What it is
The Sun is the largest energy source in the solar system, which the authors say warrants considering how future AI infrastructure could most efficiently tap into that power. They explore a scalable machine-learning compute system in space, made of fleets of satellites equipped with solar arrays, free-space optical links between satellites and Google tensor processing unit (TPU) accelerator chips. For high-bandwidth, low-latency communication between satellites, the satellites would fly in close proximity; the authors illustrate the basic approach to formation flight with an 81-satellite cluster of 1 km radius and describe an approach for high-precision, ML-enhanced models to control large constellations.
Why it matters
The paper takes on two questions such a system depends on. Google's Trillium TPUs were radiation tested: they survived a total ionizing dose equivalent to a 5-year mission life without permanent failures, and their bit-flip errors were characterized. Launch is critical to the overall system cost, and a learning-curve analysis suggests launch to low Earth orbit may reach $200/kg or less by the mid-2030s.
Every metric behind this entry is listed, with its source, under Sources and data below.
Filed underSatellite Communication Systems, Spacecraft Dynamics and Control, Spacecraft Design and Technology
