English
Back
Open Account
半导体行业观察
wrote a column · Mar 3, 2025 13:53

The first AI ASIC power system chip for outputting array signals constrained by physical quantities has been released.

Recently, China Science (Shenzhen) Wireless Semiconductor Co., Ltd. launched a power system chip for embodied robots. It outputs array PWM control physical quantity signals through an 'Edge Physical Model' and is equipped with FPGA (AI ASIC) + array Gallium Nitride GaN drivers, allowing the high-frequency neural reflex system frequency to exceed 250Hz and reducing the latency in the perception --> decision --> execution link to less than 5ms, close to human neural reflex speed (about 30-100ms). This chip generates 3D virtual models and posture coordinates based on AI physical models and, through the chip's built-in 'edge physical posture model' (which refers to integrating physical equations constraints into deep learning with physical laws, energy conservation, sensors, and vision in neural networks through data-driven methods), combines multimodal data fusion from inertial measurement units (IMU) and visual sensors to output array PWM signals that control over 100 simulated muscles and Servo Motor systems to complete complex atomic operations and dynamic balance. A single posture movement exceeds 32 degrees of freedom, enabling the robot to perform complex body movements and physical constraints similar to humans. This is also the first power system chip that constrains physical quantities and outputs array current signals based on AI models. The company's research team has been deeply engaged in this field for many years, and the technology of this chip has obtained multiple invention patents and has won second and third prizes in the ICCV International AI Competition for nearly two years, while also achieving the championship in the 2024 International AI Vision Tactile CVPR.
Recently, China Science (Shenzhen) Wireless Semiconductor Co., Ltd. launched a power system chip for embodied robots. It outputs array PWM control physical quantity signals through an 'Edge Physical Model' and is equipped with FPGA (AI ASIC) + array Gallium Nitride GaN drivers, allowing the high-frequency neural reflex system frequency to exceed 250Hz and reducing the latency in the perception --> decision --> execution link to less than 5ms, close to human neural reflex speed (about 30-100ms). This chip generates 3D virtual models and posture coordinates based on AI physical models and, through the chip's built-in 'edge physical posture model' (which refers to integrating physical equations constraints into deep learning with physical laws, energy conservation, sensors, and vision in neural networks through data-driven methods), combines multimodal data fusion from inertial measurement units (IMU) and visual sensors to output array PWM signals that control over 100 simulated muscles and Servo Motor systems to complete complex atomic operations and dynamic balance. A single posture movement exceeds 32 degrees of freedom, enabling the robot to perform complex body movements and physical constraints similar to humans. This is also the first power system chip that constrains physical quantities and outputs array current signals based on AI models.
Recently, China Science (Shenzhen) Wireless Semiconductor Co., Ltd. launched a power system chip for embodied robots. It outputs array PWM control physical quantity signals through an 'Edge Physical Model' and is equipped with FPGA (AI ASIC) + array Gallium Nitride GaN drivers, allowing the high-frequency neural reflex system frequency to exceed 250Hz and reducing the latency in the perception --> decision --> execution link to less than 5ms, close to human neural reflex speed (about 30-100ms). This chip generates 3D virtual models and posture coordinates based on AI physical models and, through the chip's built-in 'edge physical posture model' (which refers to integrating physical equations constraints into deep learning with physical laws, energy conservation, sensors, and vision in neural networks through data-driven methods), combines multimodal data fusion from inertial measurement units (IMU) and visual sensors to output array PWM signals that control over 100 simulated muscles and Servo Motor systems to complete complex atomic operations and dynamic balance. A single posture movement exceeds 32 degrees of freedom, enabling the robot to perform complex body movements and physical constraints similar to humans. This is also the first power system chip that constrains physical quantities and outputs array current signals based on AI models. The company's research team has been deeply engaged in this field for many years, and the technology of this chip has obtained multiple invention patents and has won second and third prizes in the ICCV International AI Competition for nearly two years, while also achieving the championship in the 2024 International AI Vision Tactile CVPR.
The company's research team has been deeply engaged in this field for many years, and the technology of this chip has obtained multiple invention patents and has won second and third prizes in the ICCV International AI Competition for nearly two years, while also achieving the championship in the 2024 International AI Vision Tactile CVPR.
With the large-scale commercial use of large models (such as ChatGPT, open AI, DeepSeek), autonomous driving, Edge Computing, and other scenarios, the Technology Sector continues to thrive, and DeepSeek concept stocks are once again active, boosting overall market investment sentiment and heralding the arrival of the 'AI application phase.' Traditional general-purpose chips (GPUs) are difficult to meet the demand, and the demand for specialized AI ASIC chips has surged. Countries view AI chips as strategic high points, with the USA investing 52.7 billion dollars in the 'Chip Act,' and China's '14th Five-Year Plan' emphasizing semiconductor self-sufficiency. HTSC points out that rapid developments in AI technology will bring immense market opportunities, and by 2025, the Global AI market size will exceed 500 billion dollars. HTSC suggests investors pay attention to AI Hardware (such as GPUs, AI ASIC) and Software. This time, the power system chip for general embodied robots launched by China Science Semiconductor will significantly reduce the development costs and posture learning time for embodied robots, undoubtedly accelerating the commercialization of the embodied robot industry and filling the gap in the specialized chips for robot joints and 'bionic cerebellum.'
Risk Disclaimer: The above content only represents the author's view. It does not represent any position or investment advice of Futu. Futu makes no representation or warranty.Read more
4322 Views
Report
Comments
Write a Comment...