Reconfigurable architecture technology will fly into space

Today, technological advances have made low-orbit flights increasingly commercially viable. Minal Sawant, a system architect for Xilinx space products, said that one of the main challenges in the aerospace industry is the architecture used for various projects, and a flexible architecture platform for multiple tasks is a challenge today. "Learning and processing data in space is a new technology, and we will see the introduction of machine learning (ML) technology into this field in the future." ML is used for scientific analysis, object detection, and image classification, and improving processing efficiency will reduce waiting time for on-orbit and ground operations.

This type of technology requires reliable components, but also size, weight, power and cost (SWaP) considerations. Xilinx recently launched the 20nm radiation-tolerant (RT) Kintex UltraScale XQRKU060 FPGA, which provides reconfigurable capabilities for on-orbit products and a 10x increase in the number of DSPs, thereby improving throughput and bandwidth for all orbits and deep space processing.

Designed to mitigate single event effects (SEE), the product can be used in low Earth orbit (LEO), medium Earth orbit (MEO), geosynchronous and deep space orbits, as well as on deep space missions.

The company said it has 2760 UltraScale DSP Slices and provides up to 1.6 TeraMACs of signal processing computing power, which is more than 10 times the previous generation. To meet the increased bandwidth, there are 32 high-speed transceivers (SerDes) that can run up to 12.5Gbps, providing a total bandwidth of 400Gbps. The FPGA provides 5.7 INT8 TOPS performance, which is nearly 25 times the deep learning performance of the previous generation.

On-orbit reconfiguration, real-time onboard processing and ML acceleration will allow satellites to update in real time, provide video on demand and compute in real time. The FPGA’s adaptive computing architecture allows for unlimited on-orbit reconfiguration before launch and while in orbit as protocols and applications change.

For ML development, tools for the TensorFlow and PyTorch frameworks support FPGAs to enable neural network inference acceleration for real-time onboard processing in space.

The 40mm x 40mm ceramic package is able to withstand the vibrations during launch as well as the effects of radiation in orbit.

To accommodate longer design cycles, Sawant explained that prototypes can be created and included in the board layout prior to certification, then replaced with qualified parts to speed up the design cycle, reducing the original five to six years to two to three years.

使用该公司的Vivado Design Suite编程重新设计的路由架构可以解决常见的行业瓶颈，显著消除路由拥塞，可实现90％的负载下性能不降低。该公司的Vitis统一软件平台还可用于支持三重模块化冗余（TMR）的MicroBlaze软处理器上的嵌入式软件开发。未来的扩展将增加对Vitis AI的支持，Vitis AI是在赛灵思AI推理的统一软件平台。

Starting in September 2020, Xilinx will offer space-grade products in B and Y grades certified to MIL-PRF-38535, with prototypes available now.