JPS62237307A - Dimension measuring instrument - Google Patents

Abstract

PURPOSE:To automatically measure the dimensions of a micro-pattern formed on a semiconductor wafer with high accuracy by using a scan type electron microscope (SEM). CONSTITUTION:A dimension measurement part 2 is electrically connected with a SEM main body part 1. Then, the SEM main body part 1 scans the electron beam flux 4 on the measured data forming a pattern whose dimensions are to be measured and outputs an image indicating the pattern and displays it on a CRT 14. Next, a picture signal is converted into A/D converted 17 and the image data are stored in an image memory part 18 for display and an image memory part 19 for measuring the length. These image data are processed differentially and the image data are divided into plural areas with a smoothing memory part 22 a based on a differentiated result and the weighting is differed for every area and the smoothing process is performed separately respectively. Then, an arithmetic processing part 22 calculates the distance between the edge parts based on the smoothed image data. In this way, the dimensions of the micro-pattern formed on the semiconductor wafer can be automatically measured with high accuracy.

Description

DETAILED DESCRIPTION OF THE INVENTION [Object of the Invention] (Field of Oncological Application) The present invention relates to a dimension measuring device that automatically measures the dimension of a minute pattern formed on, for example, a semiconductor wafer.

(Prior Art) Conventionally, methods for measuring the pattern width of semiconductor wafers include a micrometer using an optical microscope and an industrial television set (I).
TV) Electronic measuring device that combines a camera and an optical microscope,
In most cases, conventional methods have been used, such as optically enlarging the image using a measuring device that combines laser reflected light and a precision moving stage, or reducing the beam diameter to improve resolution. Furthermore, by applying a scale to the enlarged image obtained using a scanning electron microscope and measuring the dimensions by converting it from the magnification at that time, or by dividing the image into multiple pixels and generating a cursor on one image, There was a method in which the measurer placed a cursor on the pattern edge and obtained the dimensions from the number of pixels between the cursors and the magnification. However, recently.

As LSIs and VLSIs become more highly integrated, patterns are becoming finer and more precise, and in response to this, pattern width measuring instruments are also required to have a resolution of 0.1 μm or less.

However, conventional optical means have limitations in terms of magnification.
Its resolution is also about 1/4 of the wavelength, and it is impossible to obtain a resolution of 0.1 μm or less. Also.

In a measurement method that detects edges using laser reflected light, if the cross-sectional shape of the pattern changes (such as a difference between the shape of the resist and the shape after etching), the measurement results will vary, making it impossible to measure with high precision. moreover.

When using a scanning 1D electron microscope, the adjustment of the magnification is insufficient, when measuring with a scale there is a reading error, and when the cursor is placed on a pattern edge, there are variations in the way the cursor is placed on the edge of the pattern. This has made it difficult to measure with high precision.

Therefore, we connected an electric circuit that allows the electron microscope to measure dimensions, and based on the secondary electron signal of the pattern output from the electron microscope, we determined that the level of the secondary electron signal is different between the inside and outside of the pattern. There was a method of approximating the image signal inside the edge of the pattern with a straight line, finding the average value of the outside image signal, and finding the edge of the pattern from the intersection of these to measure the pattern dimensions. for example.

Japanese Patent Application No. 1988-88455 (Specification # Diagnosis).

However, the above image has large noise and cannot be used as is for dimension measurement.

Therefore, conventionally, noise removal has been performed using an averaging method, a Fourier transform method, a smoothing method, or the like. but,
The averaging method described above repeatedly irradiates the same location on the sample with an electron beam, adds secondary electron signals, and cancels and removes noise, which may cause damage to the sample.

Furthermore, the Fourier transform method has the disadvantage that important components of the waveform may be erased depending on the setting of the cutting frequency. Furthermore, the smoothing method averages out the entire image in the same way.

It has the disadvantage that the distinction between the pattern part and the base part is unclear.

(Problems to be Solved by the Invention) The present invention has been made in consideration of the above circumstances, and is based on a scanning electron microscope (hereinafter referred to as SEM).
It is called Electron Microscope. ), it is an object of the present invention to provide a dimension measuring device that can automatically and highly accurately measure the dimensions of a minute pattern formed on a semiconductor wafer, for example.

[Structure of the invention]

(Means and effects for solving the problem) A dimension measuring section is electrically connected to the SEM main body, and the dimension measuring section
Based on the image signal output from the M main body section and indicating the pattern whose dimensions are to be measured.

Using the fact that the level of the zero image signal changes between the inside and outside of the edge of the pattern, the edge indicating the outline of the pattern is determined, and the intersection point between the two is determined. In a system that automatically calculates the distance between multiple edges, the image signal representing the pattern is divided into multiple regions based on the rate of change, and each region is smoothed with different weights. It is designed to remove noise.

(Example) The present invention will be described below in detail based on an example with reference to the drawings.

FIG. 1 is a block diagram of a dimension measuring device according to one embodiment. This dimensional double-legged device is suitable for scanning electron microscopy (5cannin
II Electron Microscope; hereinafter simply abbreviated as SgM. ) main body g (11) and a dimension measuring section (2) that measures the dimensions of a specific part captured by this 8HM main body section (1). Outputs an electron gun (3) that emits electrons from a power source 1, a condenser lens (5) that reduces the electron beam flux (4) emitted from this electron gun (31), and the Croffico PS that serves as a reference. and a sweep signal generation unit that generates a sweep signal SS for scanning the electron beam (4) based on the clock signal PS output from the reference signal generation unit (6). Control compensation +4F is applied to the scanning coil section (8), which will be described later, in combination with the sweep signal SS output from the sweep signal generating section (force) by setting a magnification changeover switch (not shown).
A magnification switching unit (9) that outputs C8I and the control signal C
a scanning coil unit (8) that controls the scanning direction and width of the electron beam flux (4) based on 8I, and an objective lens aυ that further reduces the electron beam flux (4) and irradiates the electron beam flux (4) onto the measurement sample αQ. , a secondary electron detector αda that collects secondary electrons emitted from the measurement sample αω, and this secondary electron detector (1
an amplifying section a3 that amplifies the signal from the amplifying section Q3;
The image signal Is output from the CRT (C
athode Ray Tube) (1 and 4 image signal amplifiers for displaying images.

The CRT 04 displays an enlarged image of a specific part of the measurement sample αl held on a mounting table (not shown). On the other hand, the 1-dimensional measurement section (2) includes a timing control section uQ that outputs a control signal ST based on a clock signal PS from a reference signal generation section (6) and a sweep signal SS from a sweep signal generation section (force); An analog-to-digital (A/D) converter qη that converts the image signal Is into a digital value using the control signal ST output from the timing control unit (1e), and a digital value output from the ~■ conversion unit (17>). The display image memory section α stores the control signal 5I) output from the timing control section ae, and the control signal output from the timing control section αe stores the digital value output from the A/D conversion section η. A length measurement image memory section (11) stores data based on the signal 8M.

CRT (14) A graphic image memory section for displaying multiple cursors superimposed on an image from the SEM main body section (1) or a still image based on the image data stored in the display image memory section αυ. ), a cursor setting section (2η) that is connected to the input side of this graphic image memory section and which moves the cursor in various ways, and an image memory section for length measurement and a graphic image memory section (support).
A CPU section T24 that reads out data from the computer and has various arithmetic processing functions and storage functions, a smoothing coefficient memory section (22a) in which values of a smoothing coefficient table described later are stored, and a length measurement image memory section ■1 Display image memory section or graphic image memory section A display control section (c) performs various controls for displaying data on a CRT<14), and a length measurement image memory section (II1 display image memory section α mark). , is connected to the output side of the graphic image memory section and the display control section C23 and converts the digital data into analog data.
Q4) for output 8 display digital-analog (
D/A) It consists of a converting section (goods). The display control unit Q outputs a control signal to display the image data stored in the 1-display image memory unit αQ on the 1-image signal amplifier (CRT Q4, which is also connected to Li2S). There is.

Next, the operation of the dimension measuring device configured as described above will be described in detail.

First, a measurement sample (1) such as a semiconductor wafer on which a pattern such as L8I is formed is placed on the mounting table of the SEM main body (1).
Place υ. Thus, the electron beam flux (4) emitted from the electron gun (3) is reduced by the condenser lens (5)... ) is scanned in the X-Y direction, further reduced by the objective lens αυ, and irradiated onto the measurement sample (1G). Secondary electrons are then emitted from the measurement sample α [surface. The secondary electrons are collected by the secondary electron detector α and converted into an electric signal.The electric signal output from the secondary electron detector α is amplified by the amplifier (13) and one image signal Is This image signal amplifier (I!) is output to the image signal amplifier (151) as
19, the sweep signal SS outputted from the sweep signal generator (7) and the image signal Is are combined and sent to the CRT (
14 to display it as an image. On the other hand, one image signal Is is
The signal ST output from the timing control unit (161) is A/D converted by the A/D conversion unit αη, and first
The image for one screen is stored in the display image memory section α mark. The display image memory section is 9. For example, as shown in FIG. 2, 5
It has a capacity of approximately 12 x 512 x 8 bits. The stored data is transmitted through the signals MC and M of the display control section.
By D, D/A converter c! Then, it is converted into an analog signal, and the control signal is sent to the CRT (14)
displayed as a still image. Next, operate the cursor setting section CD for the pattern P that is displayed, and select the signal K.
Output S to the ca'r (14) via the graphic image memory section (a) and the D/A conversion section 041, and set the putter 7P to be sandwiched between the two cursors (25a) and (25b). (See Figure 3). At this time, the cursor (25
a).

The position of (25b) is 8 in response to the signal output from the cursor setting section, and is written to the graphic image memo IJ g (2) now, and is displayed completely overlapping with the optical display image memory section α barrel. . On the other hand, the CPU section (22 is a cursor (25a), (
25b) and sends the address data to the sweep signal generator (7) using the signal SA. The sweep signal generating section (7) is as follows: car: / # (25a),
Only the area specified in (25b) is scanned with the electron beam flux (4). Same thing this time.

The image signal Is is converted into a digital value by the A/D converter σD and stored in the length measurement image memory unit a. In this image memory unit u9 for length measurement, for the - scanning line.

Display image memory section (stores data at a density several to several tens of times higher than that of one barrel. For example, per one scanning line.

Collect about 2048 points. However, data is not collected over the entire screen; Kano/I/(25a), (
25b). However,
The image data of the part specified with the cursors (25a) and (25b) is stored in the length measurement image memory section (lLj) in sequence.
se) If the signal has a poor ratio, fftg processing is performed (block Ce in FIG. 4).

□□□). This is a method in which the same location is repeatedly scanned, data at the same point is sequentially added, and finally, the data is normalized by dividing by the number of additions.

On the other hand, for signals with a good S/N ratio, the integration process is not performed and sampling is completed once (block (2f9) in Figure 4).Next, the obtained waveform W (see Figure 5)
Perform noise removal on. In addition.

The actual waveform W is much more disordered due to noise, as shown in FIG. Smoothing processing is usually performed on such waveforms, but what is often used is Golay's smoothing method ("Analytical Chemistry").
J Volume 36, No. 8.

See pages 1627-1639. ). but.

In this method, all areas are uniformly smoothed to the same degree, so that both the required steep rise waveform and the flat, relatively slow undulation waveform are processed in the same way. Therefore, the waveform W shown in FIG. 5(A) is differentiated to obtain a differential waveform WD as shown in FIG. 5(B) (block 041 in FIG. 4). Next, from positive to negative,
Further, inflection points PII to PI6 from negative to positive are determined.

These inflection points PII to PI6 are points at which the one-dimensional waveform W changes significantly. Next, the inflection points PII to PI6
The waveform penalty is divided into seven regions, DA, ~, and DG, with . Then, for each area DA, to, DG, the number of pixels is different.

Smoothing is performed using the method of 5avitzky & Golay (@block country in Figure 4). For example, area D
A, DG are 17 pixels, area DB, DC, DE, D
Smoothing is performed using 5 pixels for F and 11 pixels for area DD. Incidentally, FIG. 7(B) shows the waveform of FIG. 7(A) subjected to 5-pixel smoothing processing. In other words, for areas DB, DC, DB, and DF that rise sharply, it is better not to apply smoothing too strongly (reduce the number of pixels), so the number of pixels should be reduced.On the other hand, for base areas DA and DG, Stronger smoothing (increasing the number of pixels) is preferable because a flat portion can be obtained. In this case, for example.

The calculation formula for smoothing regarding point D4 in FIG. 7(A) is expressed by the following formula.

However, W1=-3, W2=12. W3=17.
W4=12. W5=-3°No.M=35. In other words, in this case, when the number of pixels is 5, the smoothing coefficient NORM =
35, this smoothing coefficient is stored in the aforementioned smoothing coefficient memory section (22a). In short, for steep waveforms, try to keep the original shape as much as possible, and for flat noisy signals, make them smooth.

Smoothing each area separately makes it easier to measure dimensions later. Note that FIG. 8 shows a waveform obtained by subjecting the waveform of FIG. 6 to the smoothing process of this embodiment. Then, for the waveform from which the noise has been removed, as shown in Figure 9, a determination area whose dimensions are to be measured is specified using the cursors (25a) and (25b) (block (to) in Figure 4). The discrimination area can be identified by the fact that the voltage value of the waveform corresponding to the one-dimensional measurement area, that is, the pattern area (region P in FIG. 3) is larger than that of other areas. Then, within the discrimination area, the maximum value l! at one side edge shown in FIG. 51J and the minimum value 6 are determined (block 041 in FIG. 4). Therefore, any 2 υ between the maximum value 6υ and the minimum value 6
The point fi, 57) is selected and the range for linear approximation is specified (block c39 in Figure 4). Next, these two points (to)
, 69, regression line 6 is calculated using the least squares method.
Find 9 (block (to) in Figure 4). Furthermore, the data between the minimum value 6a and the point 6D, that is, the data of the flat portion, is approximated to a quadratic curve 6υ by the least squares method (block C37 in FIG. 4). Next, the intersection point 53) of the 1 regression line 61 and the quadratic curve 6υ is determined. Similarly, the regression line σ at the other side edge and the quadratic curve at the base (
c) and calculate their intersection σe (block C38 in Figure 4). The position of the above intersection −, σe is CBT 0
4) Display the trowel and the distance between the two (number of pixels)
is calculated, multiplied by the size per pixel determined by the magnification switching unit (9), and converted to the size (block (II) in Figure 4).
). Then, when the cursors (25a) and (25b) extend over a plurality of scanning lines, the same process is repeated for another line (block (40) in FIG. 4).

Therefore, @D of the pattern P obtained for each line
Based on the dimensions shown, statistical processing for each cage, such as average value calculation and standard deviation calculation, is performed (block (4υ) in Figure 4).Finally, the results of these calculations are displayed and recorded on a display unit such as a monitor or printer. (Figure 4 Block system.

Thus, according to the non-embodiment dimension measuring device.

For example, for measuring minute objects such as semiconductor patterns.

With a high resolution of 0.01 μm, it can be determined automatically and with high precision. In particular, in this example, the smoothing process is performed for each region of the obtained waveform representing the semiconductor pattern, depending on the degree of change in each region, so that the underlying portion and the pattern portion can be clearly distinguished. Because you will be able to make a clear distinction.

High precision dimension measurement becomes possible. In other words, in areas that rise steeply, the number of zero pixels is reduced to suppress smoothing.

On the other hand, for a flat base area, the number of pixels is increased to perform strong smoothing, so that the two can be clearly distinguished and one dimension measurement can be performed accurately.

In the above example, the dimension measurement of the pattern width in the horizontal direction is shown, but the pattern width in the vertical direction can also be measured by the same method by performing a 90-degree scan in the scanning direction of the electron beam beam. becomes. Moreover, the measurement is not limited to the pattern width, but two patterns P, as shown in FIG. 10(a).

Each internal 6ζ cursor (25a) of P2, (2
5b), the waveform QL as shown in Fig. 10 Φ) is obtained.
From I, the 1 regression line can, (2) and the quadratic surface ffa (to) of the base are obtained in the same manner as in the above embodiment, and from their intersection s*, the pattern P1. P! It is also possible to find the interval between Furthermore, the pattern P shown in FIG. 11(a)
The pitches of s and P can also be determined. That is, the cursor (25a).

(25b) sandwich the left (right) side edge of pattern P3,
Pinch the left (15) side edge of pattern P with the cursor (25C) and (zsd). However, in the same manner as in the above embodiment.

From the waveform (c) shown in FIG. 11(b), the regression line (to),
After finding 0υ, the quadratic curve (a) of the base, and the vessel,
These intersection points (g), (pattern P from 94J, , P
, we can find the pitch of . Furthermore, in the above embodiment, a semiconductor wafer for LSI is used as the measurement sample αQ, but the dimension measuring apparatus of the present invention can be applied to any dimension measurement on the order of 1 μm.

In addition, if data is stored in one display image memory section, one
To increase the scanning speed of the sub beam bundle, the image memory section for length measurement (1
! When storing data in J, the hill running speed may be slowed down in order to capture a signal with as good a S ratio as possible. Furthermore, before performing the smoothing processing method, which is one of the noise removal methods in the above embodiment, one image memory section αl is divided into 1024 pixels per line using a control kite No. 11, and is taken in multiple times and averaged. You may also do so. Furthermore, the present invention is not limited to output signals from SEMs: CCDs.

It can also be applied to image signals from image sensors such as ITVs.

〔Effect of the invention〕

The dimension measuring device of the present invention has a dimension measuring section connected to the SEM body that automatically measures the dimensions of a specific part of the measurement sample based on the image signal output from the 8EM body, so that it is possible to , the reading error is eliminated,
With a high resolution of 0.01 μm or less, precise measurements can be performed quickly and with high precision.

In particular, in the present invention, the obtained detection signal is divided into a plurality of regions by differential processing, and smoothing processing is performed with different weighting for each region, so that it is easy to distinguish between the base part and the pattern part. Since it can be performed clearly, it is effective when the secondary electron signal from the underlying portion near the pattern is changing significantly. Therefore, when the dimension measuring device of the present invention is applied to a semiconductor manufacturing process for LSI, VLSI, etc., product evaluation and inspection can be performed easily and with a high degree of reliability. As a result, it is possible to improve the quality and yield of semiconductor products. In addition, it can make a significant contribution to the development of various fabrication techniques and the determination of process conditions for high integration.

[Brief explanation of drawings]

FIG. 1 is an overall configuration diagram of a dimension measuring device according to an embodiment of the present invention, FIG. 2 is an explanatory diagram showing the division of an image signal into pixels obtained by the dimension measuring device of FIG. 1, and FIG. FIG. 4 is a flowchart showing the procedure for measuring dimensions using the dimension measuring device shown in FIG. 1, and FIG. Graphs shown in Figure 6 are graphs showing actually obtained waveforms, Figures 7 (A) and (B) are 5avi
Figure 8 is a graph showing the waveform after smoothing the waveform in Figure 6. Figure 9 is a graph to explain how to find the pattern width. 11 through 11 are diagrams for explaining various dimension measurements by the dimension measuring device shown in FIG. 1. (1) : SEM main body part, (2) : Dimension measurement part. (4): @child beam flux, (1[): measurement sample. α41=CRT (display unit), αTen: display image memory unit. u9: Image memory section for length measurement. Qυ: Cursor setting section. 4: CPU section (calculation control section). (22a): Smoothing coefficient memory section. (25a), (25b): Car crane. Agent: Patent Attorney Noriyoshi Chika Borrowed from Kikuo Takehana Figure 3 @ 4 Figure (A) CB) Figure 7 Figure A element in-z Figure 6 Figure f) 4 Standing Figure 8 Figure 9

Claims (1)

[Scope of Claims] A dimension measuring device characterized by having the following configuration. (b) A scanning type electronic device that scans a measurement sample on which a pattern to be dimensioned is formed with an electron beam and outputs an image signal indicating the pattern, and also has a display section and displays an image of the pattern on the display section. Microscope main body. (b) an image signal storage unit that stores the image signal as digitized image data; and a differential processing unit that processes the image data and divides the image data into a plurality of regions based on the differentiation results, and divides the image data into a plurality of regions for each region. and a calculation control section that performs smoothing processing on each side with different weights and calculates the distance between these edges based on the smoothed image data.