Google plans first test of AI chips in space under Project Suncatcher

Project Suncatcher is an ambitious Google initiative aimed at testing whether AI hardware can thrive in the extreme conditions of space. The project is launching a prototype satellite to evaluate how Google's Tensor Processing Units (TPUs) perform in orbit. This mission is a crucial part of Google's long-term plan to construct powerful AI systems in space, with potential applications in various fields, from healthcare to scientific discovery.

The satellite will ride along with the Transporter-18 rideshare mission on a SpaceX rocket. This initial test will expose the TPUs to extreme radiation, intense vibrations, and unique cooling challenges in a vacuum. Although the rocket journey is only 10 minutes long, the spacecraft experiences intense acceleration forces up to 10 times the force of gravity, or g-force. The TPUs can withstand even greater forces, up to 50 to 100 g, during these tests.

Another significant challenge the team faces is radiation exposure. High-energy particles in space can cause bit flips in electronics, potentially disrupting AI workloads. To address this, the team tested TPUs in a proton beam facility at UC Davis's Crocker Nuclear Laboratory while running AI workloads. Preliminary results indicate that Google's Trillium TPUs can survive a significantly higher total ionizing dose than they would during a five-year space mission.

Cooling is another critical aspect of operating AI hardware in space. In low Earth orbit, satellites can harness near-constant sunlight, providing up to eight times more solar power than on Earth. However, without airflow, new cooling methods must be devised. The team is exploring various approaches, including using a combination of heat pipes and radiators to dissipate heat safely.

Communication between satellites is also vital for maintaining the bandwidth necessary for AI processing. Satellites will communicate via lasers, which require extraordinary precision due to the high bandwidth over extremely short distances. This technology already exists, but for Project Suncatcher, it's being optimized for high-bandwidth operations in space.

As the project progresses, the team will test the cooling system in space and refine their designs based on learnings. By 2027, when multiple satellites are in orbit, the Project Suncatcher team aims to demonstrate the viability of scalable AI compute in space. This moonshot represents Google's commitment to pushing the boundaries of technology and making AI's benefits accessible to everyone, far into the future.