In a recent blog post, distinguished Google hardware engineer Norm Jouppi has shared details about an 'accelerator chip' called Tensor Processing Unit (TPU), which the company has been using since April 2015, to speed up its machine learning systems.

The Google TPU accelerator chip is named after the TensorFlow software which is used by the company for its machine learning programs.

Though the TPU accelerator chip - which has the capability to speed up specific tasks - has been in development at Google for the last three years, the company has largely kept it under the wraps till now. Over the last one year, Google has been using the chip for accelerating artificial-intelligence (AI) applications, because the significance of 'machine learning' software techniques to the company's core businesses has increased of late.

According to the details shared by Jouppi in the blog post, the TPU accelerator chip is ten-fold faster than the other alternatives which were considered by Google for speeding up AI applications. The new chip - which Google believes will give it a seven-year advantage - uses machine learning in over 100 programs for applications including search, voice recognition, and self-driving cars.

Jouppi also said that the TPU chip also is used in Google search ranking, photo processing, speech recognition, and language translation. He further added that Google plans to make the TPU chips available as part of the company's Google Cloud Platform computing-on-demand service.