The challenge for the upcoming full-chip CD uniformity (CDU) control at 32nm and 22nm nodes is unprecedented with expected specifications never before attempted in semiconductor manufacturing. To achieve these requirements, OPC models not only must be accurate for full-chip process window characterization for fine-tuning and matching of the existing processes and exposure tools, but also be trust-worthy and predictive to enable processes to be developed in advance of next-generation photomasks, exposure tools, and resists. This new OPC requirement extends beyond the intended application scope for behavior-lumped models. Instead, separable OPC models are better suited, such that each model stage represents the physics and chemistry more completely in order to maintain reliable prediction accuracy. The resist, imaging tool, and mask models must each stand independently, allowing existing resist and mask models to be combined with new optics models based on exposure settings other than the one calibrated previously.In this paper, we assess multiple sets of experimental data that demonstrate the ability of the Tachyon™ FEM (focus and exposure modeling) to separate the modeling of mask, optics, and resists. We examine the predictability improvements of using 3D mask models to replace thin mask model and the use of measured illumination source versus top-hat types. Our experimental wafer printing results show that OPC models calibrated in FEM to one optical setting can be extrapolated to different optical settings, with prediction accuracy commensurate with the calibration accuracy. We see up to 45% improvement with the measured illumination source, and up to 30% improvement with 3D mask. Additionally, we observe evidence of thin mask resist models that are compensating for 3D mask effect in our wafer data by as much as 60%.