The Multimodal fusion perception AI chip "Xiaoxi X1" is an intelligent SoC chip for fusion processing of vision, inertial navigation, vibration, and other sensors. It supports hardware synchronization of multiple sensors, acceleration of information processing, and it has built-in processing unit for neural network computing. It can complete the fusion processing of multi-mode data such as video, inertial navigation, audio, etc. on the chip, and improve the perception ability and intelligent analysis ability of the device end.
The X1 chip integrates a high-performance ISP processor with biomimetic visual processing capabilities developed by the company team, which can support up to 4K video input and processing. It also integrates the team's self-developed Image Processing Acceleration Engine, including multi-sensor hardware synchronization, sub-pixel level video distortion correction, video key feature extraction, stereo vision perception, and other functions. In addition, the chip integrates H264/265 codec that supports 4K video, a 4-core CPU processor with a maximum main frequency up to 1.5GHz, RISCV real-time processor with a main frequency of up to 400MHz, and an NPU neural network accelerated processing unit with 4TOPs computing power. At the same time, the chip integrates multiple commonly used control interfaces, achieving an integrated design of perception, computation, and control.
This chip architecture simulates the processing mechanism of the human brain, combining multi-sensor data collection with high-performance processing, and has the ability of "perception calculation control". It can improve the intelligence, integration, and real-time autonomous information processing level of robots and other devices, and can serve application scenarios such as intelligent robots, drones, and intelligent security.
The chip integrates various general-purpose processors and multiple algorithm acceleration units for visual processing, which can be used to achieve various functions such as 3D reconstruction, autonomous positioning, and object recognition, providing low-power chip level solutions for applications such as intelligent robots.
（1）Real-time 3D reconstruction and ranging
The chip integrates a self-developed distortion correction module (LDC) and stereo matching module (SGM), which can complete sub-pixel correction and real-time depth calculation of chip images, and complete target ranging and 3D reconstruction.
（2）Real-time positioning and navigation
The chip integrates acceleration operators for visual SLAM applications, supporting nanosecond hardware synchronization of sensors such as vision and inertial navigation, and can be applied to real-time self-pose estimation and positioning of devices such as robots and drones on the chip.
（3）Real-time positioning and navigation target recognition and tracking
Built-in NPU processor, supporting more than 120 acceleration operators, can complete on-chip acceleration inference for custom networks, and achieve functions such as target recognition and tracking.
Flip-chip 21*21mm 771pin
4-core Cortex-A55 operating frequency 1.5GHz
L1 Cache：I 32KB, D 32KB. L2 Cache：512KB
Maximum operating frequency:400MHz RAM: 256KB
Achieving computing power 4Tops@8Bits
Self developed IP
Supports 2-channel video access（MIPI/LVDS IN）
Can be configured as monocular (up to 4K), binocular (up to 2K), monocular+binocular, binocular+binocular
Support filters (noise reduction, sharpness, blur);
Programmable gamma correction for sensor adaptation;
Sensor crosstalk compensation;
Enhanced Colorimetric Noise Reduction (CNR);
Supports 3A functionality
Supports real-time subpixel distortion correction
Supports real-time feature extraction and depth calculation
Encoding and decoding
Supports 1 channel 1080P@30FPS （MIPI/LVDS OUT）
Support USB2.0 OTG，USB3.0 Device
Support 10M/100M/1000M mode
Supports MII/GMII/RMII/RGMII/RGMII interfaces
LPDDR5/4,Compatible with single/dual channels
Up to 6.4Gbps
Support for EMMC 5.0
Supports SDIO 3.0
Internal control module
Communication and mutual transmission terminals between CPUs, temperature sensing control
Power on, clock, and reset control module
6-way independent Timer
Pin reuse control module
Security control module
4, supporting up to 1.5Mbps
4 can access 8-way voice signals
More than 80
5-way parallel sigma delta ADC (Audio Codec)
1 SARADC (4 Lane)