Computing with Light – Inside a Resonator

When we think of computing, we usually imagine silicon chips and electricity – the standard architecture powering most of today’s technology. As computational demands skyrocket, the energy efficiency and parallel-processing capability of optics become increasingly compelling. Could we complement electronics with light, using the physics of diffraction and lasers as a computational resource? In fact, you’ve probably heard of a simple optical computer – an optical lens.

When a beam of light passes through a lens, the lens naturally transforms the image into its spatial frequency components, performing a spatial Fourier transform. These spatial frequency components capture how rapidly or smoothly the image varies across space. This property can be used to clean up noisy images. By encoding a noisy image in a  light filed and passing the light through two lenses with a pinhole between them, the high-frequency components of the image will be filtered out, producing a clean output image.

This example highlights the core advantage of optical computing: letting physics do the heavy lifting. Unlike digital computers, which break down operations into discrete electronic steps, optical computing leverages natural phenomena directly. It excels at problems that would benefit from rapid parallel processing and low-energy consumption. Unlike electronic circuits, which often hit data movement bottlenecks, and also heat up as traffic grows, optical systems remain efficient due to their high parallelism and low heat generation. 

While spatial filtering is a useful computational task, it is also an example of a single-pass operation: the light passes through the system once, and the computation is complete. However, most interesting computational tasks involve “many lines of code,” using the output of one calculation as the input for the next. With a single-pass optical system, every iteration demands capturing the light, converting it into a digital signal, processing it electronically, and finally converting back to optical signals if another optical step is required.

The Optical Resonator: Continuous Computation with Light

This is where LightSolver’s Laser Processing Unit (LPU), designed around a degenerate optical resonator, changes the game. Rather than performing a single pass, we trap light inside a programmable optical resonator, forcing it to circulate continuously in a closed loop.

Within the resonator, the light makes thousands of round trips. Each round trip could be thought of as an effective clock cycle taking mere nanoseconds – dictated by the resonator length, and independent of the problem size, since all degrees of freedom update in parallel. The resonator also preserves the state of the light between iterations, eliminating the memory transfer bottleneck. There is no need for moving bits between the memory and the processing unit, since it all happens in the same place! 

In each round trip through the resonator, a programmable linear operation is performed by a spatial light modulator, and nonlinear operations are induced by the gain and loss rate equations of the resonator. These nonlinearities expand the complexity of computations possible within the resonator, enabling the LPU to efficiently tackle complex optimization and simulation tasks at unprecedented speeds. 

A particular class of tasks that the LPU excels at is simulating partial differential equations (PDEs). PDEs underpin countless real-world phenomena, from predicting climate patterns and financial market behavior to designing advanced materials and optimizing aerodynamic performance in vehicles. The LPU’s ability to rapidly simulate PDEs can accelerate research and development across these sectors, reducing computation times by orders of magnitude and enabling real-time decision-making in industries that rely heavily on predictive modeling.

In future posts, we will explain both the linear and nonlinear dynamics happening in the LPU and explain how we map real computational problems into the LPU. And don’t worry – there is even a nice python frontend to program your problems into the LPU! 

Stay tuned for more insights into optical computing, and please contact us if you are interested in hearing more about how computing with light could help in your specialized application.