Controlling Light with AI

Spectro-temporal representation of femtosecond pulse patterns, prepared by a photonic chip to seed the generation of supercontinuum. The patterns are optimized via machine-learning to select and enhance desired properties in the output supercontinuum. (Source: B. Wetzel, INRS)

Optical systems make it possible to shape and control the charac­teristics of laser light, but with certain limitations for the user. Today, researchers are able to minia­turize optical tools and assemble optical chips. No bigger than a fingertip, optical chips are like an obstacle course for the light that passes through them, changing its properties in the process. Improve­ments in optical chip design, developed by INRS-researcher Roberto Moran­dotti and an inter­national team, open the door to a wealth of new scientific and tech­nological possi­bilities.

Until now, however, control over lasers has been limited to certain pulse charac­teristics and has required expensive and cumbersome devices that are not easily scalable. Many theo­retical proposals have been made in view of overcoming these limi­tations. One of these involves the use of combined light pulses. Unfor­tunately, there is a signi­ficant operational obstacle to using combined pulses effi­ciently: the number of possible combi­nations is simply too high and too complex to be dealt with by means of a con­ventional numerical or experi­mental approach.

Despite this obstacle, this avenue of research holds too much potential to be ignored. Benjamin Wetzel from INRS has developed a way to divide and recombine laser pulses and thus control their individual charac­teristics in a unique way. In doing so, he has opened up a whole new range of possi­bilities: “Using this concep­tually simple approach, we have the possi­bility of exponen­tially expanding the possible combi­nations of system para­meters we can control,” Wetzel explained. “By modifying a handful of variables on the chip, we can obtain more than 1036 pulse shape confi­gurations to control our optical system.”

These astro­nomical numbers illustrate the hurdle repre­sented by the computing power requirements this approach entails. To surmount it, the international team turned to arti­ficial intel­ligence. An automatic learning algorithm was used to help determine the best parameter combinations for obtaining precisely the type of light desired. As proof of concept, light was manipulated to obtain super­continua. These are extended light spectra obtained from intense inter­actions between light and matter. As the results demonstrate, the integ­rated optical chip used in conjunction with the learning algo­rithm produces an optimal pulse pattern that provides researchers with the complex physical dynamics they require.

These exciting results will have an impact in numerous areas of basic and applied research, as many of the optical systems currently in use depend on the same pheno­mena as those deployed for super­continuum generation. In addition, the proposed system is inexpensive and highly compact and can be scaled into more complex systems.

The research team’s work could lead to the develop­ment of other intelligent optical systems using self-optimi­zation techniques. These include the control of optical frequency combs for metro­logy appli­cations, self-adjusting lasers, pulse processing and amplifi­cation, as well as the execution of more funda­mental approaches to intel­ligent learning such as systems based on photonic neural networks. (Source: INRS)

Reference: B. Wetzel et al.: Customizing supercontinuum generation via on-chip adaptive temporal pulse-splitting, Nat. Commun. 9, 4884 (2018); DOI: 10.1038/s41467-018-07141-w

Link: Nonlinear Photonics Group, INRS-EMT, Montreal, Canada

Speak Your Mind

*