California Institute of Technology

University of California

Magic Leap, Inc.

Gregory C Roberts

Andrei Faraon

Philip Camayd-Munoz

Conner Ballew

Ashkan Aliabadi

Athanasios Nenes

Methods and devices to split electromagnetic waves across broad bandwidths in correspondence with predefined polarization state vectors are disclosed. The described methods can be used cameras or image sensors measuring directly the polarization states of an incident electromagnetic waves. The devices include three-dimensional (3D) scattering structures made of dielectric pillars and using existing CMOS processes and direct write lithography techniques. Performance metrics based on the intensity and contrast of the split electromagnetic waves are also disclosed.

Broadband polarization splitting based on volumetric meta-optics

Methods and devices to build and use multi-functional scattering structures. The disclosed methods and devices account for multiple target functions and can be implemented using fabrication methods based on two-photon polymerization or multi-layer lithography. Exemplary devices functioning as wave splitters are also described. Results confirming the performance and benefits of the disclosed teachings are also described.

Color and multi-spectral image sensor based on 3D engineered material

Methods to build multi-functional scattering structures while respecting tight requirements imposed by manufacturing processes are described. The described methods and devices are based on etching away wire networks embedded in 3D structures to form voids in order to perform a target function. Optimization algorithms for designing binarized devices that meet manufacturing requirements are also disclosed.

CMOS color image sensors with metamaterial color splitting

Systems and methods for efficient implementation of a convolutional layer of a convolutional neural network are disclosed. In one aspect, weight values of kernels in a kernel stack of a convolutional layer can be reordered into a tile layout with tiles of runnels. Pixel values of input activation maps of the convolutional layer can be reordered into an interleaved layout comprising a plurality of clusters of input activation map pixels. The output activation maps can be determined using the clusters of the input activation map pixels and kernels tile by tile.

Efficient data layouts for convolutional neural networks

A cloud condensation nuclei instrument including an aerosol flow line () to carry a desired sample to the CNN chamber () and a temperature control unit (A,B) to provide a monotonically increasing temperature profile along the direction of flow.

Growing community of inventors

San Diego, CA, United States of America

Gregory C Roberts