Artificial intelligence has become an important part of our modern lives, as it powers different applications, from voice assistants to autonomous vehicles. However, this technology is limited by the speed at which AI computations occur.
The Need for Efficient Computing Systems
More than half a century ago, experts introduced supercomputers, computing devices with a higher performance level than general-purpose computers. Using high-performing mainframe systems, these machines can perform quintillion computations per second.
However, supercomputers still depend on the same principles from the earliest days of computing in the 1960s. This poses a problem, considering that they are used for applications requiring massive databases and are expected to do much computation.
A potential alternative to supercomputers is quantum computers, which utilize the laws of quantum mechanics to create more efficient and robust systems. However, this technology is still in its infancy and faces many challenges. There also has been a rise in AI models to meet the huge demands for computers that can process large sets of information. However, current quantum computing systems are inefficient and consume a lot of energy.
One potential avenue for developing computers that can go beyond the limitations of current chips is the interaction of light waves with matter. It is based on the same principles as chips from the earliest stages of the computing revolution.
Developing Light Speed AI Training
At the University of Pennsylvania, a team of engineers has developed a state-of-the-art chip that utilizes light waves instead of electricity in carrying out complex calculations for AI training. This chip can accelerate computing devices' processing speed while reducing their energy consumption.
H. Nedwill Ramsey Professor Nader Engheta led the pioneering research in manipulating materials at the nanoscale to conduct complex tasks using light. His team developed a silicon-photonic chip with capabilities discussed in the paper "Inverse-designed low-index-contrast structures on a silicon photonics platform for vector-matrix multiplication."
In this study, Engheta's group collaborated with that of Firooz Aflatouni, who pioneered silicon device development. They aim to develop a platform for performing vector-matrix multiplication, which is a core mathematical task in the development and function of neural networks.
Instead of using a silicon wafer of uniform heights, the team made the silicon thinner, up to 150 nanometers, in specific regions. The variations in heights provide a way to control light propagation through the chip. It can also be distributed to cause scattering of light in specific patterns, enabling the chip to conduct mathematical functions at the speed of light.
According to Aflatouni, their new design is ready for commercial applications and can even be adapted for graphics processing units (GPUs). There has been a rise in the demand for graphics processing units due to the increasing interest in developing new AI systems.
Aside from faster speed and less energy consumption, the silicon-photonic chip also provides privacy advantages. Since many computations can occur simultaneously, storing sensitive information in a computer's working memory will not be needed. As a result, future computers powered by such a technology will be virtually unhackable.
RELATED ARTICLE: Photonic Chip Developed To Calculate the Optimal Shape of Light, Offers Key to Removing Barriers in Optical Wireless Technology
Check out more news and information on Silicon Chip in Science Times.
