Photonic chips for low-power neural networks
A study by the Politecnico di Milano in Science
A study by the Politecnico di Milano and Stanford University, published in the journal Science, shows that it is possible to create extremely efficient neural networks using photonic chips.
Neural networks are distributed computing structures inspired by the structure of a biological brain and aim to achieve cognitive performance comparable to that of humans. They are used in many areas, such as speech and image recognition and synthesis, autonomous driving and augmented reality systems, bioinformatics, genetic and molecular sequencing, and high-performance computing technologies.
Neural networks are trained with a large amount of known information, on the basis of which they become able to adapt their behaviour, working autonomously. However, their training is an extremely energy-intensive process.
Researchers from the Politecnico's Photonic Devices Lab and Polifab, the university's micro- and nano-technology centre, in collaboration with researchers from Stanford University, have sought a solution and developed a silicon microchip just a few square millimetres in size with an integrated photonic accelerator that allows calculations to be performed very quickly - in less than a billionth of a second - and efficiently. Thanks to this photonic chip, neural network operations take place with considerable energy savings.
In addition to neural networks, it will be possible to use this device as a computing unit for multiple applications where high computational efficiency is required, e.g., for graphics accelerators, mathematical coprocessors, data mining, cryptography and quantum computers.
For more information
The study published online in Science
H.P.C. & Quantum @polimi