JPH0897114A - Alignment method - Google Patents

Abstract

PURPOSE: To maintain alignment accuracy even if an alignment mark is partially abnormal by obtaining each chip arrangement from a mark position by the least squares method, adjusting weight according to the residue between a detection mark position and the mark position obtained from the chip arrangement, repeating the least squares method, and obtaining the chip arrangement again. CONSTITUTION: An alignment optical system 9 measures the intensity change of reflection light from a mark due to the travel of a stage 7 using laser beams 10 etc., to detect the mark position in xy directions of a wafer 8. The arrangement of each chip on the wafer 8 determined from the mark position obtained by the alignment detection signal is obtained by the least square method. Then, weight is adjusted by the residue between the mark position obtained by the alignment detection signal and the mark position on the obtained chip arrangement and the least squares method is repeated to obtain the chip arrangement again, thus eliminating error data even if the alignment measurement data partially have an error or reducing the weight for improving alignment accuracy.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to improvement of position detection accuracy of an exposure apparatus and a pattern position measuring apparatus used in a manufacturing process of semiconductor devices such as IC and LSI.

[0002]

2. Description of the Related Art In recent years, with the miniaturization of circuit patterns such as LSI, an optical reduction projection exposure apparatus having high resolution has been widely used as a pattern transfer means. However, in order to manufacture a semiconductor device having a pattern dimension of 0.2 μm or less, the conventional optical reduction exposure apparatus has a limit in pattern resolution. Therefore, the electron beam exposure apparatus is receiving attention. When writing a device pattern with an electron beam exposure apparatus, it is necessary to align the wafer with high accuracy prior to writing. The alignment detects the mark position formed on the wafer. This wafer mark measuring method can be roughly classified into a global method and a die-by-die method. In the die-by-die method, since alignment is performed for each chip on the wafer, highly accurate alignment is possible. However, in the actual production, the global alignment has an advantage that the number of times the mark position is detected is small and the time required for the alignment is short.

In the global alignment method, several alignment marks formed on a wafer are measured by using an alignment optical system, and the alignment of chips is corrected from the obtained mark positions to perform exposure.

The alignment accuracy of the exposure apparatus as described above is affected by, for example, the coating state of the resist on the wafer and the state of the marks formed on the wafer. A positioning method for reducing the influence of the mark shape is reported in, for example, Japanese Patent Laid-Open No. 2-294015.

In this alignment method, the reliability of each measurement position data is determined from the state of the detection signal (mark signal) of the mark provided on the chip or the state of the measurement position data for each chip position. The array (correction position data) of each chip on the wafer is determined from the value related to each position measurement data weighted according to the reliability.

Further, in this alignment method, when the determined chip arrangement is represented by an approximate function, the actual exposure position indicated by the measured position data and the square or absolute of the residual of the exposure position indicated by the approximation function. The value is weighted according to the reliability of each measurement position data, and the least squares method is applied to determine the approximation function so as to minimize the sum of the weighted residuals.

[0007]

However, the reliability of each measurement position data is determined from the state of the mark detection signal (mark signal) provided on the chip or the state of the measurement position data for each chip position. The method had the following problems. FIG. 5 shows an example in which the alignment detection position changes due to factors such as the asymmetry of the resist 51 coated with the alignment mark. (A) shows the case where there is no mark asymmetry, and (b) shows the case where there is mark asymmetry. In the alignment detection signal of (b), the waveform of (a) is often parallel-shifted and output. In this way, when the shape of the alignment detection signal does not change and is parallel-shifted, the conventional positioning method determines that (a) and (b) the mark position measurement values have the same reliability.

The present invention has been made in view of the above circumstances, and an object thereof is data processing in which the alignment accuracy is not affected by the abnormal data even when a part of the detected alignment mark is abnormal. Is proposed.

[0009]

In order to achieve the above object, according to the present invention, on the wafer, the state of the detection signal of the mark provided for each chip area or the measurement position data for each chip area is used. The array of each chip is obtained by the least squares method, the weight is adjusted by the residual between the mark position obtained from the alignment detection signal and the obtained mark position on the chip array, and the least squares method is repeated to re-create the chip array. Looking for. As a result, even when a part of the alignment measurement data has an error, it is possible to improve the alignment accuracy by removing the error data or reducing the weight.

[0010]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS The details of the present invention will be described below with reference to an embodiment in which an electron beam drawing apparatus for forming a circuit pattern on a wafer is taken as an example. FIG. 1 is a schematic diagram showing the configuration of an electron beam drawing apparatus which is an embodiment of a charged beam drawing apparatus according to the present invention. As shown in this figure, an electron gun 2 is installed in the upper part of the lens barrel 1a, and an electron beam 3 is emitted from the tip of the electron gun 2. The irradiated electron beam 3 is focused by a focusing lens 4, passes through a beam scanning deflector 5, is focused again by an objective lens 6, and is imaged on a wafer surface 8.

On the other hand, there is a sample chamber 1b in the lower part of the lens barrel 1a, and at the bottom of the sample chamber 1b, there is xy by means of moving means not shown.
A stage 7 on which a wafer 8 movable on a plane is placed is arranged. The irradiation position of the electron beam 3 on the surface of the wafer 8 is corrected by the deflector 5 and the deflector control system 12. Further, it is composed of an alignment optical system 9 for detecting marks formed on the wafer, a calculation unit 13 for calculating the obtained measurement values, and a control system for correcting the irradiation position of the electron beam. The alignment optical system 9 detects the mark position in the xy direction of the wafer 8 by measuring the intensity change of the reflected light from the mark with the movement of the stage 7 using the laser light 10 and the like. In the present invention, the method of the detection optical system 9 is not particularly limited.

FIG. 2 shows an arrangement example of the alignment marks 16 previously formed on the wafer 8. Prior to drawing the wafer 8 with an electron beam, the alignment optical system 9 measures the position of the alignment mark 16 while the wafer 8 is mounted on the stage 7. Wafer chip position error is shown in Figure 3.
, The linear error (parallel shift 31, rotation 32,
It can be separated into a magnification 33, an orthogonality 34) and a random error 35. Further, when the difference between the theoretical mark position (x, y) and the mark position obtained by measurement is (dx, dy), the deformation of the wafer is expressed by the following equation.

[0013]

## EQU00001 ## dx = .alpha.x-(. Theta.c + .theta.o) y + Mx.multidot.x + Sx ₍₁₎ dy = .alpha.y + .theta.c.x + My.y + Sy where (.alpha.x, .alpha.y) is the parallel shift error in the xy direction,
(Mx, My) represents a magnification error, (θc) represents an error in the rotational direction, (θo) represents an orthogonality error, and (Sx, Sy) represents the remaining random error. These linear error coefficients (αx, αy
, Θc, θo, Mx, My) are calculated by the least squares method in the calculation unit 13. Although the derivation of the first-order linear component has been described here, the higher-order component can be obtained by the same method.

However, the least-squares method is premised on that the measured values are not biased, the error has a normal distribution, and the model has no approximation error. However, these assumptions are ideal conditions that can be approximately satisfied only in the final stage of data analysis, and cannot be satisfied in the initial stage of data analysis. The actual measurement value may have a large error due to the shape error of the alignment mark. Mechanically applying the least squares method to such incomplete data leads to erroneous measurements and leads to totally erroneous values. When a person makes a graph or the like and directly fits it, he / she can immediately notice such an abnormality and take some kind of treatment. However, there are times when a computer does not realize that there is a problem because the calculation is done mechanically. As a reflection of such experience, we anticipate imperfections in the data and models, and assume that the majority of the data is reliable, although the small number of unspecified data may be erroneous and the fitting is performed. . The goal is
(A) It is difficult for the estimated values of the parameters to shift even when there is an error in a part of the data. (B) Not only when the error distribution is a normal distribution, but also various nonlinear types with wide tails Also for the error distribution, the variance of the parameter estimates is sufficiently small. One of the estimation methods that realizes such a goal is Tukey (National Ac).
ademyof Sciences, Washington
on (1974), pp. 3-14) et al.

Now, n measurement values y _i (i = 1 to n) are given, and the probability distribution of the error under each measurement condition is P () using the true value y _i ⁰ and the variance σ _i ^2. y _i ; y _i ⁰ , σ
It is assumed that _i ² ) = P ((y _i −y _i ⁰ ) / σ _i ² ) is assumed. The likelihood for the parameter estimate x is

[0016]

[Equation 2] Is. The maximum likelihood condition is expressed in the form of log-likelihood,

[0017]

[Equation 3] Becomes Differentiating this with the parameter x _j (j = 1 to n), the equation

[0018]

[Equation 4] Is obtained. Here, v _i is the residual (= y _i −f
(X)), A _ij are Jacobian coefficients f _i / x _i , Ψ (v
_i ) is

[0019]

Ψ (z) = − d (log P (z) / dz) (5) If the error distribution is a normal distribution,

[0020]

## EQU6 ## Ψ (z) = z (6), and the equation (4) is the same as the equation (7) of the least square method.

[0021]

[Equation 7] On the other hand, if we consider an error distribution other than the normal distribution,
Expression (4) is a basic expression of the maximum likelihood estimation method for the error distribution, and is different from the least squares method.

Further generalizing the idea, the error distribution P
Taking any function Ψ (z) apart from (z),
Consider equation (4) as a basic equation for fitting.
If equation (4) is written so as to correspond to equation (7),

[0023]

[Equation 8] Therefore, instead of the least-squares weight ω _i = σ _o ² / σ _i ² , the effective weight

[0024]

Ω _i ^eff = [Ψ (v _i / σ _i ) / (v _i / σ _i )] · σ _o ² / σ _i ² = [Ψ (v _i / σ _i ) / (v _i / σ _i )] · ω _i is equivalent. Here, it is equivalent to using z _i = v _i / σ _i . However, since this effective weight depends on the residual, the effective weight must also be changed according to the change in the estimated value of the parameter. By thus adjusting the effective weights and repeating the weighted least squares method, it is possible to realize an arbitrary fitting method based on the maximum likelihood method. The weight adjustment factor ω _i ^eff is given as follows.

(I) When the standardized residual | z _i | is small,
It has the same weight as the least squares method. (ii) When the standardized residual | z _i | is extremely large, the weight is set to zero. (iii) In the middle, the weight changes continuously.

FIG. 4 is an example showing an alignment measurement error when the alignment is performed. The vertical axis 41 in the figure is
It shows the difference between the theoretical value of the alignment mark seat and the position where the measured value is grid point corrected. The horizontal axis 42 represents the alignment X mark position when the center of the wafer is the origin. From this figure, 6 and No. It can be seen that the mark 10 is off by about 0.1 μm. In this way, when the mark detection error is large, it is recognized that it is abnormal, and the abnormal value may be removed by the above method when calculating the grid point correction. Although the present invention has been described using the electron beam apparatus, it can be naturally applied to the alignment of the optical reduction projection exposure apparatus and can be applied to any position measuring apparatus.

[0027]

As described above, according to the present invention, even if a large alignment error occurs due to an abnormality in the alignment shape, the alignment can be performed without being affected by the abnormal data. It becomes possible to do.

[Brief description of drawings]

FIG. 1 is a schematic configuration diagram of an embodiment of the present invention.

FIG. 2 is a plan view showing alignment marks formed on a wafer.

FIG. 3 is a plan view showing classification of errors due to alignment.

FIG. 4 is a diagram showing an example of alignment measurement error.

FIG. 5 is a cross-sectional view showing a change in a detection signal when the alignment mark is asymmetric.

[Explanation of symbols]

 1a ... Lens barrel 1b ... Sample chamber 2 ... Electron gun 3 ... Electron beam 4 ... Focusing lens 5 ... Deflector 6 ... Objective lens 7 ... Stage 8 ... Mask 9 ... Position detector 10 ... Laser optical axis 11 ... Position detector control System 12 ... Deflector control system 13 ... Computing unit 16 ... Alignment mark

Claims (2)

[Claims] 1. A positioning method for aligning a region on a sample to a desired position, the mark positions of several regions on the sample are detected, the detected mark positions are obtained, and the detected mark is detected. The array of each chip is obtained from the position by the least squares method, the weight is adjusted by the residual between the detected mark position and the mark position obtained by the array of the chips, and the chip array is reiterated by repeating the least squares method. Positioning method characterized by seeking. 2. A method of adjusting a weight by a residual between a detected mark position and a mark position obtained by the arrangement of the chips, wherein when the residual is small, the weight is equivalent to that of the least squares method. The alignment method according to claim 1, further comprising means for adjusting the weight to zero when the residual is extremely large, and adjusting the weight so as to continuously change between the small residual and the large residual. .