Astrum

AI Chips for Small Satellites

An Introduction

In our analytics-driven society, we leverage observations made by satellites in space of natural and man-made activities to predict events, set market prices and develop strategies that affect us all.

With the recent commercialization of space, we are finally able to move away from the exclusivity of these very expensive large satellites that provided a limited amount of information to a select few. Instead, we are now increasingly relying on large constellations of small satellites. A constellation not only provides better coverage of the Earth at a fraction of the cost but it also provides redundancy if some satellites fail.

All these small satellites capture huge amounts of information that gets sent back down to Earth to run analytics. Unfortunately, the download link is very slow, and these small satellites are spinning around the earth really fast – so there’s only a small window when a satellite is within range of a ground station, which collects and analyzes the information. It takes many passes and/or lots of groundstations to download data from just one satellite. Additionally, data needs to be collected from all groundstations spread throughout the world before it can be processed and analyzed.

Modern Hardware Limitations

Approximately 65% of the satellite market is considered “small satellites”. These satellites reside in lower Earth orbit and remain active for 1.5 years on average, before falling back to Earth. During their lifespan, they collect large amounts of data that is sent back to Earth for further processing. Since 2010, the cost to launch a small satellite has gone down by 45%. This has led to an increase in launches. Currently, companies are planning huge constellations of 1,000 satellites – each! With a greater number of satellites in space, more data is collected and processed. The need for ground stations increases, as does the need to quickly process large amounts of data.

The development of high-performance, energy-efficient solutions for modern applications such as machine learning and classification is challenged by the limitations of the available hardware. Existing solutions are custom built and come with high-risks and high-costs. As a result, current solutions are not well-suited for treating information when collected. As a result, any analysis required is delayed, ultimately resulting in higher expenses due to the larger amounts of data and processing required. Addressing this short-coming necessitates a new type of processing architecture and approach to data processing.

AstrumTM provides a Direct-to-user Analytics straight from the small satellite. AstrumTM leverages technologies that make it Space Robust and Low Power for small satellites. With AstrumTM equipped small satellites, ground stations will only need to download a much smaller payload of the analytics – so a single pass over a single ground station would suffice.

Not only will this stave off the need for huge networks of ground stations, but it also means that analytics can become available in real time to an analyst anywhere in the world. This capability will become crucial for strategists who deal with emergencies and need to save lives.

Real-Time Analytics

AstrumTM provides real-time processed data enabling strategists and those working in the field to utilize actionable insights derived from geospatial data. Current solutions require data to be sent to a number of ground stations and processed over time. This typically takes hours or even days and presents challenges for defense and public health when seconds matter. With AstrumTM, satellites can complete the processing prior to sending it back to Earth within seconds. Lucid Circuit is designing a microchip for small satellites that will address this problem by delivering real-time analytics to strategists who need to save lives.

Onboard Machine Learning

AstrumTM processors leverage an adaptable low-power, high-performance compute fabric that combines runtime-reconfigurable architectures and support for heterogeneous integration. The analysis capabilities of AstrumTM enables geospatial analytics in real-time, making it possible to process data as it is collected and before it is sent back to Earth.

Resilient Data Integrity

AstrumTM is specifically designed using a patented technology to withstand the harsh conditions of space. This design prevents chip failures and ensures data collection is not disrupted before 1) onboard analytics and 2) data transmission back to Earth. AstrumTM also supports the highest standards of end-to-end encryption, ensuring that data sent remains private and secure.

Information Traceability

The immutable in-silicon digital fingerprinting feature allows each AstrumTM chip to specifically tag each piece of data with a key that is unique to each chip. This makes it possible to trace any recorded data entry all the way back to the specific AstrumTM Processor microchip within the specific piece of equipment and (date) instance (when) it was created.