Min menu

Pages

"Impenetrable." Developing a light-powered computer chip

 



"Impenetrable." 

Developing a light-powered computer chip


University of Pennsylvania scientists have developed a new computer chip that

uses light instead of electricity, in an important step towards improving

the training of AI models by enhancing the speed of 

data transmission more efficiently.


The recent :

"widespread" deployment of AI models in technology has increased

the demand for computers that can process large sets of information.

However

inefficient computing systems lead to high energy consumption.


Led by Nader Ingata :

Professor at the School of Engineering and Applied Sciences at the University of

Pennsylvania, the team designed a silicon photon chip (SiPh)

that can perform mathematical calculations using light.


Scientists said that light is the fastest way to transmit data known to humanity. 

The extensive use of silicone can ensure that

the technology can be scaled up quickly.


The research team aims to design a chip that can perform "vector matrix"

processes (a common mathematical calculation), which is widely used in

the development and operation of neural networks that

are critical when developing the operating structure of existing AI models.


Scientists have reduced the height of

the chip in specific areas to control how light spreads inside. 

However

the elevation variations control process confirmed that

light travels within the chip only in a straight line.


Frose Platone :

an associate professor of electrical and systems engineering at the university

said the chip could replace the GPUs used by companies to

train and classify their AI models.


It recommends that SiPh be an addition to

the existing infrastructure used by AI companies.


SiPh can address data privacy concerns

  1. as well as perform calculations
  2. faster with minimal
  3. electricity consumption.


The chip's ability to perform multiple calculations in parallel eliminates

the need to store information in working memory during calculations.


"No one can penetrate a memory that doesn't exist to access your information."