What’s new from Google”Dreamcatcher Project“aims to launch Tensor Processing Units (TPUs) into space, Creation of a satellite-based and solar-powered artificial intelligence network.
capable of scaling machine learning beyond the limits of the Earth. Google says a “solar panel can be up to 8 times more productive than on Earth” for nearly continuous power using a “low Earth orbit synchronized with the sun between dawn and dusk” that reduces the need for batteries and other types of power generation. 9to5Google reports: These satellites would be connected via free-space optical links, with large-scale machine learning workloads “distributing tasks across numerous accelerators with high-bandwidth, low-latency connections.” To match data centers on Earth, the connection between satellites would have to be tens of terabits per second and they would have to fly in “very close formation (kilometers or less).”
Google has already carried out radiation tests on TPU (Trillium, v6e), with “promising” results: “While the high-bandwidth memory (HBM) subsystems were the most sensitive component, they only began to show irregularities after a cumulative dose of 2 krad(Si), almost three times the expected (protected) dose of the five-year mission of 750 rad(Si). No serious failure was attributed to the TID up to the maximum tested dose of 15 krad(Si) on a single chip, indicating that Trillium TPUs are surprisingly radiation resistant for space applications.”
Finally, Google believes launch costs “will fall to less than $200 per kilo by the mid-2030s.” At that point, “the cost of launching and operating a space data center could become roughly comparable to the reported energy costs of an equivalent terrestrial data center per kilowatt/year.”
