Although edge AI chips are inferior to cloud AI chips in terms of computing performance and technological advancement, they still have irreplaceable unique advantages, which are roughly as follows:
- Protect data security and privacy
- Still usable on occasions with poor internet connection. For example, AI chips with integrated vision processing units (VPUs) can be used to efficiently analyze and preprocess digital images. Cameras embedded with this AI chip can perform data analysis in real time, and only transmit relevant data to the cloud while ignoring unnecessary data, thus reducing storage and bandwidth requirements.
- Reduced power consumption - edge AI chips consume much less power than cloud AI chips, and AI calculations can be performed on many battery-powered devices with extremely low power consumption.
- Low Latency Data Transfer
- Low-cost deployment—Edge and end-side devices are generally installed in a large amount. AI chips embedded in such devices are required to be lower in power consumption and cost than cloud AI chips, so that AI functions can be deployed on a large scale.
The AspenCore analyst team selected 15 domestic edge/device AI chips from 15 domestic AI chip manufacturers, including: Rockchip, Quanzhi, Qingwei Smart, Kuxinwei, Yizhi Electronics, Shishi Technology .
Rockchip RK3588 AIoT chip
Allwinner V535 Intelligent Driving Vision Processing Chip
V535 is a new generation of intelligent driving dedicated processor developed by Allwinner Technology. It is a high-performance, highly integrated, and highly stable industrial-grade chip that integrates image and video processing and AI vision; V535 has achieved multiple breakthroughs in response to customer needs. Innovative design and optimization, it can encode and decode multi-channel images in real time, and at the same time have AI (such as: vehicles, people) detection and recognition functions.
Core technologies include:
- High-efficiency combined computing power. Single-core A7 1.2GHz with the largest 0.5T professional neural network processing unit, deeply optimizes system efficiency for typical AI application scenarios, and provides complete AI middleware and supporting model conversion tools, which can quickly adapt to customer-owned or third-party algorithms, helping The customer's AI products are quickly implemented.
- High integration and scalability. Integrating a professional star-level image processor and H265/H264 codec, it can realize the maximum 5 million-level real-time camera image processing and multi-channel time-sharing encoding. At the same time, it can support external expansion of various analog high-definition camera input and screen output interfaces, and Support low-power sleep wake-up, satisfying the expansion function of vehicle driving category.
- High specification quality assurance. After rigorous reliability testing experiments such as uHast, TCT, HTOL, PCT, HTSL, etc., the working temperature of -40~85 degrees is industrial grade, FDPPM<200PPM, and the chip life is as high as 10 years.
Main products and applications: smart driving records and driving behavior detection products, including AI monitoring and early warning devices for vehicle blind spots, driver behavior detectors, etc.
Qingwei Intelligent TX510 Intelligent Vision Processing Chip
Kuxinwei AR9341
Yizhi Electronics SV823 End-to-side Reasoning AI SoC Chip
Shishi Technology Speck series "integration of sense and calculation" dynamic vision intelligent SoC
Jiutian Ruixin ADA200 "Integration of Sense, Storage and Computing" Chip
Hangzhou Guoxin GX8002 low-power AI voice interaction chip
Zhicun Technology WTM2101 storage and calculation integrated SoC chip
Timesformer Intelligent Processor AT1611
Shencong AI dedicated voice chip TH1520
Love Core Yuanzhi AX620A
Lingbi Technology Brain-Inspired Chip KA200
Qiying Tailun intelligent voice AI chip CI1122