Lithography, transferring chip design masks tothe silicon wafer, is the most important phase inmodern semiconductor manufacturing flow. Dueto the limitations of lithography systems, Extensive design optimizations are required to tacklethe design and silicon mismatch. Inverse lithography technology (ILT) is one of the promisingsolutions to perform pre-fabrication optimization,termed mask optimization. Because of mask optimization problems’ constrained non-convexity,numerical ILT solvers rely heavily on good initialization to avoid getting stuck on sub-optimalsolutions. Machine learning (ML) techniques arehence proposed to generate mask initializationfor ILT solvers with one-shot inference, targetingfaster and better convergence during ILT. Thispaper addresses the question of whether ML models can directly generate high-quality optimizedmasks without engaging ILT solvers in the loop.We propose an implicit learning ILT framework:ILILT, which leverages the implicit layer learning method and lithography-conditioned inputs toground the model. Trained to understand the ILToptimization procedure, ILILT can outperform thestate-of-the-art machine learning solutions, significantly improving efficiency and quality.