JPH10306387A - Method for estimating etching quantity by neural network - Google Patents

Abstract

PROBLEM TO BE SOLVED: To estimate etching amount at the arbitrary point on patterns without actually executing etching for any given resist patterns. SOLUTION: At the time of estimating the etching amts. at the arbitrary measurement point set on the resist patterns, the six parameters consisting of at least (1) a metal exposure rate near the measurement point, (2) an angle formed by the adjacent two sides at the measurement point, (3) a distance from the measurement point to a forward characteristic point, (4) a distance from the measurement point to a backward characteristic point, (5) an angle formed by the adjacent two characteristic sides at the forward characteristic point and (6) an angle formed by the adjacent two sides at the backward characteristic point are made inputs. The neural network having the erosion direction of the angle formed by either of (A) an erosion rate which is a distance from the measurement point to the shortest actual object point and (B) the adjacent two sides at the measurement point and the side heading toward the shortest actual object point from the measurement point as output is built. The etching amts. (A), (B) are estimated with respect to the arbitrary measurement point by using the neural network N.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method for estimating an etching amount using a neural network, and more particularly to a method for estimating an etching amount using a neural network, which is suitable for designing a resist pattern to be used as a mask during etching.

[0002]

2. Description of the Related Art Generally, an etching product is manufactured by etching a metal plate made of a material using a resist pattern formed on the surface of the metal plate as a mask. As such an etching product, for example, there is a so-called lead frame used as a wiring of a semiconductor device.

When a new lead frame is manufactured, it is extremely important to accurately design a resist pattern used for the manufacture. As a method of newly designing such a resist pattern, first, a correction margin corresponding to a dimension expected to be excessively etched with respect to a target lead frame shape (product pattern) is added according to an empirical rule. Tentatively design a resist pattern, make a prototype by actually using the resist pattern on the production line, measure the prototype with measuring equipment, compare the actual measurement with the target value, and set the actual measurement target If it is far from the value,
There is a method of correcting the correction allowance and using it as a design resist pattern.

Another method is to perform a plurality of etching experiments on an arbitrary resist pattern, collect several types of experimental data, and analyze the data mathematically using a statistical method. There is a method of estimating and designing a correction allowance set for the resist pattern.

[0005]

However, in designing a resist pattern, a method of actually using a resist pattern temporarily designed as described above in a production line to create a prototype requires trial and error. Therefore, it takes a long time to obtain the results, and it is not economical because of the cost of material and other factors.In addition, when a new product pattern is determined, and when an unknown resist pattern is designed for that, past data is used. There is a problem that cannot be used as it is. In addition, the method of designing a resist pattern by estimating the correction allowance using the mathematical method has a problem in the accuracy of the estimated shape (correction allowance).

SUMMARY OF THE INVENTION The present invention has been made to solve the above-mentioned conventional problems.
It is possible to estimate an etching amount at an arbitrary point on a pattern without actually performing etching, and therefore, to design a resist pattern used for manufacturing an etching product of an arbitrary pattern with high accuracy and in a short time. It is an object of the present invention to provide a method of estimating an etching amount by a neural network which can be applied in such a case.

[0007]

According to the present invention, when an arbitrary measurement point is selected on a resist pattern and an etching amount is estimated for the measurement point, at least (1) a metal exposure rate near the measurement point, 2) the angle between two adjacent sides at the measurement point, (3) the distance from the measurement point to the front feature point, (4)
Six parameters consisting of an angle formed by two adjacent sides at the front feature point, (5) a distance from the measurement point to the rear feature point, and (6) an angle formed by the two adjacent sides at the rear feature point are input.
This problem has been solved by constructing a neural network that outputs an etching amount and estimating an etching amount at an arbitrary measurement point using the neural network.

[0008] In general, the advantage of using a neural network is that, as long as the relationship between the input and the output is known, various correlations existing therebetween can be taught to the network through "learning". The acquired network has the ability to estimate even unknown data. Even in a neural network having such features, the network cannot be converged through the learning based on the actually measured values unless an input value and an output value that sufficiently express the problem to be modeled thereon are found.

As a result of various studies for modeling the estimation calculation processing of the etching amount on the neural network, the present inventors set arbitrarily measurement points on the resist pattern, and set (1) ) To (6) are measured or calculated, and these values are used as input values, and include, for example, (A) the amount of erosion and (B) the direction of erosion which are actually measured at the measurement point. By learning the neural network so that the etching amount becomes an output value, it has been found that a process of estimating the etching amount can be modeled on the neural network with high accuracy.

The present invention has been made on the basis of the above findings. For example, by using the above-described neural network constructed on a computer, the etching amount of an unknown resist pattern can be accurately determined at an arbitrary measurement point. Can be estimated, and as a result, it is possible to estimate the product shape after etching with high accuracy based on the etching amount at each measurement point.

[0011]

Embodiments of the present invention will be described below in detail with reference to the drawings.

FIG. 1 is a block diagram conceptually showing a neural network applied to an etching amount estimation method according to an embodiment of the present invention.

Although not shown, the neural network N used in the present embodiment has an input layer composed of six terminals, an output layer composed of two terminals, and a 1-layer composed of, for example, six neurons between these input / output layers. The intermediate layer of the layers and the synaptic connection between each of the input layer terminal, the intermediate layer neuron, and the output layer terminal are configured by software on a computer.

In this embodiment, (1) the metal exposure rate near the measurement point, (2) the angle between two adjacent sides at the measurement point,
(3) the distance from the measurement point to the front feature point; (4) the angle between two adjacent sides at the front feature point; (5) the distance from the measurement point to the back feature point; and (6) the two adjacent sides at the back feature point. And (A) the erosion amount that is the distance from the measurement point to the shortest real point, (B) one of two adjacent sides at the measurement point, and the measurement point The above-mentioned neural network N is constructed to output the erosion direction which is the angle formed by the edge toward the shortest real point. This neural network N
The etching is actually performed using the resist pattern, an arbitrary measurement point on the resist pattern is selected, and specific values of the input parameters (1) to (6) are input for the measurement point, It is constructed by performing the learning so that the above (A) and (B) actually measured for the real thing are output.

In the present embodiment, six geometrical values that characterize the measurement point composed of the above parameters (1) to (6) are input and the above (A) and (B) that characterize the etching amount are input.
Are output as two geometric values. Therefore, these inputs and outputs will be described first.

FIG. 2 conceptually shows a part of a resist pattern as a design drawing for explaining input parameters. The resist pattern R has a coordinate value defining each refraction point and a direction of the pattern. Since the data is input as CAD data composed of the following, it is described by a straight line (vector) having the direction of the pattern shown by the thick solid line and the arrow in the figure. The resist pattern R is defined such that the direction of the pattern (the direction of the arrow) is forward, the right side of the straight line is the resist part, and the left side is the exposed metal part (non-resist part).

Description will be made on the assumption that a refraction point in the middle of the resist pattern R is a measurement point.

(1) Metal exposure rate near the measurement point. This is the ratio of the area (shaded area) located on the metal part to the total area of a circle having a predetermined radius centered on the measurement point indicated by the symbol M in the figure. This is substantially equal to the following “(2) angle between two adjacent sides (angle on the metal side) at the measurement point” with respect to 360 ° in the case of a simple pattern as shown in FIG. As shown in FIG. 3 (A), in the case of a thin lead tip, or as shown in FIG. 3 (B), when the distance between adjacent leads is narrow, it cannot be defined by the angle formed by the two sides. The area ratio is as follows.
The radius of the circle used in this calculation can be, for example, の of the thickness of the metal material used.

(2) The angle between two adjacent sides at the measurement point. This is an angle formed by two sides adjacent to each other at the measurement point, and is shown in FIG. 2 as "an angle formed by the measurement point".

(3) Distance from measurement point to front feature point. Here, the characteristic point means a point representing the characteristic of the pattern in the vicinity of the target measurement point. Usually, the nearest refraction point is selected as this characteristic point. In the case of a long time, feature points may be set on a straight line as needed. Since FIG. 2 shows a case where the feature point is a refraction point where two sides intersect, the front feature point is the refraction point on the right side in the figure, which is in the front, that is, in the direction of the pattern direction. Therefore, in this example, the distance to the front feature point corresponds to the “distance to the previous refraction point” shown in FIG.

(4) An angle between two adjacent sides at the front feature point. This corresponds to the “angle formed by the previous refraction point” in the case of FIG.

(5) Distance from measurement point to rear feature point. In FIG. 2, the rear feature point is a refraction point on the left side contrary to the above (3), and the distance to this rear feature point is
This corresponds to the “distance to the rear refraction point” shown in FIG.

(6) The angle between two adjacent sides at the rear feature point. This corresponds to “the angle formed by the rear refraction point” in the case of FIG. 2 as in (4).

When the characteristic points are set on a straight line as described above, the angle between the two adjacent sides in (4) and (6) is, of course, 180 °.

Next, the amount of etching which is the output of the neural network N will be described with reference to FIG.

As a result of actually applying the resist pattern partially shown in FIG. 2 to a metal material and performing etching,
It is assumed that a real pattern (denoted as real data in the figure) indicated by a thick solid line in FIG. 4 is obtained. As shown by a thin solid line in the same resist pattern R as in FIG. 2, for the measurement point M on the resist pattern R, the erosion amount is shown in FIG. 2A and the erosion direction is shown in FIG.

That is, (A) the amount of erosion is the shortest distance from the measurement point M to the actual object, and (B) the erosion direction is the direction from the measurement point M to the shortest point on the actual object (vector indicated by an arrow). ) And a straight line (vector representing the direction of the pattern) ahead of the measurement point M.

Next, the operation of the present embodiment will be described.

First, the neural network shown in FIG. 1 is created on a computer by software.
The learning of the neural network N is performed for each of the measurement points M1 to M9 on the resist pattern (design pattern) R shown in FIG. 5 corresponding to a part of one lead. This was performed using learning data composed of data and measured data.

Briefly describing the input data shown in FIG. 6, in the case of the measurement point M5, the metal exposure rate is 70.03%, and the angle between two adjacent sides is 180 °.
Since the front feature point is M6, the distance to this point is 0.
10, the angle between two adjacent sides at this point is 270 °, and similarly, since the rear feature point is M4, the distance to this point is 0.10, and the angle between two adjacent sides at this point is 270 °
It turns out that. For other measurement points, the input data of (1) to (6) can be obtained in the same way.

Here, the radius of the circle set for calculating the metal exposure rate is set to 0.25 mm (250 μm) exceeding the resist pattern width of 0.20 mm at the tip of the lead.
Therefore, the measurement point M5 has a metal exposure rate of 70.03, although the angle between two adjacent sides is 180 °.
%It has become. As can be seen from the relationship between the other resist pattern R shown in FIG. 7 and the dashed circles A and B for calculating the metal exposure ratio at the two measurement points M1 and M5, respectively, The rate is maximum at M5, M1 (M9
Also the same).

When the distance to the feature point exceeds 1 mm, it is assumed that the distance is long enough for the estimation calculation, and 1 mm is set here in all cases. Accordingly, in the case of the measurement points M1 to M4, the rear feature points are all Ma, and all are 1 mm assuming that they are sufficiently separated from each other as shown by an extended portion of the design pattern by a two-dot chain line with a middle part omitted. Also in the case of the measurement points M6 to M4, the front feature points are all Mb, and are set to 1 mm similarly, assuming that they are sufficiently separated.

As a result of learning the neural network N using the above data, by inputting the input data of FIG. 6 to the neural network N, output value data shown in the table of FIG. 8 can be obtained.

In the columns of output value and actual value corresponding to each measurement point shown in the table of FIG. 8, the upper left side shows the erosion amount (μm), the right side shows the erosion direction (°), and the lower side shows the erosion direction (°). The left side shows the x coordinate value converted using the erosion amount and the erosion direction, and the right side shows the y coordinate value.

As described above, the learning converges, and a neural net is constructed by modeling the process of estimating the etching amount by the learning.
The resist pattern is designed according to the procedure shown in (1).

In step 1, when a new product pattern of the lead frame is generated (for example, an order is received), in the next step 2, a correction margin for widening the width of the product pattern is set. In this first operation,
Set a value that is considered appropriate based on conventional empirical rules.

Next, in step 3, a provisionally designed resist pattern is generated by adding the above-mentioned correction allowance to the above-mentioned product pattern, and the etching amount ((A)) of the arbitrary measurement point M on the resist pattern is calculated by the neural network N. The amount of intrusion, (B) the direction of intrusion) is estimated (step 4). As the measurement point M, a point necessary for setting a resist pattern, such as the refraction point, is selected.

Next, the actual pattern estimated from the etching amount estimated in the above step 4 is compared with the product pattern generated in the step 1 to evaluate whether or not the error is within an allowable range (step 5). .

If the result of the evaluation in step 5 indicates that the error is out of the allowable range, the process returns to step 2 to correct the correction allowance for resetting the correction allowance. Is OK, the correction allowance is determined as a design value.

The above-described design operation is performed for all the measurement points necessary for designing the resist pattern R, and the overall correction allowance is determined, and the resist pattern design is completed. And the process proceeds to an etching operation using the designed resist pattern.

As described in detail above, in the present embodiment, by setting the above six parameters (1) to (6) as inputs, the resist pattern used for manufacturing an etching product and the actual product after etching can be obtained. The relationship between the shapes can be modeled on the neural network N. As a result, the neural network N constructed on the computer makes it possible to highly accurately estimate the actual shape of an unknown resist pattern after etching.

According to the present embodiment, the following effects can be further obtained.

(I) By changing the setting of the interval between measurement points, it is possible to perform a detailed estimation to a rough estimation, so that it is possible to design a resist pattern with arbitrary accuracy. (Ii) Since the resist pattern at each measurement point is characterized by a geometric value, the data amount of input data necessary for estimating the etching amount by a neural network can be reduced. (Iii) Since the process of estimating the etching amount can be performed on a computer, the estimation result can be obtained in a short time. (Iv) Since the amount of etching can be estimated on a computer, no cost such as material cost is required. (V) Since the etching amount can be estimated in a short time, defects on the resist pattern can be quickly found,
The correction can be made quickly.

As described above, the present invention has been specifically described. However, the present invention is not limited to the above-described embodiment, and can be variously modified without departing from the gist thereof.

For example, the input parameters are not limited to the six types shown in the above embodiment. For example, the distance from the measurement point may include not only the distance to the nearest feature point, but also the distance to another feature point on the pattern preceding the feature point, or the angle formed by two adjacent sides. May also include angles of feature points other than the most recent feature point. Further, etching conditions such as temperature may be included in the input parameters.

[0046]

As described above, according to the present invention,
With respect to an arbitrary resist pattern, an etching amount at an arbitrary point on the pattern can be estimated without actually performing etching. Accordingly, it is possible to design a resist pattern used for manufacturing an etching product having an arbitrary pattern with high accuracy and in a short time.

[Brief description of the drawings]

FIG. 1 is a block diagram conceptually showing a neural network applied to an embodiment according to the present invention.

FIG. 2 is an explanatory diagram conceptually showing input parameters of a neural network.

FIG. 3 is an explanatory view showing a concept of a metal exposure rate.

FIG. 4 is an explanatory diagram conceptually showing an output of a neural network.

FIG. 5 is an explanatory diagram showing a specific example of measurement points and their input parameters.

FIG. 6 is a table showing an example of learning data of a neural network.

FIG. 7 is a diagram for explaining a difference in a metal exposure rate at each measurement point.

FIG. 8 is a chart showing an output result by a neural network.

FIG. 9 is a flowchart showing a procedure for designing a resist pattern.

[Explanation of symbols]

 N: Neural network M: Measurement point R: Resist pattern

──────────────────────────────────────────────────の Continued on the front page (51) Int.Cl. ⁶ Identification code FI G06F 15/60 658Z (72) Inventor Satoshi Watanabe 1-1-1 Ichigaya-Kagacho, Shinjuku-ku, Tokyo Inside Dai Nippon Printing Co., Ltd. 72) Inventor Yuichiro Shimoto 1-1-1, Ichigaya-Kaga-cho, Shinjuku-ku, Tokyo Within Dai Nippon Printing Co., Ltd. (72) Inventor Yuji 1-1-1, Ichigaya-Kaga-cho, Shinjuku-ku, Tokyo Dai Nippon Printing Co., Ltd. In-house (72) Inventor Hiroshi Suzuki 1-1-1 Ichigaya-Kagacho, Shinjuku-ku, Tokyo Dai Nippon Printing Co., Ltd.

Claims (2)

[Claims] When an arbitrary measurement point is selected on a resist pattern and an etching amount is estimated for the measurement point, at least (1) a metal exposure rate near the measurement point, and (2) two adjacent sides at the measurement point. (3) the distance from the measurement point to the front feature point, (4) the angle between two adjacent sides at the front feature point, (5) the distance from the measurement point to the back feature point,
(6) 6 consisting of the angle between two adjacent sides at the rear feature point
A method for estimating an etching amount by a neural network, comprising constructing a neural network having two parameters as inputs and an etching amount as an output, and estimating an etching amount at an arbitrary measurement point using the neural network. 2. The method according to claim 1, wherein the etching amount is: (A) an erosion amount that is a distance from a measurement point to a shortest real point; (B) one of two adjacent sides at the measurement point; A method for estimating an etching amount by a neural network, wherein the erosion direction is an angle formed by a straight line drawn to the shortest real point.