In science fiction works, science fiction writers will depict futuristic cars in a myriad of strange ways, and it is difficult to find similar futuristic cars in different works. But the amazing thing is that science fiction writers have a high degree of imagination for the cockpit of the car, such as having an intelligent assistant that can communicate with each other, the cockpit can recognize the driver's identity and provide personalized settings, and can perceive all kinds of information inside and outside the car......
From today's point of view, the future intelligent cockpit described by science fiction writers seems to be the direction of the industry's efforts, but there is still a big gap in the experience of the products we can access now. After all, it has only been about a year since the "generative AI explosion" that claims to be able to change the entire industry, and most car companies are still thinking about how to implement it smoothly.
However, this does not prevent us from observing some cases in the industry to perceive the experience changes brought about by the combination of cockpit and generative AI, such as the improvement of the interactive ability of voice assistants, more personalized driving experience, interactive AIGC intelligent entertainment, etc., just as the industry generally discusses "AI-defined cars", "AI-defined cockpit" has become a fairly obvious general trend.
In addition to OEMs, the upstream supply chain also delivers ammunition for the "AI-defined cockpit". On April 26, MediaTek released new products of the Dimensity automotive platform, including the Dimensity automotive cockpit platform CT-X1, CT-Y1 and CT-Y0, which are built using 3nm (CT-X1) and 4nm (CT-Y1 and CT-Y0) advanced processes, not only powerful performance, but also a highlight of the support for generative AI, just like Intel redefined the PC with X86 chips, MediaTek may have to play the same role this time.
AI is the soul of the cockpit
The AI cockpit is considered to be the next stage of development of the "intelligent cockpit", and the reason why it is popular in the industry is that "AI" naturally plays the most central role. In particular, the addition of device-side AI has changed our inherent cognition from many aspects such as human-computer interaction, driving experience, entertainment/social/office application scenarios, and AI has redefined all aspects of the entire cockpit.
I think the core change is in interaction, the "AI cockpit" that integrates the device-side large model will change the cockpit interaction mode that relied on buttons + command voice in the past, and replace it with multi-modal natural interaction. Zhang Yutai, deputy general manager of MediaTek and general manager of the Automotive Platform Business Unit, gave us a simple example of how the AI cockpit in the future can sense our needs based on voice, gestures, movements, emotions and other information, which is the so-called "multimodal interaction".
At this stage, the voice assistant combined with the large model AI has indeed achieved a further improvement in perception ability, for example, it no longer needs the wake word to achieve full-time standby, but it can accurately distinguish which of our conversations in the cockpit are setting commands and which are irrelevant commands;
Zhang Yutai, Deputy General Manager of MediaTek and General Manager of Automotive Platform Division
The second experience change is mainly focused on the driving and riding level, and MediaTek has given us some scenarios: through the combination of cloud and device-side computing, the cockpit can provide a personalized driving experience, such as destination navigation recommendations, and can also improve the in-cabin DMS (Driver Monitor System) capability, which can monitor the user's health status in real time.
The functions that can already be realized include the smart car user manual, and you can directly communicate with the voice assistant when you encounter car problems in the future, instead of looking through e-books or paper manuals.
Xiaolei believes that the AI cockpit has stronger potential, so imagine this: when the cockpit system fully integrates the sensors inside and outside the car, the system will use the camera outside the car to identify the user when someone approaches the vehicle, and if it is identified as the owner, the vehicle will be unlocked, and the personalized settings such as seats, air conditioning, music, and navigation will be automatically adjusted, waiting for the owner to use the car, and if it is a stranger, it will enter the warning state, etc.
The biggest feature of the cockpit system empowered by generative AI is that it is "personalized", and the sense of exclusivity will become stronger, so as to transform the car from a simple "four-wheel vehicle" into a smart space, and at the same time improve our car efficiency.
The third change is focused on the personal application level, that is, the entertainment, social and office scenarios. According to MediaTek, the AI cockpit can provide users with personalized services and entertainment, including personalized news recommendations and summaries, personalized music recommendations, etc., and can also use interactive AIGC capabilities to provide interactive games or graphic generation services.
As for social and office scenes, we can use generative AI to quickly edit the images captured by the in-vehicle camera, and after completing a remote meeting in the car, the on-board AI can generate a meeting summary and highlight the key points for us.
In addition to providing richer gameplay, the information of the device-side large model is only processed locally, so it also has the advantages of low latency and privacy protection. As more and more personal information is trained on the end-side of the AI cockpit, its understanding of users will become more and more profound, and it will even become an "intelligent lifeform". Among them, AI will be the soul of this "lifeform", and the one that makes the "soul" ability appear high and low is naturally the cockpit chip.
AI cockpit capability looks at the "core situation"
In essence, the process of the car cockpit from "intelligence" to "AI" means that more parts need to be called, and friends who are familiar with all kinds of digital electronic products should know that when the chip needs to drive more and more parts, the requirements for comprehensive performance are not simple addition but multiplication.
Specifically, the requirements of AI cockpit for chip performance can be divided into three levels: process, computing power, and large model support.
Let's take a look at the "process" first, we know that traditional automotive chips almost do not have too high requirements for the chip manufacturing process, and the mainstream Tier 1 chips such as NXP and Texas Instruments are mostly 28nm or 16nm. The reason is that in the era of EEA (electronic and electrical architecture) is still distributed and domain controlled, most functions are controlled by independent chips, because the number of controlled functions is small and the scenarios are not complex, so the computing power requirements are not high.
However, when it comes to the most advanced "central integration" architecture, manufacturers hope that one or more chips dominate the functions of the whole vehicle, and the advantages of low energy consumption, high energy efficiency and strong performance of advanced process chips are highlighted. At this time, the advanced process is "the more, the better", and it is quite powerful to be able to provide 7 nanometers and 5 nanometers, as for 4 nanometers and 3 nanometers, it can only be "far ahead".
Zhang Yutai said that the advantages of advanced process vehicle chips are also reflected in performance redundancy. The life cycle of automobiles is very long, and now the iterative improvement of computing power and algorithms is very rapid, and car companies often continue to bring software upgrades to the vehicle through OTA. To put it simply, according to Andy Beale's Law, hardware performance is easily insufficient in the face of software requirements, so stronger performance reserves are needed to prepare for future software upgrades.
Taking MediaTek's just-released Dimensity automotive cockpit platforms CT-X1, CT-Y1 and CT-Y0 as examples, shortly after the debut, Antutu officially announced the 4nm CT-Y1 running score data, with a score of 1.07 million points and the strongest mass-produced cockpit chip at the moment. The 3nm flagship chip CT-X1 has more explosive performance, and the computing performance is 30% stronger than that of CT-Y1, which is more than enough to cope with future software upgrades.
At the same time, thanks to the advanced manufacturing process and the leading Armv9 architecture, the new Dimensity automotive cockpit platform has built-in AI computing units and end-to-end generative AI lightweight technology, CT-X1 can support AI large language models with 13 billion parameters, and CT-Y1 and CT-Y0 can also support 7 billion parameter AI large language models. In comparison, the most advanced cockpit chips now support only about 1 billion parameters for large models.
For large model AI, the number of parameters can be directly equated with "IQ", and a higher number of parameters means stronger perception, understanding, and learning capabilities. Many technology companies, including Lenovo, regard the 7B (7 billion) parameter as the lower limit of the amount of large model AI parameters, especially in the process of device-side deployment, it is difficult for large model AI below this parameter to provide a satisfactory experience. In other words, the emergence of MediaTek's new generation of cockpit chips has raised the competitive starting line supported by large model parameters up a lot.
After the comprehensive performance, the multimedia capability of the cockpit chip will become the second key factor that determines the AI cockpit experience. This corresponds to the development trend of multi-screen display and multi-sensor, and the number of parts that need to be connected in the car is so large that it can only be said to be a "stupid method" by relying on high computing power to "vigorously produce miracles". Just as heterogeneous chip architectures are more efficient in handling AI applications, MediaTek Dimensity automotive cockpit platform supports technologies such as one-core multi-screen and MediaTek MiraVision display enhancement, and also optimizes the resource occupation of multi-camera access in advance, providing more efficient solutions in the face of specific needs.
Zhang Yutai is very confident in this, he said, "Our head guests decided to use our (product) after evaluation, that is, they found that after using MediaTek's chips, so many screens and so many cameras, with our CPU computing power is much better than that of competitors." ”
As important as multimedia capabilities is the connection performance of chips, Xiong Jian, deputy general manager of MediaTek's automotive electronics business department, has a specific explanation: when the number of screens running at the same time in the car increases, the pressure on network resources increases sharply, and the bandwidth of the 4G network is difficult to meet the simultaneous work of multiple screens. Just imagine, playing a video at the same time and playing 4 videos at the same time have different bandwidth requirements, and in the future, based on the development trend of multi-screen in the cockpit, 4-screen simultaneous broadcast or even 6-screen simultaneous broadcast will be the norm, and the advantages of 5G network will be highlighted.
The stability of the network cannot be ignored, although a lot of work can be handled locally with a large model on the device side, but the cloud + end will still be the most mainstream way of working, and the basic application scenario of navigation also needs a network connection.
MediaTek's solution can be understood as a "highly integrated stack" - the new cockpit platform has added Ku-band 5G NTN satellite broadband technology, on-board 3GPP 5G R17 Modem, etc., on the one hand, the world's leading mobile Internet connection capabilities are moved to the car, and based on the requirements of vehicle specifications to ensure its stability.
The last thing to discuss is the issue of cost, which of course is MediaTek's traditional advantage. MediaTek is the global chip leader, with chip shipments in several fields such as smartphones and smart TVs ranking first in the world, plus the global shipments of the Dimensity automotive cockpit platform have exceeded 20 million sets - the scale effect is enough to bring cost advantages to MediaTek.
Weizhi Yu, deputy general manager of MediaTek Dimensity Automotive Connectivity Platform, said that the 3nm and 4nm platforms can achieve "19-in-1" functional integration, including ISP, DSP, Wi-Fi, Bluetooth, etc. However, if it is replaced by other chips with low integration, these peripheral functions need to be realized by purchasing other chips, which is very high-integration, which can help car manufacturers greatly reduce costs in terms of hardware cost.
Weizhi Yu, Vice President of MediaTek Dimensity Connectivity Platform
Assuming that MediaTek's platform only needs 8G memory to easily run all functions, while the platform of competitors requires more than 16GB, the former can obviously save more money for OEMs. In addition, there is also the cost of development, if the development support is perfect enough, the OEM does not need to invest a large number of developers to develop related applications, and there is also a large amount of debugging and optimization in the later stage.
It is conceivable that the four dimensions of performance, multimedia optimization, connectivity and cost will become an important reference for car manufacturers and consumers to evaluate the "AI cockpit" in the future, and it is not surprising if you say that this is exactly what MediaTek excels at. Whether in terms of computing power, AI support or peripheral configuration, MediaTek Dimensity automotive cockpit platform has the strength to define the rules of the game, and the competitive logic of the automotive intelligent cockpit will eventually be on par with smartphones and even the semiconductor industry - winner takes all.
The chip is powerful, and the cockpit will be rewritten by AI
From the interior materials and screen size in the past to the intelligent experience and AI capabilities, it is no wonder that some people jokingly call the smart cockpit a "large mobile phone", because the focus of competition ultimately falls on the chip. Objectively speaking, this is the inevitable result of the car embarking on the road of intelligence, and the final competition of any scientific and technological products will return to semiconductors, after all, in the real world, only chips and computing power are the real "magic".
This has a significant impact on the entire automotive industry, especially for product decision-makers, who must better understand chips, large models, and computing power, otherwise they will be completely left behind in the wave of industry transformation. In addition, advanced process chips have received attention due to the central integrated EEA architecture, which in turn will further promote the central integration of the EEA, and eventually smart cars may only have one chip and one "brain".
For the entire automotive supply chain, this is not a major reshuffle. The rapid rise of advanced process semiconductor companies, such as MediaTek and other semiconductor companies with the advantages of advanced processes and large-scale tape-out, is likely to replace traditional semiconductor companies in a short period of time and wipe out their share.
In this regard, You Renjie, senior vice president of MediaTek and general manager of the Computing Unicom Metaverse Business Group, said, "In the medium and long term, IC design companies with advanced process capabilities can have a foothold in the intelligent new energy vehicle industry defined by generative AI in the future, which is one of the values that MediaTek can provide immediately in the market."
You Renjie, Senior Vice President of MediaTek and General Manager of Computing Unicom Metaverse Business Group
When a semiconductor company's product process is more advanced, the performance is more powerful, the periphery is more complete, the solution is more reasonable, and even the cost is more advantageous—I only think of the four big words "dimensionality reduction strike".
Having said that, this is an opportunity for some traditional car companies with less intelligent capabilities than "corner overtaking" - this is not contradictory to the above, MediaTek Dimensity automotive cockpit platform CT-X1, CT-Y1 and CT-Y0 The three brothers already have a high degree of integration, MediaTek itself can also provide a series of solutions, coupled with help in application development, which means that traditional car companies can quickly obtain a set of industry-leading cockpit intelligent solutions, so as to stride into the era of AI cockpit.
From this point of view, perhaps this is the mission of semiconductor technology companies with advanced processes in the new era, to be a real technology enabler, or to help the industry achieve "science and technology inclusiveness". The era of "AI-defined cockpit" is an era of cruelty and tragedy, and it is also an era full of opportunities, and for OEMs, choice and effort are equally important.