Tensor Processing Unit (TPU) is the custom chip on which Google has been working secretly for several years to speed up multifarious machine learning tasks. Sundar Pichai, CEO of Google disclosed regarding this chip in I/O developer conference 2016.
Application Specific Integrated Circuit (ASIC)
Since a year, there is a custom application specific integrated circuit (ASIC) which was running in the data centre racks of Google. It powers various application at Google like
RankBrain: To improve the relevancy of the search results in Google Search
Google Street View: To improve the accuracy along with the quality of maps as well as navigation
AlphaGo Artificial Intelligence (AI): AI is powered Go player which beats the top rank of Go player Lee Sedol in current year.
Apart from all these application, the services related to voice recognition and Cloud Machine Learning services of the company also runs on TPU chips.
What is Tensor Processing Unit (TPU)?
TPU is a custom built for machine learning application as it is tolerant of reduced computational precision and requires fewer transistors per operation. As a result of it, more operations per second can be lined up into the chip and powerful machine learning models can be applied effectively.
Google’s Further Plans
Google is planning to disclose more machine learning APIs and make the power of these chips available to the developers which will help them to build applications for customers.
Google is not only the tech company which is working on its own chips to power machine learning and artificial intelligence.
Field Programmable Gate Arrays – Microsoft Product
Field Programmable Gate Arrays is been used by Microsoft to accelerate the computations of AI which is also being used by Bing Search.
TrueNorth – IBM Product
TrueNorth is being designed by IBM which is their won brain inspired chip. This chip is under the testing at Lawrence Livermore National Laboratory.
Last but not the least Nvidia Corporation which is famous for its graphical processing units (GPUs) that are used for gamming has also been testing them for AI and machine learning applications. In fact, Google also used Nvidia GPUs for early testing of the software named as AlphaGo.