JPH05281155A - Inspection apparatus for defect of pattern - Google Patents

Abstract

PURPOSE:To obtain a pattern defect inspection apparatus, which can returns the measured signal, wherein diffuseness is generated by the passing of the signal through a measuring optical system, into the signal, which does not contain any diffuseness and can detect the defect of the pattern by using this signal. CONSTITUTION:The measured pattern data, which are read out of a line sensor 36, are introduced into an equalizing circuit 39 through an A/D converter 37 and a buffer circuit 38. The equalizing circuit 39 has a digital filter, which processes the input signal in two dimensions and removes diffuseness. The measured data, which are processed in the equalizing circuit 39, are introduced into a comparing and judging circuit 43 together with the design data.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an inspection apparatus for inspecting a reticle pattern used for manufacturing a semiconductor integrated circuit for defects.

[0002]

2. Description of the Related Art One of the major causes of a decrease in yield in manufacturing a large scale integrated circuit (LSI) is a defect of a mask used in manufacturing by a photolithography technique. Since a mask defect causes a great deal of damage, it is necessary to perform a highly accurate inspection in advance.

As the demands on inspections increase,
The accuracy of the inspection is subject to the possible physical limitations of the optical device. In other words, the optical system including the light source, the lens, and the image pickup device used must have a sufficiently wide spatial frequency band in order to be able to reliably detect a very small defect by the inspection device.

FIG. 12 shows the principle of a conventional inspection device.

Observation light emitted from the light source 1 is imaged by the lenses 2 and 3 on the pattern 5 to be inspected on the reticle mask 4 and on the line sensor 6 composed of a CCD.
Now, suppose that the pattern 5 to be inspected is, for example, the letter "A" as shown in the drawing, and the defect 7 is located immediately to the right of it.

The reticle mask 4 is scanned by an XY table (not shown) as it advances in a direction perpendicular to the direction of the broken line arrow, and this scan is used to inspect the entire pattern. The line sensor 6 is configured by arranging hundreds to thousands of photo arrays, and its output signal passes through the A / D converter 8 and is given to the comparison / determination circuit 9 as multivalued measurement data. ..

On the other hand, original design data is stored in the disk 10 or magnetic tape of the data management computer. Its form is, for example, as an element figure based on a trapezoid.
"A" is completed. This pattern is read by the data expansion circuit 11 and converted from black and white binary data into multivalued design data corresponding to the expected value of the sensor output.

The comparison / determination circuit 9 uses the measured data and the design data.
Data and outputs the result. In the above case, since the pattern 5 to be inspected has a defect 7, the information is recorded in a memory or the like (not shown) and displayed on the monitor 12.

In the device thus constructed, the light source 1,
In the image pickup optical system including the lenses 2 and 3 and the line sensor 6, image deterioration, that is, a "blurring" phenomenon occurs due to the influence of the numerical aperture of the lens itself, the characteristics of the CCD line sensor, and the like.

Of these, the MTF of the CCD line sensor
(Modulation Transfer Function) characteristics are as follows:
Transfer efficiency, (2) degree of lateral diffusion of optical signal carriers generated inside the substrate, (3) arrangement of photoelectric conversion unit elements (pixels) and aperture width.

The CCD line sensor is shown in FIG.
As shown in (a), the storage electrode 20 and the transfer electrode 21 are provided, but the photoexcited carriers also flow in from a position other than the position directly above the storage electrode 20. Therefore, its sensitivity distribution is shown in Fig. 1.
It becomes as shown in 3 (b). For example, when the dimensions are defined as shown in FIG. 13 (c), the aperture characteristic of the sensor is given by the following equation.

F (ω) = A (d ₁ + d ₂ ) · [{sin ω (d ₁ + d ₂ ) / 2} / {ω (d ₁ + d ₂ ) / 2}] ・ [{sin ω (d ₁ −d ₂ ) / 2} / {ω (d ₁ −d ₂ ) / 2}] FIG. 13 (c) shows the transfer characteristics of the trapezoidal aperture.

When the amount of light incident on the line sensor is abundant, the S / N ratio of the sensor is improved and the accumulated exposure time can be shortened, so that the inspection speed can be increased. On the other hand, the higher the magnification of the lenses 2 and 3, the higher the resolution, and the easier it is to find a defect. However, on the other hand, the amount of light per pixel of the line sensor decreases, and the S / N ratio deteriorates. Therefore, while maintaining a certain resolution, the S / N of the sensor
There is a trade-off relationship in which the light amount is determined in consideration of the ratio.

When the lens numerical aperture (NA) is increased, the optical band is widened, but there is a limit to increase the NA while ensuring the characteristic of the depth of focus to some extent. Further, although the optical band can be widened even if the wavelength of light is shortened, the sensor sensitivity is attenuated in the short wavelength range, so that the usable wavelength is limited.

For the above reasons, the conventional inspection apparatus treats the measurement data acquired from the measurement optical system as including "blurring". That is, a two-dimensional blurring function is derived by simulating deterioration caused by "blurring" due to the lens MTF, and design data that is not deteriorated by "blurring" and the two-dimensional blurring function are convoluted, The expected value of the sensor output is calculated in anticipation of the deterioration of the measurement data.

In such a conventional inspection apparatus, the measurement data is
Since the data itself is blurred, a minute defect that should be pointed out as an original defect may be overlooked. To avoid this overlooking, the load of the defect determination algorithm equipped in the comparison determination circuit 9 becomes large, and the device becomes complicated. There was a problem to do.

Further, with respect to the measuring optical system, the design for simultaneously realizing high magnification and high NA becomes complicated, and there is a problem to be improved.

Further, in the line sensor 6 composed of CCD, there is interference between adjacent pixels. Therefore, in reality, the blurring characteristics of the inspected pattern in the x direction and the y direction are not equal. That is, although mutual interference acts in the direction of adjacent pixels (x direction) of the sensor array, the sensor movement direction (y
In the direction), the data is observed at different sensor scan cycles, so the interference component is a small amount of the residual capacitor of the last scan.

Further, inspection is performed by changing the observation magnification at the time of measurement or changing the table traveling speed. In this case as well, the aperture characteristics, interference between adjacent pixels, etc. are x, y. The direction is affected separately. With conventional inspection equipment,
This asymmetry is not taken into consideration, and when calculating the expected value of the sensor output from the design data, the configuration that handles the blur calculation as point symmetric data suitable for the circuit configuration is adopted. There is a phenomenon that the level of defects that can be detected differs between the x edge and the y edge of the pattern to be inspected.

[0020]

Therefore, according to the present invention, it is possible to restore the image data in a substantially undegraded form in response to the deterioration of the image in the measuring means, thereby improving the resolution at the time of measurement and performing inspection. It is an object of the present invention to provide a pattern defect inspection device capable of improving accuracy.

[0021]

In the pattern inspection apparatus according to the present invention, the measurement data of a pattern to be inspected, which may contain fine defect data and is deteriorated by the measurement optical system, is recorded in a two-dimensional digital format. The asymmetry in the x and y directions is also corrected by using a filter, and it is returned to a substantially undegraded form before being used for comparison / judgment.

[0022]

In the pattern defect inspection apparatus according to the present invention, signals including "blurring" that have passed through the measurement optical system are corrected and compared, so that the transfer characteristic of the measurement optical system can be equivalently widened. As a result, it is possible to detect minute defects existing in the pattern to be inspected.

[0023]

1 is a schematic view of a pattern defect inspection apparatus according to an embodiment of the present invention.

In this inspection apparatus, the observation light emitted from the light source 31 is first passed through the lenses 32 and 33 to the reticle mask 3.
An image is formed on the pattern 35 to be inspected and the line sensor 36. The reticle mask 34 is scanned in a two-dimensional direction with an XY table (not shown), and the pattern 3 to be inspected is obtained.
A full inspection of 5 is performed.

The line sensor 36 is composed of a CCD line sensor in which hundreds to thousands of photosensors are arranged in a line. Hereinafter, the number of arrays will be described as s. In order to speed up the signal extraction from the line sensor 36, a signal extraction port is prepared for each of several tens of sensor array groups, and the data for one line is extracted in parallel to the A / D converter 37. To be done. The A / D converter 37 also has a number of converters corresponding to the parallel number.

The output of the A / D converter 37 is given to the buffer circuit 38. The buffer circuit 38 inputs in the above parallel number and rearranges the data for one line of the line sensor 36.

The data rearranged by the buffer circuit 38 into the data for one line s is input to the equalization circuit 39. The equalization circuit 39 performs a weighted product-sum operation on the input, and outputs the operation result in a pipeline after a predetermined processing delay time.

On the other hand, the design data used when forming the pattern to be inspected 35 is recorded in the magnetic disk device or the magnetic tape in the data management computer 40, and this design data is inspected. Data read out in accordance with the inspection of the pattern 35, expanded into binary pattern data by the data expansion circuit 41, and then subjected to blur correction by the point spread function unit 42 to show the expected value of the sensor output value. Is output as.

The design data corrected through the point spread function unit 42 and the measurement data output from the equalization circuit 39 are introduced into the comparison / determination circuit 43. The comparison / determination circuit 43 compares and collates both data and determines the presence / absence of a defect on the inspected pattern 35.

The equalization circuit 39 is a two-dimensional FIR (Finite I
mpulse Response) with a built-in filter, and is specifically configured as shown in FIG.

FIG. 2 shows a two-dimensional FIR having a coefficient distribution of 3 × 3.
It is shown. In the figure, Z ^-1 _x is a delay latch for one clock, and Z ^-1 _y is a row delay register for applying data delay for one line of the line sensor 36.

In the buffer circuit 38, 1 of the line sensor 36
The data 101 arranged in line is input to the first column processing unit 112 and the row delay register 110. In the column processing unit 112, first, the data 101
Are simultaneously input to the integrating circuits 140 to 142,
The weighting coefficient preset for each pixel position is multiplied. Next, the respective integrated outputs are added by the addition circuit 130-
A summation operation is performed by a cumulative addition circuit including 132 and delay latches 120 to 122. One of the adder circuits 130 is held at "0" in order to give the initial value "0" of cumulative addition.

In the second column processing unit 113, the data delayed by the row delay register 110, ie, the first data
The data 102 that is exactly one line past is input from the input data to the column processing unit 112. The input data is subjected to multiply-accumulate operation in the same manner as in the first column processing unit 112, but one of the adders 133 at the head of the second column processing unit 113 receives the second data as a cascade input. The output 104 of the first column processing unit 112 is input.

Due to the function of the second row delay register 111, the third column processing unit 114 receives data one line past from the input data to the second column processing unit 113. That is, 2 from the first line
The past data for the line is input. The input data is multiplied by the weighting coefficient for each pixel position as in the second column processing unit 113.

Then, the third column processing unit 114 outputs the cumulative addition result 107 to the output result 105 of the second column processing unit 113. The final output 107 of this cumulative addition is output to the comparison and determination circuit 43 as the processing result of the equalization circuit 39.

As described above, in this embodiment, the measurement data measured by the line sensor 36 is processed by the 3 × 3 two-dimensional FIR filter, and the processed data is introduced into the comparison / determination circuit 43. There is. Therefore, the line sensor 36
From the time when the "pixel data at a certain position" of the measurement data measured in step 6 is input to the input terminal of the equalization circuit 39 via the buffer circuit 38, the conversion result by the filter of the pixel is equalized by the equalization circuit 39. The delay time to reach the output end of (2
It is a period of (line + 2 clocks) minutes.

The filter having the configuration shown in FIG. 2 can be considered as a concept in which the window 50 for convolutionally calculating the weighting coefficient moves in the two-dimensional spread of the measurement data as shown in FIG. FIG. 3 also shows a case where the coefficient distribution is 3 × 3, corresponding to FIG.

The line sensor 36 advances downward in the figure. The position of the line sensor 36 and the position of the window 50 do not match because the buffer circuit 38 is interposed. At present, the window 50 is at the position shown by the solid line, and the window moves by one address in the arrow direction at the next clock. After that,
Similarly, the window 50 horizontally moves by one address for each clock. At this time, the data 101 which reaches the beginning of the first line of the window 50 corresponds to FIG. The data 51 supplied to the second row of the window 50 corresponds to the output of the delay latch 110 shown in FIG.
The data 52 supplied to the third row of 0 corresponds to the output of the delay latch 111 in FIG.

In FIG. 3, the data processing window 5
0 advances to the position indicated by the dotted line after a certain clock time.
At this time, the data captured in the data processing window 50
The data will be the data for the first pixel of the new line. That is, it is the position of the window 50 '.

The number of elements in the horizontal direction of the window 50 shown in FIG. 3 corresponds to the number of arrays of the line sensor 36. When the number of arrays is 20, the delay latches 110 and 11 shown in FIG.
The internal delay amount of 1 is also 20 steps.

The horizontal direction of the window 50 shown in FIG. 3 has a length dimension whose element unit is the size of the array of the line sensor 36, and is also handled in the unit of time because it is sampled at a constant sensor clock frequency. be able to. The lengthwise direction of the window 50 corresponds to the length dimension and the time dimension of the acquisition cycle when the table on which the reticle 34 is placed is moved relative to the line sensor 36 at a constant speed to acquire one line of data. Either can be handled.

The "blurring" caused by the measuring means is the loss of the signal in the high frequency region at the spatial frequency.
The standardized transfer characteristics in dimensions are as shown by a in FIG.
It has the characteristics of a pass filter. By multiplying the characteristic indicated by b in FIG. 4 by the equalization circuit 39 as the characteristic opposite to the characteristic indicated by a, the characteristic indicated by the broken line c is obtained.

By performing inverse Fourier transform on the spatial frequency characteristic shown by b in FIG. 4, the characteristic on the time axis as shown in FIG. 5 can be obtained. Although it is shown in one dimension here, conversion for two-dimensional processing is performed based on this design, and the equalization circuit 3
Weighting is set in the nine integrating circuits 140 to 143. The method of designing by decomposing these two-dimensional required characteristics into one dimension is described in "The Institute of Electronics, Information and Communication Engineers, A, Vol, J72-A, No. 9, p.
p.1392-1399, September 1989 ”, etc.,
I will not go into detail here. Here, for the sake of convenience, the characteristics shown in FIGS. 4 to 5 have been described without limiting the x direction and the y direction. However, as described above, in the actual inspection apparatus, “blurring” occurs in the scanning direction and the moving direction of the line sensor 36. , The weighting coefficient is also xy.
It is necessary to define each pixel position in the two-dimensional direction.

In this embodiment, the coefficient data preset in the integrating circuits 140 to 143 of the equalizing circuit 39 is stored in the coefficient table memory 59 as shown in FIG. From the coefficient table memory 59, the product-sum operation circuits 112, 113, 11
4 has been adopted. The coefficient table memory 59 is provided with a plurality of sets of weighting coefficients.
A weighting coefficient set is appropriately selected according to the state of the reticle to be inspected (type of resist, exposure amount, inspection speed, etc.) and set in the coefficient table in the product-sum calculation circuit.

For example, in the coefficient table memory 59,
Coefficients for realizing the characteristics a to c in FIG. 7 are prepared. Normally, the characteristic indicated by b in FIG. 7 is used, but when the finish condition of the reticle to be inspected is bad, the correction characteristic indicated by a in FIG. 7 is used to correct it. The command for changing the coefficient set is performed by receiving a command from the host control computer that controls the entire inspection apparatus via the interface 61. This upper control computer can also rewrite the contents of the coefficient table memory 59. In this case, a predetermined command and data are transmitted via the interface 61. Figure 6
The ROM / RAM 62 stores the processing program of the microprocessor 60. In addition, 63, 6 in FIG.
Reference numeral 4 indicates a data bus.

The data read from the line sensor 36 is superposed with an offset corresponding to the dark current of the sensor and electrical noise generated in the circuit. These noises can be attenuated to a certain level that does not affect the inspection threshold value by examining the filter characteristics of the equalization circuit 39. However, as shown in FIG. 8, a clamp circuit 65 is provided as shown in FIG. 9 so that the output is zero when the input data is below a certain threshold value K, and the noise of the signal is cut in advance. By performing a predetermined equalization operation from, the observation optical system can be compensated more effectively. The threshold value K itself may be modified by an instruction from the host control computer, like the weighting coefficient described above.

In general, in a CCD line sensor having a large number of pixels, in order to speed up signal extraction, for example, the pixels are classified into even-numbered pixels and odd-numbered pixels, and the signals are extracted in parallel. In this case, the buffer circuit 38 shown in FIG.
Is replaced with the buffer circuit 38a shown in FIG. 10, receives the parallel output of the line sensor 36, and re-edits it as parallel data for each sensor line.

In FIG. 10, for convenience of explanation, the A / D converter 3 is used.
7 is omitted. Assuming that the sensor output is an even / odd parallel extraction, even-numbered pixel data is input to the input buffer 66, and odd-numbered pixel data is input to the input buffer 67. The number of clocks required for input from the line sensor 36 to the input buffers 66 and 67 is s / 2 clocks, where s is the number of sensor pixels. The data stored in the input buffers 66 and 67 is input to the output buffer 69 via the data selector 68. At this time,
The data selector 68 uses the input buffer 6 for each even / odd number.
6, 67 are alternately selected to reproduce the array of sensor pixels. The number of clocks required to set data in the output buffer 69 is s clocks.

The line sensor 36 moves (actually, the reticle mask 34 moves), the data of the next line is input to the input buffers 70 and 71, and the data is transferred to the second output buffer 73 via the data selector 72. After waiting for the data to be set, the output buffers 69 and 73 output the data simultaneously in parallel. According to this procedure, the number of clocks required to fetch data from the line sensor 36 to the input buffers 66, 67, 70, 71 is (s / 2) × 2 lines = s [clock] The number of clocks required to output from 69 and 73 is s clocks, and the input / output speeds are balanced. However, since the buffer cannot be input / output at the same time, specifically, the buffer has two systems as described above, and the input / output is performed alternately.

FIG. 11 shows a two-dimensional filter means corresponding to the buffer circuit 38a shown in FIG. 10, that is, an equalization circuit 39a. The two-dimensional digital filter means is configured to receive the output of the parallel-lined buffer circuit 38a, perform a predetermined calculation, and output the parallel output again. The row delay means 74 and the sum-of-products calculation means 75 are prepared for each parallel configuration.

In the examples shown in FIGS. 10 and 11, the two parallel configuration is adopted, but by configuring with a larger number of parallels, equivalently higher speed processing can be realized.

As a method of designing the weighting coefficient that satisfies these various correction items, any general method such as the window function method can be adopted, and any result obtained is effective for the inspection apparatus of the present invention. In the embodiment, the case where the configuration of the weighting coefficient is 3 × 3 or 5 × 5 has been described, but the present invention can be applied as long as the number of configurations is 3 or more. In this case, the two-dimensional x
The number of components in the direction may be different from that in the y direction.

[0053]

As described above, according to the present invention, signals including "blurring" due to passing through the measurement optical system are corrected and compared, so that the transfer characteristics of the measurement optical system are equivalently wideband. Can be converted. As a result, it is possible to detect fine defects existing in a fine pattern to be inspected.

[Brief description of drawings]

FIG. 1 is a schematic configuration diagram of a pattern defect inspection apparatus according to an embodiment of the present invention,

FIG. 2 is a block configuration diagram of a product-sum operation circuit in an equalization circuit incorporated in the device,

FIG. 3 is a conceptual diagram explaining two-dimensional data processing,

FIG. 4 is a frequency characteristic diagram for explaining a characteristic to be corrected by an equalization circuit,

FIG. 5 is a diagram showing an example of filter characteristics preset in the equalization circuit;

FIG. 6 is a block diagram showing a hardware configuration of an equalization circuit,

FIG. 7 is a frequency characteristic diagram for explaining a filter characteristic prepared in the memory.

FIG. 8 is an input / output characteristic diagram of a clamp circuit,

FIG. 9 is a block configuration diagram of an equalization circuit in a pattern defect inspection apparatus according to another embodiment of the present invention,

FIG. 10 is a block diagram for explaining a modification of the buffer circuit,

11 is a block diagram of an equalization circuit suitable for combination with the buffer circuit shown in FIG.

FIG. 12 is a diagram for explaining a conventional pattern defect inspection inspection;

FIG. 13 is a diagram for explaining aperture characteristics of a CCD line sensor.

[Explanation of symbols]

31 ... Light source 32, 33 ... Lens 34 ... Reticle mask 35 ... Inspected pattern 36 ... Line sensor 37 ... A / D converter 38, 38a ... Buffer circuit 39, 39a ... Equalization circuit 40 ... Data management computer 41 ... Data expansion circuit 42 ... Point spread function unit 43 ... Comparison determination circuit 59 ... Coefficient table memory 60 ... Microprocessor 101 ... Equalization circuit input 107 ... Equalization circuit output 110, 111 ... Line delay registers 120-127 ... Delay Latches 130 to 138 ... Addition circuit 140 to 148 ... Weighted integration circuit

─────────────────────────────────────────────────── ─── Continuation of the front page (51) Int.Cl. ⁵ Identification number Office reference number FI technical display location G06F 15/68 400 A 9191-5L H01L 21/027

Claims (2)

[Claims] 1. A measurement pattern data acquisition unit including a line sensor for irradiating a sample on which a pattern is formed with light and receiving an optical image of the pattern to perform photoelectric conversion, and when forming a pattern on the sample. A storage unit that stores the used pattern design data, a bit expansion unit that expands the pattern design data read from the storage unit into bit data, and a filter process is applied to the bit data expanded by this unit. In a pattern defect inspecting apparatus equipped with a determining unit for comparing the obtained data with the measured pattern data to determine the presence / absence of a defect in the pattern formed on the sample, a pattern defect inspection apparatus is produced in the system for obtaining the measured pattern data. A two-dimensional digital filter means for correcting image deterioration is provided, and the two-dimensional digital filter means has m stages in a column direction and a row direction. A finite impulse response type of n stages (m and n are each an odd number of 3 or more), (n-1) row delay means for applying data delay for one line of the line sensor, and each row are provided. There are provided n number of integrating means for integrating the input data and the weighting coefficient characterized for each pixel position, and m adding means for cumulatively adding the output of each integrating means by delaying the clock. It is characterized by comprising a column processing unit and means for cascading the output of the final stage adder of the column processing unit located in the previous stage to the first stage adder of the column processing unit located in the next stage. Pattern defect inspection system. 2. The two-dimensional digital filter means has the constituent elements of m stages in the column direction and n stages in the row direction, and is provided with a plurality of systems for simultaneously performing product-sum operations in parallel. A pattern defect inspection apparatus characterized by: