Methodology Characterizes Data-to-Decision Process

While tighter process windows and process tolerances spur the need for more measurements throughout the patterning process, most metrology data analysis is done offline. This creates a predicament for the fab: Additional measurements and measurement types mean more data to process and speed yield learning. The conundrum is turning these data into information that quickly assists in taking the correct action.

An outcome of advanced lithography's requirements (tighter tolerances and shrinking windows) is that monitoring first-order effects like CD and overlay error is no longer sufficient; previously unchecked parameters show critical yield correlation. These include feature profile, critical shape, intrafield spatial distribution, focus/exposure error, etc., which must be monitored and characterized in real time, requiring an on-tool engine to model them for accurate feed-forward and feedback advanced process control (APC). For example, knowing that measured CDs are 96.2 ±0.9 nm, 94.4 ±0.8 nm, etc., is insufficient for optimal process control. Analysis showing that isolated 90 nm lines systematically vary by 6.5 nm across the slit for a lot, and that a 175 nm focal-tilt adjustment would reduce systematic variation across the slit from 6.5 to 2.1 nm helps facilitate critical decision-making. Knowing this in real time, before the next lot, would improve yield; however, most fabs lack the analytical machinery and best-known methods to timely employ this parametric data to achieve tighter lithography and etch cell process control.

KLA-Tencor (San Jose) has developed a data-to-decision methodology for parametric analysis, in addition to software and algorithms for its lithography overlay metrology tool to do real-time data analysis, as well as a server to store and process data for offline analysis. Named the Archer Analyzer, this centerpiece of its analyzer product family provides overlay correctables, overlay and focus/dose disposition, overlay and focus/dose troubleshooting, and engineering analysis functions. Similar functionality will be available for KLA's CD-SEM and optical CD platforms, enabling real-time, on-tool, on-product analysis across the lithography and etch cell.

In the overlay area, there are feedback and feed-forward capabilities. Conventional state-of-the-art uses four-corner sampling with large feature targets, such as a box-in-box, and simple linear correctable models. Another step is required for high-order effects, using grating-based targets, increased intrafield sampling, and the analyzer providing higher-order overlay correctable models on the platform. These provide the enhanced information needed to improve APC and advanced equipment control, including lot and tool disposition, run time and preventive maintenance feedback and feed-forward, root cause analysis, and automated fault detection. The enhancements make it possible to go after currently uncharacterized, unmodeled systematics, improving feedback and feed-forward for APC. Effects like pattern placement error — scanner aberration overlay variations — are traceable. These have been beyond simple linear measurements and analysis' reach.

Yield ramp delays are severely limiting ROI in 300 mm factories. Increasingly complex designs interact with shrinking process windows, causing delays, slower ramps and decreasing yield. (Source: KLA-Tencor)

The situation for CD is very similar. Today, the state-of-the-art is linear dose feedback and feed-forward based on mean CD. The wafer is measured on a CD-SEM, and APC feedback/feed-forward is done based on a linear dose model for the mean CD. Advanced analysis enables manufacturers to use higher-order effects in real time. By measuring profile information, critical shape information, top-metal bottom, CD sidewall angle and height, as well as corner rounding or critical shape, and putting these measurements into run-time production use through advanced analysis, it becomes possible to look not only at linear dose, but also second-order effects such as focus, responsible for 80% of CD variation.

As process windows decrease, resolution enhancement techniques, immersion lithography, design-for-manufacturing (DFM) and other methods are intended to delay this challenge. Metrology's goal is to characterize, quantify and center process windows across multiple dimensions, including spatial variation. If focus/exposure can be characterized at multiple points in the field, it can be modeled as a tilt or a tilt curvature; the usable depth of focus is actually smaller than supposed, based on simple sampling in one or two locations. There is also density- or proximity-based observation, so isolated, dense and critical-shape features have independent process windows and must be compared in real time for process characterization and yield.

As process complexity increases, additional measurements and measurement types are needed on more and more critical layers. This data increase must be handled automatically.

Find more information on inspection, measurement and test.