auto show, Intel exhibited products such as the latest - generation cockpit SOC chip, with stronger performance. Compared with the previous generation, the generative and multi - modal AI performance can be increased by up to 10 times, and the graphics performance can be increased by up to 3 times.

Moreover, the second - generation cockpit chip adopts the Chiplet (pre - manufactured wafers with specific functions that can be combined and integrated) architecture.

Jack Weast, vice president of Intel Corporation and general manager of the automotive business unit, introduced that Chiplet is a more optimal cost structure.

"Advanced process nodes are very costly. If the entire chip uses the most advanced process nodes, the cost will be the highest." If the Chiplet architecture is adopted, different chip processes can be used in different areas of the chip. For relatively mature functions, older process nodes can be used, which also reduces the cost.

Based on the Chiplet architecture, Intel said that its second - generation cockpit SOC can match each functional module with a chip with excellent and appropriate performance. Automakers can customize computing, graphics, and AI functions according to their own needs, reducing development costs and shortening the time to market.

In terms of mass - production progress, Intel said that its second - generation cockpit SOC will be officially installed in mass - produced models in 2026.

Another notable product line of Intel is the domain controller.

Twenty years ago, Intel used hardware virtualization technology in data centers such as CPUs to allocate computing resources, which Intel called the "software - defined architecture."

Now, Intel wants to replicate the capabilities of this "software - defined architecture" in the automotive field.

Jack Weast explained that if a car originally had 100 electronic control units (ECUs), after integrating the entertainment system and the ADAS system, the number of ECUs only decreased from 100 to 99, but the goal of automakers may be to streamline it to 50.

How to continuously reduce the remaining ECU units? Intel believes that the integration of workloads can be achieved through the software - defined domain controller.

For example, users don't turn on the air - conditioner and the seat heating function at the same time when driving. Intel can put these two workloads on a single device. Originally, two chips were needed, but now only one chip is required. And how to allocate resources in the most optimal way on this chip is a task that Intel is good at.

"We can leverage our experience in data centers and other technologies. One is the pooling of computing resources, which has been done in the cloud for a long time; the other is the dynamic migration of applications. Regardless of the underlying hardware, if it is turned into a virtual machine, the computing power can be dynamically allocated internally, using the most needed part of the computing power. This is the real software - defined vehicle." said Cloud Li, the sales director of Intel's automotive business in China.

However, the allocation of computing resources also requires in - depth involvement in the underlying aspects such as the definition of the electronic and electrical architecture of automakers. In the current trend where Chinese automakers tend to develop their own operating systems and master the underlying capabilities of vehicles, how Intel can persuade automakers will be another challenge.

The following is an edited conversation between 36Kr and others with Jack Weast, vice president of Intel Corporation and general manager of the automotive business unit, and Cloud Li, sales director of Intel's automotive business in China

Question: Why did Intel cooperate with Black Sesame Technologies in ADAS instead of Mobileye?

Jack Weast: We hope to build an open platform and are willing to cooperate with any ADAS manufacturer. Some customers like to use Mobileye's products, while others like to use Black Sesame's products.

This time, we launched the cockpit - driving integration with Black Sesame Technologies, but this concept is not only for Black Sesame. We provide chips based on the Chiplet architecture, and ADAS manufacturers can integrate their Chiplets into our products. In the future, we will cooperate with more ADAS manufacturers.

Question: Is the cockpit - driving integrated platform cooperated with Black Sesame Technologies based on Chiplet?

Jack Weast: Currently, it is not based on Chiplet. Intel's is the cockpit chip. The external cameras will first enter the ADAS chip and then flow into the Intel chip through PCIe. This is the current integration method, but in the future, an architecture based on Chiplet will be built.

Cloud: Currently, the two chips in the cockpit - driving solution mainly rely on Ethernet. The problem is that the cockpit end cannot see the data from the driving - end cameras. Many cameras are redundant to some extent because some are connected to the cockpit end and some are on the driving end. Intel connects the two chips through PCIe to ensure real - time performance. Automakers can come up with more ways to use the ADAS cameras at the cockpit end, and can use multi - modal AI at the cockpit end to recognize the external environment of the vehicle and have more interactions.

Question: In Intel's view, what are the main obstacles preventing the automotive industry from fully transitioning to a centralized electronic architecture?

Jack Weast: Intel adopts the "software - defined vehicle" architecture, which comes from our data centers. However, traditional automakers don't have data centers, and their previous understanding of "software - defined" was off - target.

They think it's just about continuously updating the software, but in fact, the "software - defined" architecture is a new concept and a new way of thinking. It requires completely separating software and hardware, not only in terms of CPU and memory but also in I/O. When a workload comes in, resource evaluation should be carried out at the software level to allocate resources in the most optimal way. This is different from the traditional embedded ECU architecture.

If the entertainment system and the ADAS system are integrated, the 100 ECUs will only become 99, but this is not the best solution and may also increase the complexity of the system, including the complexity of safety - related systems.

What we want to do is to integrate the ECUs and finally reduce the number from 100 to 50. This involves not only the entertainment system and the ADAS system but also systems such as air - conditioners, seats, and door locks.

For example, a customer, an electric vehicle company, has adopted Intel's vehicle - wide architecture. There is a "sentry mode" in the car. After the electric vehicle is turned off, the "sentry mode" can monitor the situation around the car. If this function is built into the cockpit system, it will consume 30W - 40W of electricity.

But we put this workload in the Linux Container, that is, another unit instead of the central computing unit, which only consumes 4W of electricity. Once an emergency is detected, the large - scale system is awakened, and then more electricity will be consumed.

Cloud: To add, Intel views the vehicle as a whole rather than just looking at a single domain. Integrating the cockpit and assisted driving only reduces one domain controller. Our goal is to reduce 100 domain controllers to 50.

We can leverage our experience in data centers and other technologies. One is the pooling of computing resources, which has been done in the cloud for a long time; the other is the dynamic migration of applications. Regardless of the underlying hardware, if it is turned into a virtual machine, the computing power can be dynamically allocated internally, using the most needed part of the computing power. This is the real software - defined vehicle.

Jack Weast: There is great potential in this. For example, now one ECU is for many functions, but in fact, we don't turn on the air - conditioner and the seat heating function at the same time when driving. We can put these two workloads on a single device. This means that originally two chips were needed, but now only one chip is required. Intel helps automakers significantly reduce costs in this way.

Question: Is Intel's current development focus on the cockpit, which means it has given up the intelligent driving market?

Jack Weast: This comes back to what the real software - defined architecture is. In such an architecture, everything is just a workload or an application, whether it's ADAS, infotainment, air - conditioning, karaoke, or anything else.

Since the chip is separated from the software, there is no need to build a dedicated ADAS chip or a dedicated infotainment chip. Instead, a general - purpose, high - performance computing platform is built. This software - defined architecture has been adopted by many industries such as telecommunications and industrial manufacturing. When high - performance general - purpose computing becomes powerful enough, the demand for customization will gradually disappear.

Question: Intel entered the automotive market a bit late. Now the competition landscape is very fierce. Does this mean that Intel has to make some compromises on price? Are there other ways to help Intel get more partners?

Jack Weast: Intel has been in the automotive field for a long time. We have an infotainment product used in more than 30 million vehicles globally, including Chinese brands. Now we have strengthened our roadmap and added new product lines.

The differentiating advantage of our products compared with other manufacturers is the architecture based on Chiplet.

Because the industry generally provides a large chip. Whether it's a high - end vehicle or an entry - level vehicle, the chips are the same. The difference is that for entry - level models, 50% of the chip's functions are turned off, but the chip manufacturing cost remains the same. So for many manufacturers, a large chip is still a relatively costly solution.

Using Chiplet is a more optimal cost structure. The same is true for multi - node Chiplet. We are the first in the automotive industry to support the multi - node architecture.

Because advanced process nodes are very costly. If the entire chip uses the most advanced process nodes, the cost will be the highest. We can split it up, and use older process nodes for a small part, which is lower in cost and more suitable for these functions.

Question: What percentage of Intel's total revenue does the automotive business account for?

Jack Weast: Currently, the automotive revenue is not presented as a separate percentage. But the automotive business is part of Intel's Client Computing Group, the same department as Intel's PC product business. This means that the automotive business can better leverage our huge investment in AI PCs and apply it to the automotive field.

We believe that the automotive business is one of Intel's growth opportunities, and the future of the global automotive industry is being shaped by China today. We hope to be a part of it.