Method of Analyzing Metrology Data

Abstract

The preferred embodiments are directed to a metrology method used, for example, in recess analysis in semiconductor fabrication that includes using atomic force microscopy (AFM) data of a sample having an array of 2D-periodic features to generate a sample image, and calculating a periodicity of the features. The method identifies the peaks in the periodicity to determine a feature period and a lattice angle, and constructs a lattice mask that is registered to the image to perform an alignment calculation. The mask is offset, and alignment calculation made, to optimize cost.

Claims

1. A metrology method comprising the steps of: using atomic force microscopy (AFM) data of a sample having an array of periodic features to generate a sample image having feature pixels and background pixels; calculating a periodicity of the features; identifying peaks in the periodicity to determine a feature period and a lattice angle; constructing a lattice mask template using the feature period and the lattice angle; overlaying the image with the lattice mask template; performing an alignment calculation to determine a cost; applying an offset of the lattice mask template to the image and recalculating the cost; and repeating the applying and the recalculating steps to determine an alignment between the lattice mask template and the image.

2. The method of claim 1, wherein the performing step includes at least one of a) calculating a standard deviation of the background pixels and setting the standard deviation as the cost value, and b) calculating a median of the background pixels and the feature pixels, and setting the median as a cost value.

3. The method of claim 2, further comprising determining the offset of the lattice that establishes a minimum cost value if the standard deviation is calculated, and the offset of the lattice that establishes a maximum cost value if the median is calculated.

4. The method of claim 1, further comprising extracting data with respect to the features after applying the alignment.

5. The method of claim 4, wherein the data corresponds to at least one of feature characteristic including height, depth, shape, uniformity, variance and slope.

6. The method of claim 5, further comprising comparing the at least one feature characteristic to a known model to determine feature quality.

7. The method of claim 6, wherein the comparing step is used in semi conducting fabrication recess analysis.

8. The method of claim 1, wherein the features are 2D-periodic features and identifying peaks in the periodicity step begins at a center of the sample image and continues radially outwardly.

9. The method of claim 8, further comprising: iterating over 2D model types including at least two of square, rectangular, hexagonal, and oblique; and selecting the periodicity of the lattice type that produces the smallest deviation between the model lattice type and the acquired data.

10. The method of claim 1, wherein the calculating the periodicity step is performed using a Fast Fourier Transform (FFT) algorithm.

11. The method of claim 1, wherein the lattice mask template is hexagonal.

12. The method of claim 1, further comprising applying an adaptive flattening algorithm to the sample image.

13. A metrology method comprising the steps of: generating an image of a sample using atomic force microscopy (AFM) data; calculating a periodicity of features of the image; searching for at least one peak in the periodicity; obtaining a feature period and a lattice angle; constructing a lattice mask template using the feature period and the lattice angle; overlaying the image with the lattice mask template; performing an alignment calculation to determine a cost; applying an offset of the lattice mask template to the image and recalculating the cost; and repeating the applying and the recalculating steps to determine an alignment between the lattice mask template and the image.

14. The metrology method of claim 13, wherein the cost is calculated over an entire area of one unit cell.

15. The metrology method of claim 13, further comprising a step of downsampling the image for faster calculation of the cost.

16. The metrology method of claim 13, wherein the searching for at least one peak in periodicity step begins from a center of the image and continues radially outwardly.

17. The metrology method of claim 13, wherein the calculating periodicity step is accomplished by using Fast Fourier Transform (FFT) algorithm.

18. An AFM for collecting data of a sample AFM comprising: a probe that interacts with a surface of the sample; a controller that controls the probe-sample interaction and collect atomic force microscopy (AFM) data of a sample having an array of periodic features; and wherein the controller: uses the AFM data to generate a sample image having feature pixels and background pixels; calculates a periodicity of the features; identifies peaks in the periodicity to determine a feature period and a lattice angle; constructs a lattice mask template using the feature period and the lattice angle; overlays the image with the lattice mask template; performs an alignment calculation to determine a cost; applying an offset of the lattice mask template to the image and recalculating the cost; and repeats the applying and the recalculating steps to determine an alignment between the lattice mask template and the image.

19. The AFM of claim 18, wherein the controller performs the alignment step by at least one of a) calculating a standard deviation of the background pixels and setting the standard deviation as the cost value, and b) calculating a median of the background pixels and the feature pixels, and setting the median as a cost value.

20. The AFM of claim 19, wherein the controller further determines the offset of the lattice that establishes a minimum cost value if the standard deviation is calculated, and determines the offset of the lattice that establishes a maximum cost value if the median is calculated.

Description

BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Preferred exemplary embodiments of the invention are illustrated in the accompanying drawings in which like reference numerals represent like parts throughout, and in which:

[0028] FIG. 1 is an image of the scanning probe microscope system;

[0029] FIG. 2A is an image of raw AFM data;

[0030] FIG. 2B is a schematic drawing showing the step of detecting periodicity in the raw AFM data;

[0031] FIG. 2C is a schematic drawing showing the step of analyzing rings of periodicity in the AFM data;

[0032] FIG. 2D is a schematic drawing showing the step of generating a lattice from the AFM data;

[0033] FIG. 3 is a flow diagram of the present metrology method of the preferred embodiments;

[0034] FIG. 4A is an image of raw AFM data;

[0035] FIG. 4B is an image of the raw AFM data of FIG. 4A, after the application of an adaptive flatten to remove image tilt;

[0036] FIG. 4C is an image of a periodicity map acquired with Fast Fourier Transform autocorrelation of the flattened AFM data of FIG. 4B;

[0037] FIG. 4D is an image illustrating the peaks in periodicity of FIG. 4C;

[0038] FIG. 4E is an image of a lattice mask template; and

[0039] FIG. 5 is sketch of a hexagonal lattice showing the features of interest and differences in height between them.

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENT

[0040] The preferred embodiments are directed to a metrology method for analyzing spatial and topographical data of 2D elements/features in a lattice from raw atomic force microscopy (AFM) data. The methods described herein combine a lattice detection algorithm with a novel lattice alignment technique. After applying both detection and alignment steps, the found lattice is used to analyze each feature's localized depth, variance, slope and more. This invention helps to satisfy the need for users to check the quality of their samples quickly and with minimal user intervention.

[0041] Turning first to FIG. 1, a scanning probe microscope instrument 150 (e.g., AFM) according to a preferred embodiment is shown. In this embodiment, a probe 152, having a tip 154 extending from a distal end of a cantilever 155 is held by a probe holder (not shown) supported by piezoelectric tube scanner 156. Scanner 156 may be a “Z” or vertical scanner responsive to sample properties in the closed loop control system to position the tip 154 relative to a sample 158 during AFM imaging. Tube scanner 156 is coupled to an XY scanner 160, preferably also a piezoelectric tube, that is used to raster the probe tip 154 relative to the sample 158 surface during AFM operation. Notably, a scanned sample may be employed as an alternative. A mechanical Z-stage 162 is included for providing large movement in Z between tip 154 and sample 158, for example, during AFM image acquisition start-up to engage tip 154 and sample 158.

[0042] Sample 158 is mounted on an XY stage 164 that primarily provides coarse XY motion to position probe 152 at a region of interest of sample 158. An XY stage controller 166 controls stage 164 to locate the probe/sample at that region of interest. Again, however, stage 164 may be configured to provide relative scanning motion (e.g., raster) between tip 154 and sample 158 at a selected scan speed. Controller 166 is also responsive to AFM controller 174 to position the image scan at a region of interest. Controllers 166, 174 are implemented by a computer 180.

[0043] In operation, after tip 154 is engaged with sample 158, a high speed scan of the sample is initiated with XY scanner 160 in an AFM mode of operation (e.g., PFT mode), as discussed previously. As tip 154 interacts with the surface of sample 158, the probe 152 deflects and this deflection is measured by an optical beam-bounce deflection detection apparatus 168. Apparatus 168 includes a laser 170 that directs a beam “L” off the backside of cantilever 155 and toward a photodetector 172 which transmits the deflection signal to, for example, a DSP 176 of AFM controller 174 for high speed processing of the deflection signal.

[0044] AFM controller 174 continuously determines a control signal according to the AFM

[0045] operating mode, and transmits that signal to the piezo tube scanner 156 to maintain the Z position of probe 152 relative to sample 158, and more specifically, to maintain deflection of the probe at the feedback set point.

[0046] Turning to FIGS. 2A-2D, a series of schematic drawings depicting the progression

[0047] of analysis of the AFM data according to the present metrology method are shown. In FIG. 2A the raw AFM data is shown with image 200. This raw AFM data is generated by the above described scanning probe microscope instrument and methodology. The data consists of features 202 and background 204. Next, FIG. 2B is a schematic representation 206 of detecting the periodicity in the raw AFM data. This preferably is achieved by Fast Fourier Transform (FFT) autocorrelation to find the peaks and periodicity in the raw AFM data. The image is correlated with itself, and the peaks represent points 208 where the image is symmetrical with itself. Further, in FIG. 2C, a schematic drawing of a ring of periodicity 210 found in the data is shown. Each point 208 in FIG. 2C represents a peak in periodicity. The locations of these peaks relative to the center and relative to each other in terms of distance and 2D-angle is quantified and used to construct different lattices (described in more detail below). FIG. 2D is a schematic 212 representation of a lattice that may be generated for use as a mask in extracting sample information, in regions of features 202 adjacent regions of background 204.

[0048] Now turning to FIG. 3, a simplified diagram of the present metrology method 300 is shown. At Step 302, the raw AFM surface data is collected from the sample. Next, at Step 304, the periodicity of the image is calculated using Fast Fourier Transform (FFT). At Step 306, rings in periodicity (see FIG. 2C) are found by searching radially outward from the center of the image. In the case of a hexagonal lattice, collections of four may be used in finding the rings. When using different lattices (rectangular, triangular, octagonal, etc.), other collections may be sought. These rings in periodicity provide information about the location of the peaks relative to the center in terms of distance and also provide information about the distance and angle of the peaks relative to each other. Again, multiple rings in periodicity are found moving radially outward from the center. Then, at Step 308 the circular shells of peaks in periodicity are quantified and possible lattice periods and angles are obtained. At Step 310, the image may be down sampled for faster cost calculation. Then, at Step 312, a lattice mask is constructed using the previously acquired lattice periods and angles. Several of these lattices, like that shown in FIG. 2D, are generated.

[0049] At Step 314, the lattice mask is overlaid on top of the image, allowing the algorithm to distinguish feature pixels from background pixels. Here the mask matrix is added/multiplied with the image matrix to extract feature pixels. The mask 212 (FIG. 2D) breaks up the image into black area and white area pixels. The white area pixels represent the sample features, while the black area pixels represent the background. Then, at Step 316, an alignment calculation user input parameter is applied. The user may choose between using a standard deviation calculation or a median calculation. Standard deviation is likely to be applied when the background is rough, while median is likely to be applied when the background is smooth.

[0050] Depending on which parameter the user chooses, the next step may differ. If the user chooses standard deviation, the standard deviation of the background pixels (black region) is calculated and that standard deviation value is set as the cost at Step 318. Then, at step 322, an offset of the lattice mask overlay is applied, and the cost is recalculated. The cost is calculated at each offset in preferably, a 1.2 period range to cover all alignment options. This is an exhaustive search over the area of one unit cell so that all possible offsets are tested. Finally, at step 324, the offset that gives minimum cost is found and set as the final lattice alignment.

[0051] The method varies if the user chooses the median as the input parameter. In this case, the next step following step 316 is step 320, in which the difference in median between the background pixels (black region) and the feature pixels (white region) in FIG. 2D is calculated. The difference between the median of the background pixels 204 and the feature pixels 202 is calculated and set as the cost. Next, at Step 322, an offset of the lattice mask overlay is applied, and the cost is recalculated, as in the standard deviation case. The cost is calculated at each offset in a 1.2 period range to cover all alignment options. This is an exhaustive search over the area of one unit cell so that all possible offsets are tested. Finally, at Step 326, the offset that gives the maximum cost value is found and set as the final lattice alignment.

[0052] Once the cost is properly calculated, the final lattice alignment is determined, and the design of features is established. For example, if it is established that the features are a series of concentric rectangles, pixels can be extracted from the AFM image corresponding to every area of the rectangle and specific pixels can be analyzed corresponding to specific parts of the features.

[0053] Note that when the 2D lattice type is unknown, one can iterate over all possible lattice types in 2D: square, rectangular, hexagonal, or oblique (see https://mwikipediamrglwikilBravais lattice), and select periodicity of the lattice type which results in the smallest deviation between the model lattice and the acquired data. Here, the smallest deviation corresponds to the best cost of alignment.

[0054] Turning now to FIGS. 4A-4E, a series of images showing the progression of

[0055] analysis of the AFM data according to the present metrology method is shown. In FIG. 4A the raw AFM data is shown. The image 400 shown in FIG. 4A would be generated at step 302 of the above described metrology method. The data consists of features 402 and background 404. Defects 406 may also be present in the sample and resulting data. Next, FIG. 4B is an image of the AFM data after application of an “adaptive flatten” to remove image tilt. FIG. 4C is an image showing the periodicity map acquired with Fast Fourier Transform autocorrelation. The points 408 in the image represent the peaks in the periodicity.

[0056] FIG. 4C corresponds to step 304 of the above described metrology method. FIG. 4D is an image showing a ring of periodicity 410. FIG. 4D corresponds to Step 306 from the above described metrology method. Further yet, FIG. 4E is an image showing a lattice mask template 412 generated using the distribution of the peaks. This mask is generated at Step 312 above. This lattice mask template 412 is then overlaid on top of the AFM image. In doing so, the mask matrix is multiplied with the image matrix so only certain pixels are analyzed. This corresponds to Step 314 described above.

[0057] Turning to FIG. 5, a sketch of a hexagonal lattice 500 showing the features of

[0058] interest and differences in height between them is shown. Rather than a square lattice, a hexagonal lattice is shown here. The dark squares 502 represent the location of design features of interest 502 (e.g., 402 of FIG. 4A) as found by the present method. Once the features of interest 502 are identified, they can be analyzed to quantify the distribution of height between the features 502. The shading 504 over the features 502 represents differences in height between the features 502. FIG. 5 is a sketch of the results that would be acquired at step 324 or 326 above. The center of FIG. 5, where the squares have no shading, indicates that the features in this portion of the sample may be missing or are barely printed on the wafer. When the wafer is printed, it is always periodic, so the features 502 in theory should exist at every square in the image. This information regarding each feature's 502 localized depth, variance, slope height and more is of great importance to the quality of the sample and its functionality.

[0059] The preferred embodiments are particularly useful in semiconductor manufacturing.

[0060] Recess analysis, for example, enables critical metrology for IC manufacturing processes in which two semiconductor wafers with patterned surfaces are bonded together. This wafer-to-wafer bonding requires highly accurate topographical knowledge of the post polished (CMP) wafer surfaces that consist of metal pads surrounded by dielectric material. The effectiveness of the bonding requires a very flat surface. Recess analysis calculates the height difference, known as dishing, of the metal pads with respect to the surrounding dielectric, the local slopes of the dielectric material in proximity of the metal pads, as well as global planarity over the entire field of view.

[0061] The output of the recess analysis permits the IC manufacturer to make critical process decisions based on the percent of out of specification roughness and slope regions.

[0062] Although the best mode contemplated by the inventors of carrying out the present invention is disclosed above, practice of the above invention is not limited thereto. It will be manifest that various additions, modifications and rearrangements of the features of the present invention may be made without deviating from the spirit and the scope of the underlying inventive concept.