JPH02236938A - Image restoration method, scanning electron microscope, pattern appearance inspection device, and scanning image detection device - Google Patents

Abstract

(57) [Summary] This bulletin contains application data before electronic filing, so abstract data is not recorded.

Description

Detailed Description of the Invention [Field of Industrial Application] The present invention relates to an image restoration method for restoring signal waveform deterioration due to response delay of an image detection system, a scanning electron microscope, a pattern appearance inspection device, and a scanning image detection device. .

[Conventional technology]

Conventional electron beam detectors for scanning electron microscopes, for example, use a scintillator and a photomultiplier tube as described in Japanese Patent Application Laid-open No. 62-260335, which has improved detection speed and S.
It is widely used because it is superior to other detectors in terms of /N. This T-son beam detector using a scintillator and a photomultiplier tube has a response characteristic (afterglow time) of the scintillator.
This is because the frequency bandwidth is limited to about 4 MHz at most due to the response characteristics (rise time) of the photomultiplier tube. If detection is performed at a sampling frequency exceeding about 10 MHz, the signal waveform will become dull and only degraded images will be obtained. LSI wafer visual inspection equipment, X-ray mask visual inspection equipment, and the like that use such scanning electron microscopes need to detect patterns at high speed in order to shorten inspection time. However, as mentioned above, the frequency bandwidth of the electron beam detector is limited to about 4 MHz, so if high-speed detection is performed at a sampling frequency of 10 MHz or higher, minute defects cannot be detected. The detection speed was slow and the detection time was long. [Problems to be Solved by the Invention] The above-mentioned conventional technology has a problem in that, for example, the frequency band width of the electron beam detector is narrow, so that when the detection speed is increased, the signal waveform becomes dull and the image deteriorates.

The object of the present invention is to provide an image restoration method for restoring signal waveform deterioration due to response delay in an image detection system, and to prevent deterioration even when high-speed detection exceeds the frequency bandwidth of an electron beam detector using the image restoration method. scanning electron microscope, which can obtain images without
and pattern visual inspection equipment using a scanning electron microscope that has a fast detection speed and short inspection time, and laser scanning microscopes, visual inspection equipment, or foreign matter inspection equipment that uses a photomultiplier tube or the like as a detector with increased detection speed. The present invention provides a scanning image detection device.

[Means for Solving the Problem] In order to achieve the above object, the image restoration method, scanning electron microscope, and pattern appearance inspection apparatus of the present invention restore degraded images whose waveforms have become dull due to response delay of the image detection system. In order to do this, we first model the transfer function of the image detection system, and here we calculate the accuracy of the components of the electron beam detection system, such as scintillators, photomultiplier tubes, and semiconductor detectors, by the sum of first-order return elements or first-order delay elements. Since it can be approximated well, the transfer function of the electron beam detection system can be described as the sum and product of first-order lag elements, and the detected waveform can be determined by signal processing according to the differential equation obtained by inverse Laplace transform of the inverse function of these transfer functions. Restoring dullness and deterioration. In addition, this signal processing increases noise in the image, so in order to suppress the increase in noise due to restoration of the detected waveform, smoothing processing is applied to the detected image in advance, so that most of the noise in the detected image is removed by the image detection system. Because the noise is generated in the front, the high-frequency component of the noise is small and the high-frequency cut filter cannot remove the noise. Therefore, focusing on the small amplitude of the noise, a smoothing filter is applied when the fluctuation of the signal level within the local area is small. When the signal level changes are large, smoothing processing can be applied to the detected image to output the original signal as is.By restoring the detected waveform after these smoothing processes, a restored image with less noise can be created. get. Furthermore, a scanning image detection device such as a laser scanning microscope, visual inspection device, or foreign object inspection device using the point sensor imaging means of the present invention similarly uses the transfer function of an image detection system including a photodetector or a photomultiplier tube. is determined in advance, and the inverse function is Laplace-transformed to perform signal processing according to the differential equation obtained, thereby restoring the dullness and deterioration of the detected waveform. Further, in order to remove noise from the detection signal, the structure is such that a smoothing process is performed and then a restoration process is performed.

[Effect]

In the above image restoration method, scanning electron microscope, and pattern appearance inspection device, the transfer function of the image detection system is expressed as G(s).
Then, the Laplace transform Y of the degraded image signal y (t)
(s) is the Laplace transform X of the original image signal x (t)
Using (s), Y(s) = G(s) ・X(s)
It is expressed as (1). By using the inverse function 1/G(s) of G(s), the original image can be restored from the degraded image. This restoration process is an operation on the Laplace transform surface (Fourier transform surface). 1/G(s) is called an inverse filter. Restoration using an inverse filter requires Fourier transform of the image, applying the inverse filter, and then inverse Fourier transform to restore the image, so even though the amount of calculation is large, it is not suitable for real-time processing. In addition to inverse filters, many filters such as Wiener filters have been proposed, but all of them involve calculations on the Fourier transform surface, which requires a large amount of calculation. Image restoration using an iterative method has also been proposed, but since it is a successive approximation method, the amount of calculation is large and real-time processing is difficult. Equation (2) generally cannot be subjected to an analytical inverse Laplace transform, but it can be inversely Laplace transformed if G(s) is expressed in a specific form such as an integral element, a differential element, or a first-order lag element. Therefore, for example, if the transfer function of a scintillator, photomultiplier tube, or semiconductor electron beam detector, which is a key component of an electron beam detection system, is approximated by the sum of first-order delay elements, the transfer function of the entire electron beam detection system is 1. It is the sum and product of the next lag elements. As a simple example, G (s) = , τ: time constant (
3) Assuming τ S ÷ 1 and a first-order lag element, when formula (2) is inversely Fourier transformed, x (t) = τy' (t) + y (t)
(4) and the sum of the differential signal of the degraded image and the signal of the degraded image. The differential of this signal can be calculated as a difference, which requires less calculation and can be processed in real time. Furthermore, since the restoration using Equation (4) involves differential processing, noise increases, so in order to suppress the increase in noise, it is effective to perform smoothing processing on the degraded image in advance. If the entire screen of the degraded image is smoothed using an average value filter or a median value filter, effective signals as well as noise will be weakened, so the restoration process cannot completely restore the original image without deterioration. In addition, with a high frequency cut filter, most of the noise is generated in front of the detection system, and the high frequency components of the noise are small, so the noise cannot be removed. For this reason, focusing on the fact that the amplitude of the noise is smaller than the amplitude of the effective signal, when the fluctuations in the signal level within the local region are small, a smoothing filter such as an average value filter is applied to reduce the fluctuation of the signal level within the local region. When is large, nonlinear smoothing processing that outputs the original signal as is becomes effective. In this way, by performing nonlinear smoothing processing on the electron beam detection signal of the scanning electron microscope and pattern visual inspection device, and then performing restoration processing represented by equation (4), a restored image with less noise can be obtained. High-speed detection becomes possible. In addition, in scanning image detection devices such as laser scanning microscopes, visual inspection devices, and foreign object inspection devices that use the imaging means (detection means) of the point sensor described above, similar nonlinear smoothing processing and restoration processing are performed to reduce noise. A restored image is obtained and high-speed detection is possible.

〔Example〕

Embodiments of the present invention will be described below with reference to FIGS. 1 to 20.

FIG. 1 is a configuration diagram showing an embodiment of a scanning electron microscope using the image restoration method according to the present invention. In FIG. 1, 1 is an electron gun, 2 is a scanning coil, 3 is an electron lens, 4 is a sample, 5 is a scintillator, 6 is a light guide, 7 is a photomultiplier tube, 8 is an amplifier, 9 is a deflection circuit, 10 11 is a synchronization circuit, 1l is an AD converter, 12 is a smoothing circuit, 13 is a restoration circuit, 14 is a display circuit, and 15 is an electron beam. This embodiment is an example of a scanning transmission electron microscope using the image restoration method of the present invention, in which an electron beam 15 generated and accelerated by an electron gun 1 is focused (narrowly narrowed) onto a sample 4 by an electron lens 3. At the same time, the scanning coil 2 driven by the scanning signal of the deflection circuit 9 synchronized with the synchronization circuit 10 causes the microscope field of view on the sample 4 to be deflected and scanned. The electron beam transmitted through the sample 4 is converted into light by a scintillator 5, guided by a light guide 6 to the light receiving surface of a photomultiplier tube 7, and further converted into an electrical signal by this photomultiplier tube 7. This electron beam detector is a type that uses a scintillator 5 and a photomultiplier tube 7 for detection, but is not limited thereto, and for example, a semiconductor detector may also be used. An electric signal from the photomultiplier tube 7 is amplified by an amplifier 8, and is quantized by an AD converter 1l while being synchronized with the scanning signal from the deflection circuit 9 by a synchronization circuit 10 to obtain a scanning transmission electron image. In this scanning transmission electron image, the image flows in the scanning direction mainly due to the response delay of the scintillator 5 and the photomultiplier tube 7. The gist of the image restoration method of the present invention is to restore the flow of this electronic image (image) (the dullness of the signal waveform). To do this, first, a signal quantized by an AD converter 11 is subjected to noise reduction in a smoothing circuit 12, and then the output signal of this smoothing circuit 12 is inputted to a restoration circuit 13 to restore the deterioration of the signal waveform. The signal restored by the restoration circuit 13 is converted from digital to analog by the display circuit 14 and displayed as an image on a cathode ray tube. FIG. 2 is a functional block diagram showing the operating procedure of the smoothing circuit 12 of FIG. 1. In Fig. 2, 20 is an input image. 21 is a local area extraction circuit, 22 is a level fluctuation detection circuit, 23 is an average value filter, and 24 is an output image. In FIG. 2, the output signal of the AD converter 11 shown in FIG. A signal level fluctuation within this local area is detected by a level fluctuation detector 22, and this level fluctuation detection circuit 22 calculates the difference between the brightest density value and the darkest density value within the local area. When this density difference is smaller than a predetermined value, the signal level in the local area is smoothed by the average value filter 23, and when the density difference is larger than a certain predetermined value, the density (signal level) of the input image 20 is smoothed. By outputting the level) as is, a nonlinearly smoothed output image 24 is obtained. This average value filter 23 may be any other filter, such as a median filter, as long as it has a smoothing effect.

The processing contents of the smoothing circuit 12 described above are summarized as follows. A=max(in(i-1,j-1)win(i L
jLin(i Lj+ILtn(it.i-t)t
in(i, j), in(i, j+1), in(1
+1,j-1), in(i+1,j), in(i+
l, j + 1)) B = min (same as above)
}if(A-B≦N) t
hen C=max(in(i−LjL in(iJ+ in(
t+t,j))1++in(in(i-Lj),in(
iyjL in(i+1,j))D=max(in(i
, j-1), tn(i, j), in(i, j+1)
)−win(in(Lj t), in(iJ+ i
n(i,j+1))if(C≦D.AND.C≦N)
thenelse if(C>D.AND.D<
:N) thenellse out (i, j)=1n (ipj) end if end if However. in(i,j) is the input image, out(i,j
) is the output image, and N is a constant (for example, N=8). Third
The figure is a functional block diagram showing the operating procedure of the restoration circuit 13 shown in FIG. In Fig. 3, 30 is an input signal, 31
is a second-order differential circuit, 32 is a first-order differential circuit, 33 is a convolution circuit, 34, 35, 36. 37 is a multiplier, 38 is an adder, and 39 is an output signal. The output signal of the smoothing circuit 12 of FIG. 1 is used as the input signal 30 of the restoration circuit l3 of FIG.
The first-order differential circuit 32 calculates the first differential of the input signal 30.
A convolution circuit 33 calculates a convolution of the input signal 30, and a multiplier 34 calculates a second derivative. The multiplier 34 multiplies the first derivative by k
2, and the convolution is multiplied by k4 by multiplier 36,
The input signal 30 is multiplied by k3 in the multiplier 37, and these four signals are added in the adder 38 to obtain the output signal 39 as follows.

The restoration of the deterioration of the detected waveform by the restoration circuit 13 described above uses an inverse function of the transfer function of the electron beam detection system. When a YAP single crystal is used as the scintillator 5 in Fig. 1 and a head-on type (Hamamatsu Photonics IR 269) is used as the photomultiplier tube 7, the transfer function G (s) of the electron beam detection system is expressed by the following equation. Obtain what is given by experiment. However, AI = 0.2
9, A4=0.24,τ, =0.02μS,τ2=
0.07μs, τ3=1.5μS, τp=0.15μS
Therefore, the restoration transfer function, that is, the inverse function of the transfer function G (s) of the electron beam detection system (signal transfer system) H (s) = 1
/G(s) is obtained as. However, I(s)=(Azt+t2+(1-AI-A4)tz
ts+A+tst+)s”+((A1+A2)tt+(
1-AI)tz+(1-A2)ts)s+1. Substituting a numerical value into this results in . Signal y (t) of the degraded image of this input signal 30
Let Y(s) be the Laplace transform of the output signal 39, and let X(g) be the Laplace transform of the signal x(t) of the restored image of the output signal 39.
X(s)=H(s) ・Y(s)
(7) holds true. Substituting equation (6) into equation (7) and performing inverse Labrass transformation, h (t)=-0.223e-'°870t +11.
60-19°7t (8)/ is obtained. Equation (8) is the restoration equation required. In order to make Equation (8) into a discrete form, we convert the differential to a difference and the integral to a sum of products. +ksy(t) 10 years h(ti●Δt)・y(i●Δt
)・Δtl:O where Δt: signal sampling interval, n=−Δt. Here, the sampling frequency is assumed to be 15 MHz, that is, the sampling interval Δt=0.067 μs. Fourth
The figure is a graph of the function h (t) of equation (8)' in the restoration equation (8) of the restoration circuit 13 in FIG. 3, where the horizontal axis is time t [μS] and the vertical axis is h ( t). In Figure 4, the value of this function h (t) approaches zero as t increases, so when t exceeds a certain value, h (t)
= It can be approximated as O. Therefore, the convolution of the fourth term in restoration equation (8) can limit the integral period. Therefore, equation (9) can be further written as follows. k,=0.00540, k2=0.275
, k3=0.665, the output of the first-order differentiation circuit 32 is multiplied by k2 by the multiplier 35, and the output of the convolution circuit 33 can be transformed into the multiplier 36. Also, if we omit the convolution term and further simplify 1P, we get + k3y (t)
(11). The above restoration formula (10).
(1.1), the degraded waveform y (t) is transformed into the restored waveform x (
t).

According to the above restoration formula (10), the operation of the restoration circuit 13 shown in FIG.
The first differential of the signal y (t) is calculated using the first term of equation (10), and the first differential of the signal y (t) is calculated using the first term of equation (10).
The convolution circuit 33 calculates the signal y
The convolution of (t) is the third term of equation (10) (
10), the output of the second-order differentiation circuit 31 is converted to k by the multiplier 34.
．． The adder 38 multiplies these four signals by k3.
By adding the restored signal x (t
) is obtained as the output signal 39.

FIG. 5 shows the degraded waveform y (t) of the input signal 30 and its output signal 39 with respect to the step response of the restoration circuit 13 in FIG.
It is a graph showing the restored waveform x (t) of . In FIG. 5, the degraded waveform y (t) is considerably restored by the restored waveform x (t) according to the above restoration formula (to), (11). The signal x(t) restored by the restoration circuit 13 is converted into DA by the display circuit 14 and displayed as an image on the cathode ray tube. FIG. 6 is a circuit diagram of an embodiment of the smoothing circuit l2 of FIG. 1 (FIG. 2). In FIG. 6, 40 is an input terminal;
1a, 4lb are shift registers. 42a, 4
2b, 42c, 42d, 42e, 42f, 42g, 4
2h, 42i are registers, 43 is an adder, 44 is a divider, 45 is a subtracter, 47 is an absolute value calculator, 48 is an adder,
49 is a memory, 50 is a comparator, 51 is a switch, and 52 is an output terminal, which are connected as shown in the figure. From the input signal from the AD converter 11 to the input terminal 40 in FIG. 6, shift registers 41a, 4lb and registers 42a to
A local region of 3×3 pixels is cut out using 42i. This 3
In order to obtain the average value of the signal value y of ×3 pixels, the adder 4
3 calculates the total sum of the outputs of the registers 42a to 42i, and a divider 44 calculates the total sum and a constant K.3 stored in the memory 45.
Find the quotient with =9. In the explanation of Fig. 2 above, the signal level fluctuation was evaluated by the difference between the maximum value and the minimum value in the local area, but here it is evaluated by the variance σ', and the variance σ' is expressed as: σ'=year ly+yl 1 = 1. From the outputs of the registers 42a to 42i in FIG. 6 and the output y of the divider 44, a subtracter 46 and an absolute value calculator 47
and the adder 48 calculates the variance σ' of equation (12).

A comparator 50 compares this division σ' with a predetermined constant K2 stored in the memory 49. The comparison result is σ
'>K2, the switch 51 is switched so that the contents of the register 42e are output as they are to the output terminal 52 by the shift register 49a, and when σ'≦K2, the average value y of the local area of the divider 44 is outputted to the output terminal 52 by the shift register 49b. The switch 51 is switched to output to the output terminal 52. These shift registers 49a and 49b are inserted to set the timing between the output of the comparator 50 and the output of the register 42e or the output of the divider 44. The operation of this circuit configuration allows nonlinear smoothing processing to be performed in real time.

FIG. 7 is a circuit diagram of one embodiment of the restoration circuit 13 of FIG. 1 (FIG. 3). In FIG. 7, 60 is an input terminal, 61
a, 6lb, 61c are registers, 62a,
62b, 62c are subtracters, 63 is an adder, 64a,
64b and 64c are memories. 65a, 65b
, 65c is a multiplier, 66a, 66b, 66
Q T - + 66m is memory, 67a, 67b
, 67c, -67m are multipliers, 68a, 6
8b, -, 68m-+ are registers, 69a,
69b, ++, 69m-1 are adders, 70a, 70
b, 70c, 70d are shift registers, 71 is an adder,
72 is an output terminal, which is connected as shown in the figure. The restoration circuit 13 in FIG. 7 is based on the above restoration formula (10).
When expressed by a signal sequence, x (i) = A+ (y(i+1) − 2y(i
) +y(i-1)) + A2 (y(i+1)-
y(i-1)) However, AH = kI/Δt', A2=
k2/2・Δt* A3=k3y, and this restoration formula (
This is a circuit that performs restoration processing using 13). This restoration circuit 1
A signal y (i
) and input the signal y to registers 61a, 6lb, 61c.
(x+1) ty (1) ey b-1) is stored. From this, the subtracters 62a, 62b and the adder 63 calculate y (i+1) −2y(i
) +y(i-1) is calculated, and further multiplied by the constant A1 stored in the memory 64a and the multiplier 65a, and the result is ↓
Term A+ (y(i÷1)-2y(i) +y(i-1
)) is obtained. Similarly, the second term A2 (y(i+1
) y(i-1)) and set the constant A in the memory 64c.
3 and the multiplier 65c to calculate the third term As y (1). Further, the convolution path of the input signal y (i) is formed by a memory 66 (668 to 66m) storing a constant K I (K g − K ra ) and a multiplier 67 (67 a
~67m) and register 68 (68a ~6g+a-
1) and an adder 69 (69a to 69m-1). The outputs of these multipliers 65a to 65c and adders 69ms are transferred to shift registers 70a to 70d.
When the timing is determined by the adder 71 and the sum is performed by the adder 71, the restored signal x (i) of the restoration formula (l3) is outputted to the output terminal 72. FIG. 8 is a block diagram showing a method for measuring the response characteristics of the electron beam detection system without the scintillator 5 shown in FIG. In Figure 8, the same symbols as in Figure 1 indicate corresponding parts,
16 is a light emitting diode, l7 is a power supply, 18 is a blinking circuit,
19 is memory. Among the components of the electron beam detection system shown in FIG. 1, the scintillator 5 and photomultiplier tube 7 have poor response characteristics. Therefore, we first measured the response characteristics of the electron beam detection system without the scintillator 5. To do this, as shown in FIG. 8, a light emitting diode 16 installed on the light receiving surface of the photomultiplier tube 7 is blinked by a power source 17 and a blinking circuit 18. A lamp may be used instead of the light emitting diode 16, or a shutter may be inserted between the light emitting diode 16 and the photomultiplier tube 7 and the shutter may be opened and closed instead of blinking the light emitting diode 16. The output of the photomultiplier tube 7 at this time is amplified by an amplifier 8, quantized at a predetermined sampling frequency by an AD converter l1, and taken into a memory 19. The signal is read from this memory 19 and a transfer function G(3) that accurately approximates the signal waveform of the step response is determined.

Head-on type photomultiplier tube 7 (Hamamatsu Photonics R)
269), it was found that it can be approximated by a first-order delay element with a time constant τ, = 0.15 μS. Therefore, the transfer function GPMT is. FIG. 9 is a block diagram showing a method for measuring the response characteristics of the electron beam detection system with the scintillator 5 shown in FIG. 1 attached.
In Figure 9, 29 is a knife edge. A knife edge 29 is installed in front of the scintillator 5 shown in FIG. tt
Scan the t-satellite line 15. Since it is difficult to obtain a knife edge 29 with a sufficiently sharp edge at the resolution level of an electron microscope, by making the scanning speed of the electron beam 15 faster than the normal scanning speed, the electron beam 15 incident on the scintillator 5 can be Makes a sharp step input. The incident electron beam l5 is converted into light by the scintillator 5, guided to the light receiving surface of the photomultiplier tube 7 by the light guide 6, converted to an electric signal by the photomultiplier tube 7, amplified by the amplifier 8, and then
The D converter 11 quantizes at a predetermined sampling frequency,
Import into memory 19. The signals read from this memory 19 are sent to the scintillator 5 and photomultiplier tube 7 of the electron beam detection system.
The waveform is dull due to the response characteristics of the amplifier 8 and AD converter 11. Here, the response delay due to the photomultiplier tube 7, amplifier 8, and AD converter 1l can be approximated by a first-order delay element (Equation (14)), so the overall response characteristic of the electron beam detection system is the transfer function of the scintillator 5. It is the product of first-order lag elements. When a single crystal of YAP is used as the scintillator 5, the response characteristic of this electron beam detection system is expressed by the transfer function G(8) of the above equation (5).
It can be approximated by , as in . FIG. 10 is a graph showing the response curve of the electron beam detection system including the scintillator 5, photomultiplier tube 7, amplifier 8, and AD converter 11 in FIG. 1 (FIG. 9), and the horizontal axis is the time [ μ
S], and the vertical axis is the response output (relative value). In Figure 10, the step response characteristics (circles) based on experimental values and the equation (
Figures 11 (a), (b), (C), and (d) are in good agreement with the response curve (solid line) due to the transfer function G (s) of 5).
It is a waveform diagram showing the electron beam detection image (pattern waveform) shown in the figure. FIG. 11(a) shows a pattern in the scanning line direction of the X-ray mask sample 4, where 54 is a membrane, 55 is a pattern, and 5
6 is a black spot defect, 57 is a white spot defect, and 58 is a scanning line.
11(b) is the output waveform of the AD converter 11, FIG. 11(c) is the restored waveform of the output of the AD converter 11 by the restoration circuit 13, and FIG. 11(d) is the output of the smoothing circuit l2. The restored waveforms by the restoration circuit 13 are shown respectively. As shown in FIG. 11(a), the X-ray mask is used as sample 4, and the membrane 54 is
When scanning the electron beam 15 in the direction of the scanning line 58 across the pattern 55, for example, there is a black spot defect 56 on the membrane 54 and a white spot defect 57 on the pattern 55.
There is. In this case, the output waveform of the AD converter l1 of the electron beam detection system obtained by scanning the electron beam 15 along the scanning line 58 is as shown in FIG. Therefore, the waveform is dull. Therefore, the detection signal amplitudes of the black spot defect 56 and the white spot defect 57 are small, and the detection signal amplitude of the black spot defect 56 and the white spot defect 57 is small. 5
7 cannot be detected sufficiently. In contrast, the detection image (waveform) restoration method of the present invention allows the transfer function G of the electron beam detection system to be
The waveform restored by the restoration circuit 13 using the inverse function of (s) has no waveform dullness as shown in FIG.
6. The detection amplitude of 57 is also large, but since it is a differential process, the noise increases significantly. Therefore, the waveform obtained by further processing the output of the AD converter 11 of the electron beam detection system through nonlinear smoothing processing using the smoothing circuit 12 and performing restoration processing using the inverse function of the transfer function G (s) using the restoration circuit 13 is As shown in FIG. 11(d), the increase in noise is suppressed and the waveform dullness is restored, so that defect 56. 57 can be sufficiently detected without being buried in noise. FIG. 12 is a configuration diagram showing an embodiment of a pattern appearance inspection apparatus using a scanning electron microscope according to the present invention. FIG. 12 shows an example of an X-ray mask pattern visual inspection apparatus using the scanning transmission electron microscope shown in FIG.
The same symbols as in the figure indicate corresponding parts, 75 is an electronic detector, 76 is a stage, 77 is a stage control circuit, 78
are gray level conversion circuits, 79a, 79b, 79c
, 79d, 79e are binarization circuits, 80a, 8
0b is a positional deviation detection circuit. 81a and 8lb are image shift circuits, 82 is a storage device fi (CAD data), 82a,
82b is a pixel size conversion circuit, 83 is a readout circuit, 83
83a and 83b are screen dividing circuits; 84 is an image signal generator;
5 is a corner rounding circuit, 86 is a multi-level conversion circuit, 87 is an image comparison circuit, and 88 is a memory (defect), which are connected as shown in the figure. The pattern appearance inspection apparatus of this embodiment is a sample 4 detected by a scanning transmission electron microscope shown in FIG.
CAD which is the electron beam image of the ray mask pattern and the drawing data of the X-ray mask pattern stored in the storage device 82
This is a design pattern comparison method that compares an ideal pattern generated from data and depicts mismatched areas as defects. The electron gun 1, scanning coil 2, electron lens 3, and f! shown in FIG. ! A sub beam detector 75, an amplifier 8, a deflection circuit 9, a synchronization circuit 10, . The AD converter l1 is a component of a normal scanning transmission electron microscope, and is as described in FIG. 1 together with the smoothing circuit 12 and the restoration circuit 13 according to the present invention. In addition, the field of view of the scanning transmission electron microscope is sample 4.
Since the inspection area is narrower than the inspection area of the X-ray mask, the sample 4 is placed on the stage 76 and the stage control circuit 77 inspects the front of the inspection area in a step-and-repeat manner. The electron beam image output from the restoration circuit 13 of the scanning transmission electron microscope described above is input to the gradation conversion circuit 78, and the density is adjusted so that the average brightness of the pattern part and membrane part of the X-ray mask becomes the reference value. Convert the gradation. The average density value of the pattern part of this image is p, its reference value is p', the average density value of the membrane part is m, its reference value is m', the density value of the original image is 2, and the converted image Assuming that the density value of is 2', the density gradation conversion is given by the following equation. In 22, the average density value p of the pattern part and the average density value m of the membrane part can be easily obtained from the histogram of the detected image. The reading circuit 83 reads out the CAD data for pattern drawing stored in one of the storage devices 82, and the image signal generator 84 generates a reference image ( create an ideal image). The mask pattern of the actual sample 4 has rounded corners, and in order to prevent these rounded corners from being detected as defects, the reference image from the image signal generator 84 is input to a corner rounding circuit 85 to round the corners. Furthermore, in order to compare the detected image and the grayscale image, the binary image is converted into a grayscale image by the multivalue conversion circuit 86. For this conversion process, it is best to use a blurring filter that assumes a Gaussian distribution for the point spread function PSF. In this case, if the input image is f(x*y), the output image is g(x*y), and the PSF is h(xyy), the conversion formula is σl: variance. In order to convert this reference image (binary image) into a grayscale image with smooth density changes, it is recommended to convert the reference image with a pixel size smaller than the detection pixel size, and then sample the pixels to match the detection pixel size. ．． Next, the detected image that is the output of the grayscale gradation conversion circuit 78 and the reference image that is the output of the multilevel conversion circuit 86 are converted into binary images by the binarization circuits 79a and 9b, respectively.
0a calculates the amount of positional deviation between the two images, and the image shift circuit 81a aligns the detected images. In order to detect further minute defects, pixel size conversion circuits 82a and 82
b reduces the pixel size by half (for example, 0.05 μm/pi
x to 0.025μs/pix). This conversion process divides one pixel into four pixels, and the processing formula is a = (9A + 3B + 3C + D) / 16 b = (9 B + 3 A + 3 D + C) / 16
(16)c=(9C+3D+3A+B)
/16 b=(9D+3C:+38+A)/16 However. A, B
, C, D: Brightness of pixel before conversion a g b @ Q
g d: Brightness of the pixel after conversion. Since there is image distortion in the image detected by the scanning transmission electron microscope, we need to align it again. At that time, if there is double elemental distortion, even if the entire screen is aligned, the entire screen will only be aligned on average, and each part of the screen can be aligned individually, so it is possible to divide the screen. Then, each divided screen is binarized again and aligned. From this, the detected image and the reference image are screen-divided by screen division circuits 83a and 83b, respectively, and binarized by binarization circuits 79a and 79d,
A displacement detection circuit 80b calculates the amount of displacement between the two images, and an image shift circuit 8lb aligns the detected images. The image comparison circuit 87 compares the detected image and the reference image aligned as described above, and the difference image is binarized by the binarization circuit 79e, and the coordinates of the mismatched part are stored in the memory 88 as a defect. do. The local perturbation pattern matching method is suitable for the grayscale image comparison algorithm of the image comparison circuit 87. In the local perturbation pattern matching method, the detected image is divided into x for each local area with respect to the reference image
This is an algorithm that performs matching in the y-plane and brightness direction, and extracts parts that cannot be matched as defects.
*(xsy), the detected image is "("y y),
If the reference image is R (x ey), then,
S+-Sa(xyy)=I(xyy)-R(x"Ly"
x), (i, j)-(-1.0), (1.0)-(0,
-1), (0.1)SsNSs(xyy)=I(x-y
)−(()lΣ:1)H(x−y)÷R(x+x,y+
j))/)lku*(itj)=(-L-t), (1t,t)tQ,-t)*Q,oSs~S+o(xpy)
= I (x,y)-R(Xty). It is given by ±α. FIG. 13 is a block diagram showing another embodiment of a pattern appearance inspection apparatus using a scanning transmission electron microscope according to the present invention. FIG. 13 shows another example of the X-ray mask pattern visual inspection apparatus using the scanning transmission electron microscope shown in FIG. 1, where the same reference numerals as in FIG. be. The difference between the pattern appearance inspection apparatus of this embodiment and the embodiment shown in FIG. 12 is that in FIG.
In contrast, in Fig. 13, the image detected by the scanning transmission electron microscope is not restored but is stored in the storage device 82.
The response characteristic adding circuit 89 adds the response characteristic of the electron beam detection system of the scanning transmission electron microscope to the reference image (ideal image) generated from the CAD data stored in the CAD data. This response characteristic adding circuit 89 dulls the waveform of the reference image according to the transfer function G(S) of the response characteristic of the electron beam detection system including the electron beam detector 75, amplifier 8, and AD converter 11 described above. This transfer function G(s
), that is, the impulse response g (t)
becomes . The output signal y (t) can be expressed by convolution with the input signal x (t) and the impulse response g (t). When this is made into a discrete form, Δt: signal sampling interval. Therefore, the response characteristic adding circuit 89 may have a circuit configuration that realizes equation (21), and may be, for example, a digital signal processor DSP that calculates a convolution sum or the circuit shown in FIG. 14. The rest is the same as in Figure 12. FIG. 14 is a circuit diagram for calculating the convolution sum of the response characteristic addition circuit 89 in FIG. 13. In FIG. 14, 101 is an input terminal, 102a, 102b,
102c, -, 102n are memories, 103a, 1
03b, 103c, -103n are multipliers, 10
4a, 104b, -, 104, -1 are registers, 105a, 105b, -・-, io
so- is an adder, and 106 is an output terminal. The signal x (t) input to the input terminal 101 in FIG. 14 is the coefficient K. =Δt−g(i・Δt) is multiplied by the coefficients KG to Kn by the multipliers 103 (103a to 103n), and then delayed by the registers 104 (104a to 104n−+) to the adder 105 (105a to L
O51+-+), the convolution sum signal y(t) according to equation (21) is output to the output terminal 106. FIG. 15 is an elliptical diagram showing an embodiment of a scanning image detection device using the image restoration method according to the present invention. In FIG. 15, an example of a laser scanning microscope of a scanning image detection device using the image restoration method of the present invention is shown, in which the same reference numerals as in FIG. 1 indicate corresponding parts, 90 is a laser, 91a is a deflector, 92 is a drive circuit, 93 is a half mirror, 94 is a lens, and 95 is a photodetector. 1st
The laser beam output from the laser 90 in FIG.
The image is narrowed down to a spot and scanned. The laser beam reflected by the sample 4 is guided by a half mirror 93 to a photodetector 95, where it is converted into an electrical signal. This photodetector 95 is, for example, a photodiode, a phototransistor, a CDS,
Or a photomultiplier tube, etc. The converted electrical signal is amplified by an amplifier 8, quantized by an AD converter 1l,
A smoothing circuit 12 reduces noise, a restoring circuit 13 restores waveform dullness caused by response delay in the image detection system, and the resulting image is displayed on a display circuit 14. At this time, a two-dimensional image is created by scanning the laser beam deflected by a deflector 91a driven by a drive circuit 92 in synchronization with the synchronization circuit 10 and displaying it on the display circuit 14 in synchronization with the synchronization circuit 10. can get. The transfer function G (s) of the image detection system consisting of the photodetector 95, amplifier 8, and AD converter 11 is 1
With the next lag element, it can be approximated as G(s)=, Ps+1(22). Therefore, the restoration formula is: X(s) = (τps + 1)Y(s)
(23), and by inverse Laplace transform, x(t)=τpy'(t)+y(t)
(24). Therefore, the restoration circuit 13 is
), (b).

FIGS. 16(a) and 16(b) are circuit diagrams of an embodiment of the restoration circuit l3 in FIG. 15, where 1lO is an input terminal, 111, ll
la, lllb are registers, 112 is a subtracter, 1l3
is a memory, 114 is a multiplier, 115 is a shift register,
l16 is an adder, and 117 is an output terminal. Figure 16 (
a) is an example of a restoration circuit l3 that discretizes a first-order differential into a forward (backward) difference; is selected, and the subtracter 112 calculates the difference S
k+I Sk is calculated and multiplied by the constant τP/Δt stored in the memory 13 in the multiplier 114. The signal sk synchronized with the above circuit in one shift register 115 and the output of the multiplier 114 are added in an adder 116, and the output terminal 1]7 is (τP/Δt)(Sk++−S
k)+Sk is output. FIG. 16(b) is an example of a restoration circuit l3 that discretizes the first-order differential using a central difference.
In 11l b, the k-1st signal Sk=1, the k-th signal sk, and the k+1st signal S k++ are selected, and the subtracter 112 calculates the difference Sk+1-St+-t, which is stored in the memory 113. The constant τP/(2・Δt
) is multiplied by the multiplier 114. The signal sk, which is timed with the above circuit in one shift register 115, and the output of the multiplier 114 are added in an adder 116, and the output terminal 1
17 (τp/(2・Δt)) (Sk+I-St-t
)+Sk is output. Furthermore, the smoothing circuit l2 may have the circuit configuration shown in FIG. FIG. 17 is a block diagram showing another embodiment of a scanning image detection device using the image restoration method according to the present invention. FIG. 17 shows an example of a foreign object inspection device of a scanning image detection device using the image restoration method of the present invention.
The same reference numerals as in the figure, Fig. 12, Fig. 15, etc. indicate corresponding parts, 9lb is a deflector, 96 is an integrating sphere, and 97 is a 2
Value conversion circuit. 100 is a calculator. A laser beam outputted from a laser 90 in FIG. 17 is focused to a spot on the sample 4 and scanned by a deflector (for example, a polygon mirror) 9lb and a lens 94. The integrating sphere 9
6, the photomultiplier tube 7 converts it into an electrical signal, the amplifier 8 amplifies it, the smoothing circuit 12 reduces noise, and the restoration circuit 13 restores the waveform dullness caused by the response delay of the image detection system. If there is a foreign object on the sample 4, the scattered light from the foreign object is strong, so the waveform restored by the restoration circuit 13 is converted to the binarization circuit 97.
Foreign objects can be detected by binarizing with an appropriate threshold. The coordinates of the detected foreign objects are recorded in the computer 100, and the number and coordinates of the foreign objects can be known. In order to inspect the entire surface of the sample 4, the sample 4 is placed on the stage 76, and the stage is moved by the stage control circuit 77. At this time, the entire surface of the sample 4 is inspected by controlling the drive circuit 92 of the deflector 9lb and the stage control circuit 77 by the computer 100. The smoothing circuit 12 and the restoration circuit 13 have the circuit configurations shown in FIG. 6 and FIGS. 16(a) and (b), respectively. FIG. 18 is a block diagram of another embodiment of the degraded image restoration circuit l3 of the image detection system according to the present invention. In FIG. 18, an example of a restoration circuit l3 using an inverse filter is shown, where 120 is an input image, 121 is a smoothing circuit, 122 is an FFT circuit,
123 is an inverse filter, 124 is an FFT circuit, and 125 is an output image. Since the inverse filter 123 in FIG. 18 is sensitive to noise, noise is removed from the input image 120 by a smoothing circuit 121 in order to restore the degraded waveform.
This smoothing circuit 121 may be, for example, the one shown in FIG. 2 or FIG. 6. Here, the image with noise removed is FFT
A circuit 122 performs Fourier transform, and an inverse filter 123 restores the image. The FFT circuit 124 performs inverse Fourier transform to obtain a restored output image 125. Here, if the deterioration function (responsiveness of the image detection system) is G, then the inverse filter is 1/G.

FIG. 19 is a block diagram of still another embodiment of the degraded image restoration circuit 13 of the image detection system according to the present invention. FIG. 19 shows an example of the restoration circuit 13 using a Wiener filter, where the same reference numerals as in FIG. 18 indicate corresponding parts, and 126 is a Wiener filter. Degraded input image 1 in Figure 19
20 is subjected to Fourier transformation in an FFT circuit 122, image restoration is performed in a Wiener filter 126, and inverse Fourier transformation is performed in an FFT circuit 124 to obtain a restored output image 125. Here, the degradation function is G, and the power spectrum of the image without degradation is W.
f, and the power spectrum of the noise is Wl+, then the Wiener filter 126 has G / (l G l' + w
n/Wr). (Wn/wf)=l- and expressed as a constant L. M may be abbreviated as G/(IG+'+r). Also, instead of the Wiener filter 126, a homomorphic filter,
Various restoration filters such as a general inverse filter, restricted least squares filter, parametric Wiener filter, projection filter, and parametric projection filter may be used.
FIG. 20 is a block diagram of still another embodiment of the degraded image restoration circuit 13 of the image detection system according to the present invention.

FIG. 20 shows an example of the restoration circuit 13 using the iterative method, and 127 is a computer. A restored output image 125 is obtained by repeatedly calculating the degraded input image 120 in FIG. 20 using a computer 127. In addition to the method of restoring image deterioration due to response delay of the image detection system, the image restoration method of the above embodiment approximates the response characteristics of the signal transmission system with an appropriate transfer function, and converts the inverse function of the transfer function into an inverse Laplace. It can also be applied to signal restoration methods that restore signal degradation caused by signal transmission systems by processing signals according to the converted equations. Further, the scanning electron microscope of the above embodiment can be applied not only to a scanning transmission electron microscope that detects transmitted electrons but also to a scanning electron microscope that detects reflected electrons, secondary electrons, and the like. Furthermore, the pattern appearance inspection apparatus using the scanning electron microscope of the above embodiment is not limited to the design pattern comparison method described above.

Further, the scanning image detection device of the above embodiment is not limited to the above laser scanning microscope, visual inspection device, foreign matter inspection device, or the like.

〔Effect of the invention〕

According to the present invention, it is possible to restore the waveform deterioration caused by the response delay of the image detection system, so scanning image detection can be performed using a scanning electron microscope, a pattern inspection device using the scanning electron microscope, a laser scanning microscope, a foreign object inspection device, etc. This has the effect of improving the detection speed of the device and shortening inspection time.

[Brief explanation of drawings]

Fig. 1 is an elliptical diagram showing an embodiment of a scanning electron microscope according to the present invention, Fig. 2 is a functional block diagram of the smoothing circuit shown in Fig. 1, and Fig. 3 is a functional block diagram of the restoration circuit shown in Fig. 1. Figure, 4th
The figure is a graph of the function h (t) of the restoration equation in Figure 1, Figure 5 is a graph of the restored waveform in Figure 1, Figure 6 is a circuit diagram of an embodiment of the smoothing circuit in Figure 1, and Figure 5 is a graph of the restored waveform in Figure 1. Fig. 7 is a circuit diagram of an embodiment of the restoration circuit shown in Fig. 1, Fig. 8 is a block diagram of a method for measuring the response characteristics of the detection system without a scintillator shown in Fig. 1, and Fig. 9 is a circuit diagram of an embodiment of the restoration circuit shown in Fig. 1.
Fig. 10 is a graph of the response curve of the detection system shown in Fig. 1, Fig. 11 (a), (b), (c), ( d) is a waveform diagram of the detected image in FIG. 1, FIG. 12 is a configuration diagram showing one embodiment of the pattern visual inspection device according to the present invention, and FIG. 13 is a diagram showing another embodiment of the pattern visual inspection device according to the present invention. 14 is a circuit diagram of the response characteristic addition circuit shown in FIG. 13, FIG. 15 is a block diagram showing an embodiment of the scanning image detection device according to the present invention, and FIGS. 16(a) and (b) 15 is a circuit diagram of an embodiment of the restoration circuit, FIG. 17 is a schematic diagram showing another embodiment of the scanning image detection device according to the present invention, and FIG. 18 is a circuit diagram of another embodiment of the restoration circuit according to the present invention. FIG. 19 is a block diagram of still another embodiment of the restoration circuit according to the present invention.
FIG. 20 is a block diagram of still another embodiment of the restoration circuit according to the present invention. DESCRIPTION OF SYMBOLS 1...Electron gun, 2...Scanning coil, 3...Electron lens, 4...Sample, 5...Scintillator, 6...
Light guide, 7... Photomultiplier tube, 8... Amplifier, 9... Deflection circuit, 10... Synchronization circuit, 11...
AD converter, 12... Smoothing circuit, 13... Restoration circuit, 14... Display circuit, 16... Light emitting diode,
18... Blinking circuit, 21... Local area extraction circuit, 2
2... Level fluctuation detection circuit, 23... Average value filter, 29... Knife edge, 31... Second order differentiation circuit, 32... First order differentiation, 33... Convolution circuit, 34 ~37... Multiplier, 38... Adder, 8
2... Storage device (CAD data), 84... Image signal generator, 87... Image comparison circuit, 89... Response characteristic addition circuit, 90... Laser light, 91a, 9lb... Deflector, 92... Drive circuit, 93... Half mirror, 94... Lens, 95... Photodetector, 96... Integrating sphere, 100... Computer. Agent Masami Akimoto (voice S) (voice S) (voice S) (voice S) (a) (d) (a)

Claims (1)

[Claims] 1. Approximate the response characteristics of the image detection system with an appropriate transfer function,
An image restoration method that processes an image detection signal according to an expression obtained by inverse Laplace transform of the inverse function of the transfer function, and restores deterioration of the image detection signal caused by the image detection system. 2. A finely focused electron beam is scanned over the sample to detect reflected electrons and
In a scanning electron microscope that obtains an electron image by detecting secondary electrons or transmission electrons, the scanning electron microscope has a restoring means for restoring deterioration of an electron beam detection waveform due to a response delay of an electron beam detection means for converting an electron beam into an electric signal. A scanning electron microscope featuring: 3. A finely focused electron beam is scanned over the sample to detect reflected electrons and
In a scanning electron microscope that obtains an electron image by detecting secondary electrons or transmission electrons, the response characteristics of the electron beam detection means that converts the electron beam into an electric signal are approximated by an appropriate transfer function.
1. A scanning electron microscope characterized by comprising a restoring means for processing an electron beam detection signal according to an expression obtained by inverse Laplace transform of the inverse function of the transfer function and restoring deterioration of the electron beam detection signal. 4. Scan the sample with a narrowly focused electron beam to detect reflected electrons and
In a scanning electron microscope that obtains an electron image by detecting secondary electrons or transmission electrons, there is provided a noise removing means for removing noise in an electron beam detection signal of an electron beam detection means for converting an electron beam into an electric signal; The response characteristic of the detection means is approximated by an appropriate transfer function, and the electron beam detection signal from which noise has been removed by the noise removal means is processed in accordance with a formula obtained by inverse Laplace transform of the inverse function of the transfer function, thereby eliminating deterioration of the electron beam detection signal. A scanning electron microscope characterized in that it has a restoring means for restoring. 5. The scanning electronic device according to claim 3 or 4, wherein the response characteristic of the electron beam detection means for converting the electron beam into an electric signal is approximated by the sum or product of first-order delay elements, or the sum and product of first-order delay elements. microscope. 6. The scanning electron microscope according to claim 3 or 4, wherein the restoring means includes processing means for performing nth (n; natural number) differential processing and convolution processing. 7. The noise removing means for removing noise in the electron beam detection signal operates a smoothing filter when the fluctuation in the signal level within the local region of the electron beam detection signal is small, and when the fluctuation in the signal level within the local region is large. 5. The scanning electron microscope according to claim 4, wherein the scanning electron microscope is a noise removing means that sometimes outputs the electron beam detection signal as it is. 8. A pattern appearance inspection apparatus comprising the scanning electron microscope according to claim 3 or 4, and inspecting a pattern of a sample using the electron beam detection signal restored by the restoration means. 9. In a pattern appearance inspection device that inspects the pattern of a sample by comparing and inspecting an electron beam detected image that does not use a restoring means from an electron beam detection means of a scanning electron microscope and an ideal image generated from design data, the above-mentioned A pattern appearance inspection apparatus comprising means for adding response characteristics of the electron beam detection means to an ideal image. 10. In a scanning image detection device such as a laser scanning microscope, a visual inspection device, or a foreign object detection device using an imaging means of a point sensor, a noise removing means for removing noise from a scanning image detection signal of the imaging means; and the imaging means. approximating the response characteristics of by an appropriate transfer function, and processing the noise-removed scanning image detection signal of the noise removing means in accordance with a formula obtained by inverse Laplace transform of the inverse function of the transfer function, and restoring the deterioration of the scanning image detection signal. A scanning image detection device comprising: means. 11. The scanning image detection device according to claim 10, wherein the imaging means of the point sensor is a photomultiplier tube.