Hyperspectral imaging is a fast-emerging technique, with applications in sensors, remote imaging, bio-optics, and virtual reality systems. Current hyperspectral imaging system rely on traditional glass optics, elements that have undesirable inherent limitations. A lens made from a metamaterial with a negative index of refraction would be useful for imaging systems. A metamaterial is formed from an array of structures whose size is less than a wavelength. Besides the sub wavelength resolution, the optical component can be quite thin (on the order of a wavelength) as well as much lighter. Also, metamaterials can have indices of refraction larger than unity, leading to a larger numerical aperture and tighter focus. Thus, a metamaterial lens can provide superior performance at a tremendous savings in weight and size. We propose using neural network methods to solve the inverse problem: given a desired intensity profile with subwavelength features, what is the optimal set of design parameters? Inverse problems are notoriously hard to solve. Traditionally, given a design of an optical system, it is straightforward to calculate the resulting intensity pattern; there is commercial and open-source software for such tasks utilizing partial differential equation solvers. However, the inverse problem generally requires solution of integral equations, which is much more difficult. Our solution will rely on neural nets to create efficient inverse problem solutions, a method demonstrated to be over a million times faster than traditional methods. The neural nets will have constraints placed on parameters to restrict them to materials that can be fabricated easily, this will be done via cost functions that will restrain the design parameters to realistic values. This is well within the capability of the Azimuth Autonomy, Artificial Intelligence and Machine Learning Lab (AAIM Lab) high-performance computing cluster. At the end of the Phase I option, we will have a monolithic design with a layer that acts analogously to an antireflection coating, a lens doublet or triplet, and a layer for diffraction. Further the focal plane will be tunable and will include several metamaterial layers. This is to be accomplished by placing the metamaterials on an elastomer which can be stretched and contracted by applied voltages. We will design the lens to be nearly achromatic, to reduce the needed tuning range needed. This design will improve the overall detection efficiency of the imaging system. It will have a thickness on the order of a micron, and will occupy less space, and be lighter than current glass systems. We will make the device as polarization insensitive as possible.
Benefit: The Inverse Hyperspectral Lens Design tool we will develop has potential application in numerous markets: Bioimaging; Virtual reality; Remote Sensing; Sensors; and Space Systems. Metamaterial optics will generate a paradigm shift in device design. Unlike conventional lenses, metalenses based on the metasurface optics with a negative index of refraction are truly flat and compact and exhibit superior performance. These markets can all benefit from one or more technical benefits of metamaterial lenses, whether they are multi-spectral or hyperspectral. Any portable device with an imaging system, cell phone, tablet, endoscope, will benefit from a smaller lens, tighter focus, and a lightweight solution. Greater efficiency will result in reduced illumination levels to achieve results available. Endoscopy and surgical cameras will be less invasive to human subjects. Virtual reality systems and helmet based displays will similarly benefit. Any portable or flight-based systems may enhance performance with improved SWaP-C, as will space based systems. Improved SWaP-C will result, and notably reduce system complexity and secondary support requirements (e.g. vibration and thermal concerns). Fabrication of these devices is anticipated to become simpler and cheaper, this is thanks in part to the technology's compatibility with microprocessor chip and memory foundries, including CMOS chips where it could potentially be mass-produced.
Keywords: Metamaterial Optics, Metamaterial Optics, nanophotonics, Hyperspectral Imaging, Materials Design, Machine Learning, Inverse Design, metalens, Neural networks
