‘Virtually unhackable’ chip could make GPU more energy efficient and much faster in AI – by combining light and silicon for a fundamental mathematical operation
A groundbreaking new chip developed by Penn Engineers uses light waves instead of electricity for AI calculations, marking a potentially significant leap in processing speeds and energy efficiency.
The chip, based on silicon photonics (SiPh), combines Penn Professor Nader Engheta’s esteemed research into manipulating nanoscale materials to perform mathematical calculations using light, with the SiPh platform using silicon, a cheap and abundant element in computer chips.
This revolutionary approach to chip design could potentially surpass the limitations of current chips SciTechDaily points out that they still work according to the principles of the beginning of the computer revolution in the 1960s.
Mathematical calculations at the speed of light
The chip was developed in collaboration with Firooz Aflatouni, associate professor in Electrical and Systems Engineering, whose research group is at the forefront of nanoscale silicon devices. The team wanted to create a platform for vector matrix multiplication, a fundamental mathematical operation in neural networks, which are the backbone of modern AI tools.
The chip’s design involves varying the height of the silicon wafer in specific areas, which allows light to be scattered in specific patterns and allows the chip to perform mathematical calculations at the speed of light.
The design is reportedly ready for commercial applications and could potentially be adapted for use in GPUs, where demand has surged due to growing interest in developing AI systems.
In addition to improved speed and lower energy consumption, the chip also offers privacy benefits. Because many calculations can take place simultaneously, there is no need to store sensitive information in a computer’s working memory, making a future computer powered by such technology virtually unhackable.
The design of the new chip was detailed in an article published in Nature Photonics.