Google is putting its first Project Suncatcher hardware into orbit, launching a prototype satellite to test whether its Tensor Processing Unit (TPU) AI chips can survive and operate in space, the company said in a blog post from Travis Beals, Senior Director of Paradigms of Intelligence at Google Research. Project Suncatcher, announced last year, is a long-term research effort exploring whether space could eventually host scalable machine learning infrastructure, since satellites in low Earth orbit can access near-constant sunlight and generate up to eight times more solar power than on the ground.
The mission is flying aboard the upcoming Transporter-18 rideshare mission with SpaceX and was developed in partnership with Planet. A rocket trip to low Earth orbit lasts about 10 minutes, during which a spacecraft experiences vibration and sustained acceleration loads of up to 10 times the force of gravity, with individual components such as TPU chips experiencing forces of up to 50 to 100 g. Google said it shook the satellite on all three axes to mimic launch conditions and was "pleasantly surprised" the hardware held up.
To evaluate radiation exposure, Google tested TPUs in a proton beam facility at UC Davis's Crocker Nuclear Laboratory while running AI workloads, monitoring for errors such as bitflips. According to the company, its Trillium TPUs "hold up remarkably well" and can survive a total ionizing radiation dose greater than what they would receive during a five-year space mission.
Because there is no airflow in space, heat from the TPUs can only be diffused through radiators rather than conventional cooling systems. Google says it is testing a combination of heat pipes and radiators, which it has so far validated in a thermal vacuum chamber that simulates the thermal and vacuum conditions of space. The company's longer-term vision involves satellites carrying dozens of TPU chips each, connected in orbiting clusters via high-bandwidth laser links; a two-satellite test of that interconnectivity is planned for 2027.
"This first launch is about seeing what works, identifying points of failure, and applying those findings to future missions," Beals wrote.