Google reported at its I/O dev conference on May 18, Wednesday it has developed computer chips to power its artificial intelligence (AI) tech known as deep neural networks. CEO Sundar Pichai shared that the Alphabet company had developed an application-specific integrated circuit (ASIC) for deep neural nets. The software and hardware networks use machine learning to pick up particular tasks by analyzing large amounts of data such as for Google's facial recognition tech, voice recognition features on Android phones, and search engine.
Google's neural nets have many other applications. They include object recognition and text translation from one language to another, according to Wired.
Google reports that it has been running the chips for around one year and developed them a short time before that.
The tech giant has named its chip Tensor Processing Unit (TPU). That is because it supports Google's software engine TensorFlow that powers its deep machine learning services.
Google released TensorFlow last fall under an open-source license so anyone can modify the software engine.
It is unclear if Google will share TPU's designs. However, anyone can access the company's machine learning hardware/software through its cloud services.
TPU boards go into the same slots as hard drives on huge hardware racks in Google's data centers. Google also points out that its chips provide better performance (per watt) for machine learning.
The company explains TPU chips require fewer transistors for each operation. That allows it to add more operations per second into it, and use more powerful machine learning models faster. The results are thus more intelligent and quicker.
Google is not the only company that is adding deep learning into different Internet services. They include Microsoft, Facebook, and Twitter, which use graphics processing units (GPUs) to power their neural nets. The OS giant is also programming chips known as field programmable arrays (FPGA) for certain tasks.
In related news, IBM announced on May 17, Tuesday that it has made a big improvement to a tech innovation called phase-change memory (PCM), according to CNET. The technology has been developed for years but Big Blue claims it will now be cheap enough to be added to mobile devices.
PCM works by heating a small amount of a glassy material in the chip.
Here's a video on Intel chip names: