Identifying systematic critical features using silicon diagnosis data

Deriving Feature Fail Rate from Silicon Volume Diagnostics Data

“…This is true for both the dedicated test structure as well as the data derived from scan diagnosis results. Note that the FFR derived from the scan diagnosis represents an upper bound as each and every diagnosis callout matching the design location of a given critical feature is counted as a ''hit, '' even if the true root cause is something else [2]. It is possible, however, to apply some filtering on the diagnosis results to remove some of the noise in the scan diagnosis data, a technique used in the volume diagnosis analysis today [15].…”

“…Furthermore, dedicated test structures cost time and money in terms of silicon wafers and fab cycle time. An alternate approach to identify critical features involves leveraging volume diagnostic data, as discussed in detail in [2]- [4] and [14]- [17]. This approach can identify previously unknown critical features which are representative of product designs as well.…”

On improving estimation of root cause distribution of volume diagnosis

“…At the point that the minimal feature size of a layout becomes smaller than the lithographic wavelength, certain layout features become hard to fabricate correctly and are more likely to cause failures than other features. This kind of prone-to-fail feature is called critical feature in some literature [43]. Besides limitations of lithography, improperly validated Design for Manufacture (DFM) rules can also lead to defects during the manufacturing process of a critical feature.…”

“…A design feature [40] is the characteristic of a cell, metal layer, via or layout shape. Recently, as manufacturing processes have advanced, specific layout patterns in a design that are hard to fabricate have been identified as key causes of systematic defects [42], [43], [51], [54], [55], [58]. Layout shape can be characterized by pixel image of the entire shape, polygons or just center lines [51], [57].…”

“…Layout shape can be characterized by pixel image of the entire shape, polygons or just center lines [51], [57]. Figure 2-3 shows images of a proneto-fail layout structure and defects caused by this structure [43].…”