🤖 AI Summary
This work addresses the limitations of conventional on-orbit computing satellites, which struggle to support large-scale AI inference due to low computational power per unit mass, poor thermal dissipation, and bulky power systems. The authors propose an integrated distributed satellite architecture that unifies structural, power, computing, and thermal management subsystems. By incorporating large-area vapor chamber radiators, custom space-grade chips, high-density photovoltaics, and modular subarray communications, the design achieves exceptional thermal regulation and specific power. The resulting system delivers over 100 kW/ton of computational capacity and a specific power of 500 W/kg. A single 150 kg satellite can thus provide 16 MW of compute power, concurrently supporting 31 subarrays and enabling more than 7,900 large-model inference sessions, each achieving a throughput of 553 tokens per second.
📝 Abstract
We describe and analyze a distributed compute architecture for SSO computational satellites that can potentially provide >100 kW compute power per launched metric ton (including deployment and station keeping mass). The architecture co-locates and integrates the solar cells, radiator, and compute functions into multiple small panels arranged in a large array. The resultant large vapor chamber radiator area per panel should permit ICs to operate at junction temperatures near 40*C with benefits in compute efficiency and reliability. Using the structure of the radiator to support the solar cells may also yield a specific power of about 500 W/kg compared to less than 100 for existing conventional implementations. Assuming development of custom solutions for all components, a 16 MW computation, 150 ton satellite comprising a 20 m x 2200 m grid of 16,000 panels can fit in a single Starship hold. The concept is scalable to much larger satellites with higher mass payloads or using on-orbit assembly. We consider panel sizes from 1 to 4 m2 to allow trading vapor chamber heat transport with compute efficiency and inter-panel communication. Assuming a 1 kW/panel design, 512-panel subarrays of the satellite can run a representative inference-only LLM with 500,000 token context window and 128 attention blocks, at a rate of 553 tokens/sec/session, across 256 simultaneous in-flight sessions. A full satellite could support 31 such subarrays, for >7900 inferences at a time.