AI in space is taking off

This article originally appeared in the August'22 magazine issue of Electronic Specifier Design – see ES's Magazine Archives for more featured publications.

Today’s world is Edge-hungry, predicted by 2025 to have 75% of all data produced and processed at the Edge. This is exacerbated by the proliferation of sensors, connected devices and wireless networks, meaning data is produced at an astronomical rate. Electronic Specifier’s Kiera Sowery explores further.

A collaboration between Teledyne e2v, Klepsydra Technologies and Beyond Gravity enables space designers to manufacture systems and solutions that were not feasible in the past.

“Teledyne e2v has been ensuring that its platforms bring the right level of maturity and quality and are perfectly capable of working together with third parties and product implementations,” said Thomas Guillemain, Marketing & Business Development, Data Processing Solutions.

He continued: “There is a key challenge to implement lots of reliability together with lots of computing in space, also known as Edge computing, to treat more data in space, rather than sending it to the ground.”

What are the drivers?

Space equipment is difficult to build as everything must be launched into space, and it therefore needs to withstand the force, vibrations and shock involved in the launch. Once equipment is launched into space, it faces further problems including having to serve a long operation time with little to no maintenance, micrometeorites, extreme temperatures, and radiation.

Traditionally, the most performant computers aren’t necessarily used in space as its more important to have a reliable system. For instance, the more performative a processor is the more prone it is to any kind of particle radiation. However, there is now a desire for better performance for space applications, especially those related to AI. Spacecraft autonomy, AI for Earth observation, and Edge processing are three key drivers for such a development.

Spacecraft autonomy

Both the private and public space sector have already launched satellite constellations into space, including Star Link. If you have hundreds of satellites, you cannot control them by a team on the ground individually. Therefore, those satellites need a higher degree of autonomy to work.

Also, as a nation we want to travel back to the Moon! We want to explore Mars and the rest of space as and when that becomes possible. This also requires more autonomous spacecrafts; after all, in deep space, any signal could travel for minutes or hours.

AI for Earth observation

Many applications for imagery on Earth would have useful applications in space, including natural disaster early-warning systems, automatic recognition of forest fires and glacier movements, automatic tracking of ships and aircrafts and smart farming. Other applications are also likely to emerge over the next few years.

Edge processing

We now have a ‘data bottleneck’. Traditionally, the images created in space were sent to dedicated data centres to be processed. However, now that we have better sensors, larger datasets can be created in space, but they cannot be sent to data centres fast and efficiently.

Edge processing offers a solution to this. Edge processing means that the data is processed at the Edge (close to where it is generated) – in this case, the satellite or spacecraft.

Beyond Gravity provides a processing platform to maximise the value of the data generated in space.

This solution, called LYNX, is based on LS1046-Space (a quad-core ARM Cortex-A72 processor by Teledyne e2v). It features over 30k DMIPS, is available in two form factors, has a power consumption of up to 40W, and is radiation tolerant up to 100krad(Si) TID. Future generations are planned, including REFLEX, an FPGA-based product, and ALEX, which is GPU based.

Beyond Gravity develops and manufactures products for satellites, launch vehicles, and other spacecraft. In particular, it provides Edge computing platforms hardware, meanwhile parts of the software framework may come from partners, which is where Klepsydra Technologies comes in. Klepsydra Technology is a software company developing on-board data processing solutions for real-time applications, providing Edge computing software for today’s data-hungry world. Its software has been deeply optimised for Beyond Gravity’s Lynx computer, therefore enabling real AI onboard.

Our data hungry world

Space computers are undergoing an important challenge as the number of sensors and data sources inside satellites continue to grow, and more data needs to be processed. We are reaching a point where it’s common to lose some data during onboard processing, which can lead to critical failures that cannot be ignored, including collision, mission failure, and crashes across the globe.

There was a solution created in the early 2010s within the financial industry. The solution was trading software using cutting-edge lock-free programming techniques; however few developers have the required skills.

Klepsydra’s software can improve computer resources by eight fold, reduce power consumption by 50%, and provide no data losses. The company created its own AI software based on this solution using a high-performance, advanced parallelisation method.

The solutions cover a range of uses for latency throughout the CPU.

The software is already used by ESA for Earth orbit testing and sun imagery weather classification. Performance benchmarks also show that Klepsydra AI on the Lynx computer performs up to four times better than others, making space missions more reliable and successful.

QlevER Sat: deforestation

Teledyne e2v provides radiation-tolerant compute-intensive solutions, solving the key problems discussed earlier. These include LS1046-Space, DDR4T04G72M and QLS1046-Space. There are three key pillars to the success of Teledyne’s radiationtolerant computing solutions, including a robust and proven space manufacturing flow, radiation characterisations and agile, highly-skilled R&D and technical support.

The following trends and innovations are enabled by the solutions:

• on-board processing, autonomous/AI image processing
• new space platforms and landers
• ground-sat and sat-2-sat communications
• heterogenous data analytics

The collaboration between the three companies allows projects to take place that wouldn’t otherwise be possible.

This includes a smart satellite featuring a high-resolution image sensor from Teledyne e2v with a compute-intensive processing module running on innovative space AI algorithms.

The system is designed for 6U Earth observation CubeSats, reducing link bandwidth by processing images into simplified binary maps before transmission to the ground. Deforestation is the primary use case; however, it could also be used to monitor volcanic activity and glacial movement, evaluate damage caused by natural disasters and also performs defence related tasks.

Processing is in a 1.8Ghz 64bit Teledyne quad-core Qormino QLS1046-Space module, which has ARM Cortex A72 cores alongside 4Gbyte of DDR4 memory. Imaging data is captured using a 16Mpixel Emerald CMOS sensor, also from Teledyne.

“To ensure that Qormino is robust enough to deliver long-term operation when exposed to space conditions, and mitigate the threat of errors occurring, our components are subject to extensive Space High Reliability qualification, radiation characterisation and mitigations,” said Guillemain.

He continued: “These cover both the processor and the accompanying memory. The modules can cope with total ionising dose levels above 100kRad and have single event latch-up resilience beyond 60MeV. cm2/mg. An operational temperature range of -55°C to +125°C is supported.”

What’s next?

Teledyne aims to further develop its computing-intensive solutions to serve Edge computing in space, with higher CPU capabilities, improved power efficiency, bigger DDR4 densities and integration in systems in packages. Also, increasing activities on reference designs and key companion chips selection for platforms are all key to ease system developments and reduce time to market for customers.

“To be able to this, we need to gain extensive testing and proof of evidence that it will work in the expected way,” said Guillemain.

“What we can expect in the future is more capabilities and computation, making faster decisions and actions in different areas, providing that the foundation has been demonstrated in advance, which is where we invest.”