KR100916884B1 - Pattern positioning method in micro lithographic system - Google Patents

Abstract

The present invention relates to a method of determining the coordinates of a pattern arbitrarily formed in a deflector system. This method basically consists of the following steps.

Moving the pattern in the first direction (X-direction),

Calculating the position of the edge of the pattern by counting the number of micro sweeps performed in a second direction (Y-direction) perpendicular to the first direction until the edge of the pattern is detected,

Determining a correction function for the surface that reflects the change in a third direction (Z direction) perpendicular to the first and second directions, and

Determining coordinates by associating the number of counted micro sweeps with the speed of movement of the pattern to compensate for the change in the third direction using the calibration function.

The invention also relates to software for implementing this method and a computer readable recording medium having recorded the software.

Description

Pattern positioning method in micro lithographic system {A METHOD FOR MEASURING THE POSITION OF A MARK IN A MICRO LITHOGRAPHIC DEFLECTOR SYSTEM}

The invention relates to a method for determining the coordinates of an arbitrary shaped pattern on the surface of a deflector system as described in claims 1 and 14. The invention also relates to software (or computer readable recording medium having recorded a program) that implements a method of determining the coordinates of an arbitrary shaped pattern on the surface of a deflector system as defined in claim 22.

The method used to measure time in deflector systems has long been used. So far there has been little change in this algorithm. Only patterns used for many different types of calibration have been modified. Today we are implementing the experimentally confirmed repeatability of this method in the 10-15 nm range on a 800x800 mm ² surface. 10-15 nm means measurement overlay.

One disadvantage of this method is that it can only measure in the same direction as the micro sweep. To measure the X coordinate, we must use a specific pattern with a 45 degree bar.

The method according to the known art is briefly introduced. This is because it is important for understanding the present invention.

According to the known art, it is difficult to measure time with high accuracy. For example, if you want to measure pulses with a resolution of 1 ns, you will need a measurement clock with a frequency of 1 GHz when traditional frequency measurement methods are used. In the known art system described there is no need to measure a single shot of a pulse. By using the scanning beam during the measurement, you will obtain several one-dimensional images of one bar or several bars. Only the CD (critical dimension) of the bars or the average position of the sides are of interest. The measurement system only shows the average result with the sigma. It is important to remember that this measurement system is good enough if this sigma is less than the system's natural noise. This natural noise can be summarized as laser noise, electrical noise, and mechanical noise. The noise from this measurement system itself can be calculated theoretically or can be actually confirmed using a known reference signal. It is also possible to obtain a figure of the measurement system noise by simulation. CD or bar position measurements will therefore have errors.

Error _tot = √ ((Error _natural ) ² + (Error _measurment ) ² )

When measuring time we use a random phase method. This means that the unit of measurement itself has nothing to do with the phase of the signal we want to measure. Since the signal phase is arbitrary with respect to the measurement clock phase, we use the measurement clock frequency corresponding to a much lower frequency and use the averaging effect instead of achieving accuracy.

In Figure 1, the measured clock phase is shown for the reference signal SOS. Note that the input signal bar is synchronized with the reference signal. This is because the input signal is generated from the micro sweep itself. The upper row of the clock in FIG. 1 is the ruler displayed in measuring the clock increment. It is next to be noted where the positive leading edge 10 of the input signal appears relative to our reference signal. Of course, negative leading edge 11 is also of interest. However, the same method can be used to locate any edge.

Consider the cycle time tm of the measurement clock. Since the input signal is the result from a micro sweep, we can accurately know the relationship between the pixel clock period in time and the corresponding value in nanometers. Here we can introduce a pixel clock period tp in nanoseconds. In addition, the pixel clock period in nanometer units may be set to pp. Accordingly, the scaling equation is expressed as follows.

pm (nm) = tm (ns) * pp (nm) / tp (ns)

pm represents each measurement clock period in nanometers. 1, we can see that the schematic position of the first edge 10 corresponds to eight pixel clocks. By using one measurement, i.e. using only one of the six measurements (1-6), it can be seen that this edge is within the range of 8-10 measurement clocks. This accuracy is expressed as 2 * tm. Using the above scaling equation, this can also be expressed in nanometers.

Below are some realistic values.

tm = 1/40 = 25 ns

tp = 1 / 46.7 = 21.413 ns

pp = 250 nm

The result is pm = 291.86 nm.

When counting measurement clock ticks by resetting the counter by the reference signal, we will only count 8 or 9 ticks. No other count is possible in this example. The edge position with respect to the phase of the measurement clock will be rectangular distributed in this way in tm. Thus, the average position can be calculated by adding together counts from several measurements and dividing this number by the number of measurements. In this example, we get (8 + 8 + 8 + 8 + 9 + 9) / 6 = 8.33 counts as average values. Therefore, the estimate of the edge position can be calculated as follows.

8.33 * 291.86 = 2432 nm

Now, it is not enough to use six measurements in this example. It is common to use thousands of measurements (the three sigmas in the detailed description are from a theoretical point of view).

Moreover, no method has been disclosed so far that can compensate for variations in the thickness of an object when the coordinates of the randomly pattern arranged on the surface of the object are measured for calibration purposes in the deflector system.

It is an object of the present invention to provide a method of determining coordinates, in particular two-dimensional coordinates, in a deflector system using any kind of pattern to compensate for unevenness of the patterned surface.

Solutions are provided in the features recited in claims 1 and 14.

It is another object of the present invention to provide software for performing the methods provided in the features stipulated in claim 22.

An advantage of the present invention is that it is possible to create an image of a pattern without using any other detection method than is used today. This is because the present invention is similar to the known method, except that it rotates 90 degrees with superior accuracy compared to existing systems.

Another advantage is that no new hardware is needed because the invention is implemented in software.

1 is a schematic diagram illustrating a method of measuring time in the same direction as a micro sweep in accordance with known techniques.

2 is an image diagram of a star-mark that may be used for time and position measurements in accordance with the present invention.

3 is an enlarged view of a portion of the image of FIG. 2;

4 shows a principle measurement technique of a horizontal bar according to the invention.

5 shows a principle measurement technique of a vertical bar according to the invention.

6 is a diagram of a preferred method for obtaining X coordinates using a Y-direction sweep in accordance with the present invention.

7 is an image obtained using the method according to the invention.

8 is an enlarged view of the image of FIG.

9 is a cursor diagram applied to the image shown in FIG.

10 is an enlarged view of the cursor of FIG.

11 is a graph showing the average speed for one measurement.

12A is a diagram of the statistical principles underlying the arbitrary phase measurement, used in a preferred embodiment according to the present invention.

12B is a view of an exposure case.

FIG. 13 is a diagram of plate bending effect for computing offset used in the present invention. FIG.

14A and 14B show a glass plate having a flat top and a machined bottom and a plate bending effect on the insertion of a reference surface when arranged on a flat support.

15A and 15B illustrate a glass plate having a machined top and a bottom parallel to the plate bending effect on insertion of a reference surface when arranged on a flat support.

16A and 16B show a glass plate having a flat plate top and a flat bottom and a plate bending effect on the insertion of a reference surface when arranged on a processed support.

So far we have only used this method to measure along microsweep (ie one dimensionally). Nevertheless, it is possible to measure this method in two dimensions. In doing this, we are actually generating images of the pattern we measure.

When talking about an image, we can think of it as a set of pixels. Each pixel has a predetermined "gray-level" that indicates the intensity of the pixel.

When handling a CCD image, each pixel is fixed positionally with a given raster (or grid). When analyzing a CCD image to find the position of one edge, information about pixel position and gray level should be used. Several other intuitive methods can be used to estimate the edge position of the image. The accuracy of the position estimate depends on the calibration of the CCD array. That is, it depends on where the pixels are located in the array, how sensitive the pixels are to light, and how well the images can be placed in the array without distortion. Light distribution to the CCD and several other types of optical distortions will contribute to the error in position estimation. A significant number of these errors can be overcome when calibrating the measurement system based on known criteria.

When using the method according to the invention we also refer to the pixel as well. However, the pixels we refer to are not fixed positionally in a given grid. If we take a snapshot of the pattern by measuring it all at once, we will get the information with very rough resolution (or rough accuracy). It is important to realize that the information we use is pixel location. We do not use gray-level information at all. Of course, it is also possible to use gray-level information by recording the pattern using different "trig" levels in hardware. This is done if you are interested in beam-shaping as in focus measurement. However, we are only interested in measuring the position of one or several bars, so we can calculate the center of gravity and the critical dimension (CD).

When measuring the center of gravity and the CD, we have no interest in the exact location of one single pixel. In general we are only interested in the average of the positions of several pixels. In the CD measurement, we use the cursor to define the number of pixels to be used for this average value. Even in the center of gravity estimation, we even out the noises from the edges. This noise may be from the pattern itself or may be noise from the measurement system. The same is true when using a CCD image as an input.

In the method presented in this way, we use the micro sweep itself as a light source (or ruler). It is hard to find a ruler that is more accurate than this. We already have ways to calibrate this ruler very accurately in terms of power and linearity.

In FIG. 2, we captured an image 20 as part of our star-mark. This image shows the position of the pixels 21 in a grid of 316 x 250 nm. Nothing has been done to the image except to show the pixels in this grid. Image 20 shows the events within this area. Marks were scanned with a 100um hardware cursor. Positive advancing edge 22 is represented by a white pixel, and negative advancing edge 23 (chromium-glass transition) is represented by a black pixel. By just observing the image, it can be seen that the mark rotates slightly counterclockwise. The number of black pixels in the lower Y-parallel bars 24 clearly indicates this fact when compared to the number of white pixels.

In order to represent the actual grid we are using, and to show how the pixels are distributed in the grid, we can refer to FIG.

Here, we can see a partial enlarged view of the image 20. The "hard copy" of this image clearly indicates where we found the event. How to sharpen this image will be presented below. The scale of this image is accurate given that one pixel is 316 nm (vertical scale) in the X-direction and 250 nm (horizontal scale) in the Y-direction.

X-coordinate estimation

As described in the Background section herein, there are very accurate methods for estimating the Y-coordinate of an event. Microsweep is used as the ruler and the measurement clock, which has an arbitrary relationship to the phase of the ruler. This measurement clock will give a rough resolution of tm (292 nm) in a single shot measurement. If we use several measurements and build an average value, we will get a higher resolution (see below). In practice, we can select the accuracy by selecting the length of the cursor to be used and the number of measurements. So far, the above statement in Y-coordinate estimation has been true. The problem is how we can estimate the X-coordinate.

It is difficult to obtain an X-value from the data derived by scanning the beam in the Y-direction. The preceding step is that it should be practically possible to retrieve this information with the same accuracy as the Y-coordinate. But to do this we must introduce another signal lambda / 2 X-signal.

In the prior art, when measuring the 45 degree bar of a pattern as in the case of a star-mark, we use the X-lambda / 2 signal as a mark in the X-direction to define the X-cursor. Inside the cursor we record the lambda / 2 signal at the same time we count the measurement clock. But since we measure about 45 degree bar, we actually get the X-coordinate using only Y information. In combination with lambda / 2 information, we can compute X-coordinates with very high accuracy. The drawback of this method is that it is not possible to measure any kind of pattern. In particular, we cannot make measurements on bars that are parallel to the ruler. If we extend this method in the way we are using it in the Y-direction, we will easily see that the problem to be solved is exactly the same as we have in the Y-direction but rotated 90 degrees. When we changed our measurement clock for our Start of Sweep (SOS) and using the Lambda / 2 signal as the reference signal, we rotated this problem 90 degrees.

When doing the rotation of this problem, we need to recalculate our parameters. In the Y-direction our resolution was one measurement clock corresponding to 292 nm. During one run for the pattern of interest, we scanned the pattern at a frequency of approximately 30 kHz. The question is how far we move in the X-direction during the scan. If we set the speed as low as possible, we will be able to inject a pattern 8-10 times per lambda / 2 cycle. Since one lambda / 2 cycle corresponds to 316 nm, we will have a resolution in the range of 30-40 nm in the X-direction. This is why we scan the pattern at a frequency of 30 kHz during the movement in the X-direction. When we use the lambda / 2 signal as the reference signal, we have one clock with a spatial resolution of 30-40 nm in the X-direction. This is much higher than the resolution in the Y-direction. However, this is important. This is because the motion in the X-direction will not be able to obtain multiple samples in the X-direction as in the Y-direction. This fact is represented in FIG. 4.

The situation in the X-direction is shown in FIG. 4. One bar 40 is scanned in one stroke and generates only one event of six scans 41. Thus, when moving relative to the bar at one time, we can know the position of the bar with an accuracy of +/- 40 nm in the X-direction. The Y-coordinate of the bar position can be seen with an accuracy of +/- 292 nm (for each of the two edges 42, 43). In FIG. 4, the CD in the X-direction of the bar is lower than the 40 nm measurement grid we are using in the X-direction. Thus, one injection of a pattern may miss the fact that there is at least one bar.

This is natural because of the lower resolution than the CD to be measured. In order to measure the bar with high resolution, it is necessary to make several scans in an arbitrary phase on the pattern.

A comparison of the situation in the Y-direction is shown in FIG. 5. Here, the bar 50 is scanned at the same length in the Y-direction. The resolution in the Y-direction in each scan 51 is 292 nm, but for bars of equal length we will make seven scans.

If we isolate this problem, we can say that one pixel can be resolved with a resolution of 40 nm in the X-direction and 290 nm in the Y-direction in one scan.

algorithm

So far we have described the key principles in the Y and X-directions. We rotated the problem in the Y-direction 90 degrees in the X-direction. In the Y-direction, we have two processes that are random with respect to each other. That is, it has a measurement clock and SOS (or any signal correlated to SOS). In the X-direction, the measurement clock corresponds to the SOS signal and the reference signal is a lambda / 2 signal. These signals (or processes) are also uncorrelated. We have different resolutions in different directions. But the accuracy turned out to be about the same.

In FIG. 6, the principle of acquiring the X-coordinate of the bar 60 is described. In FIG. 6, the bar 60 can be seen parallel to the ruler and micro sweep directions. The reference signal corresponds to the lambda / 2 position signal. In each lambda / 2 interval, we scan the pattern 61 with a micro sweep. Movement for the pattern in the X-direction is performed at a much slower speed than the speed used for pattern exposure. In this example, we get about eight injections 61 in the lambda / 2 interval. If we start counting SOS in this interval, we will get a situation very similar to that described above. If you count the total number of scans within this interval, this will be a speed measurement within this interval. Since speed cannot be assumed to be the same in all sections, it is important to perform this velocity calculation to ensure accurate "weight" of events in the section in the X-direction. In this example, we count two SOSs in the first run 62 and three SOSs in the second run 63, and so on, one Y-event (transition from glass to chrome). Will get By adding the index of the event within one interval, and dividing this number by the total number of SOS in the interval, an estimate of the X-coordinate of one event in the interval may be obtained. Above we will get the approximate location of the first Y-event. You will get it after three runs on the mark, and the value will be (2 + 3 + 2) / 3 = 2.3 SOS ticks in this interval. To calculate this value in nanometers, we multiply this number by the local resolution.

Here, we get 2.3 * 316/8 = 92.2 nm. This corresponds to the local coordinate 64 for the edge of the bar 60 in the first interval. Local resolution depends on the total number of SOS in the interval, ie speed. If you run the system more slowly, this resolution will be better. However, you may be able to increase the resolution by injecting the bar in multiple runs. The accuracy of the average position estimate is described below.

As can be seen from above, we can compute the X-coordinate from the data imported from the scan sweep in the Y-direction. It is our job to take advantage of the fact that every time we cross the section boundary 65 we know exactly where it is located in the X-direction. It is to be assumed that the velocity is constant within one interval. This does not mean that the speed must be constant for all sections. In practice, we make several scans between patterns in both directions and record the Y-event and lambda / 2 positions simultaneously. Thus we can calculate local speed with high accuracy by using information from all runs.

The method described above is suitable for laser lithography systems or e-beam lithography systems.

Filter

The next thing we are interested in is not the exact location of individual pixels. So far, the point is that the positional accuracy of a single pixel depends on how many times the pattern has been recorded and the resolution we used during recording. If we scan the pattern a certain number of times, we can choose the accuracy. This is possible because we have complete control over the measurement process. When we make this accuracy choice, we also need to consider our cursors. As mentioned above, the cursor is also just one way of defining multiple pixels for use in averaging.

There are many ways to apply filters to this kind of data. The obvious approach is to conform to the line using the standard regression technique. Although these techniques are applied, they do not produce optimal results in this case. The key reason is that the pixel data we handle does not follow the Gaussian distribution. The data we have to deal with has a rectangular distribution. When using the regression technique, we will "overweight" pixels close to the boundary of the lambda / 2 interval or in the tm interval of the Y-case. A better use is a simpler "area" estimation method. This method is more accurate for this kind of data than regression techniques. To align the lines with the edges, you can split the database in half, or two. The data we have in this case is the x and y coordinates. Compute the average value of all coordinates in each half. In this way we can get two x and y points. These two points represent the line that will be used for further calculations.

Realistic results

In FIG. 7, data is filtered using the algorithm described above. So far we have not applied any cursors. Only the average positions of the pixels were calculated. The image 70 shown is from four runs on the mark. A small rectangle 71 in image 7 is magnified in FIG. 8. Here, we used the previous algorithm and some filtering to sharpen the data. Each pixel in this image is the result of all four runs for this pattern.

Cursor( cursors )

We will now apply a cursor to the data to measure the CD and the center of gravity of the cross. The center of gravity of the cross is measured using four cursor pairs. These cursors are shown in FIG.

Each line 90, 91 of the cursors is computed based on the data from the edge of the cross. This line is computed using the simple "area" estimation method mentioned above.

In Figures 10A and 10B, portions of the X and Y bars are expanded.

10A shows a portion of the upper left edge. The computed cursor is an accurate estimate of the position of the edge in the X-direction.

10B is a portion of the upper right edge of the cross. The position of this line 91 defines the edge position in the Y-direction.

The reason why the white pixels and the black pixels are mixed along the Y-bar in FIG. 10A is described. In this example, only one event can be retriggered after two clock cycles of the measurement clock. This means that if we have a positive transition and a negative transition within this time period, we will miss one of the events. This is one of the reasons why pixels are bit spread in the Y-direction. Then, because of the noise, the hardware will randomly trigger a negative or positive transition. This is not really a problem at all because it is not important information whether we have a positive transition or a negative transition. What counts is where the transition occurs. To know the direction of the edge, several transitions can be used or other kinds of logic decisions can be used to see what kind of transition occurs.

In the table below, center of gravity and CD are shown for the cursor. The table below shows the results of four cursor pairs individually.

The center position (Xcenter, Ycenter) of the mark may be calculated as the average value of the Y-cursor centers and the X-cursor centers.

Secondary effect ( Second order effects )

So far we have discussed the key principles of this algorithm. We will now discuss two vivid corrections that should be performed on the data, which are the secondary effects from this method.

First, we need to correct the ultimate azimuth of the data. When using a writer, we already have a pre-misallignment between the X-motion direction and the ruler. This azimuth angle α can be expressed as follows.

vx is the exposure speed of the system, and vy is the speed of the micro sweep.

This angle operation can be simplified to the following equation.

In this case, Sos_rate is the total number of pixel clock periods between two SOS.

Another effect to consider is the effect of X-movement during the measurement. Here we introduce azimuth error. Even if we implement the same number of positive and negative strokes, we will not be able to completely eliminate this error. The reason is that this error proceeds with the speed difference for the positive stroke and the negative stroke. For a stroke in one direction, we will get an error that can be expressed in degrees β.

This angle β can be expressed as follows.

In this case, xInc is lambda / 2 (nm), and speed is the total number of start of sweeps (SOS) in the xInc section. If we divide β by α, we can get the relationship between the two angles.

Inserting a substantial value, xInc = 316 nm, Speed = 8 Sos / interval, nbeams = 9 beams, yPix = 250 nm, thus yielding the following results.

If you calculate the error caused by α for a distance of 100um, you will get

alpha_error = 100 * 9/1435 = 0.6272 um

Sos_rate is the value obtained from the TFT3 system parameters. Since β = 0.0175 * α, we can calculate the error caused by movement during measurement as

0.0175 * 627.2 (nm) = 11 nm

This is a significant level of error that cannot be ignored. This error will change the sign according to the direction of movement. If the measurement is made during the same number of positive and negative strokes and the local velocity is the same for both strokes, this error will be completely cleared. Not really. Therefore, we will get a small net error due to this fact.

In the graph shown in FIG. 11, the average speed is shown for one measurement. Two forward strokes and two reverse strokes were used. The hardware cursor was 99.22um (314 lambda / 2 intervals). From this, it can be seen that there is a difference in local speed for the forward and reverse strokes.

Arbitrary Phase Measurements Random phase measurement )

When using an arbitrary clock for the measurement, we will observe this as a statistical problem. In FIG. 12A, a measurement situation is shown. What we want to measure is tp, the time difference between t1 and t0. This signal is synchronized with the reference signal.

You can rewrite time tp as

tp = (k + d) * tm

Where k is an integer and d is the fractional part of tm. In this case, d may be a number between intervals [0, 1]. It will be explained later why this formula is a reasonable one for tp.

We now introduce a measurement clock with an arbitrary phase relative to the reference signal. We also introduce a counter that counts the positive going flank of this clock. If we reset this counter to the reference signal, we will know that we count k flanks or k + 1 flanks. No other count is possible. We introduce a discrete stochastic variable K to get the values of k and k + 1.

Now another random variable A is introduced with reference to FIG. 12A. Variable A represents the phase of the clock relative to the reference signal. The sample point of A (i) will be the number in the interval [0, 1]. A is a continuous random variable.

In FIG. 12A, we can identify the following important facts.

If ∝> d, then the sample point of K will be k.

If ∝ <d, the sample point of K will be k + 1.

Now we need to compute the probabilities for sample points k and k + 1. For this purpose, we should use the frequency function shown in Figure 12a. Since all phases have the same probability, this becomes a rectangular distribution function.

Thus, the probability of obtaining sample point k + 1 from K will be d, and the probability of obtaining sample point k from K will be 1-d.

If we add the clock count for each measurement and then divide by n, we will estimate the mean value for the random variable K.

The estimated average value can be expressed as follows.

Here we have two possible sample points. Thus, we have the following result.

If we rescale this to nanoseconds, we get:

According to this result, if we get the average value of the counter ticks and then scale this value to tm, we get the time we will be next.

Sigma

To calculate the accuracy of the mean value E (K), we must find a change in K.

The change in distribution can be expressed as follows.

This can be expressed as follows.

Accordingly, we get the following result.

And

The change function is actually of great interest. We can see that V (K) = 0 when d = 0, that is, it has no fractional part. If d is very close to 1, we can see that V (K) = 0. When d = 0.5, this change has a maximum value. In this case, the change is 0.25. Sigma will therefore be 0.5 at its maximum.

To interpret this, we can think of the following: If d is zero, we will count k ticks from the counter. Suppose we count one tick if the positive leading edge from the clock matches the reference signal. Since we are counting k ticks regardless of the phase of the measurement clock, the variance from the mean value will be zero. This is because the change is a measure of the providing distance from the estimated average value (see equation (1)).

Let's look at the physical meaning of these equations and interpretations.

Let me give you a practical example first.

In the case of measuring the signal of the integral part k = 2 and the fractional part d = 0.01, the probability of counting 3 in one measurement is 0.01. This probability is the same for each measurement. If we compute the mean of 100 measurements, we will probably add 99 samples of 2 and 1 sample of 3 (case 1). However, 100 samples of 2 and 0 samples of 3 may be added (case 2). The actual error we get from the mean is

Thus, after 100 measurements in case 1, we obtain

In case 2, it would be 2.00 +/- 0.005.

There is another interesting way to confirm the physical conclusion of this case when d = 0.

Suppose you want to measure a signal that is exactly k * tm. In this case, the source portion is zero. If we add counter ticks, we should always count k ticks. Otherwise, this is important, and we can never get the exact average of k in this case. In other words, we can never count k + 1 ticks. If this is the case, the average value we compute will not be k. For this reason, the change must be zero. Note that only two values, k and k + 1, can be counted in Iran. Thus k-1 can never be counted. In other words, a count of k + 1 cannot be compensated for by k-1, so we get the correct average anyway.

Since we don't know tp in advance, the worst case scenario is when we assume an error. In other words, the error caused by this method is as follows.

Error (K) = 0.5 * tm [ns]

This is the maximum value of the function d * (1-d) as mentioned above. If we want to use symmetric error instead, we can express the result of this method as

The error of this method will be small if we use a large number of measurements. We can express this error as

This equation can be scaled in nanometers as follows.

Where rs is the actual resolution for the actual direction. If we substitute some values, rs = 291nm in the Y-direction and rs = 40 (316/8) nm in the X-direction. Therefore, the estimation error of the pixel position in the X or Y-direction can be approximated as follows.

azimuth

In FIG. 12B, the case of exposure is shown. Between the two sweep start points, we move nbeams * dy [um] in the X-direction. dy is the pixel size. Here, assume a square pixel. At the same time, we move the distance of N * dy [um] in the Y-direction.

The angle α can be expressed as atan (vx / vy). If you calculate this angle, you get the following result.

sos_time can be expressed as N * pixel_clock_time. N is the total number of pixels between the two sweep start points. Finally, we can express angle α as

It should be noted that this angle α is a constant "compensation" that is preferably removed from the database.

z-calibration correction )

The previously described method of determining coordinates of an arbitrary shape pattern on a surface in the deflector system assumes that the surface is flat. But in reality it is not. Height variations in the Z-direction perpendicular to the X-Y plane will appear slightly for all surfaces. This is disclosed in International Patent Application No. PCT / SE2004 / 001270, filed September 3, 2004. The contents of which are incorporated herein by reference. The method of determining any form of pattern on the surface is preferably combined with the method of determining a correction function that compensates for changes in height Hz.

An essential part of the present invention is to determine the reference surface and calculate the difference in height Hz against it. This difference is denoted by H and is explained in connection with FIG. The reference surface can have any desired shape as long as the shape of the reference surface remains unchanged. It is preferred that the shape of the reference surface is planar.

If possible, it would be desirable to use a free (weightless) form. That is, it would be desirable to use the center line of the plate as the reference surface, but this is difficult to implement in practice. The underside of the plate is not seen as a good alternative to the reference surface. This is because a stepper or aligner uses the top side as a reference surface.

On the other hand, when the upper surface is used as the reference surface, it is also necessary to know the shape of the support and the lower surface of the plate. Although the form of the support can be obtained, it is very difficult to actually know about the bottom surface. The top side can be measured without knowing the bottom side. Large glass plates placed on the three feet will deform due to the plate weight. However, knowing the thickness of the plate, the material, and the three-foot configuration, the deformation function for the perfect plate can be calculated. Incomplete measurement of the glass plate will result in measurement of the deformed plate when placed on three-foot. The shape of the top surface can be obtained by knowing the deformation function calculated for the perfect plate from the measurements of the deformed plate.

The top side of the glass plate is relatively even compared to the bottom side. That is, the height change with respect to the centerline is small. The best result is that the top of the plate is the reference surface. However, it is not clear that the top side is the best choice due to the deformation of the glass plate during the next step of the exposure system. If the top surface 113 of the glass plate shows a change when it is close to the position on which it rests on the support, the pattern on the top surface 113 will be distorted near the support.

However, any surface can be used as the reference surface. It is of course preferred that the top surface is the reference surface.

13 shows the plate bending effect on the glass plate 111 having a thickness T. Reference surface 130 is determined. In this example the reference surface is flat, the glass plate is divided into several measuring points 131, and the height Hz is measured at each measuring point by means known from the known art. The height H between the reference surface 130 and the deformed surface 113 of the glass plate 111 is determined from the height Hz measured for the surface 113 of the glass plate 111 by the device, from the reference surface at the measurement point ( It is easily calculated by knowing the height of 130).

The local offset d is then calculated (as a function of x and y) for each measurement point, which depends on three variables. That is, it depends on the thickness T of the glass plate, the distance P between adjacent points, and the measurement height H between the reference surface 130 and the surface 113 of the glass plate 111. The local offset should be considered positional deflection from the location where the pattern is to be recorded with respect to the reference surface, as described with reference to FIGS. 14-16. The pitch P on the surface of the plate is different from the nominal pitch Pnom on the reference surface.

The distance between adjacent measuring points shall not exceed the specified distance. This depends on the accuracy required for the measurement in order to obtain good and reasonable results from the measurement. An example of the maximum distance between adjacent measuring points is 50 mm when the thickness of the glass plate 111 is 10 mm and the glass plate material is quartz. The distance between adjacent measuring points also varies with the thickness of the glass plate to achieve the same measurement accuracy. The thickness change of the glass plate is 10-15 μm, but may be larger. The measuring points can be arbitrarily distributed between the surfaces 113. However, it is preferred to arrange in a grid structure with pitches that are not necessarily the same in the x and Y-directions, i.e., the specified distance between each point.

The local offset is a function of the X and Y-directions at each measurement point and can be calculated using a very simple equation.

If the distance P between two adjacent measuring points 131 is known, the angle α can be calculated from the measured height H.

If the angle α is small,

Furthermore, when α is small, the local offset d can be calculated using the following equation.

However, the above formula for calculating the local offset d is just one example of the operation for determining the offset d. The slope at each measurement point can be measured directly by the system and the local offset is proportional to this slope and plate thickness.

As described above, FIG. 13 shows the bending effect in one dimension. However, the local offset d is a two-dimensional function of the derivatives (dx, dy) at each measurement point.

As a non-limiting example, we assume that the distance between two adjacent points 131 is 40 mm, the glass plate thickness is 10 mm, and the measured height H is 1 μm. This will be represented by a one-dimensional local offset d of 125 nm.

14A and 14B illustrate a glass plate 141 having a flat upper surface 143 and an uneven lower surface 142 and a flat reference surface 144 inserted in this example when supported by the flat support 145. Present the plate bending effect of.

When the glass plate 141 is arranged on the flat support 145, the shape of the top surface 143 will change and the bottom surface 142 will follow the shape of the flat support 145. As a result, the generated patterns represented by the points 146 on the top surface expand to derive the correct reference surface.

15A and 15B illustrate a glass plate 151 having an uneven top surface 153 and a flat lower surface 152, and a flat reference surface 144 inserted in this example when arranged on a flat support 145. Present the plate bending effect of.

When the glass plate 151 is arranged on the flat support 145, the shape of the top surface 153 remains unchanged and the bottom surface 152 will follow the flat support 145. The pattern generated is represented by dots 155 on the top surface, which must be expanded to obtain an accurate reference surface. This is because the top surface becomes flat when placed in the typical exposure equipment known in the art in the vicinity of the support. The portion of the glass plate located right between the supports will be deformed. Moreover, the support will modify the pattern of the glass plate if the support form does not follow the reference surface form.

16A and 16B show the plate bending effect on the flat top surface 143 and the flat bottom surface 152 when the flat reference surface 144 is inserted in this example when arranged on the uneven support 162. do.

When the glass plate 161 is arranged on the uneven support 162, the shape of the top surface 143 will change and the shape of the bottom surface 142 will follow the uneven support 162. The pattern of occurrences represented by the dots 164 on the top surface is expanded to obtain an accurate reference surface. This is because the top surface is flattened when the exposure equipment is located.

14A, 14B, 15A, 15B, 16A, and 16B illustrate extreme situations, in which all three changes appear in the process of writing a pattern on a glass plate.

However, the overall error is much smaller because all errors from the top, support surface, and contaminants are eliminated or reduced.

Glass plates have been used herein as illustrative examples, but the scope of the present application is not limited to plates made of glass.

The process of determining the appropriate calibration function for the surface of the object may be performed at any time before, during or after the process of determining the coordinates of any type of pattern on the surface. The object is then used to determine the position of the mark on the object for calibration purposes. The calibration function will improve measurement accuracy and improve the calibration process.

Claims (22)

A method of determining the coordinates of a randomly formed pattern on a surface in a deflector system, the method comprising: a) selecting a reference clock signal (lambda / 2) that defines movement in the first direction (X-direction), b) providing a micro sweep that repeatedly scans the surface in a second direction (Y-direction) perpendicular to the first direction (X-direction), c) selecting a measurement clock signal SOS related to the signal used to initiate each micro sweep in the second direction (Y-direction), d) performing a first run consisting of d1) to d4) below, d1) starting the first micro sweep at the starting position, d2) detecting one or more edges of the randomly formed pattern as the pattern moves in a first direction with respect to the deflector system, d3) generating one or more events when an edge of the pattern is detected, d4) counting the number of micro sweeps performed until each event occurs, e) calculating the coordinates of the edge for each event in the first direction (X-direction) using the number of micro sweeps performed In this case, the arbitrarily formed coordinate determination may be performed by using a signal used to start each micro sweep in a second direction as a reference signal and using a pixel clock signal as a measurement signal (Y). Direction of the pattern). 2. A run according to claim 1, wherein the run defined in step d) is performed two or more times, and for each run, the starting position defined in step d1) is arbitrarily selected to produce a microsweep distributed randomly between each run. Pattern coordinate determination method characterized in that the generation. 3. The method of claim 2, wherein the average value of the edges is calculated in step e) to increase the accuracy of the pattern coordinates in the first direction. 4. A method according to any one of the preceding claims, wherein the reference clock signal selected in step a) comprises a known position of the system in the first direction (X-direction). The method of claim 4, wherein the reference clock signal selected in step a) is divided into sections, wherein each section corresponds to a lambda / 2 period, and the measurement clock signal selected in step c) is 8-10 of the pattern in each section. The pattern coordinate determination method characterized by having a period corresponding to a time scan. 4. Compensation for the inserted azimuth error when the micro sweep scans in the second direction (Y-direction) during movement of the surface in the first direction (X-direction). Pattern coordinate determination method comprising the step of performing. The method of claim 6, wherein the compensation is a constant compensation. The method of claim 1, f) determining the calibration function for this surface before, during, or after the steps a) -e) have been performed, so that the two-dimensional local offset of the xy plane with respect to the measurement points on the surface (d) And compensating for the change in the third direction (Z direction) perpendicular to the first and second directions, g) calculating the coordinates of the randomly formed pattern using the determined calibration function Pattern coordinate determination method comprising a. A method according to any of the preceding claims, wherein the method is used in a laser lithography system or an e-beam lithography system. The method of claim 8, Placing an object with thickness T on the surface on the stage of the measuring device, Dividing the surface of the object into a plurality of measuring points, wherein two adjacent measuring points are spaced apart from each other by a distance not exceeding a specified maximum distance, Determining the slope of the surface at each measuring point, Calculating a two-dimensional local offset (d) of the x-y plane for each measurement point as a function of the thickness (T) and the slope of the object, and Determining a calibration function for the surface using the two-dimensional local offset (d) computed for each measurement point Pattern coordinate determination method comprising a. 11. The method of claim 10, wherein determining the slope of the surface comprises measuring a change in height of the surface at each measurement point. 12. The method of claim 11, wherein measuring the height change of the surface comprises: Determining a reference surface, Measuring the height H between the object's surface and the reference surface at each measurement point, whereby the two-dimensional local offset d of the xy plane is determined from the measured height H, from one or more adjacent measurement points. Characterized in that it can be calculated as a function of the distance P and the thickness T of the object, Pattern coordinate determination method comprising a. The method of claim 10, wherein the object is a reference object, and the surface is provided with a mark at each measurement point. A method for determining coordinates of a pattern randomly formed in a deflector system, the method comprising: Moving the pattern in the first direction (X-direction), Calculating the position of the edge of the pattern by counting the number of micro sweeps performed in a second direction (Y-direction) perpendicular to the first direction until the edge of the pattern is detected, Determining a correction function for the surface that reflects the change in a third direction (Z direction) perpendicular to the first and second directions, Determining coordinates by associating the number of counted micro sweeps with the speed of movement of the pattern to compensate for the change in the third direction using the calibration function. Wherein the calibration function determination comprises measuring the physical properties of the surface of the object having a thickness T, which step comprises: Determining the slope of the surface at the partitioned measuring points, Calculating a two-dimensional local offset (d) of the x-y plane for each measurement point as a function of the thickness (T) and the slope of the object, and Determining a calibration function for the surface using the computed two-dimensional local offset d for each measurement point. Pattern coordinate determination method comprising a. The method of claim 14, wherein the method is Adjusting the speed of movement in the first direction (X-direction) to determine the distance between the start of each micro sweep, wherein the speed of movement of the pattern is correlated with the number of micro sweeps performed; Pattern coordinate determination method. 16. The pattern of claim 14 or 15 wherein the pattern is scanned a number of times (called a run) and the offset in the first direction relative to the first micro sweep is randomly selected for each run. Pattern coordinate determination method characterized by the above-mentioned. 17. The method of claim 16, wherein the position of the edge is obtained from an average value from all runs. delete 16. The method of claim 14 or 15, wherein determining the slope of the surface comprises measuring a change in height of the surface at each measurement point. The method of claim 19, wherein measuring the change in height of the surface comprises: Determining a reference surface, Measuring the height H between the object's surface and the reference surface at each measurement point, whereby the two-dimensional local offset d of the xy plane is measured height H, from each adjacent measurement point. Computed as a function of distance (P) and thickness of object (T) Pattern coordinate determination method comprising a. 16. The method of claim 14 or 15, wherein the object is a reference object and the surface is provided with a mark at each measurement point. A computer-readable recording medium having recorded thereon a program for performing the pattern coordinate determining method according to claim 1 or 14, which is used in a deflector system to determine the coordinates of a pattern arbitrarily formed in the deflector system.

Images