A new silicon photonics platform that can do mathematical operations far more efficiently than previous designs has been unveiled by Nader Engheta and colleagues at the University of Pennsylvania. The US-based team hopes that its system will accelerate progress in optical computing.
Analogue optical computers can do certain calculations more efficiently than conventional digital computers. They work by encoding information into light signals and then sending the signals through optical components that process the information. Applications include optical imaging, signal processing and equation solving.
Some of these components can be made from photonic metamaterials, which contain arrays of structures with sizes that are on par, or smaller, than the wavelength of light. By carefully controlling the size and distribution of these structures, various information-processing components can be made.
Unlike the bulky lenses and filters that were used to create the first analogue optical computers, devices based on photonic metamaterials are smaller and easier to integrate into compact circuits.
Mathematical operations
Over the past decade, Engheta’s team have made several important contributions to the development of such components. Starting in 2014, they showed that photonic metamaterials can be used to perform mathematical operations on light signals.
They have since expanded on this research. “In 2019, we introduced the idea of metamaterials that can solve equations,” Engheta says. “Then in 2021, we extended this idea to structures that can solve more than one equation at the same time.” In 2023, the team developed a new approach for fabricating ultrathin optical metagratings.
Engheta and colleagues have now set their sights on vector–matrix multiplication, which is a vital operation for the artificial neural networks used in some artificial intelligence systems. The team has created the first photonic nanostructure capable of doing vector–matrix multiplication. The material was made using a silicon photonics (SiPh) platform that integrates optical components onto a silicon substrate.
Inverse design
The researchers also used an inverse design procedure. Instead of taking a known nanostructure and determining if it has the correct optical properties, inverse design begins with a set of desired optical properties. Then, a photonic structure is reverse-engineered to have those properties. Using this approach, the team designed a highly compact material that is suited to doing vector-matrix multiplications with light.
Polarization switch makes ultrafast photonic computer
“By combining the inverse design method with the SiPh platform, we could design structures with sizes on the order of 10-30 micron, with a silicon thickness ranging between 150–220 nm,” Engheta explains.
The team says that its new photonic platform can do vector–matrix multiplication far more efficiently than existing technologies. Engheta also points out that the platform is also more secure than existing systems. “Since this vector-matrix multiplication computation is done optically and simultaneously, one does not need to store the intermediate-stage information. Therefore, the results and processes are less vulnerable to hacking.”
The team anticipates that their approach will have important implications for how artificial intelligence is implemented.
The research is described in Nature Photonics.
