JP3721110B2 - Defect inspection apparatus and defect inspection method - Google Patents

Description

[0001]
BACKGROUND OF THE INVENTION
The present invention relates to a defect inspection apparatus and a defect inspection method, and more particularly to an apparatus and method for inspecting defects on a pattern such as a photomask, a semiconductor chip, a printed board, and a liquid crystal substrate for semiconductor manufacturing.
[0002]
[Prior art]
Current semiconductor devices are manufactured by repeating a process of exposing a circuit pattern formed on a mask serving as an original by photolithography to a wafer. Here, the mask is a mask blank obtained by evaporating a light absorber on a quartz glass substrate. After applying an electron beam resist and exposing it with an electron beam lithography system, etc., development and etching are performed to form a desired circuit pattern. It is a thing.
[0003]
If there are defects such as pinholes and pindots in the mask, these defects will be transferred to all wafers. Therefore, unacceptable defects must be detected and corrected. Further, as the circuit pattern becomes finer, the defect size of the mask to be detected becomes smaller, and it is required to improve the defect detection sensitivity.
[0004]
Defect inspection techniques are roughly classified into a die-to-database method that compares design data and a mask, and a die-to-die method that compares adjacent patterns. The die-to-die method is a simple method for comparing sensor data, but there is a possibility that a defect common to the patterns to be compared may be missed.
[0005]
On the other hand, since the die-to-die method is compared with design data, a reliable defect inspection is possible. However, in order to increase the sensitivity, it is necessary to make the sensor data and the reference data generated from the design data more coincident.
[0006]
However, there is a limit in alignment accuracy due to uneven stage speed, yawing, etc., and this has been compensated by image processing techniques such as pattern matching. However, as side effects of this image processing, there are problems such as (1) indistinguishable from mask pattern drawing deviation itself, and (2) high cost of the image processing circuit.
[0007]
[Problems to be solved by the invention]
With the miniaturization of circuit patterns, mask defects to be detected have become smaller, and it has been required to improve detection sensitivity. Therefore, it is necessary to improve the contrast of the defect image by shortening the inspection optical wavelength.
[0008]
However, when receiving short-wavelength light, photons pass only to the shallow surface of the photocathode, so the hole-electron pair immediately recombines, making it difficult to effectively collect the charge due to diffusion. In addition, the sensitivity of the CCD (Charge-Coupled Device) sensor is lowered due to the fact that the photons are absorbed by the transparent electrode and do not easily reach the photocathode, so that high-throughput inspection becomes difficult.
[0009]
As a method for solving this problem, a time delay and integration type CCD sensor has come to be used. This type of CCD sensor is composed of a two-dimensional light receiving element of M pixels × N lines, and sequentially transfers charges for each pixel to pixels on the next line in a direction opposite to the stage movement in a fixed cycle. Thus, an electric charge amplified N times is obtained. By doing so, practical sensor sensitivity can be obtained.
[0010]
However, the contrast of the sensor image is lowered or the position is shifted due to uneven speed of the stage or yawing. For this reason, there is a possibility that the sensitivity of practical defect inspection is lowered and pseudo defects are caused.
[0011]
On the other hand, for example, instead of synchronizing with a fixed sensor scan time, a method of reading the coordinates of the stage and synchronizing at a fixed pitch can be considered. Certainly, only the traveling direction can alleviate the influence of the positional deviation, but no effect is obtained for the positional deviation in the direction orthogonal thereto.
[0012]
It is an object of the present invention to prevent detection of pseudo defects caused by uneven speed and yawing mismatch when using a time delay integration type CCD sensor for defect inspection, and to improve sensitivity of defect detection.
[0013]
[Means for Solving the Problems]
In order to solve the above-described problems, the defect inspection apparatus of the present invention uses a time delay integration type sensor that optically reads a two-dimensional pattern on an inspection substrate to obtain electrical image information, and detects detected image data. A sensor data input unit for converting to, a position coordinate input unit for reading the coordinates of the reading position of the two-dimensional pattern in synchronization with the optical reading of the two-dimensional pattern, and the time delay of the coordinates of the reading position An average processing unit for adding and averaging the number of integration stages of the integration sensor, and generating the second reference image data by correcting the position coordinates of the first reference image data by the coordinates averaged by the average processing unit And a comparison unit that compares the detected image data input from the sensor data input unit with the second reference image data and detects a different portion as a defect. Characterized by including the.
[0014]
Further, the defect inspection method of the present invention optically reads a two-dimensional pattern on an inspection substrate using a time delay integration type sensor, converts the two-dimensional pattern into detection image data which is electrical image information, In synchronization with the optical reading, the coordinates of the reading position of the two-dimensional pattern are read, the coordinates of the reading position are added after the number of integration stages of the time delay integration sensor, and averaged. The second reference image data is generated by correcting the position coordinates of the first reference image data, and the detected image data input from the sensor data input unit is compared with the second reference image data. The method is characterized in that a different part is detected as a defect.
[0015]
DETAILED DESCRIPTION OF THE INVENTION
Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.
A time delay integration sensor (hereinafter referred to as a TDI sensor) integrates charges by the number of integration stages in the running direction, thereby improving sensitivity proportional to the number of stages and reducing noise such as variations in elements and fluctuations in illumination brightness. . At that time, it is necessary that the moving speed of the stage and the transfer speed of the electric charge in the traveling direction completely coincide with each other. The mismatch between the two causes deterioration of MTF (Modulation Transfer Function) and displacement.
[0016]
FIG. 8 is a diagram for explaining a TDI sensor, and shows how a dot image picked up by an arbitrary pixel moves from above to below on an integration line (stage) of the TDI sensor. The sensor is composed of M pixels and N lines.
[0017]
FIG. 8A shows a case where the amount of movement of the stage and the transfer rate of charges in the traveling direction are completely the same. The dots are picked up at the same relative position in each line, and even if the final CCD output based on the charge amount integrated in all the lines is seen, there is no positional shift and there is little blurring of the image.
[0018]
FIG. 8B shows a case where there is a mismatch between the two speeds described above. The relative positions of the dots fluctuate in each line, and the final CCD output overlaps the images of all lines. It can be seen that when there is a speed mismatch, positional deviation and image blurring occur. Since misregistration has an adverse effect on defect inspection as well as image blur, it is required to improve these.
[0019]
First, the influence of speed variation on MTF is reported in the technical document “High Sensitivity Line Scan Primer and Selection Guide” of DALSA INC. According to this, if the accumulation stage number N and the speed fluctuation ratio with respect to the average speed are δV / V, the MTF deteriorates in proportion to NδV / V. That is, the MTF deteriorates as the speed unevenness increases, and the MTF deteriorates as the number of stages N increases even if the speed unevenness is the same. According to this document, it is reported that when N = 100, if the speed unevenness is 2% or less, the MTF is not affected.
[0020]
Next, positional displacement will be described. In a defect inspection apparatus, accurate alignment of reference data and sensor data is important not only for preventing false defects but also for maintaining practical defect inspection sensitivity. As a conventional example, a stage coordinate is input and reference data is generated while correcting a stage speed variation or the like.
[0021]
Even in that case, when using the time delay integration type sensor, it is most advantageous to use the position coordinates at the intermediate point of the integration time. However, even in such a case, in order to suppress the positional deviation to, for example, ½ pixel or less, it is necessary to suppress the speed unevenness to at least 1% or less when the number of integration stages N = 128. This is not an easy level even in view of the current level of stage control. When the number of integration stages is further increased, it is necessary to further reduce the speed unevenness.
[0022]
A model is set up below to consider the relationship between the positional deviation between the reference data image and the sensor data image and the speed fluctuation. And the standard deviation sigma _delta misalignment delta, standard deviation sigma _v of the velocity v has been found to be expressed by the following equation.
[0023]
σ _Δ = PEF · σ _v (1)
In the above equation, the sensor scan time is set to unit time 1 for simplicity. The PEF (Placement Error Enhancement Factor) indicates the degree of influence that the stage speed variation has on the positional deviation. The larger this value is, the more the position shift becomes sensitive to the stage speed fluctuation.
[0024]
FIG. 9 is a characteristic graph showing the relationship between speed fluctuation and PEF in a conventional apparatus. An autocorrelation function with a Gaussian distribution was assumed as a statistical model of the velocity unevenness from the average velocity. The vertical axis shows the PEF when the number of integration stages N is 128.256,512. The horizontal axis represents the variance (T) of the Gaussian distribution. The right side of the horizontal axis corresponds to a low-frequency speed fluctuation, and the left side corresponds to a high-frequency speed fluctuation. It can be seen that the TDI sensor is easily affected by the low-frequency speed fluctuation. In addition, PEF has half the number of integration stages N, and it can be seen that the positional deviation increases in proportion to the number N of storage stages, similarly to the deterioration of MTF.
[0025]
Determining the optimum number of integration stages requires the following considerations. First, it is also effective to increase the number of integration stages when the high-frequency vibration is large, while suppressing the number of integration stages when the low-frequency vibration is large. Further, the sensor sensitivity decreases as the inspection wavelength becomes shorter, and the amount of light decreases as the optical magnification increases. Therefore, increasing the number of integration stages is effective for maintaining the necessary sensor sensitivity.
[0026]
In order to solve these problems, the sensor data and the reference data are aligned by pattern matching as disclosed in, for example, Japanese Patent Application Laid-Open No. 2001-5961, “Fine Defect Inspection Apparatus and Method, and Misalignment Calculation Circuit”. Things have been done. However, problems such as (1) indistinguishable from the mask pattern drawing misalignment itself, (2) a large-scale image processing circuit is required, and there is a risk of high costs, etc. are expected.
[0027]
FIG. 1 is a block diagram of a defect inspection apparatus according to the first embodiment of the present invention based on die-to-database comparison. The operation of this configuration will be described below.
[0028]
The sensor data input unit 1 inputs a two-dimensional pattern from the sensor. The position coordinate input unit 2 inputs the position coordinates of the imaging part simultaneously with the sensor data. Further, the position coordinates are averaged by the averaging processing unit 3 over the whole period during which the sensor data is captured at all the pixels in one line.
[0029]
In a conventional defect inspection apparatus using a TDI sensor, if the TDI sensor has N line integration stages, the position coordinates in the N / 2 (or (N + 1) / 2) th line are simply used as the average coordinates. However, as shown in FIG. 2, when the stage speed increases in the latter half of the number of integration stages, a deviation occurs between the average coordinate / X and the correct average coordinate / X ′.
[0030]
In this embodiment, since the X and Y coordinates are measured in all lines (integration stages) and averaged, the above-described divergence does not occur. In addition, since the X and Y coordinates are measured, it is possible to correct not only the speed unevenness but also the yawing (position shift in the direction perpendicular to the stage feed direction).
[0031]
The reference data generation unit 5a inputs the design data of the sample simultaneously with the position coordinates, and generates reference data that should become reference data for the sensor data. A method for generating reference data is disclosed in, for example, registration number 1459387 “Pattern Inspection Device”.
[0032]
Finally, the comparison unit 7 compares the sensor data that has passed through the delay circuit 6 and the reference data to detect a mismatched portion. The delay circuit 6 corrects the calculation delay time in the reference data generator 5a.
[0033]
FIG. 3 is a block diagram of a defect inspection apparatus according to the second embodiment of the present invention based on die-to-die comparison. The die-to-die method is a method of comparing sensor data. The type that compares sensor data captured by two types of optical systems at the same time and the sensor data for the reference pattern are stored prior to inspection. There is a type to compare with sensor data for. In the former type, the speed unevenness mismatch is canceled out, so that the problem of misalignment due to the speed unevenness does not occur. Here, only the latter type will be considered.
[0034]
First, reference data serving as a reference is input. That is, the sensor image data of the reference pattern input from the sensor data input unit 1 and the position coordinates that are simultaneously input from the position coordinate input unit 2 and averaged through the average processing unit 3 are input to the memory unit 8.
[0035]
In the inspection process of the inspection board, the switch SW is switched, the position coordinates input from the position coordinate input unit 2 are input to the reference data generation unit 5b through the average processing unit 5, and further stored in the memory unit 8 above. The reference sensor image data and the position coordinates are input, and reference image data corrected by the difference between the position coordinates is generated. Finally, the comparison unit 7 compares the sensor image data of the inspection board that has passed through the delay circuit 6 and the reference image data to detect a mismatched portion.
[0036]
FIG. 4 is a block diagram of the average processing unit 3 that realizes the average processing of the X and Y coordinates. A digital value representing the X coordinate is input, and the sensor scan clock is transferred as a synchronous clock by the DQ flip-flops QX1 to QXN. The sensor scan clock is generated once every time one line of sensor data starts to be transferred. The number of DQ flip-flops is the same as the number of stages of the TDI sensor.
[0037]
As the average calculation of the X coordinate, the difference between the outputs of the first stage QX1 and the last stage QXN of the DQ flip-flop is calculated for each sensor scan clock using the subtractor UX1, and this is calculated using the DQ flip-flop QX0 and the adder AX1. In this way, the circuit scale is reduced by cumulative addition. Finally, division is performed by the cumulative stage number N using the divider DX1, and the average coordinates are calculated. The calculation of the Y coordinate is performed in the same manner.
[0038]
However, considering the trade-off between position correction accuracy and circuit scale, the number of DQ flip-flops can be reduced if the circuit scale is to be suppressed, and can be increased if position correction precision is to be improved. . The selectors SX1 and SX2 (or SY1 and SY2) select whether the number of stages N is N1 or N2, and are selected by a selection signal.
[0039]
FIG. 5 is a time chart of the coordinate averaging process. The horizontal axis represents time, and the vertical axis represents the operation status of each line sensed by the line sensor. Sensor data is sensor-scanned through N stages of integration (integrated value is output). Position coordinates are input in synchronization with each integration process, and an average value of the coordinates is output simultaneously with the sensor scan.
[0040]
FIG. 6 is a block diagram of the reference data generator 5b for die-to-die comparison. In the die-to-die comparison, as described above, since there are dies in which the same pattern is drawn on one mask, one is used as a reference reference pattern and the other is used as a pattern to be inspected to compare the two. In the following description, the X direction is the continuous movement direction of the stage, and the Y direction is the step movement direction.
[0041]
The X and Y offsets are for setting the relative positions between the two dies of the reference pattern and the pattern to be inspected, but may be used for other purposes such as mask rotation correction and expansion / contraction correction. The reference X and Y coordinates read from the memory unit 8 recorded prior to the inspection and X and Y offsets added by the adders AX2 and AY2, respectively, are used as reference data reference positions.
[0042]
The relative position shift between the reference data and the sensor data can be calculated by subtracting the reference position coordinates of the reference data subjected to the offset correction from the X and Y coordinates for inspection using the subtracters UX2 and UY2. it can.
[0043]
As will be described later, the relative positional deviation is further calculated in units of pixels or smaller sub-pixels, then aligned in the subsequent position correction and pixel alignment unit 11 and transferred to the comparison unit 7. .
[0044]
In the pixel unit alignment, the determination unit 12 compares the position coordinates for inspection and the reference position coordinates of the reference data. If this difference is larger than the pixel size, a read request is issued to the memory unit 8 and the next request is made. This is done by reading position coordinates and sensor data. Sub-pixel alignment is performed by interpolating the reference sensor data and calculating the positional deviation.
[0045]
FIG. 7 is a characteristic graph of the apparatus according to the first embodiment of the present invention. Under the same conditions as in FIG. 9, the vertical axis indicates the PEF when the number of integration stages N is 128, 256, and 512 stages. The maximum value of PEF is 1 or less in any of the integration stages of 128, 256 and 512, and the positional deviation is remarkably small.
[0046]
Further, the difference from the conventional example is that the positional deviation becomes smaller as the number of integration stages of the TDI sensor increases. This indicates that the positional deviation between the reference data and the sensor data due to uneven stage speed can be greatly reduced, and the occurrence of pseudo defects can be suppressed and the practical defect detection sensitivity can be improved. Similar results are obtained in the configuration of the second embodiment.
[0047]
【The invention's effect】
According to the present invention, it is possible to improve the sensitivity of defect detection by correcting the positional deviation caused by speed unevenness or yawing using stage coordinates and creating reference data, and to prevent false detection of pseudo defects.
[Brief description of the drawings]
FIG. 1 is a block diagram of a defect inspection apparatus according to a first embodiment of the present invention. FIG. 2 is a diagram for explaining averaging of position coordinates. FIG. 3 is a defect according to a second embodiment of the present invention. FIG. 4 is a block diagram of a coordinate averaging processing unit. FIG. 5 is a time chart of coordinate averaging processing. FIG. 6 is a block diagram of a reference data generating unit for die-to-die comparison. FIG. FIG. 8 is a graph illustrating the operation of a time delay integration (TDI) sensor. FIG. 9 is a graph showing the relationship between the speed variation of a conventional defect inspection apparatus and PEF. Characteristic graph shown [Explanation of symbols]
DESCRIPTION OF SYMBOLS 1 ... Sensor data input part 2 ... Position coordinate input part 3 ... Average process part 4 ... Design data input part 5a, 5b ... Reference data generation part 6 ... Delay circuit 7 ... Comparison part 8 ... Memory part 11 ... Position correction and image position Matching section 12 ... determiner

Claims (5)

A sensor data input unit that optically reads a two-dimensional pattern on an inspection board and converts it into detection image data using a time delay integration type sensor that is electrical image information;
A position coordinate input unit that reads the coordinates of the reading position of the two-dimensional pattern in synchronization with the optical reading of the two-dimensional pattern;
An average processing unit that averages the coordinates of the reading position by cumulatively adding the difference between the output of the first stage and the last stage of the time delay integration sensor by the number of integration stages and dividing by the number of integration stages ,
A reference data generation unit that generates second reference image data by correcting the position coordinates of the first reference image data based on the coordinates averaged by the averaging processing unit;
A comparison unit that compares the detected image data input from the sensor data input unit with the second reference image data and detects a different location as a defect;
A defect inspection apparatus comprising:   The defect inspection apparatus according to claim 1, further comprising a design data input unit that inputs the design data when the first reference image data is design data. When the first reference image data is image data based on a pattern other than the inspection pattern in the inspection substrate, the first sensor image of the reference pattern is input from the data input unit, and A first position coordinate of the reference pattern that is synchronously input from the position coordinate input unit is input after being averaged by the average processing unit, and further storing a memory unit;
When the second sensor image of the inspection pattern is input to the sensor data input unit, the second of the inspection pattern that is input from the position coordinate input unit and averaged by the average processing unit in synchronization therewith. Position coordinates are input to the reference data generation unit together with the first sensor image of the reference pattern and the first position coordinates stored in the memory unit, and a difference between the second and first position coordinates is obtained. The defect inspection apparatus according to claim 1, wherein the corrected second reference image data is generated and compared with the second sensor image of the inspection pattern in the comparison unit.   The sensor data input unit increases the number of integration stages of the time delay integration when the substrate moving speed at the time delay integration fluctuates at a high frequency, and when the substrate moving speed fluctuates at a low frequency, In order to suppress the number of integration stages of the time delay integration and maintain the sensitivity of the sensor, there is provided means for making the number of integration stages variable so as to increase the number of integration stages of the time delay integration in accordance with a change in light amount. The defect inspection apparatus according to claim 1. A two-dimensional pattern on the inspection board is optically read using a time delay integration type sensor, and converted into detection image data which is electrical image information.
In synchronization with optical reading of the two-dimensional pattern, the coordinates of the reading position of the two-dimensional pattern are read,
The coordinates of the reading position are averaged by cumulatively adding the difference between the output of the first stage and the last stage of the time delay integration sensor by the number of integration stages and dividing by the number of integration stages ,
Correcting the position coordinates of the first reference image data according to the averaged coordinates to generate second reference image data;
A defect inspection method comprising: comparing the detected image data input from the sensor data input unit with the second reference image data to detect a different portion as a defect.

Images