Photonic devices are increasingly being used in the microelectronics sector for faster communication and faster and new types of computing, including neuromorphic circuits for deep learning. Their energy efficiency further provides promise of more energy efficient computing. Designing these devices with the required performance is challenging: they are computationally very large and sensitive to multiple parameters and manufacturing variations. Existing commercial solutions do not scale and do not provide efficient optimization, nor can they take advantage of the new forms of heterogeneous computing coming on line. There is a need for a robust simulation tool that would combine high performance computingwith multiple compute devices and efficient optimization and sensitivity analysis with an easy-to-use interface, accessible to industry engineers. Tech-X will create a high-performance application for sensitivity analysis and optimization of photonic devices crystals, fibers, gratings, couplers, transceivers, optical phased arrays, photonic integrated circuits, etc.) based on the adjoint method, cut-cell subpixel) grid for multi-material shapes ability to model dispersoin, and able to be used in new heterogeneous computing environments. Adjoint methods allow for the accurate gradient calculation needed for sensitivity analysis and for use in efficient optimization algorithms. Cut-cell algorithm provides the second order convergence and accurate representation of the material boundaries needed for shape optimization. Ability to run in a wider array of heterogeneous compute environments will make this tool more performant and increase its availability. This tool will have a graphical interface with multiple templates for simulating common passive photonics devices. The proposed development will be based on an existing software base TxSim) that can do forward calculations of the photonics devices using cut-cell/sub-pixel methods with the Finite Difference Time Domain FDTD) method for electromagnetics and photonics. In Phase I, we will augment TxSim withthe prototype implementation of adjoint optimization and dispersive embedded boundaries, and we will begin to port it to a broader array of GPUs. The proposed tool will streamline the design of new photonic devices with best performance: high sensitivity, strong coupling, high Q, low losses, low cross-talk, and clear signal filtering etc., while enabling use with the latest compute devices. This tool will also facilitate prediction and mitigation of the manufacturing variations effects. Example of photonics applications effecting energy and environment are: sensors for oil/gas exploration, medicine, environment and defense, datacenter CPU to memory interconnects, optical computing, self-driving cars, and Internet of Things IOT), telecommunications. The techniques developed here will be applicable to simulations for DOE mission areas, including radiofrequency heating of fusion plasmas and particle accelerator design.
