At the 2025 Shanghai Auto Show, ThinkSilicon Technology officially launched its newly developed AI cockpit chip, the X10. This chip utilizes an advanced 4-nanometer process, boasting powerful computing capabilities that enable the local deployment of a 7B-parameter multimodal large model on the device. This marks a significant breakthrough for ThinkSilicon in the intelligent cockpit chip field and is expected to significantly enhance the intelligent driving experience.

QQ_1745719021167.png

From a technical specifications standpoint, the X10 chip features a 200K DMIPS Arm v9.2 architecture CPU, a 1.8 TFLOPS GPU, and a 40 TOPS NPU, ensuring high efficiency in handling complex computational tasks. The chip also supports 128-bit wide 9600MT/s LPDDR5x memory, resulting in a system memory bandwidth of 154GB/s—significantly exceeding that of other similar products currently on the market.

Thanks to this outstanding memory bandwidth, the X10 chip can flexibly schedule multiple smaller AI models while running large AI models, supporting the collaborative processing of various AI inference tasks. This highly efficient performance configuration allows the in-vehicle system to more intelligently process diverse information from both inside and outside the vehicle.

Beyond its powerful computing performance, the X10 chip also integrates an Image Signal Processor (ISP), an Audio Digital Signal Processor (DSP), a 4Kp120 video encoder, and an 8K display engine. It also supports various interface standards, including UFS4.0, PCIe5.0, and USB3.1/2.0, ensuring efficient connectivity with other devices.

The X10 also integrates a wealth of sensor interfaces. In addition to conventional voice recognition capabilities, it can also support the perception of in-vehicle occupant status and the monitoring of the external environment. Combined with vehicle network access, it obtains real-time vehicle status and location information. These functions provide comprehensive information input for the multimodal AI large model, greatly enriching the application scenarios of the intelligent cockpit.

ThinkSilicon Technology stated that the X10 series chips are planned for mass production in 2026. In the future, they will provide strong technical support for automotive intelligence, autonomous driving, and other fields, driving the digital transformation of the automotive industry.

Key Highlights:

🧠 ThinkSilicon's X10 chip uses a 4-nanometer process and supports the local deployment of a 7B-parameter multimodal AI model.

💪 The X10 chip features high-performance CPU, GPU, and NPU, with a memory bandwidth of 154GB/s, far exceeding competing products.

🚗 The chip integrates rich sensor interfaces, enhancing the diverse application scenarios of the intelligent cockpit, with mass production expected in 2026.