Defect-Aware Adaptive Metallization Framework for Perovskite Solar Cells Using FPGA Edge-AI

Microscopic pinholes in perovskite solar cells can create local shunting paths when the metallization grid crosses the defective regions. This problem becomes more important in large-area and roll-to-roll fabrication, where fixed metal grids cannot avoid local film defects. In this work, a defect-aware adaptive metallization framework is proposed for perovskite solar cells using FPGA edge-AI. SEM images are used for pinhole sensing, and each 416 × 416 image is divided into 32 × 32 patches by the MicroBlaze processor on a Xilinx Artix-7 FPGA. The patches are processed using an INT8-quantized TinySegNet_FPGA_v3 accelerator synthesized through hls4ml and stored within on-chip BRAM. The segmented defect map is then converted into C-swerve routing coordinates to avoid pinhole regions during metallization. The model processes each 32 × 32 patch in 5.647 ms at 100 MHz, and the complete 416 × 416 image is processed in 954.3 ms. The electrical effect of the proposed routing is evaluated using a distributed electrical model that measures the actual finger-defect overlapping area directly from the routed metallization geometry, rather than assuming a fixed shunted-area fraction. For a representative 25% nominal defect-area case, the simulated power conversion efficiency improves from 19.95% under conventional straight-line metallization to 20.34% under adaptive C-swerve routing, with the benefit growing at higher defect severities. This result was extended to a full parameter sweep that co-varies finger width (10-80 µm) with defect-area fraction (0-40%), re-optimizing the cell geometry at each finger width. Adaptive routing improves PCE at every tested condition, with the recovery plateauing at +0.045-0.046 percentage points across a 40-80 µm finger-width window at 25% defect area identifying this range as the practical manufacturing window where shunt suppression consistently outweighs the added series-resistance cost of the detour, while the benefit grows further with defect severity at every width. These results suggest that FPGA edge-AI can provide a possible route to connect defect sensing, adaptive metallization, and geometry-aware electrical modelling for scalable perovskite solar-cell manufacturing.