In recent years, the application of AI at the edge has become an unstoppable trend, and embedded AI refers to this application acceleration mode that can efficiently process various deep learning neural networks at the edge. Embedded AI can enable products to play
intelligent detection, identification, classification and other functions at the equipment level, so it has become a hot spot for intelligent product development or industrial upgrading.
However, embedded AI not only involves a series of integration problems such as algorithm design and tool design at the software level, but also chip design at the hardware level, so it is also a field of high threshold and complexity.
Main application markets and trends of embedded AI/ML
ADAS and autonomous driving, smart cities, smart healthcare and industrial vision applications are typical scenarios for embedded AI applications, and are also market areas where Xilinx has been focusing.
Automotive field
In the automotive field, Xilinx mainly focuses on forward, rear, circumnavigation, blind spot detection, automatic parking, intelligent cockpit, sensor fusion and domain control, covering ADAS and autonomous driving from L2 to L5 levels. We know that the automobile
is a special scenario, which requires the real-time and correctness of the entire ADAS or autonomous driving system, which requires embedded chips with superior AI acceleration capabilities, flexible chip architectures, and the industry's highest reliability testing
standards.
It is worth mentioning that in order to ensure the speed and correctness of decision-making and control in vehicle driving, multi-sensor fusion is an inevitable trend. Nowadays, cars are equipped with more and more sensors, such as cameras, radar, liDAR, millimeter
wave radar, ultrasound, and more, and Xilinx brings great value to this multi-sensor architecture. These play an important role in automotive multi-sensor solutions, and even 5G combined with AI vehicle-road collaboration.
Smart city field
Xilinx products have been widely used in the field of smart city, such as smart transportation, smart retail, smart buildings, etc. In addition to flexible I/O, support for MIPI, LVDS and a variety of sensor interfaces, flexible ISP support policies, H264/H265 video codec
processing units and other capabilities, Xilinx products also have powerful real-time AI processing capabilities.
Medical and industrial vision
The medical and industrial vision fields are also important scenes for the landing of Xilinx embedded AI. In the medical field, we can provide medical image analysis or super-resolution enhancement based on deep learning, and use AI to provide more accurate
judgments for medical personnel in endoscopy, ultrasound, nuclear magnetic, X-ray detection and other scenarios. Since the outbreak of COVID-19 in 2019, Xilinx and industry partner Spline.ai have used Vitis AI to combine MPSoC edge devices with integrated deep
learning processing unit Dpus and AWS iot services to develop scalable intelligent solutions for lung infections and COVID-19 prediction systems.
Industrial vision field
In the field of industrial vision, Xilinx products also play an important role in smart factories, intelligent industrial cameras, vision control and robots, etc. The role of AI is mainly reflected in the predictive maintenance control supported by machine learning such as
defect detection, text recognition, real-time analysis, remote diagnosis and other embedded intelligent terminals.
Embedded AI scene landing challenges and solutions
In the process of landing embedded AI products, the challenges encountered by developers are mainly:
· First, is the solution flexible and scalable to accommodate different product sizes or custom modules?
Second, for users who lack FPGA development experience, are there easy to use AI development tools to reduce the difficulty of development?
· Third, how to break through the bottleneck and maximize the acceleration performance of AI at the edge of limited computing power?
In order to meet these challenges, Xilinx officially launched the Vitis AI solution in early 2020, which is an AI development platform for Xilinx Zynq SoC, Zynq MPSoC, Alveo and Versal ACAP, which can bring users the most powerful Machine learning acceleration
performance, which is mainly achieved through the neural network acceleration engine DPU in the solution, a series of automated software tools (fixed-point processor, compiler, optimizer) and AI runtime and acceleration library.
Finally, in order to help customers achieve a breakthrough in edge performance, Xilinx integrated a large number of bottom-optimized acceleration libraries in the Vitis software development environment, and provided them to users through C++ and OpenCL library.
Therefore, in addition to DPU acceleration for AI, We speed up the entire process of end-to-end pre - and post-processing to maximize performance at the edge.
On February 14, 2022, AMD announced the completion of the acquisition of Xilinx, and former Xilinx Board members Jon Olson and Elizabeth Vanderslice have joined AMD's Board of directors.
Summarize
The above is about the relevant introduction of Xilinx chip. Hope to help you understand Xilinx chip products! More knowledge of electronic components please continue to pay attention to Joinwin ! Joinwin is a professional electronic components distributor, since its
establishment has always insisted on providing electronic inventory solutions for original factories, foundries, brands owner, agents, etc., taking environmental protection as its responsibility, to help all kinds of idle electronic components, residual materials fast
transaction! If you need to deal with electronic inventory materials, you can contact us!