Skip to content Skip to sidebar Skip to footer

Source: Le Big Data as of 27-07-2020

A team of researchers at George Washington University has created an artificial intelligence processor powered by light rather than electricity. This TPU is proving to be more efficient and efficient, and could mark a major breakthrough in the field of Machine Learning.

Innovations follow one another in the field of artificial intelligence, simplifying or speeding up the process of training algorithms. A few days ago, researchers in California unveiled  a new method for training AI on a single computer.

Now it is the researchers at George Washington University who are unveiling their discovery. By replacing electricity with light to power the calculations, they were able to make AI faster and more efficient.

Until now, the processors used for Machine Learning were limited by the power required to process the data. This hindered their ability to perform complex operations.

The more difficult the task, the more complex the data to be processed, and the greater the power required. In addition, data transmission between processor and memory is slow and further restricts neural networks.

However, by using photons within the TPUs of neural networks, the researchers have discovered that it is possible to overcome these limits to create a more powerful and energy efficient AI.

Thus, their photon-based TPU was able to perform calculations 2 to 3 times more complex than a traditional electric TPU. However, it consumes only a fraction of the energy needed for an electricTPU.
This innovation could save vast amounts of energy, improve a processor’s reaction time and reduce data center traffic. Such processors could be deployed in the field of 5G and 6G networks, or in Data Centers.

For the time being, however, this is only a scientific experiment and it will be necessary to wait a few years to see the emergence of potential commercial applications. The study by researchers at George Washington University is published  in the journal Applied Physics Reviews.

Show CommentsClose Comments

Leave a comment

News ORS © 2021. All Rights Reserved.