Geely's thousand-yuan lidar explodes, the latest high-end smart driving chip crushes Nvidia

LiDAR, who else wants to join the game at this time?

And it is none other than Geely, which is good at multi-track layout. It has just released two self-developed lidars through its subsidiary Yikatong.

One semi-solid state has 192 wires, and one all-solid state has 150 wires, which are directly approaching the industry-leading level. They will be mass-produced and launched this year, and the cost will be reduced to a thousand yuan.

Ekatong, which has already released a number of smart cabin and smart driving software and hardware solutions, why does it still need to deploy hardware sensors?

Because CEO Shen Ziyu once said that Yikatong’s business must be benchmarked: Huawei Auto BU.

Yikatong releases two lidars

Let’s take a brief look at the performance parameters of these two laser radars.

The first is a long-range semi-solid 192-line lidar with a maximum detection range of 300m and a ranging capability of 180m@10% reflectivity. It is mainly used on car roofs as the main lidar.

It is equivalent to a black tire placed on the ground 180m in front of the car, which can be identified by the lidar on the roof.

And this lidar has a field of view of 120 degrees x 25 degrees and can generate 666 pixels in the horizontal direction.

This lidar was also placed on site, and the real-time point cloud effect is as follows:

Another short-range all-solid-state lidar has a maximum detection range of 50m and can achieve a ranging capability of 40m@10% reflectivity. It is mainly used on both sides of the vehicle as a blind-filling lidar.

The field of view is 120 degrees x 50 degrees, the line bundle reaches 150 lines, and a maximum of 360 pixels are generated in the horizontal direction.

The real-time point cloud effect of this lidar was also displayed on site:

Moreover, the SPAD array, VCSEL array, and laser driver IC used in short-range lidar are all self-developed by Yikatong, which are the core detectors and laser modules of lidar.

Shen Ziyu, CEO of Yikatong, said that the company has actually invested about three or four years in these two lidars. The goal is to achieve full customization and comprehensive self-research of lidars, so that it can achieve the ultimate performance and BOM. Costs are also reduced to the extreme.

According to Yikatong, the cost of this short-range lidar can be reduced to US$100 (approximately 720 yuan), while the long-range lidar is US$200 (approximately 1,440 yuan).

High performance and low price are the key points emphasized by Yikatong lidar, and it will be mass-produced by the end of this year and will be launched on the Polestar model.

Lidar is Yikatong’s new attempt to expand its product line of smart car solutions. This release certainly did not leave out Yikatong’s core business—chips.

What else has Yikatong released?

Longying No. 1 is a chip developed by Yikatong, which is mainly used in smart cockpits. This time, Longying Smart Driving AD1000, a smart driving chip of the Longying series, was released.

It adopts 7nm technology and reaches automotive grade. The CPU computing power is 250KDMIPS and the NPU computing power is 256TOPS. Officially, it has six characteristics: high integration, high computing power, high perception capability, high ease of use, high security and rich interfaces.

Moreover, Longying Smart Driving AD1000 supports multi-chip collaboration, with one chip supporting up to L2++ advanced smart driving, two chips supporting L3, and four chips supporting L4.

The official also compared it with an international manufacturer starting with N (NVIDIA: Just give me your ID number), saying that the Longying Zhijia AD1000 has 10% higher CPU capability, 100% higher NPU capability, and 100% higher image signal than N products. The processing (ISP) capacity is 160% higher, the NPU local storage space is 185% larger, and it has higher performance.

In addition to new smart driving chips, Yikatong also released three new smart cockpit computing platforms.

The first is the Atlas computing platform based on Snapdragon SA8255P, and the Parker computing platform based on SA8295P.

Among them, the Atlas platform uses automotive-grade 5nm chips. The CPU reaches 230KDMPIS, the GPU computing power is 1.3TFLOPS, the NPU computing power is 24TOPS, and the memory capacity and bandwidth are both better than 8155.

And Ekatong said that the Atlas platform can not only be used as a smart cockpit platform, but its functions can also be extended to an integrated cabin, parking and parking platform, which supports cockpit, driving and parking domain systems, and supports up to basic L2 smart driving capabilities.

The Parker computing platform is also based on automotive-grade 5nm chips, with GPU computing power reaching 2.9TFLOPS and NPU computing power reaching 46TOPS. It has stronger graphics rendering capabilities and smart cockpit interaction capabilities.

Therefore, the Parker computing platform can support vehicle-machine immersive 3D HMI, rendering 3D scenes in real time while ensuring interactivity.

At the same time, the Parker computing platform can not only be expanded into an integrated cabin, cabin, and parking platform like the Atlas platform, but it also supports the application of large AI models and has stronger interactive capabilities.

In addition to the two platforms based on Qualcomm's cockpit chips, Ekatong also launched another higher-performance platform, the Qiogori computing platform, based on Qualcomm's recently released Snapdragon 8s Gen3 chip.

The Qiogori computing platform CPU reaches 330K DMIPS, the GPU reaches 4.2T FLOPS, and the NPU reaches 60 TOPS.

Vehicles using the Kiogori computing platform support up to 8K resolution, and the hardware supports ray tracing technology, with a frame rate of 240FPS, making the vehicle higher definition.

At the same time, the Kiogori computing platform can also support the device-side deployment of large models with 10 billion parameters, such as large language models, large visual models, and generative AI, bringing a better experience to users.

In addition, the Antola 1000 and Antola 1000Pro computing platforms previously released by Ekatong were originally mainly used for smart cockpits. Now they have been expanded into cabin-parking integrated computing platforms that support basic L2 smart driving capabilities and can reduce 20% of program cost.

In addition, Ekatong said that its intelligent driving system already has high-speed NOA capabilities, and will soon cooperate with OEMs to launch an urban NOA system that can be deployed nationwide. Four cities in Zhejiang, Jiangsu, Guangdong and Shanghai will be the first to experience it.

At this point, it is better to summarize the overall layout of Yikatong.

In terms of SoC chips, there is the Dragon Eagle series developed by Xinqing Technology, including Dragon Eagle One and Dragon Eagle Smart Drive AD1000.

In terms of smart driving hardware, there are two laser radars launched by its subsidiary Light Torch, as well as corresponding software tools.

At the same time, Yikatong has also launched a variety of smart driving and smart cabin solutions based on these chips and external cooperation, such as the Antola 1000 computing platform and this time the Kiogori computing platform.

It fully covers hardware and software, providing a variety of solutions for car companies. Officials also stated that Ekatong’s basic products have been installed in more than 6 million models around the world. It is indeed a bit like “it does not build cars, but helps car companies make cars”. It means "good car".

The key to popularizing high-end smart driving lies in lidar

Finally, let’s talk briefly about lidar.

Lidar generally includes a transmitting module, a receiving module, a scanning module and a signal processing module. According to the different scanning modules, lidar can be divided into three categories: mechanical, semi-solid and solid.

Each of these three has its own characteristics. For example, although the mechanical type can quickly provide a panoramic field of view, it is large in size and high in cost, making it difficult to meet car-level standards.

Among them, the semi-solid type and solid-state type, although the cost is lower than the mechanical type, the semi-solid type also has many shortcomings such as large size and narrow field of view, while the solid-state type has many shortcomings such as shorter detection range.

But it is undeniable that lidar is still superior to other sensors required for intelligent driving in terms of detection speed, accuracy, and anti-interference ability.

For example, 4D millimeter-wave radar is often hailed as an alternative to lidar. While its performance is comparable to lidar, it is also more advantageous in terms of price and can cost only 1/10 of the cost of lidar.

However, in similar scenarios such as underground garages, 4D millimeter wave radar will produce a lot of noise, resulting in lower detection accuracy; and from the point cloud density point of view, 4D millimeter wave radar is still lower than lidar.

△ 4D millimeter wave radar point cloud

Therefore, it is not difficult to see that the models that have been mass-produced in China and can achieve the highest level of intelligent driving capabilities (urban NOA) are still equipped with lidar, such as Huawei's Wenjie and Zhijie, as well as Xiaopeng and Jikrpton Waiting for models.

The price of these models without exception exceeds 200,000 yuan. But as we all know, the price range of domestic new energy vehicles has the highest proportion in the market below 200,000 yuan, especially the 100,000-200,000 yuan level.

However, Yikatong said that lidar is now in a "price bargaining period." A research report from Guolian Securities also stated that the cost of lidar for mainstream companies in the industry is continuing to drop, with the highest drop reaching 72.8%.

Of course, other options are evolving as well. For example, 4D millimeter-wave radar with a price as low as 1/10 of LiDAR is advancing mass production; DJI has brought a pure vision solution to reduce the urban NOA cost to 7,000 yuan, and 150,000-class passenger cars can also use high-end smart driving; Tesla Lageng has completed the North American push of pure visual FSD.