Alignment Optimization for Critical DUV Lithography

Requirements for photolithography imaging continue to increase as chip feature sizes decrease, and the reduction of critical dimensions (CDs) creates more stringent overlay requirements. These requirements have traditionally been met with improvements in stage technology and alignment systems, but development of these technologies is unable to keep up with continued demands. Therefore, new methods of controlling overlay and alignment need to be explored.

Valuable information about the quality of alignment can be obtained from the exposure tool. During wafer alignment, the positions of multiple points on the wafer are measured. Based on these measurements, an alignment model is calculated and used by the tool to step out the exposure pattern. This model can then be used to calculate the residuals from the alignment portion of the exposure process, giving an indication of the alignment quality. To meet tighter overlay tolerances, more attention needs to be placed on these residuals. The residual analysis can serve as the basis for choosing the exposure alignment strategies.

Typical wafer alignment schemes measure only grid parameters, so overall exposure shot placement is corrected at the time of exposure. However, the field alignment parameters, specifying how well each exposure shot is aligned with respect to the reticle plane, are typically not measured. The lack of shot-to-reticle alignment results can drive large overlay excursions. Understanding the influences of both the grid and the shot errors on alignment is required to extract the corresponding, acceptable residual overlay errors.

This work examines the alignment residual data in relationship to the maximum-minimum measured overlay values with linear factors corrected. Within the scope of this work, the relationships of maximum and minimum overlay vectors are examined instead of a 3 ó value characterizing the overlay (the maximum vector is very close to a 3 ó number, so the effective analysis would not be very different if a 3 ó metric had been considered). Such evaluation can lead to the understanding of what levels of alignment residuals are required to meet corresponding overlay budgets. It can be used to choose the alignment strategy or as a filter of real-time production alignment data.

The work presented here gives the process engineer alignment performance targets based on tool specifications; helps to choose the proper mark for each process layer; and allows exploration of future IC design rules, leading to the estimates of possible new overlay targets. Using real-time alignment data filtering, the engineer can set tool control limits to ensure that every wafer processed yields the overlay requirements. This becomes more critical in today's manufacturing when the practices of reduced sampling and skip-lot processing are adopted.

This work examines the relationship of alignment residuals and field parameter correction control to the overlay performance. It reveals what levels of alignment residuals and field parameter correction are required to meet the ever-tightening overlay requirements.

A Nikon scanner system was used for all exposures analyzed here. The alignment sensors used were LSA, a diffraction-based system using a HeNe laser; and FIA, a broadband, CCD camera-based system. Both of these alignment systems require specialized targets (marks) on the wafer. Each sensor offers a choice of alignment marks designed to optimize the overlay control of different levels (process layers).

For this work, a single process layer was chosen, including multiple alignment marks resulting in different alignment residual levels. Each analyzed process lot was screened prior to its exposure and four different alignment schemes were chosen. A single lot was split into different groups selected for different mark/sensor combinations.

When a lithography tool aligns a wafer, it finds the user-definable locations where the alignment targets are located. Nikon's wafer alignment system, called EGA (Enhanced Global Alignment), reads the mark location, and compares that location to the ideal location known from the reticle design and the exposure layout. This is done for several sites on the wafer. A typical alignment layout includes six to eight sites located at about 70% of the wafer radius in a circular fashion. From this data set, a linear model can be determined specifying the wafer grid and/or shot terms. The linear model quantifies the scaling, rotation and orthogonality of the wafer grid and the exposure shot terms. These model calculations quantify the alignment corrections for scaling, rotation and orthogonality of the tool. The linear model can then be subtracted from the measured alignment data resulting in alignment residuals, or non-linear errors (NLEs). The NLE is used here as a metric of the quality of the level alignment.

For this work, the wafer was aligned based on nine global alignment locations spread in a circular pattern on the wafer. For each alignment test run, the exposure tool software calculated the residual maximum vectors. The number and location of the alignment sites were not critical to the overall alignment analysis.

The same exposure tool was used for two alignment and exposure runs. The first exposure was to lay out exposure patterns and alignment marks. The overlay of the first and the second exposures was monitored by the alignment model developed by analyzing the location of the marks found during the wafer alignment just prior to the second exposure.

The alignment measurements were done on a typical IC production-type layout. The measurements were done on the four corners of the exposure field and on four exposure shots at the outer edge of the wafer. The overlay data were collected on the metrology tool and analyzed using Mono-lith alignment analysis software.

This work analyzes the relationship between the corrected or uncorrected overlay ranges and the alignment residuals in the form of the maximum non-linear vector value, NLE. The motivation for this approach is to track the overlay tradeoffs based on the performance of an easily available metric such as maximum NLE.

The resulting data set included raw data, i.e. the measured data and the resulting linear correctable terms. The data are presented in a format in which the maximum vector minus the minimum vector is considered to be the overlay range. The data set also included the alignment residuals in the form of the maximum NLE.

The raw data were used to determine the alignment model of the tool. The alignment model used in analysis of the overlay data was the standard Nikon scanner alignment model. It consisted of grid scaling, orthogonality and rotation terms; and shot scaling, orthogonality and rotation terms.

All these linear terms are considered correctable, and for each group representing different mark/sensor combinations a lot-average model was calculated by Mono-lith software. This averaged model was subtracted from the raw data to produce a set of maximum vectors representing lot-corrected data. Also, for some of the wafers measured, their correctable terms were driven to zero to see what the underlying spread was between vector minimum and maximum. This was needed because the lot-corrected data groups still have some small level of linear errors. Lot correction was done to simulate the production conditions.

The data set was analyzed to understand the overlay drivers. Because the same tool was used for the first and second prints, lens distortion — one of the potential drivers — is considered negligible.

The analysis focused on the correlation of non-linear alignment error to the resulting overlay linear and random errors. As mentioned above, the linear errors such as grid and shot scaling and orthogonality are considered correctable. The random errors are the residuals of the measured data after the linear terms are subtracted.

In production, the maximum vector overlay performance can be broken down into two parameters: 1) the average (mean) of the wafer overlay, and 2) the deviation about this mean. Here it is assumed that the required level of mean control is achievable, so the range of overlay vectors becomes the critical driver of the overlay performance.

This is based on the assertion that the mean overlay control is driven by the tool's alignment baseline control and the linear offsets based on the alignment model of the tool. In this sense, the alignment mean is not a direct result of alignment quality. The mean variation needs to be directly subtracted from the overall overlay budget. The level of mean control can be derived from the exposure tool specification and historical data. The expected mean control is subtracted to give the desired overlay range for a given overlay specification.

The range of the variation around the alignment mean represented the quality of the alignment for each mark/sensor combination. Plots of the resulting overlay ranges vs. the NLE for the raw and lot-corrected data show what the drivers for this overlay range are.

1. The results of overlay range vs. non-linear X maximum vector for the uncorrected and lot-corrected overlay. The line represents a second-order fit serving as a general fit of the uncorrected data.

2. The results of overlay range vs. non-linear Y maximum vector for the uncorrected and lot-corrected overlay. The line represents a second-order fit to serve as a general fit of the uncorrected data.

The linear error is comprised of both the wafer (grid) and the shot terms. The analysis of the relationship between the linear error and the overlay range is based on RMS estimates for grid and shot components. The RMS of these terms were combined for rotational terms (rotation and orthogonality) and scaling terms (X and Y scaling), then compared with the overlay range (Figs. 3 and 4).

Figures 3 and 4 show that the grid term RMS greatly outweighs the shot term RMS in determining the overlay range. This is expected given that the scale of the shot vectors is limited to 10 to 15 mm. This range is small compared with the grid overlay terms, which can be offset from the wafer center by as much as 100 mm. As long as there are no intentionally induced shot errors, the primary driver of overlay range comes from the grid. Note also that, in standard lithography configuration, the alignment results for grid measurement are used by the tool as the shot placement corrections. This underscores the need for grid control over shot control because the grid correction feeds into the shot correction.

3. The results of overlay range vs. RMS for the rotational terms (orthogonality and rotation) for the grid and shot. The solid and dashed lines show the grid term X overlay range and Y overlay range second-order fit.

4. The results of overlay range vs. RMS for the scaling terms (X and Y) for the grid and shot. The solid and dashed lines show the grid term X overlay range and Y overlay range second-order fit.

Figures 5 and 6 show the range of individual grid term errors vs. the average non-linear maximum vector for a given group of wafers in the experiment for the X and Y axes. The overlay vectors are statistical, so there is no expected relationship to any particular overlay component (rotational or scaling). In the figures, grid term range refers to the range of values of the correctable grid terms across the lot and the average NLE max vector is the lot average, based on the alignment non-linear maximum vectors.

5. A plot of the range of grid term error vs. the NLE average maximum vector for the X axis. The fitted lines represent a second-order fit.

The data in Figures 5 and 6 show the relationship of the range of the grid correctable terms to the alignment NLE maximum vector of an exposure lot. This data display the correlation of the grid terms and the maximum vector of alignment residuals.

6. A plot of the range of grid term error vs. the NLE average maximum vector for the Y axis. The fitted lines represent a second-order fit.

Figures 3 and 4 show a relationship of linear term RMS to overlay range. These figures can be used to assign an allowable limit of variation of the linear terms. Given the required overlay range, we can define a maximum non-linear alignment target, assuming that the linear terms are correctable. To meet the manufacturing requirements, the range of these linear terms should be about half the worst-case term limit, reducing wafer-to-wafer variability and allowing better control of the correctable terms and the overall overlay. If these limits are adopted, a correction from a send-ahead wafer is likely to reflect the lot, and the overall lot performance should meet the desired overlay performance targets.

Based on overlay specifications, a likely overlay range can be estimated. The overlay range leads to limits on linear correctable error that can be allowed on the wafer. To meet the limits of correctable error, the variations of the grid and shot parameters need to be controlled by setting an appropriate NLE limit for the process. The limits can be assured by identifying the required alignment mark/sensor combinations.

There are many factors influencing the tool alignment and overlay NLE, making any comprehensive model truly complex. In particular, there are many possible NLE sources between a first and second print. The overlay error at any given point on a wafer comes from the wafer stage three times (first and second print, and alignment), from the scan synchronization accuracy twice (first and second print), from the alignment microscope, true wafer non-linear expansion, the overlay metrology tool and the reticle fabrication. A full model showing the relationship of alignment NLE to the grid and shot term ranges would require a large amount of test data. Instead of the comprehensive alignment analysis, engineering approximations were made and presented here.

The overlay sampling here involves the entire wafer and the results are representative of alignment and overlay of the actual product. In that sense, the engineering approximations and manufacturing limits leading to alignment residual limits ensure compliance with manufacturing specifications of a sampled product. The actual values of overlay shown are based on the sampling plan used to control the overlay in production. In this case, the outermost exposure fields were used to give worst-case scenarios, but the ultimate results are a function of where on the wafer the overlay is measured. Also, the alignment NLE is a function of where the wafer is sampled. Given all this, the absolute overlay values given here do not apply in all cases, but the methodology is based on aggressive overlay targets. The trends shown apply in general.

Although this analysis did not focus on the fundamental overlay limit, the lower limit to overlay performance of the current generation of tools could be estimated to have an overlay range of ~50 nm. This figure was extracted from the overlay data when the correctable terms were set to zero, and represents the fundamental performance of the stage and the alignment system and the overlay measurement accuracy. This estimate of the overlay lower limit agrees well with the expected value known from independent analysis of overlay capabilities of current tools.

It is becoming a common practice in fabs to reduce the overall metrology for the photolithography sector. One way to do this is to measure fewer wafers per lot or to reduce the sample plan. Another option is to measure overlay on some lots and skip the others. At the same time, the overlay specifications for these wafers have become more stringent, posing new challenges.

Understanding the expected variation of overlay terms within the lots could lead to greater confidence that every wafer within a lot meets the specification. This confidence can be increased even further with software measuring and analyzing in situ the alignment residual for each wafer. In situ overlay monitoring could prevent exposure of wafers having alignment residuals greater than the limits set by the process engineer.

The information obtained from the relationship between non-linear alignment errors and the resulting overlay performance can be used to select the proper alignment strategy for a given overlay specification, and to ensure process control under a minimal metrology plan. The methodology provides a practical path to meet future overlay targets. The analysis of alignment residuals can be used to select the proper wafer alignment marks, and can serve as a guide in selecting tool limits on real-time NLE performance.

This paper was originally presented at the 1999 IEEE/SEMI Advanced Semiconductor Manufacturing Conference and Workshop (ASMC), Boston.