Software for Simulation and Design of Optical Computers and Optical Neural Nets

The Office of Science of Advanced Scientific Research (ASCR) invested heavily into various HPC software packages, which focus on the DOE mission, but need specialization and enhancements to make them deployed into high-impact industrial applications. Examples of such applications are optical computing for HPC and its subset, optical neural nets for Deep Learning. Optical computing has a potential to be much faster and more power-efficient than traditional computers. The design of optical computing systems faces two main challenges: low precision of photonic circuits due to high losses and parasitic effects; and their significant packaging problems. There is no simulation tool, which is readily available to photonic designers and capable to address these challenges. In Phases I and II, Tech-X Corporation will extend VSim software (funded by ASCR and other DOE offices) for the needs of the simulating and designing high-precision and compact optical computers and neural nets. VSim will (1) provide bi-directional integration with foundries databases; (2) have a means to generate and vary 2D and 3D layouts of photonic circuits; (3) simulate manufacturing variations, (4) be capable to create and model large circuits for optical computing, as well as accept and create parametric cells (PCells); (5) be extended for scans and optimization of imported and created geometries, (6) offer advanced diagnostics for the circuits and nets performance; (7) augment foundry design rules for optimized structures; (8) be integrated into a foundry design flow. In addition to developing new simulation capabilities, we plan to collaborate with photonic designers to do layout and fabrication of circuits designed using new software. In Phase I, Tech-X will address the feasibility to analyze and mitigate the effects of manufacturing variations on the performance of photonic circuits. In particular, Tech-X will (1) introduce a means to simulate sidewall roughness of optical components (with the statistical distribution available from the literature) and save geometries in GDS2 and STL formats; (2) enable ensemble simulations with the parametric and statistical geometry variations (distances, curvatures and roughness); and (3) test new capabilities by performing sensitivity analysis of sample photonic circuits to roughness and geometry variations. Simulation results will consist of the 3D distributions of EM fields in time domain in HDF5 format and S parameters in frequency domain in HDF5 format, visualizable in VisIt. All capabilities will be expressed to users through an easy-to-use Graphical User Interface. The proposed software will be available to multiple researchers in the national laboratories and universities performing research on photonic devices, as well as be available to industrial concerns involved in the manufacture of such devices. Commercial applications include: Artificial Intelligence for Big Data and Autonomous Vehicles, optical computing in general, telecommunications and bio-sensors.