Google’s Project Suncatcher to put ML infrastructure in space
7 hours ago
- Project Suncatcher is Google's moonshot to explore using AI hardware in space, with its first test launching a prototype satellite to evaluate TPU performance in orbit.
- The initiative aims to leverage low Earth orbit's near-constant sunlight, which could generate up to eight times more solar power than on Earth, potentially enabling scalable machine learning infrastructure and satellite constellations.
- The initial mission, in partnership with Planet and on SpaceX's Transporter-18, will gather data on how TPUs handle spaceflight stresses like vibration, radiation, and thermal extremes.
- Vibration testing showed the hardware surprisingly survived intense shaking, while proton beam tests at UC Davis indicated Trillium TPUs can withstand radiation doses beyond a five-year mission.
- Cooling in space is a key challenge due to the vacuum; the team is testing heat pipes and radiators in thermal vacuum chambers to manage TPU heat.
- Future satellite clusters will use laser communications for high-bandwidth, short-distance links, requiring precision similar to hitting a coin-size target from miles away, with tests planned for 2027.
- The project emphasizes methodical engineering, starting with proving hardware survival, and sees this launch as the first step toward identifying failures and improving future missions.