Showing results for 
Show  only  | Search instead for 
Did you mean: 

High-Performance Xilinx FPGA Platform for AI/ML Edge Computing

Xilinx Employee
Xilinx Employee
1 0 2,358

This guest post is by Thasneem Niyaz from iWaveSystems 

With the advent of IoT and the proliferation of connected devices, one of the biggest challenges in developing competitive IoT solutions is the ability to bring intelligence in edge devices. Edge computing is crucial in IoT applications as it paves the way for faster real-time inference in on-premise infrastructure. This results in a dramatic improvement in overall system reliability and performance.

With edge computing increasingly forming the foundation of next-generation connected devices, it is important to highlight the significance played by hardware accelerators in determining the efficiency of such applications. The hardware component forms the core building block, and therefore should be considered with utmost importance while developing edge solutions.

Over the years significant advancements in FPGA  technology has led to FPGA's becoming mainstream in IoT edge platforms.  FPGAs sophisticated performance coupled with their ability to deliver the highest throughput at the lowest latency makes them ideal for edge applications.



iWave’s Zynq® UltraScale+™MPSoC FPGA SOM, shown above, offers versatile hardware accelerations for intuitive deployment of functions such as, image /speech recognition, object /pose detection, etc. and a flexible platform that enables developers to continually refine features and sharpen their competitive edge. Implementing artificial neural networks in FPGAs provides the flexibility to adapt applications with changing standards and end-user demands, which in turn future proofs the designs.iWave also provides comprehensive Zynq® UltraScale+™MPSoC development platform (shown below) for immediate evaluation of AI/ML applications.1.jpg


Why Xilinx?

  • Xilinx configurable Deep Learning Processor Unit (DPU) engine for convolutional neural networks, accelerates AI inferencing.
  • Heterogeneous execution environment with high inference speed and accuracy for machine vision applications.
  • The Xilinx AI platform supports a number of industry-standard models: Caffe, Tensorflow, Darknet, MXNet frameworks.
  • Xilinx AI programming model makes it easier to develop and deploy deep learning applications on FPGA Zynq platforms.
  • Provides the scalability device family for different AI applications.
  • Xilinx unique tools for model optimization and model compression, reduce neural network model complexity by 5x to 50x and takes the performance of the AI inference to the next level.
  • AI software development tools like Xilinx SDK_AI, DNNDK (Deep neural network development Kit), Vitis speeds up the path of the development.
  • Xilinx FPGA platform provides easy interface access for USB camera, serial digital interface camera, Internet protocol camera, and Ethernet for AI Edge computing solutions.


Xilinx Vitis AI implementing AI/ML inference with iWave’s Zynq® UltraScale+MPSoC SOM



The image shows some AI/ML acceleration examples on iWave’s Zynq UltraScale+ MPSoC development kit using the Xilinx Vitis AI platform.

The Zynq UltraScale+ MPSoC SOM features an intelligent blend of MPSoC and FPGA functionality in an ARM® + Xilinx FPGA architecture. The heterogeneous ARM® multicore processors complements the edge applications with high-performance non-real-time processing such as system boot, peripherals management, server communication, etc., while offloading the FPGA to execute critical real-time tasks using Vitis AI models.

With its support for a wide range of Neural Networks, the Xilinx Vitis AI platforms are continuously evolving, integrating new and advanced algorithms for improved determinacy and inference in AI/ML applications. iWave supports a huge portfolio of Vitis AI models based on various application needs. 


Example industrial edge applications

Smart City: Intelligent platforms that perform real-time monitoring and inference using a combination of FPGA acceleration and neural networks.

Intuitive ADAS: Real-time computing platform capable of generating accurate and timely inferences with AI/ML algorithms on-board.

Industrial Automation: AI powered intelligent devices that can sense, connect and compute massive data influx, perform predictive maintenance and generate smart intuitive decisions.

Smart health care: AI/ML accelerated devices that enable real-time monitoring and diagnosis for early diagnosis of diseases.



Needless to say, edge computing continues to revolutionize the IoT ecosystem with competitive applications. iWave’s Xilinx/Deephi platforms offer high-performance hardware acceleration for AI/ML inference and accelerates innovations at optimized cost and lead time. 

For further information or inquiries please write to or contact our Regional Partners.



Thasneem Niyaz, Technical Writer, Member Technical, iWaveSystems.