OSA Incubator on Flat Optics: Reflections on Days 2 and 3
By Soon Wei (Daniel) Lim, OSA Student Member, Harvard University
Here is a quick question to start your day: in the future, when manufacturing flat optics (or any precision optics, for that matter), should one follow the model of a paper printout, poster print, or blackboard? That is, should these devices be produced at low volume but in a widely distributed fashion, thereby fully exploiting its versatile and highly customizable capabilities? After all, when you want to print out the latest photonics publication, you do not send the PDF to a specialized foundry expecting to receive eighty reams of glossy cardstock a week later. Alternatively, should flat optics be treated like a poster print: drawing upon specialized expertise in using industrial-grade equipment to produce high-quality, high-volume duplications of a much-needed structure? When flat optics attains full integration into the standard CMOS manufacturing process, it can draw upon the well-established production lines of the semiconductor industry to produce complex, multi-layered structures at scale. Or perhaps, would the most appealing aspect of flat optics be its reconfigurability, just like that of an erasable blackboard? One could write over the same underlying structure to re-use the device for multiple purposes at different times.
Over the second and third days of the OSA Incubator on Flat Optics, we zoomed into the specifics of generating, optimizing, fabricating, and applying these versatile devices to various situations. Expect to see both high-speed and extreme-scale simulations, devices at varying levels of technology readiness, and unexpected use cases for this technology platform.
Day 2: Thursday, February 27, 2020
Day two of the meeting kicked off with a trio of talks on the theme of simulation and optimization of large metasurfaces. Professor Jonathan Fan from Stanford University introduced the concept of Global Topology Optimizers based on neural networks (GLOnets), which is a technique of stably obtaining the globally optimal (or near globally optimal) structures to achieve a specified objective (such as the deflection of incident light). Instead of performing local optimization with gradient-based approaches, and thereby spending large amounts of computational resources exploring local optima in inefficient design spaces, these GLOnets train a generative neural network to map noise to a distribution of highly efficient devices. These highly efficient devices tend to cluster around the global optimum, exploring the design space efficiently and with reduced computational cost. Professor Fan ended with a challenge for the photonics community to follow the example of the computer vision and image recognition communities, within which code is open-sourced and datasets are widely distributed for benchmarking purposes. His initial venture, MetaNet, takes a step towards realizing such a future by sharing results relating to his recent work.
Next up, Dr. Zin Lin from MIT reframed the distinction between traditional ray optics and current nanophotonics by emphasizing how the latter exploits the full physics of electromagnetic waves to control optical waves. He emphasized that the most important aspect of electromagnetic design is in the formulation of the optimization problem itself; one has to carefully construct the objective function and exploit the most suitable numerical solvers for each specific problem. In particular, topology optimization techniques thrive when applied over an immense design space with multiple interacting layers (specifically in 3D nanophotonics), high numerical aperture behavior, large controllable areas, and broad functionalities. As an exemplar, Dr. Lin presented designs for thick multi-layer inverse-designed plan-achromat metasurfaces that can be adjusted to fit fabrication constraints. He envisioned a future where the millions of degrees of freedom afforded by modern 2D and 3D nanofabrication can be fully exploited by efficient exascale simulation techniques and optimization algorithms that explore the full space of possible designs. Interestingly, he suggests that this may involve a paradigm shift away from relying on traditional ray-optic concepts such as the long focal length requirements. In the domain of nanophotonics, we are able to engineer the electric field over a sensor surface directly and hence may achieve superior performance and miniaturized form-factors without requiring the long propagation distances associated with traditional optics.
Professor Owen Miller from Yale University rounded out the first session of the morning with a different perspective on optimization. Instead of working with specific device geometries or materials, which typically produce highly non-convex optimization problems, he identified physical bounds to device performance independent of material parameters and design geometries. By examining the mathematical structure of Maxwell’s equations, Professor Miller was able to derive convex constraints that serve as aspirational upper bounds. This perspective is valuable in that it allows one to select material and geometry platforms that are the most promising in terms of the maximum theoretical performance, even before performing any kind of optimization within these constraints. These bounds can even incorporate fabrication constraints, such as the minimum feature size, so as to be applicable to various manufacturing capabilities. In the context of meta-lenses, Professor Miller was able to show that state-of-the-art high numerical aperture meta-lenses could still be improved substantially since there is a sizable gap between the realized and theoretical performance limits.
Moving into a device-level perspective, Professor Harry Atwater from the California Institute of Technology opened up the next session by exhibiting a dynamically tunable metasurface platform with individually addressable meta-elements. While such phased array systems are regularly deployed in radio-frequency applications, it is exceedingly difficult to exert this level of control at visible and infrared wavelengths. Professor Atwater demonstrated how it is possible to use inverse design techniques to compensate for non-idealities in meta-elements (such as amplitude variations and incomplete phase coverage) and thereby recover near-ideal device performance. Professor Atwater’s metasurfaces were applied to beam steering systems, which are presently in high demand by the automotive industry for self-driving applications.
In the same vein of applying tunable metasurfaces for 3D optical sensing, Dr. Gleb Akselrod from the startup Lumotive provided us with a glimpse into a brand new LIDAR device based on a large liquid-crystal metamaterial array. The device is able to steer multiple beams over a wide range of angles and receive optical signals from controllable directions. Such a device would obviate the current need for mechanical rotation mounts for modern LIDAR systems. In addition, since this device was fabricated using CMOS-compatible technology, it is even possible to scale these devices down in size to fit into a smartphone and make these sensing techniques available to a wider audience. More information can be obtained from his article published in Laser Focus World.
Professor Debashis Chanda from the University of Central Florida then presented a range of innovative display systems and platforms based on nanostructured plasmonic surfaces. In a rapid-fire succession of slides, he detailed how nanostructured aluminum and 2D material surfaces could be tuned, controlled, and fabricated so as to produce high resolution thin displays, wavelength sensitive camouflage surfaces, and even room-temperature infrared sensors. Several of these technologies were bio-inspired: the color arising from nanostructured butterfly wings led to his color-control and advanced metallic paint technologies, while the room-temperature infrared sensing abilities of the pit viper led to the development of a graphene-based tunable infrared sensor.
After lunch, a total of nine speakers from various disciplines presented a wide range of metasurface conceptions, applications, and fabrication procedures in a rapid-fire oral presentation session. This space is too short to justify the achievements detailed in each of the presentations, and the reader is encouraged to check out recent publications from each of the authors below.
- Cascaded Metasurface Optics – Professor Amir Arbabi, University of Massachusetts Amherst, United States
- Photonic Inverse Design for 3D Color Splitting Applications – Gregory Roberts, California Institute of Technology, United States
- Two Metasurface Layers for Phase Gradient Imaging – Hyounghan Kwon, California Institute of Technology, United States
- Computational Imaging with Dielectric Metasurfaces – Shane Colburn, University of Washington, United States
- RGB-Achromatic Metalenses for a VR/AR System – Dr. Zhaoyi Li, Harvard University, United States
- Infrared Metasurfaces – Dr. Clara Rivero-Baleine, Lockheed Martin Corporation, United States
- Multiscale Inverse Design for Systems of Metasurface Optics – Dr. Adam Backer, Sandia National Laboratories, United States
- Metasurface Holographic Projectors for Augmented Reality on Contact Lenses – Professor Shoufeng Lan, Texas A&M University, United States
- Additively Manufactured Freeform Gradient Index Optics – James Field, Voxtel Inc., United States
In the following session, industry leaders in surface-patterning presented on the prospect of manufacturing metasurface nanostructures at high volumes. Personally, as an experimentalist working in an academic laboratory with one tiny precious sample at a time, seeing these meta-devices produced seemingly effortlessly at meter-scale or even kilometer-scale regimes was simultaneously humbling and awe-inspiring.
To start off the session, Dr. Dim Lee Kwong and representatives from the Institute of Microelectronics (IME) in the Agency for Science, Technology, and Research (A*STAR), Singapore, detailed the process development steps towards realizing large-scale CMOS-compatible fabrication of metasurfaces. They described the challenges that were successfully overcome to achieve a reproducible process with less than 5 nanometers of critical dimension variation over a 12-inch wafer, and high aspect ratio structures exceeding 20:1. The process was proven on various material platforms and devices ranging from polarization optics, to spectral filters and metalenses.
Dr. Robert Visser from Applied Materials unveiled the results from their first test run of near-infrared metalenses on a 12-inch wafer platform. The pillar array in the metalenses was stabilized and protected by an inorganic layer, which successfully filled the high aspect ratio pillar gaps using a specialized process. He recommended that the industry leverage the established semiconductor industry processes to scale up near-infrared silicon metalenses to high volumes.
Dr. Martin Wolk from 3M spoke on the semiconductor route to metasurfaces with a discussion on alternative high-volume fabrication methods for manufacturing flat optics at the kilometer-scale: nano-replication and roll-to-roll nanoimprint lithography. Both methods begin with a nanopatterned semiconductor wafer master and involve continuous molding and photocuring of functional organic resin formulations onto a film substrate in a rapid and cost-efficient roll-to-roll process. Dr. Wolk exhibited preliminary results demonstrating both routes to producing aperiodic nanohole arrays. While roll-to-roll processes are an established means of producing large area structured surfaces, the strict tolerances and small critical dimensions associated with nanostructures continues to motivate intense research and development within 3M to push the limits of continuous replication technologies. A discussion after the presentation included a question about the compatibility of the roll-to-roll process with future multilayer optical metasurface designs. Designs requiring submicron overlay accuracy will not be feasible. However, as Professor Amir Arbabi pointed out, cascaded metasurface designs with decoupled surfaces that are relatively far apart (e.g. on the opposing side of a film substrates) would require much less stringent alignment than may be obtained in a roll-to-roll process.
Mike Bülters from Temicon then provided an overview of the nanostructuring capabilities of the company, demonstrating how nanostructured moth’s eye antireflection surfaces can be produced through several fabrication techniques. In particular, the company’s laser interference lithography capability allows one to obtain resolutions under 100 nm over a meter-scale area without the need for field stitching.
In the final session of the day, we dived into the intricacies of flat optics for augmented reality (AR) applications. Dr. Pierre St. Hilaire from Magic Leap , an AR display startup, provided an in-depth view into the physics and engineering behind the development of their product, the Magic Leap One, as well as an overview of the geometry and material constraints that needed to be met. While meta-gratings are not currently incorporated into the production device due to efficiency constraints, it is still a candidate for future iterations.
Professor Hwi Kim from the IPDS Lab at Korea University then wrapped up day two of the incubator meeting by presenting a huge range of projects related to structuring light and producing color holograms using meta-optics and spatial light modulators (SLM). He proposed a design for an SLM with full amplitude and phase control by cascading a metasurface with an SLM.
Day 3: Friday, February 28, 2020
Moving into the third and final day of the incubator meeting, Reinhard Völkel from SUSS MicroOptics presented an intriguing overview of the progress of micro-optics from the dawn of life, to its first human applications in computer generated holograms, to the state-of-the-art micro-imprinting techniques for high-precision modern optics.
Dr. Amit Agrawal from the National Institutes of Standards and Technology followed up with a variety of projects enabled by metasurfaces. One project drew upon metasurfaces as an interface from the classical world to quantum systems, drawing upon the multi-functional nature of metalenses to produce high NA focused spots for different incident wavelengths and polarizations. These focused spots were then used for full 3D atomic trapping of Rubidium atoms. Another project was that of a photonic optical clock, which drew upon multiple metalenses and meta-reflectors for both focusing and polarization control of optical traps within the compact chip-scale atomic clock. Both these systems were greatly simplified and made more compact by the introduction of multifunctional metasurfaces.
Dr. Wei-Ting Chen, one of the co-hosts of the incubator, then presented a detailed study on the aberrations and efficiency of high-end metalenses, drawing quantitative comparisons of optical behaviors between metalenses, traditional refractive optics, Fresnel diffractive optics, and multi-level diffractive optics. He demonstrated how a fundamental difference between metalenses and spherical or aspheric lenses is that the metalens has no field curvature, allowing the focal plane to be truly planar. In addition, he exhibited detailed measurements of how a metalens achieves large fields of view without vignetting.
In the final session of the incubator, Dr. Paulo Dainese from Corning Inc., another co-host of the meeting, demonstrated how metalenses could help solve a key problem in digital communications: the ever-growing gap between humanity’s rapidly-growing ability to generate data and the limited transport bandwidth available to move that data around. In short, every degree of freedom of propagating light to carry information is being explored to achieve the physical limits of communication rates over single mode fibers: amplitude and phase modulation, wavelength multiplexing, polarization multiplexing and last but not least spatial division multiplexing. Corning has collaborated with Harvard University to develop folded-light devices that perform multiplexing and demultiplexing of waveguide modes in a compact device using inverse-designed metasurfaces. Such multifunctional metasurfaces are able to convert between spatial modes of an optical fiber and also launch the modes into other fibers. They may eventually be explored to increase the amount of information that can be carried in our fiber optic networks.
Finally, Dr. Pawel Latawiec and Dr. Rob Devlin from startup company Metalenz, a spin-off from Harvard University, provided a hands-on demonstration of a product built around metasurfaces: a near-infrared VGA camera module that offers a simpler architecture, higher brightness, and better or equivalent sharpness as compared to traditional camera modules. While a modern camera module contains around four specially-designed polymer lenses to compensate for various lens aberrations, the Metalenz product simplified the camera structure, removing unwanted vignetting and increasing the amount of light captured by the camera sensor. They also exhibited how the metalens camera was fully integrated in an end-to-end differentiable approach to design, which enables one to perform fabrication sensitivity analysis, yield analysis, and factor prioritization for efficient device production.
To conclude the incubator meeting, Professor Federico Capasso reiterated that the key applications of new technologies like flat optics will likely be those that are created by the technology itself, not applications that are already in existence. This incubator meeting is just one-step toward bringing a promising future enabled by flat optics and metasurfaces closer.
On a more personal note, it has become increasingly clear throughout the incubator that it may be more appropriate to consider the areas of research that would produce the most human impact from the application of flat optics and metasurfaces, instead of simply adding to the exponentially growing list of feasible applications. All the necessary ingredients are present and ready for scaling: exact and high-performance simulation platforms, powerful local and global optimization algorithms, proven control over virtually every aspect of optical waves, global market demand for high performance multifunctional optics, and proven compatibility to both mature and promising manufacturing process lines.