Calibre Curvilinear Verification

Curvilinear masks offer significant benefits for advanced technology nodes. A primary advantage is the improved process window compared to traditional rectilinear shapes. This larger process window enhances wafer lithography performance and makes the printed image less sensitive to process variations. Curvilinear inverse lithography technology (ILT) results have been demonstrated to yield the best process window metrics, such as the best depth of focus (cDOF). Overall, curvilinear masks are increasingly seen as the next generation mask technology due to their unique power in delivering high-wafer performance.

Read the technical paper "Curves ahead! IC manufacturing prepares for curvilinear masks" and learn more about the benefits of curvilinear masks.

Curvilinear masks inherently have shapes including features like sharp angles, T/Y junctions and sharp turns. Therefore, simple verification checks are often insufficient. For example, unlike rectangular shapes where you mainly measure perpendicular to straight edges, the edges of curvilinear shapes “change direction” constantly. This means that measuring minimum distance involves measuring in various directions to find the true minimum.

Curvilinear masks need complementary area checks. Minimum area checks are crucial for curvilinear shapes because they can catch conditions that pass simple dimension checks but would still fail to print due to the strong contribution of mask blur on small 2D features. In addition, curvilinear masks require mask rule checks that are unique to them, such as the maximum curvature checks that are not required for rectangular masks.

Calibre offers Curvilinear Mask Rule Check (CLMRC) verification through two approaches. One approach is to leverage Calibre OPCverify. Examples of primary checks for CLMRC in Calibre OPCverify include checking the radius of curvature, computing minimum width, computing depth-to-width ratio, measuring distance between edges, checking for spike angles and checking for spline on a multigon layer.

The other approach is Calibre RET CLMRC. It is a set of standard verification rule format (SVRF) commands designed to perform CLMRC checks on curvilinear masks. All the CLMRC checks available in Calibre OPCverify are supported in Calibre RET CLMRC.

The need for machine-learning classification arises because geometry-based error classification, traditional for Manhattan designs, has limitations for curvilinear designs due to edge angle dependency, large vertex/edge counts and grid snapping inconsistency. Machine-learning approach classifies errors based on feature vectors derived from the layout context at the error location, rather than relying solely on vertices and edges. The key benefit is a dramatic reduction in the number of unique error categories for review, making the error review process much more manageable. Machine-learning approach also allows for further sub-classification of errors based on criteria like edge type (convex, concave, straight) or geometric properties like width or space, providing flexibility for fine-tuning results.

Machine-learning based classification is particularly useful in the context of curvilinear layout processing because it enables automated, efficient and scalable identification of error types across complex patterns. Machine-learning models can be trained to classify errors in the feature vector space, providing ideal guidance for error classification for curvilinear layouts. This capability can be directly applied to Curvilinear Optical Proximity Correction (CLOPC) verification, Curvilinear Mask Process Correction (CLMPC) verification and Curvilinear Mask Rule Check (CLMRC), where quick and accurate identification of design or process issues is critical. Different from the traditional geometry-based classification, by using machine-learning based classification, users will achieve good classification ratios (fewer errors to review) while still having the ability to differentiate between error examples that are sufficiently different to warrant separation in the classification categories.