A Look at Overlay Error
Ruth DeJule, Associate Editor -- Semiconductor International, 2/1/2000
 |
Ruth DeJule,
Associate Editor
A trend toward mix-and-match lithography schemes is running head to head with tighter overlay requirements, calling for greater overlay control, increased analysis and understanding of overlay errors.
The main contributors to overlay error are the stage, alignment system and distortion signature. Errors can be broken down into stage motion or wafer alignment errors such as placement and rotation inaccuracies and field errors like errors on the reticle and in camera magnification. These errors are correctable. Pincushion or barrel distortions, third-order field errors, are not correctable but are routinely characterized for a given exposure tool.
For production, first order errors, primarily correctable, are measured, 12 to 18 points per product wafer. Mix-and-match calibration requires more detail, increasing the sample size to ~49 points and in some cases as high as 121. For design rules 0.18 µm and below, higher order errors must be tracked so the sample size may increase to 250 to 500 points.
To measure overlay error, a bright field microscope is used on ~66% of the process layers. Measurement time is on the order of 2 sec/ site. However, for polished layers, the signal-to-noise (S/N) ratio can be poor and affected by contrast variations in film thickness. Optical filters often are added to increase contrast, but some metrology tool suppliers such as KLA-Tencor (San Jose, Calif.) have gone a bit farther by adding an interferometer to the microscope. This enables phase-based measurements that can pick up subtle differences in index of refraction and topography. Also called coherence probe microscopy, CPM may be slower than traditional bright field measurements. To remedy this, KLA-Tencor introduced additional signal processing, Advanced Noise Reduction Algorithm (ANRA, see Figure), which uses a bright field image and therefore has speeds comparable to conventional measurements.
Another way of increasing S/N is with the design of the overlay target, noted Michelle Zimmerman, product marketing manager, yield management software division at KLA-Tencor. The idea is to increase the amount of information contained at the edge of the overlay target where the measurement takes place. Overlay targets often are variations of box-in-a-box. Two adjacent layers will contain box-shaped structures, one 20 µm on a side, the other 10 µm. The center of each box is calculated independently, and a vector difference between them is determined. Some metrology tools measure overlay error as a combination of linewidth measurements. To increase contrast, the boxes can be replaced with combinations of bars and frames, which add structure at the target's perimeter by providing two edges instead of one.
| Targets on CMP layers (a) can be enhanced with new algorithms (b). (Source: KLA-Tencor) |
Overlay measurement techniques used today are focused on repeatability (on the order of 2 nm), noted Scott Ashkenaz, vice president of lithography module solutions at KLA-Tencor. The problem is that there is no practical standard for overlay, he said. Therefore, a true value for any particular overlay target is not known. Some fabs may periodically look at cross sections or make comparisons to electrical parameters, but this is time consuming and relegated to characterization environment, rather than in production. Though accuracy is an unknown figure, there are several components of error that can be estimated, which, when added together, produce a figure called total measurement uncertainty (TMU). This includes estimates of accuracy, bias that comes from errors in the process, and mix-and-match. "In design rules where the total overlay error budget is 50 to 60 nm and the fab thinks they are getting 2 nm of measurement repeatability, what they may not realize is that the total uncertainty is potentially larger than that," stated Ashkenaz. Essentially, too little is being budgeted for measurement error. For the tool manufacturers, this means reducing TMU by breaking it down into components for individual analysis and optimization: improve optical assembly, new algorithms, better stability and matching across tool sets.
Processes are becoming more difficult to align to and measure. Resolution enhancement techniques such as PSM and OPC masks may further drive up errors in an already dwindling budget, especially when used in mix-and-match schemes.
