Award

In this project, AI re-engineering of printed circuit boards (AIR-PCB), Deep Analytics LLC and Sandia National Labs outline a front-to-back architecture to extract PCB bill of material, schematic, and layout data from color images and x-ray tomography data of a printed circuit board. This approach is unique in that it aims to recreate fully editable PCB design files in the free open source tool KiCAD. For the Phase I, we propose to a 3 month project to develop key algorithms to identify the connectivity parts by identifying traces, vias, and unique part identifiers from color and X-Ray tomography images supplied by the microtomography group at Sandia National Labs. The embodiment of the final deliverable is the re-creation the intermediate data product called a netlist. The netlist is a crucial component for both the layout and schematic that identifies the connectivity between inputs and outputs of PCB parts. The project will demonstrate recreating a KiCAD netlist by 1) collecting images and x-ray tomography images of PCBs made with KiCAD, 2) developing and demonstrating algorithms that identify and extract pixel locations of vias, follow traces, and extract PCB unique identifying text (e.g., R1, C1, U1, etc.), and 3). Correlating items in the database (e.g., a solder pad with a via with a trace, etc.) to identify connectivity between parts. The reconstructed netlist is then compared with the original netlist created during the PCB design and performance is measured. While the DA and Sandia team both boast significant past performance in AI-based data extraction techniques, filling out the database with parts, part text, and design rule information is left to the Phase II effort. In the Phase II a database all all PCB text, parts, vias, drill holes, planes, and traces will be used to produce the human readable KiCAD design files for the netlist, schematic, and BOM, and cross referenced part footprints, trace data, and design rule data will be used to recreate layouts (including automated detection and reproduction of complex impedance matched traces). Finally, as PCBs are complex and can contain custom unknowable parts (e.g., ASICs), AIR-PCB will provide the user a report identifying missing data, and enable the user to populate data manually (e.g., values from passives). In summary, AIR-PCB provides a front-to-back PCB re-engineering suite. At the conclusion of Phase II, DA and Sandia will delivery prototype software and a set of input data requirements for Government evaluation.