SLFAG

Abstract

The increasing trends for miniaturization and better functionality of electronic components have a significant effect to fabrication of Laser Direct Imaging (LDI) on printed circuit board (PCB). LDI uses focused laser beam to directly expose PCB panel coated by photoresist generating PCB image. The scan line fill algorithm is typically used to rasterize PCB image. However, conventional implementation of scan line fill algorithm on CPU is serial, this method is slow and has costly overhead. In this paper, we propose SLFAG, a new Scan Line Filling Algorithm for efficiently rasterizing PCB image based on GPGPU. Due to the slow transfer between CPU and GPU, we take the share memory to improve the transfer and kernel execution. Moreover, we align the threads in thread block to promote the memory access when calculating intersection points. We benchmark six groups of PCB images to evaluate SLFAG. The experimental results shows that SLFAG speeds up 9.4 times on average (and up to 20 times) on a NVIDIA Tesla K80 GPU compared with serial implementation on CPU for PCB image rasterization. We also report that the acceleration result of GPU is influenced by the number of polygon and vertex along with type of polygon.