JPH0291502A - Method for detecting position of alingment mark - Google Patents

Abstract

PURPOSE:To keep processing from being complicated by forming a wafer mark and a mask mark in such mark shapes having similarity that a wafer mark derivation signal obtained by taking a first derivative to the wafer mark and a mask mark derivation signal obtained by taking a first derivative to the mask mark are nearly overlapped with each other by horizontally moving them. CONSTITUTION:The wafer mark and the mask mark are formed in such mark shapes having the similarity that the wafer mark derivation signal obtained by taking the first derivative to the wafer mark and the mask mark derivation signal obtained by taking the first derivative to the mask mark are overlapped by being horizontally moved. Then, similar pattern matching is performed to a signal obtained by the derivation of an image input signal. Thus, the relative position between the wafer mark and the mask mark can be recognized with high accuracy with simple processing. Moreover, throughput can be improved, a producing process can be simplified and producing efficiency can be enhanced.

Description

[Detailed Description of the Invention] [Industrial Application Field] The present invention relates to a mask formed on a mask in a proximity aligner or stepper used in an exposure device such as an X-ray exposure device that prints a circuit pattern on a wafer. The present invention relates to a method of detecting an alignment mark consisting of a mark and a wafer mark formed on a wafer, and determining the relative position between a mask and a wafer with high precision from the detected waveform.

[Prior Art] Generally, two types of methods are used to accurately determine the position of an alignment mark from a detected waveform: a diffraction grating method using Fresnel zones, etc., and a pattern measurement method. Among these methods, the diffraction grating method has the disadvantage that detection accuracy is greatly affected by variations in the gap between the mask and the wafer, and alignment marks with steps cannot be detected with high accuracy. On the other hand, the accuracy of the pattern measurement method depends on the shape of the pattern of the alignment mark to be detected, so there is a drawback that the detection accuracy deteriorates due to deterioration of the pattern shape of the alignment mark.

Here, an example in which the diffraction grating method depends on the symmetry of alignment marks is
IEEE TRANSACTION ON ELECTRON DEVICES
, ECTR0N DEVICE) (7) Paper journal (
7) VOL, HD-26.

No. 4 "Automatic Alignment System for Optical Projection Printing (AutomaLIc AllgrvenL 5y
ste+g f'orOpLIeal ProJecL
It has been proposed by GIJs Bouvhuls et al. in a paper entitled lon Printing) J. Furthermore, an example in which the pattern measurement method depends on the symmetry of alignment marks is disclosed in Japanese Patent Application Laid-Open No. 62-5491.
This method is disclosed in Japanese Patent Application No. 8 and Japanese Patent Application Laid-open No. 199833/1983.

Pattern measurement methods can be further broadly classified into two types: contrast methods and edge detection methods.

Among these, the conventional example of the contrast method is JP-A-61-23
It is disclosed in Japanese Patent No. 6117. This JP-A-61-2
In the publication No. 38117, a symmetrical pattern matching process expressed by the following equation is performed.

Or, below, D (n) is a differential value, M is a comparison range, and n is a position.

The above formula is an application of pattern matching processing in ordinary image processing, and this Japanese Patent Application Laid-Open No. 61-238117 utilizes the line symmetry of the alignment mark pattern to apply this to the above formula. It is incorporated into That is,
In order to find the center line of a symmetrical figure, it is equivalent to folding the figure in half and then finding the center line.

The area where the image is folded in half is, of course, an edge, but this uses an edge processed by an edge extraction method using differentiation, which is a common image processing method.

In this way, Japanese Patent Application Laid-Open No. 61-238117 takes advantage of the symmetry of a figure and can determine its center line using a relatively simple arithmetic expression. By using this symmetric pattern matching process to detect alignment marks on a mask or wafer with a proximity aligner or a stepper, the relative positional relationship can be determined from the center line of each alignment mark.

[Problems to be Solved by the Invention] However, the method using the symmetric pattern matching process described in Japanese Patent Application Laid-Open No. 61-236117 has the following drawbacks.

That is, this method assumes that the figure is symmetrical. Ideally, mask marks (chrome formed marks) on the mask and wafer marks (resist formed marks and process formed marks) on the wafer are line symmetrical, but in reality, this symmetry can be disrupted due to various factors. .

For example, in the case of a resist formation mark, the symmetry of the alignment mark is disrupted depending on the irradiation angle of X-rays or the like when exposing the alignment mark, the resist development speed, etc. (see FIG. 2). In addition, in the case of process-formed marks, the symmetry of the alignment mark is easily destroyed, and the alignment mark is easily deformed due to thermal deformation (chemical deformation, deformation due to force) due to the process, or due to the thin film coated on the alignment mark. The symmetry of the alignment mark will be lost (see Figure 3).

As is clear from the above description, it is extremely difficult to guarantee the symmetry of wafer marks on actual wafers in all layers. Therefore, using alignment mark data that is not symmetrical for a computation that assumes symmetry as a precondition causes a contradiction, and the computation result becomes erroneous. In other words, the more the symmetry of the alignment mark breaks down, the larger the error included in the calculation result becomes.

An example of countermeasures to prevent the accuracy from decreasing due to the asymmetry of the alignment mark is the above-mentioned Japanese Patent Application Laid-Open No. 82-5491.
This method is disclosed in Japanese Patent Application Publication No. 81-199833. In Japanese Patent Application Laid-Open No. 62-54918, the film thickness distribution of photoresist coated on a wafer is measured, the wafer alignment pattern position detection error is determined from the measured film thickness distribution, and the alignment amount is determined based on the position detection error. is being corrected. However, this method has the disadvantage that processing becomes complicated. Also, JP-A-81-
In Japanese Patent No. 199633, the shape of the alignment mark itself is made so that position detection errors are less likely to occur. However, the alignment mark disclosed in JP-A-61-199 [133] has a drawback that its shape is very complicated.

[Means for Solving the Problems] A method for detecting the position of an alignment mark according to the present invention is obtained by imaging an alignment mark consisting of a wafer mark formed on a wafer and a mask mark formed on a mask using an imaging device. In the method of detecting the relative position of the wafer and the mask by converting an analog image signal into a digital image signal and processing the converted digital image signal, the wafer mark and the mask mark are The wafer mark differential signal obtained by firstly differentiating the mark and the mask mark differential signal obtained by firstly differentiating the mask mark have mark shapes that maintain similarity so that they almost overlap each other when horizontally moved, and the digital image signal processing differentiates the digital image signal, and for this differentiated signal, (where V'(j) is the differential value of the digital image signal,
R is the correlation interval, P is the start point of the correlation interval, and W is the distance between the center positions of the mask marks and the center positions of the wafer marks that are adjacent to each other. The method is characterized in that the relative position of the wafer and the mask is detected from the position where the processing result is maximum.

[Examples] Examples of the present invention will be described below with reference to the drawings.

First, the principle of the present invention will be described.

The present invention has a feature in which a part of the pattern matching method commonly applied in the field of pattern measurement is modified. The process flow is to extract the edges of a pattern by applying a differential operator to the input image, and then perform similarity pattern matching on the edge-extracted data. Note that the definition of similarity pattern matching and its method will become clearer as the following explanation progresses. Similarity pattern matching uses a partially modified formula of a well-known autocorrelation function.

In normal pattern matching or waveform data processing, the autocorrelation function is defined by the following equation.

Here, i-0,1,. . . , N-1, X(1) is a finite number of N sample values, and RX(1) is an autocorrelation function.

By applying the above equation to data or signal waveforms that are targeted for matching, it is possible to obtain correlation results for patterns or signal waveforms that serve as a certain reference. As is clear from this equation, this calculation requires a considerable number of product-sum calculations, and thus takes p time for calculation. Therefore, although the autocorrelation function is suitable for batch processing, it is not suitable for real-time processing.

For example, when using the five equations for alignment processing of an X-ray exposure device, etc., the long calculation time means that the control system, including the X/Y stage, uses the alignment processing results as a feedback signal. is also a major obstacle. Furthermore, the long computation time is the biggest cause of lowering the throughput of the entire device, which can lead to a fatal defect.

Therefore, in the present invention, after devising the shape of the alignment mark, the autocorrelation function is The equation was modified as follows.

(1) Improving the shape of the alignment mark An alignment mark shape that allows the superimposition method using a general autocorrelation function to be applied to the edge-extracted data signal by applying a first-order differential operator to the input image of the alignment mark. And so.

That is, the wafer mark formed on the wafer and the mask mark formed on the mask are shifted in parallel by a wafer mark differential signal obtained by firstly differentiating the wafer mark and a mask mark differential signal obtained by firstly differentiating the mask mark. For example, the marks are shaped so that they overlap each other and maintain similarity.

For example, in the case of a pattern as shown in FIG. 4, a superimposition method using an autocorrelation function is possible. In FIG. 4, (a) is a top view, (b) is a side view, (C) is a raster compressed image v1, and (d) is a raster compressed image 1 missing differential value V'. As is clear from FIG. 4(d), the first-order differential value V' of the raster compressed image V
If you translate it left and right, it will almost overlap itself at positions 1 and L2. Therefore, (L2+L1)
/2 allows the relative position of the mask and wafer to be measured.

Another example is shown in FIG. In FIG. 5, (a) is a top view, (b) is a side view, (C) is a raster compressed image (2), and (d) is a first differential value V' of the raster compressed image V.

In the example of Fig. 5, as in Fig. 4, (L2+L1)
/2 allows the relative position of the mask and wafer to be measured.

In the examples shown in FIGS. 4 and 5, if the wafer mark and mask mark are moved in parallel, the wafer mark differential signal obtained by first-order differentiation of the wafer mark and the mask mark differential signal obtained by first-order differentiation of the mask mark are moved in parallel. Although the mark shapes almost overlap each other and maintain similarity, the wafer mark and mask mark are divided into a wafer mark differential signal obtained by firstly differentiating the wafer mark, and a mask mark differential inversion signal obtained by firstly differentiating the mask mark and inverting the polarity. However, if the marks are moved in parallel, the marks may have a shape that maintains similarity so that they overlap each other.

(2) Transformation of autocorrelation function formula In the present invention, in order to transform the autocorrelation function formula, we focus on the features of alignment marks used in exposure equipment.
The signal (edge pattern) obtained by applying a first-order differential operator to the analog image signal (human image) obtained by imaging the alignment mark with an imaging device differs from the normal signal waveform in the following three basic ways: transform as.

[1] The distance #lW between two edge patterns for pattern matching is basically a constant value, and the variation in the distance W due to the process is extremely small.

[2] The range for pattern matching only needs to be set around the edge pattern.

[3] The position of the mask/wafer alignment mark at the detected coordinates (coordinates on the TV left camera) can be considered to be approximately at the fixed position due to coarse alignment in the alignment procedure.

Based on the features [1], [2], and [3], the process of transforming the autocorrelation function equation will be described next.

The autocorrelation function for discrete values is expressed by the following equation as described above.

However, they are i-0, 1, . . . , N-1.

Souichi below Here, to simplify the expansion, the normalization coefficient −N Omit.

The distance W is basically a constant value, and in equation (3), the width W corresponds to the delay time i.

Equation (2) is used for the data shown in FIG.

In this figure, the horizontal axis represents the pattern distance.

An image captured by a two-dimensional camera is used for inputting the image, and the discrete distance is expressed by the number of pixels (j) on the imaging tube surface of the two-dimensional camera. The vertical axis is the result of converting an analog image signal captured by a two-dimensional camera into a digital image signal, and applying a first-order differential operator to the digital image signal obtained by this conversion. ), it is expressed as v'. Therefore, in FIG. 6, equation (2) is replaced with the following equation.

From feature [1], wafer mark and mask mark (4)
The equation calculates the correlation between the distances AiW between the left and right marks in the entire region of j in FIG.

However, from feature [2], it is only necessary to set the pattern for correlation (superposition) in the area where the edge pattern exists, and as is clear from FIG. is finite.

Further, from feature [3], the position where the edge pattern exists is given P as the starting point of the correlation interval from rough position detection performed in advance.

In Fig. 6, if the range where the wafer mark exists is defined as a correlation interval R, and the starting point of the correlation interval is P, then the range in which the correlation is taken is from j-P to j-P+R in Fig. 6. Limited. This is shown in Figure 7.

From this, equation (4) is transformed as follows.

Note that equation (5) is defined as a function of the width W, but
From the feature [2] of the alignment mark, the correlation section R and the starting point P of the correlation section are also parameters of A.

Therefore, if the starting point P of the correlation interval R1 is also taken as a parameter, equation (5) is converted as follows.

From the above development, equation (2) of the autocorrelation function was finally obtained as equation (6) by considering the features [1], [2], and [3] of the alignment mark.

Equation (6) can be considered as a basic definition equation for similarity pattern matching, and the following six equations can be considered as extended equations.

In other words, A (W, P, R) does not have W, P, and R as parameters at the same time, but it uses one of the three parameters or a combination of them as a parameter. It is a formula.

However, as a result of the inventor's experiments, the function A (P,
It has been revealed that W, R) changes rapidly as W changes. In other words, for the remaining correlation starting point P and correlation interval R, it was found that A has almost no effect on the value of W, which takes the interpersonal value.

Regarding P, unless a very inappropriate value is assigned, A
The value of W, which takes the maximum value, does not change, and the correlation interval R should be a variable because the thickness of the edge changes depending on the type of process mark, but in reality it is considered to be a constant value. Almost irresistible.

By the way, according to experiments, R-32 (constant value) was used for all process marks, and the target detection accuracy of around 0.01 μm was achieved for all patterns. .

From the above, among the extended expressions (8-1) to (13-6), the expression (8-3) has a substantial meaning, and this is taken as the basic expression.

In addition, since R can be considered as a constant from the above (8-3
) can be partially transformed to obtain .

FIG. 8 shows the results obtained experimentally using equation (7).

In Figure 8, the horizontal axis is the width W1, and the depth axis is the correlation point P
, the vertical axis indicates the function A (P, W). From Figure 8, the peak position of the function A (P, W) is independent of P and has a width of W-2.
It can be seen that the value is constant at 04.

Also, in Fig. 8, the height of the peak of the function A (P, W) changes depending on the value of R, which affects the interpolation accuracy.
From the experimental results, it was confirmed that the detection of the order of 0.01 μι does not have much influence.

Next, the number of operations is compared between the autocorrelation functions (2) and (7).

Assume that the number of data is 1000. Autocorrelation function (2)
In the case of the formula, the number of additions is 999 times and the number of additions is 998 times for each data, so for all data,
The number of operations is (999+998)X100O-1997000 times. On the other hand, the number of operations of equation (7) is calculated based on the following assumption.

In equation (A), when considering a certain process-formed mark, the scale per pixel is 0.1 μm, the starting point P is 100, the correlation interval R-30 (3 μ!+), and the expected pattern width W WMIN-200 (20μIl
), WMAX -210 (21uta>),
Substitute into equation (7).

However, W −200,201,...,209,21
In the case of formula 0(8), the number of integrations is 30 times and the number of additions is 29 times for each W, so the total number of operations is (30+2
9) It becomes Xll-649 times. Therefore, the time is reduced to about 1/3077. In fact, this number of times can be calculated in less than 1ass even if a digital signal processing (DSP) filter is used.

Furthermore, for product-sum operations (MAC: mu), which has recently become popular,
If a chip (luLIplIer-accumulaLror) is used, even higher speeds are possible.

As explained above, by applying the autocorrelation function only to alignment marks, the number of calculations can be extremely reduced.

Furthermore, according to equation (7), the present invention does not assume pattern symmetry, but relies on the similarity between the mask mark and the wafer mark.

In other words, the method of the present invention is a pattern matching method that utilizes similarity regardless of the symmetry of the mask and wafer alignment marks, so that It's called ching.

In FIG. 4, the results of processing using equations (e-a) and (7) are shown in FIG. In FIG. 9, A (W,
Let W at the point where the first peak of P) is shown be Wl, and let W2 be the place where the second peak is shown. Since the true peak exists between Wl and W2, the value of W indicating the true maximum value is determined by interpolation and set as Wl-2. This results in
L 1- W +-2 is found.

Next, perform the same process for L2, and set the obtained W to W, -
4, L2-W, 4 can be found. From now on, the relative position of the mask mark and wafer mark is (L1+L2)/2
required from.

Next, examples of the present invention will be described.

In FIGS. 1 and 4, in this embodiment, if the wafer mark differential signal obtained by firstly differentiating the wafer mark, the wafer mark, and the mask mark differential signal obtained by firstly differentiating the mask mark move in parallel, The marks have similar shapes that almost overlap each other.

The two-dimensional camera (imaging device) (
(not shown) to obtain an analog image signal. The analog image signal is converted into a digital image signal by an analog-to-digital converter (not shown). Here, the number of gradations for digitization should be 6 bits or more, preferably 8 bits. By synchronously integrating this digital image signal, noise is removed, the S/N ratio is increased, and at the same time data compression is performed. Here, the number of times of integration may be any number. Next, the digital image signal that has been subjected to noise removal and data compression in this manner is differentiated to extract the edges (contours) of the alignment marks. Similarity pattern matching processing expressed by equation (8-3) or equation (7) is performed on this differentiated data.

After determining the position where the maximum value is obtained as a result of the processing, distances L1 and L2 between the marks are determined with higher accuracy by interpolation processing, and the relative position between the mask mark and the wafer mark is determined by (L1+L2)/2.

It goes without saying that the shape of the alignment mark is not limited to this embodiment, and may be of any shape as long as the data obtained by processing the input image with the differential operator maintains similarity.

By combining this method with a processing device for a "device for detecting the position of two objects separated by a minute distance," an alignment system in an exposure apparatus such as an X-ray exposure apparatus can be constructed. This alignment system makes it possible to set an arbitrary gap, and also realizes a highly accurate and high throughput alignment system with extremely low process dependence.

Further, the scope of application of the present invention is not limited to exposure apparatuses, but can also be applied to other pattern recognition application fields in which the shape and distance of a pattern are estimated in advance. In this case, by providing appropriate parameters in the similarity pattern matching formula, the calculation time can be reduced from several hundredths to several thousandths compared to cases based on conventional autocorrelation functions and cross-correlation functions. can do.

[Effects of the Invention] As is clear from the above description, according to the present invention, a wafer mark and a mask mark are divided into a wafer mark differential signal obtained by first-order differentiation of the wafer mark and a mask mark differential signal obtained by first-order differentiation of the mask mark. The relative positions of the wafer mark and mask mark can be determined with high accuracy and easily by creating a mark shape that maintains similarity so that they overlap when they move horizontally, and by performing similarity pattern matching on a signal obtained by differentiating the image input signal. It has the effect of being recognized through treatment.

Therefore, the throughput can be improved, making a significant contribution to simplifying the production process and improving production efficiency.

[Brief explanation of the drawing]

FIG. 1 is a diagram for explaining a method for detecting the position of an alignment mark according to an embodiment of the present invention, FIG. 2 is a diagram for explaining the collapse of the symmetry of a conventional resist formed mark, and FIG. 3 is a diagram for explaining a conventional method for detecting the position of an alignment mark. Figures 4 and 5 are diagrams for explaining the collapse of the symmetry of the process-formed marks, and Figures 4 and 5 are diagrams for explaining that superposition is possible using the autocorrelation function in the case of the alignment mark shape of the present invention. , FIG. 6 is a diagram showing the first differential value V' of the video signal of the alignment mark in FIG. 5 and the distance W between the mask mark and the wafer mark, and FIG. A diagram showing a correlation interval R,
FIG. 8 is a diagram showing the experimental results of the autocorrelation function A (P, W) according to the present invention, and FIG. 9 is a diagram showing the processing results of similarity pattern matching according to the present invention. Interval W: Paddle of left and right pattern v″(j): !th order differential value V of the image input value at pixel j
'(j) I4 Figure Huntrast Figure Wafer mark (line symmetry) (Asymmetric) Figure (Line symmetry) (Asymmetric) Figure Figure

Claims (1)

[Claims] 1. Converting an analog image signal obtained by imaging an alignment mark consisting of a wafer mark formed on a wafer and a mask mark formed on a mask using an imaging device into a digital image signal; In the method of detecting the relative position of the wafer and the mask by processing a converted digital image signal, the wafer mark and the mask mark are combined with a wafer mark differential signal obtained by firstly differentiating the wafer mark and the The mask mark differential signal obtained by firstly differentiating the mask mark has a mark shape that maintains similarity so that it almost overlaps with each other when horizontally moved, and the digital image signal processing differentiates the digital image signal, and the differentiated signal A method for detecting the position of an alignment mark, characterized in that the relative position of the wafer and the mask is detected from the position where the processing result is maximum. 2. The similarity pattern matching process is performed using A(W,P
)=Σ＾R_i_=_1(V'(P+j)・V'(P+
j+W)) (where, V'(j) is the differential value of the digital image signal, R is the correlation interval, P is the starting point of the correlation interval, and W is the center position of the mask mark adjacent to each other and the center position of the wafer mark. 2. The alignment mark position detection method according to claim 1, wherein the alignment mark is a distance between center positions.