Avnet Silica’s market segment manager aerospace and defence, Paul Leys, explains how the dawn of Space 2.0 has escalated demand for adaptive compute acceleration platforms
Marking a new era led by commercial interest more than governmental initiatives, Space 2.0 is catalysing rapid progression in sectors like reusable rockets, satellite internet connectivity, earth observation platforms and space tourism. As a result, the skies above us are busier than ever.
Applications such as internet connectivity services, Earth observation, navigation, avionics, defence and security require significant computing power and advanced hardware acceleration to manage dense data streams. These applications also need machine learning capabilities to enhance autonomy in crowded orbits.
Notable contributors to the proliferation of internet connectivity services are Elon Musk’s Starlink, OneWeb, and Amazon’s Project Kuiper. Starlink has already launched over 3,580 satellites into low-earth orbit (LEO). OneWeb deployed 36 satellites in March 2023 with its 18th rocket launch, completing its 618-strong constellation. Amazon’s Project Kuiper plans to deploy a constellation of 3,236 satellites over coming years.
This commercial momentum, labelled Space 2.0, imposes a significant shift with a tilt from bespoke to mass-produced satellites, reduced time-to-market and a push for greater functional integration. Electronic components need to withstand harsh space environments and provide long-term support and supply guarantees.
One method of tackling these complex challenges is programmable logic. Xilinx produced its XQR VERSAL adaptive compute acceleration platforms (ACAPs) after its 2022 acquisition by AMD. These ACAPs, comprising adaptable SoCs for space applications, are built on a 7nm process, minimising the effect of radiation-induced single event upsets (SEUs). Variants within the family can host a mixture of CPUs, DSP and ML accelerators, programmable logic, memory resources and advanced peripherals and I/O facilities. Importantly, configurability is not affected by radiation exposure, letting a payload’s processing functions be regularly updated in use in space.
The components feature a robust processor system, ML capabilities and on-chip networking facilities. The processor system employs a 64-bit dual-core Arm Cortex-A72 for managing complex algorithms and a 32-bit dual-core Arm Cortex-R5F for real time tasks. The platform management controller oversees device and power management, error handling, configuration and analogue measurements.
Each part has an AI engine that handles ML and AI workloads, with an architecture that supports various levels of parallelism. The Network on Chip (NoC) provides about 400Gbit/s of on-chip bandwidth and facilitates off-chip data access.
For space suitability, Xilinx’s space-grade parts are adapted for the harsh environment, meeting the US Department of Defense’s MIL-PRF-38535 Class B manufacturing quality standard. Rigorous qualification testing and characterisation against radiation have been conducted for the XQR VERSAL parts.
The dawn of Space 2.0 has escalated the demand for adaptive compute acceleration platforms—hardware that is powerful, flexible and reconfigurable to handle general tasks and dedicated resources for specialised tasks such as signal processing and machine learning inference. Xilinx’s XQR VERSAL parts, coupled with strong safeguards against radiation-induced errors, suit swiftly evolving Space 2.0 applications.