Breakthrough Silicon-Photonic Chip Enables AI Computing at Light Speed

Silicon-photonic chip enables AI computing at light speed

Artificial intelligence (AI) is transforming various fields of science, technology, and society, but it also poses significant challenges for computing hardware. Current computer chips rely on electricity to perform the intricate calculations needed to train and run AI models, which limits their speed and efficiency and consumes a lot of energy.

A group of engineers from the University of Pennsylvania has come up with an innovative idea. They have created a chip called a silicon-photonic chip that uses light waves instead of electricity to carry out the complicated calculations needed for AI. This chip has the potential to greatly increase the speed of computers and decrease their energy usage.

The SiPh chip’s design is the first to combine the groundbreaking research of Benjamin Franklin Medal Laureate and H. Nedwill Ramsey Professor Nader Engheta, who has pioneered the manipulation of materials at the nanoscale to perform mathematical computations using light, with the SiPh platform, which uses silicon, the cheap and abundant element used to mass-produce computer chips.

The researchers, led by Engheta and Firooz Aflatouni, Associate Professor in Electrical and Systems Engineering, describe the development of the new chip in a paper published in Nature Photonics.

How the silicon-photonic chip works

The SiPh chip’s main innovation is its ability to do vector-matrix multiplication, a key math process in neural networks, the technology behind AI tools.

Vector-matrix multiplication involves multiplying a row of numbers (a vector) by a grid of numbers (a matrix) to produce another row of numbers. This operation is repeated millions of times to train and run neural networks, which learn from data to perform tasks such as image recognition, natural language processing, and self-driving cars.

To perform vector-matrix multiplication using light waves, the researchers exploited the properties of silicon, which can bend and scatter light in different ways depending on its thickness. By creating regions of silicon with varying thicknesses on the chip, the researchers were able to control the propagation of light through the chip and use it to encode and manipulate the numbers involved in the calculation.

The result is a chip that can perform multiple vector-matrix multiplications simultaneously, at the speed of light, using very little energy.

Advantages and applications of the silicon-photonic chip

The SiPh chip offers several advantages over conventional computer chips, according to the researchers. First, it can greatly increase the processing speed of computers, as light travels faster than electricity. Second, it can reduce the energy consumption of computers, as light waves do not generate heat as electricity does. Third, it can enhance the privacy of computers, as light waves do not need to be stored in a working memory, making them immune to hacking.

The researchers say that the SiPh chip is ready for commercial applications, and could potentially be integrated with existing computer chips, such as graphics processing units (GPUs), which are widely used for AI computing. This could enable faster and more efficient training and classification of AI models, which are essential for various domains, such as health care, education, and entertainment.

“We decided to join forces,” says Engheta, who collaborated with Aflatouni, an expert in nanoscale silicon devices. “They can adopt the silicon-photonic platform as an add-on,” says Aflatouni, “and then you could speed up training and classification.”

Other co-authors of the paper include Vahid Nikkhah, Ali Pirmoradi, Farshid Ashtiani, and Brian Edwards of Penn Engineering.

The research was supported by the Office of Naval Research and the National Science Foundation.