A new optical computing system can solve very complex problems that are not available to the most powerful supercomputers.
An important class of difficult computational problems can be solved by multiplying light signals with applications in graph theory, neural networks, artificial intelligence, and error correction codes, according to researchers at the University of Cambridge and the Skolkovo Institute of Science and Technology in Russia. .
In a paper published in the journal Letters of Physical Opinion, propose a new type of computation that can overturn analog computing, dramatically reducing the number of light signals required, simplifying the search for the best mathematical solutions, enabling ultra-fast optical computers.
Optical or photonic computation uses photons generated by lasers or diodes for calculation, as opposed to classical computers that use electrons. Because photons are essentially massless and can travel faster than electrons, an optical computer is super fast, energy efficient, and capable of processing information simultaneously through multiple optical channels in time or space.
The computer element of an optical computer — an alternative to the digital computer and zeros — represents the continuous phase of the light signal, and computation is usually achieved by adding and then projecting two light waves from two different sources. result in ‘0’ or ‘1’ states.
However, real life presents problems that are very non-linear, where multiple strangers simultaneously change the values of other strangers while interacting in multiplied ways. In this case, the traditional view of optical computing that combines light waves in a linear fashion fails.
Now, Professor Natalia Berloff of the Cambridge Department of Applied Mathematics and Theoretical Physics and Nikita Stroev, PhD, at the Skolkovo Institute of Science and Technology have found that optical systems can combine light, replace functions and represent them instead of multiplying them by describing light waves. different types of connection between light waves.
This phenomenon was illustrated with quasi-particles called polaritons – which are half-light and half-matter – extending the idea to larger classes of optical systems such as light pulses in a fiber. Consistent and fast-moving polytones can produce small pulses or small motions in space and can overlap with each other in a nonlinear way, due to the component matter of the polaritons.
“We found a key ingredient in how to connect legumes to each other,” Stroev said. “If you get the coupling and the light intensity well, the light multiplies, triggering the individual pulse phases, giving you the answer to the problem. This makes it possible to use light to solve non-linear problems.”
The multiplication of wave functions to determine the phase of the light signal in each element of these optical systems comes from the nonlinearity that occurs naturally or enters the system from the outside.
“What was surprising was that we didn’t have to constantly project the light phases to the‘ 0 ’and‘ 1 ’situations needed to solve binary variable problems,” Stroev said. “Instead, the system often causes these situations to end up looking for a minimum energy configuration. This is a property that comes from multiplying light signals. In contrast, previous optical machines require a resonant excitation that fixes the phases externally to binary values.”
The authors have also proposed and implemented a way to direct the trajectories of the system towards a solution by temporarily changing the signal coupling forces.
“We should start by identifying the different classes of problems that a dedicated physical process can directly solve,” said Berloff, who also has a job at the Skolkovo Institute of Science and Technology. “Higher order binary optimization problems are one of those classes, and optical systems can be very effective at solving them.”
There are still many challenges to be faced before proving the dominance of optical computing in solving difficult problems compared to modern electronic computers: noise reduction, error correction, scalability improvement, focusing on the best true system solution.
“Changing our scope to deal directly with different types of problems can bring optical computer machines closer to solving real-world problems that classic computers can’t solve,” Berloff said.
Reference: “Discrete polynomial optimization by changing with coherent networks of complex condensates and complex couplings” by Nikita Stroev and Natalia G. Berloff. February 5, 2021, Physical Review Letters.
DOI: 10.1103 / PhysRevLett.126.050504