پروژه سان‌کچر گوگل؛ آغاز آزمایش دیتاسنترهای فضایی برای هوش مصنوعی
Information Technology

Google Suncatcher project; start of testing space datacenters for artificial intelligence

Google is preparing the first orbital test of the Suncatcher project (Suncatcher); an ambitious plan to investigate the possibility of using satellites equipped with AI chips as processing infrastructure in space. According to ITPna quoting Techspot, the project's experimental satellite named MVP is

Google is preparing the first orbital test of the Suncatcher project (Suncatcher); an ambitious plan to investigate the possibility of using satellites equipped with AI chips as processing infrastructure in space.

According to ITPna quoting Techspot, the project's experimental satellite named MVP is scheduled to be launched next month on a SpaceX Falcon 9 rocket from Vandenberg Space Force Base in California.

MVP is not a full space datacenter, but an experimental platform with four Google Tensor Processing Units (TPUs). The total processing power of these chips is roughly equivalent to a datacenter server. The solar panels also generate about one kilowatt of power so the chips can perform short AI processing tasks.

The mission's main goal is to examine two major challenges: space radiation and heat. Radiation can cause a phenomenon called a bit flip; that is, a stored value in memory changes from zero to one or vice versa, resulting in computational or software errors. Google previously exposed its chips at the Crocker Nuclear Laboratory in California to radiation roughly equivalent to five years in space. The tests showed that rebooting the chips usually can correct these errors.

Heat is another significant problem. AI chips produce a lot of heat during processing, but in space it is not possible to use cooling fans because there is no air to transfer heat. Google designed a system to move heat from the chips through layers of different thermally conductive materials and then dissipate it to space via a radiator.

However, the chips cannot operate continuously. According to Google, the processors run for about 15 minutes and then must be turned off to cool down. While active, the satellite can process short requests related to Gemini.

Google also conducted vibration tests to assess the satellite's resistance to launch stresses. In these tests, the solar panels and internal components were inspected and the chips remained undamaged.

Google emphasizes that this project is still an initial experiment and is not intended to become a commercial service in the short term. The MVP satellite is expected to operate for at least one year and may remain in orbit for up to six years.

The main idea behind Suncatcher arises from the sharp increase in power demand of AI datacenters. Placing processing hardware in orbit could provide direct access to solar energy and reduce the need for land, water, and terrestrial power infrastructure; although the cost and engineering complexity of such a system remain very high.

Google intends to launch two more satellites next year and is even exploring the idea of building a constellation of more than 80 spacecraft and a dedicated satellite the size of a football field.

Originally published by Information Technology News - ITNA

Topics in this story

  • datacenter
  • chips
  • orbit
  • google
  • infrastructure
  • artificial
  • project
  • processing
  • mvp
  • heat
  • tests
  • testing
  • california
  • suncatcher
  • solar
  • intelligence
  • power
  • satellite
  • roughly
  • short
  • radiation
  • equivalent
  • panels
  • errors
  • operate
  • experimental
  • idea
  • launch
  • next
  • remain
  • space
  • system