Abstract
Inkjet printing is considered a very promising technology in the field of organic light-emitting diode (OLED) substrate manufacturing. Compared with the vapor deposition process, inkjet printing has the advantages of a simple process, high material utilization and applicability to a wide range of display manufacturing processes. However, during the inkjet manufacturing process, the resolution of the printhead (nozzle per inch, NPI) usually does not match the pixel resolution of the substrate (pixel per inch, PPI). Therefore, the travel path of the printhead module must be planned to minimize the number of print cycles required to complete the pattern. This involves a challenging multi-objective optimization process. The difficulty intensifies in large-area OLED production, where angular alignment errors in the substrate are magnified. This results in an exponential increase in pixels requiring planning, with the total pixel pit count reaching hundreds of millions. In addition, the time complexity of the planning problem grows exponentially, denoted as O(m(n)), and the space complexity grows rapidly with the matrix dimension. This problem is NP-hard. This problem has a significant impact on the productivity of the manufacturing process. In this paper, a large-area substrate printing planning algorithm based on graph attention networks and integer programming (GIP-LASP) is established. GIP-LASP provides partitioning rules and parallel modeling methods specifically for substrate misalignment angles, and proposes imitation learning based on a multi-head graph attention network on a SCIP solver, which is applied to the solution of the printing planning problem. The planning method was implemented on a G4.5 half-size substrate with a resolution of 394 PPI, and the color filter (CF) layer was successfully printed.