Building an adaptable, intelligent world: a Q&A with Xilinx

Demand for live video streaming has skyrocketed during the pandemic with more events, classes, and meetings being conducted online than ever before. This has made things difficult for video service providers who have seen bandwidth costs increase as they try to deliver high-quality video streaming services to their users.

To learn more about how Xilinx’s new real-time video server appliances can help ease this burden and the company’s first 20nm space-grade FPGA, TechRadar Pro spoke to the company’s director of video product marketing Aaron Behman and its space systems architect Minal Sawant.

Xilinx is known as the inventor of the FPGA, programmable SoCs and the ACAP. What else can you tell us about the company? 

Founded in 1984, Xilinx has a strong heritage in innovation with over 4400 patents and over 60 industry firsts. Today, the company develops highly flexible and adaptive processing platforms and is on a mission to build the adaptable, intelligent world by increasing the speed of applications such as machine learning and artificial intelligence.

To help achieve this goal, last year Xilinx launched Vitis™ and Vitis AI. The unified software platform offers a design methodology and programming model that enables all developers, including software and algorithm engineers with no hardware design expertise, to leverage the power of Xilinx adaptive platforms for edge to cloud deployments.

Five years in the making and the work of some 800 software engineers, Vitis includes a rich set of open-source performance-optimised libraries, runtime libraries and drivers that abstract away the low-level specifics of data movement and synchronisation and comprehensive developer tools to build, analyse performance bottlenecks and debug accelerated algorithms for Xilinx platforms.

(Image credit: Xilinx)

How will the release of your new Real-Time Video Server Appliances benefit those working in industries that rely on streaming video?

The video streaming solutions offer industry leading performance to enable cost, area and power reductions. For example, for high density streaming the U30 based reference architecture enables 4x the throughput, in 25% of the space at one sixth of the cost and 20% the power of a competing Nvidia-based Tesla T4 implementation.

Can you tell us more about Xilinx’s new Alveo U30 data center acceleration cards and the Zynq UltraScale+ MPSoC that powers them?

The new Alveo U30 card is an application specific media accelerator. It supports a very dense implementation of video streams. Soon, the card will also support AI use cases.

(Image credit: Xilinx)

Xilinx recently released the first 20nm space-grade FPGA. What exactly goes into making a chip radiation tolerant?

The natural space radiation environment can damage electronic devices and the effects range from a degradation in parametric performance to a complete functional failure. These effects can result in reduced mission lifetimes and sometimes system failures. 

Xilinx implements Single Event Upset Mitigation enhancements during the silicon and architecture design phase in critical portions of the circuits. This allows our FPGAs to be tolerant to radiation in the space environment. In addition, the 20nm was designed in a ceramic package which is a robust packaging technology to handle vibration and shock during satellite launches. We test the devices to the MIL-PRF-38535, which are stringent military standards for electrical screening. This involves putting the parts through temperature testing, radiation tests, burn-in, tests for package integrity and reliability.

How will the XQRKU060 FPGA’s machine learning capabilities be used in space and what other benefits does this new FPGA bring?

With trends moving to not only process data in orbit but to analyse it as well, machine learning can be instantiated for edge inference in space. This FPGA based solution is the first of its kind. The task of processing raw sensor information is very intensive and requires rigorous signal processing techniques such as non-uniformity correction, decimation, equalisation, pulse compression, beamforming and a variety of others. The task of then rendering usable images can include additional tasks such as rotation, back-projection and ortho-rectification. Not all these tasks have traditionally been possible on-orbit, but the KU060 opens the door to making this possible. 

Cloud detection is a relatively straightforward problem – 66% of earth is covered in clouds at any given time. If the images are being captured and stored, most of them are of clouds. When the processing is done, these images are stored and sent down to earth, which is a waste of bandwidth and storage – two very valuable resources for a spacecraft. Now, if a machine learning algorithm was implemented to discard images of cloud and only store and transmit information related to everything but clouds, critical bandwidth and storage are saved! Neural networks is one such technique.

Woman using laptop in front of servers

(Image credit: Christiana Morillo / Pexels)

What’s next for Xilinx and what areas are you currently focusing on / exploring?

The data center market continues to be a key focus area for Xilinx, especially as advances in AI, increasingly complex workloads, and an explosion of unstructured data demand new requirements for modern data center infrastructure.

Focus will remain on improving performance, video quality on the live streaming use cases. In addition to that, Xilinx is actively working on enabling its AI inferencing capabilities (i.e. DPUs) on this platform.

In addition to the data center market, the transition to 5G is an area Xilinx has devoted extensive resources to and will continue to do so, meaning it can provide value propositions no one else can offer. For 5G, Xilinx technology is helping solve capacity, connectivity, and performance challenges. It also provides the flexibility to support multiple standards, multiple bands and the multiple sub-networks that enable the many diverse IoT driven applications of 5G.

In the context of three major trends: the explosion of big data, the dawn of artificial intelligence and the post-Moore’s Law era, Xilinx continues to serve its core markets. These include automotive, aerospace & defense, broadcast & A/V, industrial, medical, and test, measure & emulation. These markets are central to Xilinx and the company plans to advance innovation in these areas.

What lessons has your organization learned from the pandemic and has the global strain on the internet influenced any of your current or upcoming products?

Adaptability, collaboration and innovation have been the key takeaways for Xilinx in the face of COVID-19. Xilinx customers in the healthcare industry have displayed these attributes tremendously, adapting so they can supply equipment to the many hospitals around the world to help save patient lives. For example, in late January Xilinx supported China’s largest medical equipment maker, Mindray, with thousands of Spartan-7 FPGAs to power patient monitoring systems. Xilinx has also been working with Philips, supplying Artix-7 FPGAs and hands-on support, to help the company meet the unprecedented demand for patient monitors. Additionally, Xilinx is working with other medical providers across the globe to supply products for testing and treating COVID-19. For the team at Xilinx, it’s rewarding to know that Xilinx’s adaptive computing technology is playing a vital role with these customers and the global response and relief efforts.

While the pandemic has created a great deal of uncertainty in the global business environment, this hasn’t hindered the speed of innovation at Xilinx, with the company continuing to serve its customers and move forward with new product launches. In the past few months alone, Xilinx announced the industry’s first 20nm space-grace FPGA for satellite & space applications, unveiled the Xilinx Virtex® UltraScale+™ VU23P FPGA, and introduced two real-time computing video appliances for easy-to-scale, ultra-high-density video transcoding applications.

Source Article