CuA Optical Metrology Using Effective Medium Region Classification

The patent segments the measurement process into distinct phases: training phase where measurement models are developed using training samples with known parameters, and measurement phase where the developed models are applied to test samples. This segmentation allows the complex task of accounting for CMOS influence to be handled during training, enabling accurate measurements during production without repeating the complex model development process.

The patent performs preliminary action by developing measurement models in advance using training samples before actual production measurements. The measurement models are trained to account for the influence of underlying CMOS circuitry on optical measurements. This preliminary model development enables accurate measurements of production samples without requiring complex real-time analysis during the measurement process.

The patent applies local quality by developing separate measurement models for different spatial regions of the wafer. The wafer is divided into multiple regions, and each region has its own measurement model that accounts for the specific characteristics of underlying CMOS structures in that region. This allows the model to capture spatial variations in CMOS influence without requiring a single overly complex global model.

The patent uses parameter changes by training measurement models with varying parameters that represent different CMOS structure configurations and spatial positions. The models learn to adjust parameters to account for the influence of underlying CMOS circuitry, enabling accurate measurements across different regions without explicitly modeling every detail of the CMOS structures.

The patent performs unsupervised clustering and region identification during the training phase as a preliminary action. By completing the computationally intensive clustering analysis before production measurements, the system establishes region classifications in advance. This allows rapid application of pre-determined region models during production without repeating the time-consuming clustering process.

The patent segments the wafer into multiple spatial regions using unsupervised clustering algorithms that analyze measurement data to identify distinct areas with different CMOS structure characteristics. This segmentation enables the system to handle spatial variations systematically by applying appropriate models to each region, improving classification accuracy while managing processing complexity.