JPH11150047A - Semiconductor device manufacturing method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To effectively and successfully incorporate an experiment for controlling a semiconductor device performance into a manufacturing step. SOLUTION: An important factor selection (S18) is provided in a manufacturing experiment or a simulation step or by a computer. In addition semi- automation, is attained by having experimental data (S21 to S23) stored in the computer, performing semi-automation of the entire processing by the use of the computer and by providing a man-machine interface to process or the make judgements and statistical calculation processings with respect to the experimental result.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

[0001] The present invention relates to a method for manufacturing a semiconductor device.

[0002]

2. Description of the Related Art In recent years, miniaturization of semiconductor devices has been rapidly advanced. In this miniaturization process, manufacturing conditions in many manufacturing steps (for example, artificially variable amounts such as impurity ion dose and implantation angle) and environmental factors (for example, temperature,
Raw materials, operators of manufacturing equipment, and the like (hereinafter, referred to as factors) interact with each other to affect the electrical or physical characteristics (hereinafter, referred to as semiconductor characteristics) of the semiconductor device. In order to improve the quality control and productivity of semiconductor devices, semiconductor characteristics are strictly controlled in the manufacturing process, that is, values indicating semiconductor characteristics (for example, threshold voltage value, saturation current value, etc. Value) is always near the target value, and the deviation from the target value must be kept within an expected range. In order to control the semiconductor characteristics as described above, an optimum value of a condition value (hereinafter referred to as a condition value; for example, an actual number such as the number of ions, the number of tilt angles, and the like) of the above factors is determined in a design stage of a semiconductor device. It is necessary to reflect it in the actual manufacturing process. However, as described above, since many factors affect the characteristics of the semiconductor in a complicated manner, it is difficult to determine the optimum condition value. Further, even if an attempt is made to optimize the condition value of a factor in a manufacturing process, an error always occurs due to manufacturing equipment and environment.

In the design of semiconductor devices,
In order to determine the optimum manufacturing condition value, an experiment in which condition values are assigned to a plurality of factors or a simulation using a computer (hereinafter collectively referred to as an experiment) is performed.
However, it is practically difficult to take all the factors into the experiment. Therefore, in the past, narrowing down the factors to be taken up in the above experiments often relied on experience and intuition. In the experiment procedure, we first experiment with factors that seem to have a large influence on semiconductor characteristics empirically, and then assign condition values for other factors to the optimal condition value alone and conduct experiments. In many cases. When the conclusion of the experiment is obtained in this way, only one combination of condition values considered to be optimal is obtained.

FIG. 3 is an outline of a conventional method for manufacturing a semiconductor device. The manufacturing conditions of the semiconductor device are designed in S41, and the designed manufacturing conditions are tested in S42 as described above.
In step S44, the optimum condition value is determined. In step S44, a prototype of the semiconductor device is actually manufactured. In step S45, it is determined whether the prototype has the target characteristic value, and it is confirmed that the manufacturing conditions are appropriately determined. Here, if the manufacturing conditions are appropriate, a mass production system is entered in S46. However, if it is determined that the manufacturing conditions are not appropriate, the manufacturing conditions are changed based on empirical judgment in S47, and an experiment for confirmation is performed again in S42.

[0005]

However, in the above-described conventional configuration, not only the interaction of a plurality of factors cannot be determined correctly, but also the effect of factors that could not be conceived from experience is highly likely to be overlooked. There is. In addition, because the experiment was performed to find one combination of condition values considered to be optimal, the target characteristic value could not be obtained when the semiconductor device was actually prototyped or entered mass production. In addition, the change of the manufacturing condition value depends on empirical judgment, and the result of the experiment cannot be used. In addition, not only the man-hour for experimenting again to change the manufacturing conditions is expended, but also as a result, the trial production and mass production are repeated many times from the experiment, and the man-hour and cost are further increased.

In view of the above problems, an object of the present invention is to
A method of manufacturing a semiconductor device capable of efficiently determining, using a computer, a procedure for comprehensively determining the influence of a plurality of factors on semiconductor characteristics (including the effects of factors) and determining optimum manufacturing conditions using a computer. Is provided. The determination method used in the present invention is a data aggregation method (variance analysis, regression analysis, etc.) based mainly on statistical theory, and is hereinafter referred to as statistical calculation.

[0007]

In order to achieve this object, a method of manufacturing a semiconductor device according to the present invention sets optimum condition values for a plurality of manufacturing conditions that may affect the characteristics of a semiconductor device. A manufacturing method, in which an experiment for determining an optimum condition value of a manufacturing condition is performed, and for this experiment, data defining each condition value with respect to the manufacturing condition included in the entire semiconductor device manufacturing process. A step of preparing a certain initial condition value card, a step of selecting a manufacturing condition to be subjected to an experiment, a step of combining each two condition values of the selected manufacturing condition, and each condition value card which is data recording the combination , A step of performing an initial experiment by reflecting each condition value data on a part of the initial condition value card, and a step of selecting important conditions from manufacturing conditions based on the results of the initial experiment , Increasing the condition value for the selected important manufacturing conditions, changing or adding each condition value card for the increased condition, and performing additional experiments by reflecting each condition value data on a part of the initial condition value card And a step of expressing the characteristics of the semiconductor device by a functional expression including a condition value of a manufacturing condition based on a result of the initial experiment and the additional experiment, and a step of determining an optimum condition value based on the functional expression. I have.

[0008]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Embodiments of a method for manufacturing a semiconductor device according to the present invention will be described below with reference to the drawings.

FIG. 1 shows an outline of a method of manufacturing a semiconductor device according to the present invention. In S1, the manufacturing conditions of the semiconductor device are designed and S
In step 2, a manufacturing condition designed based on the method of the present invention is tested, and at the same time, a function expression expressing the characteristic value of the semiconductor device by the condition value of the manufacturing condition is calculated. In step S3, the optimum condition value is determined. Here, the function formula obtained in S2 can be effectively used when determining the optimum condition in S3. Because, by the function expression,
This is because the characteristic value can be accurately predicted. With the optimum condition values determined in S3, a semiconductor device is actually prototyped in S4, and it is determined in S5 whether the prototype has a target characteristic value, and the manufacturing conditions are appropriately determined. If the manufacturing conditions are appropriate, the system enters the mass production system in S6. When it is determined that the manufacturing conditions are not appropriate, in S7, in light of the function formula obtained in S2,
Manufacturing conditions can be changed. This is because the function formula can also predict a change in the characteristic value when the condition value is deviated, so that the factor that reduces the margin at the time of manufacturing becomes clear from the function formula. That is, the function expression of S2 is
It can also be used effectively to change manufacturing conditions after trial production and mass production. As described above, after the manufacturing conditions are accurately changed in S7, a prototype is manufactured for confirmation.

FIG. 2 shows a flowchart in S2 of FIG. In FIG. 2, S11 to S17 indicate data processing, S21 to S23 indicate data files on a computer storage medium, and S31 to S34 indicate determination. The data processing may be performed automatically by a computer or semi-automatically including manual operation. The data file can be used as an input / output file at the time of automatic processing, or can be manually confirmed and edited.
For the determination, it is possible to automatically perform a statistical calculation in accordance with a statistical calculation parameter set in advance, and to perform the determination manually and empirically and exploratively.

The present invention is intended to improve the efficiency and quality of the entire semiconductor device manufacturing flow by providing S2 in FIG. 1, and has a special feature in the experimental method of S2. .

An embodiment of the method of manufacturing a semiconductor device according to the present invention will be described below with reference to the flowchart of FIG.

First, in step S11, characteristics of the semiconductor device to be controlled are selected, factors that are considered to affect the characteristics are listed, and condition values of the respective factors are determined. For example, when a threshold voltage, which is an electrical characteristic of a semiconductor device, is used as a characteristic value, factors such as a channel dose and energy are given as factors, and the upper and lower limits of the channel dose are used as condition values.
Determine the upper and lower limits of the amount of energy. In the experiment referred to in S11, two condition values are set for each factor, and the condition values are combined for all factors. This experiment is often called a 2n-type factor experiment (or a part thereof), and is in line with the theory of statistical calculation, and can be used to determine the degree to which each factor affects the characteristic value. S1
Since the subsequent processing, that is, experiments and statistical calculations are performed on the factors listed in 1 above, in S11, in addition to the factors that obviously affect the characteristic value, the factors that are likely to affect the characteristic value should be taken up. It is.

Next, in S12, combinations of the respective condition values of the factors listed in S11 are performed. A description of this combination in a table format is hereinafter referred to as an experiment plan table. S23 shows this experiment plan table, and Table 1 is an example of the experiment plan table.

[0015]

[Table 1]

In Table 1, each column indicates a factor, and each row indicates one experiment, that is, a combination of condition values. Numerical values (-1, 1) in the table are numbers (hereinafter, referred to as levels) assigned to the respective condition values, and are unique among the respective factors. Table 2 shows an example of each column (A to E) and each level of Table 1, and a table in which condition values corresponding to each level are described below is referred to as a level table.

[0017]

[Table 2]

In this example, for example, the first row of Table 1 contains A to D
Is -1 and E is 1, so the actual experiment on the first line is
From the level table in Table 2, the boron ion dose (A) = 2e-1
2 / cm ^Three, Boron ion dose energy (B) = 10 eV,
Arsenic ion dose (C) = 2e-12 / cm^Three, Arsenic ion
Energy (D) = 10 eV, annealing temperature (E) = 90
Indicates that the process is performed under the condition value of ° C

The processing in S12 following S11 is for generating an experiment plan table for the first time, and is different from the addition / change described later. The experiment plan table corresponding to S11 generated in S12 is hereinafter referred to as an initial experiment plan table. As described above, S1
In the initial experiment of No. 1, two condition values are set for each factor, and according to Table 2 (the number of columns is 5, the number of rows is 16) and 5
For one factor, 16 experiments are good. In this initial experiment, the degree of influence of each factor on the characteristic value can be compared.

The manufacture of a semiconductor device is performed by setting one condition value for each of various manufacturing conditions. S
The initial condition value card 21 is data in which various manufacturing conditions are defined and one condition value is recorded for each of the manufacturing conditions. In order to perform the experiment described above, S2
In addition to using the initial condition value card of No. 1, data for variously changing condition values is used only for the factors listed in the experiment. Such data for partially changing the initial condition value card is hereinafter referred to as each condition value card. S22
Indicates each condition value card.

[0021]

[Table 3]

[0022]

[Table 4]

[0023]

[Table 5]

[0024]

[Table 6]

Table 3 is an example of an initial condition value card, and Tables 4 and 5 are examples of each condition value card corresponding to Table 3. Table 4 corresponds to the experiment in the second row of Table 1, and Table 5 corresponds to the experiment in the third row of Table 1.

For example, when Table 4 is reflected in Table 3, five condition values (boron ion dose amount, dose energy,
The arsenic ion dose, dose energy, annealing temperature) are changed, and other conditions (gate oxidation, TEOS growth in the example of Table 3) are not changed. According to the contents of the change, an experiment is performed in S14, and the result is statistically calculated in S15.

For example, in a method of performing a semiconductor process simulation using a computer as an experiment in S14, conditions corresponding to the semiconductor device manufacturing process are read from the initial condition value card in S21, and the entire semiconductor device manufacturing process is simulated. Determine the set value when performing. Further, in order to conduct experiments on various manufacturing factors, S
The condition values recorded in each condition value card of No. 22 are applied to reproduce in a manufacturing process of a plurality of semiconductor devices. When an experiment using a computer is performed in this manner, S12
Steps S15 to S15 can be performed fully automatically, and steps S21 to S23 are data on a computer, so that efficient processing can be performed. Further, the efficiency of processing is further improved by sharing the reproduction results before the manufacturing process described in each condition value card, instead of reproducing all the manufacturing processes of the semiconductor device from the beginning to the end.

There are two types of statistical calculations for the results of the experiment. One is a calculation for comparing each factor.
The other is a calculation for expressing the characteristic value by a linear combination of the condition values of the factors (hereinafter, referred to as a response function or RSF), which is performed in S16. In step S31, it is determined which of the two types of calculations is to be performed. RSF is
This is effective as a means for reflecting an experimental result in a semiconductor device manufacturing process. For example, RSF is given by the following equation.

Y = a1 + a1 × x1 + a2 × x2
+. . . + Ai × x1 × x1 + aj × x2 × x2 Here, y is a characteristic value, xn is a factor value, and an is a coefficient. This approximation formula numerically clarifies the relationship between the characteristic value and the factor, and includes the interaction between the factors. In general, central composite programming is used as a method for obtaining such an approximate expression with high accuracy. To implement this central composite programming, an experiment using a fixed combination of five condition values of each factor is required. It is. From the results of experiments according to the central composite design, the characteristic values
By approximation, it is possible to predict the characteristic value when the condition value of the factor is changed, which is useful for controlling the manufacturing process of the semiconductor device. Of course, it is possible to obtain the above-described approximation formula only based on the experimental results with two condition values for each factor, but the accuracy is not sufficient to determine the manufacturing conditions of the semiconductor device. Conceivable.

In step S32, whether the statistical calculation result in step S15 is appropriate is automatically or semi-automatically examined. A fully automatic method relies on determining output values from statistical calculations according to a procedure involving certain parameters. For example, if the average of the characteristic values obtained by the experiment differs from the target characteristic value by 10% or more, it is determined that the condition value needs to be changed. If the determination procedure is registered in the computer in advance in this way, the processes from statistical calculation to factor selection can be performed completely automatically. However, it is important to add empirical judgments made by human intelligence. This is because the results of the experiment include errors and abnormal values, and consideration should be given to problems that do not become apparent through statistical calculations.

Therefore, it is important to provide a process for performing a semi-automatic process, that is, an interactive process. By this semi-automatic processing, in addition to the change of the automatic calculation formula and the change of the parameters used for the calculation and the determination criteria, it is possible to consider a problem that does not become apparent by the statistical calculation. In this way, when it is determined in S32 that the condition value needs to be changed, the condition value is changed in a fully automatic or semi-automatic process in S17. For example, factors are selected in descending order of the degree of influence on the characteristic value, the result when the condition value of this factor is changed is predicted, and a condition value that makes the prediction sufficiently close to the target value is selected. Here, the degree of influence of the factor on the characteristic value and the expected result can be expressed numerically by statistical calculation. Therefore, as in the case of the fully automatic determination in S32, the change of the condition value in S17 can be fully automatic, and in addition, semi-automatic.

After changing the condition value in this way,
The process proceeds to S12, changes the experiment plan table, changes each condition value card for the changed portion, and performs the experiment again. The process of S12 at this time corresponds to the change of the condition value of the change content (1).

On the other hand, if it is determined in S32 that the condition value does not need to be changed, it is determined in S33 whether the RSF should be calculated.
The process proceeds to 18. In S18, the statistical calculation result in S15 is processed fully or semi-automatically in order to select an important factor. The selection here is intended to reduce the number of additional experiments required for RSF calculation. As a fully automatic processing procedure, for example, the effect of a factor on a characteristic value is expressed numerically. Assuming that this effect is the ratio of the change width of the characteristic value to the change width of the factor, it can be calculated from the experimental results of two condition values of each factor.

In S34, whether to reduce the number of factors taken up in the experiment is automatically or semi-automatically determined based on the result of S18. As a fully automatic processing procedure, for example, when the above-mentioned effect is standardized to 0 or more and 1 or less, it is assumed that a boundary value for selecting a factor is 0.5 or more. Further, it is semi-automatically examined whether the result of the automatic selection is appropriate. For example, for factors near the boundary value,
Empirical judgment is required. After the factors are selected in this way, the process proceeds to S12, the experiment plan table is changed, each condition value card is changed for the changed portion, and the experiment is performed again. The processing of S12 at this time corresponds to the addition of the condition value of the change (2) when all the factors are selected. When some of the factors are selected, the change number (3) corresponds to the reduction of the number of factors and the addition of condition values. The reason for adding the condition value is to calculate the RSF.

After the additional experiment for calculating the RSF is performed in this way, the RSF is calculated in S16, and the process ends.

For example, the experiment plan table shown in Table 1 is set as an initial experiment in S31, and it is unnecessary to change the condition values in S32.
If the number of factors is not changed, a planning table as shown in Table 7 is created again. Table 8 is a level table of Table 7.

[0037]

[Table 7]

[0038]

[Table 8]

Lines 1 to 16 of Table 7 have already been tested in the initial experiment of Table 1, and additional experiments for RSF calculation are performed only for the eleven experiments of Lines 17 to 27.
In order to calculate the RSF, in addition to the two condition values set in the initial experiment, a combination experiment of the three condition values on both sides and the center is required. Table 2 shows two levels and Table 8 shows five levels. The levels (-2, 0, 2) in Table 8 correspond to the three condition values to be added. The experiments on the 17th to 27th lines in Table 7 are experiments in which these three condition values are combined, and this additional experiment enables calculation of the RSF based on statistical theory.

Here, a method of an additional experiment using each condition value card will be described. In order to carry out the additional experiment as described above, each condition value card is required for the experiment on lines 17 to 27 in Table 7. The initial condition value card shown in Table 3 has already been created in S21, and is recorded on the computer. An experiment plan table corresponding to Table 7 is created (S23), and S1
Each condition value card is created using the initial condition value card created in step 3. At this time, it is possible to recognize from the record on the computer that the first to sixteenth rows in Table 7 have already been tested.

In the recording on the computer, for example, a unified identification number is given to a series of processes from an initial experiment on one object to an RSF calculation. By managing the various cards and managing the current processing state based on the identification numbers, it is possible to automatically recognize the difference between the previous experiment and the current experiment as described above. In this way, the difference between Table 1 and Table 7 is automatically recognized, and the condition value cards (S22) corresponding to lines 17 to 27 of Table 7 are automatically created, and S14
Can be used for additional experiments. For example, Table 6 shows condition value cards corresponding to the 27th line of Table 7. In this way, for example, when the computer simulation is performed as an experiment, the change contents of the experiment are determined in S12 and input to the computer, and the processing from the experiment to the statistical calculation, that is, S13, S14, and S15 ( Alternatively, it is possible to perform steps up to S16) fully automatically.

Although the case of the additional experiment has been described above, there may be various other experiment plan change methods depending on the result of the examination of the condition values and the factor selection. Hereinafter, another method of changing the experiment will be described. In the following changing method, automatic processing using a computer is possible in the same manner as described above.

When the experiment plan table in Table 1 is set as an initial experiment in S31, the condition value is not required to be changed in S32, and the number of factors is reduced from five to two in S19, the plan table as shown in Table 9 is re-created. create. Table 10 is a level table of Table 9.

[0044]

[Table 9]

[0045]

[Table 10]

The experiments in the first to fourth rows of Table 9 can use the results of the initial experiments in Table 1. For example, 1 in Table 9
For the row, the results of the first to fourth rows of Table 1 can be used, and for the second row of Table 9, the average values of the fifth to eighth rows of Table 1 can be used. The fifth to ninth rows in Table 9 correspond to additional experiments for RSF calculation.

In this way, by selecting factors necessary for RSF calculation in advance, compared with Table 7,
The portion corresponding to the additional experiment is reduced from 11 experiments to four experiments, and the man-hour and cost for the experiment can be saved.
This effect increases as the number of factors taken up in the initial experiment increases, and the fewer the factors that actually have a large effect on the characteristic value, the greater the effect. Therefore, it can be said that this effect is an effective method in view of the actual semiconductor design process.

When it is determined in S32 that the condition value needs to be changed, the number of invalidated experiments can be minimized. For example, if the target characteristic value is not obtained after performing all the experiments in Table 7, 27 experiments are wasted. However, in the present embodiment, the first to first rows of Table 7
At the time when the 16 experiments on the sixth line are performed, it is possible to examine whether the obtained characteristic values are appropriate, and it is only necessary to discard the results of the 16 experiments.

In the present invention, in order to summarize the experimental results for determining the manufacturing conditions of the semiconductor device into a functional formula, the manufacturing conditions are changed theoretically after entering the development / prototyping stage using this functional formula. be able to. In addition, in the experiment method, the number of experiments required for calculating the function formula can be reduced by providing a step of determining a condition value from a characteristic value obtained by an experiment and a step of selecting an important factor in advance. . In addition, the entire process can be automated using a computer, and each statistical calculation process can include a semi-automatic process.

Here, the invention described in the claims is not limited to the embodiment described in the above embodiment.

[0051]

As described above, according to the present invention, the setting and changing of the manufacturing conditions after entering the development / prototyping stage can be performed with high accuracy by using the function formula, and the manufacturing conditions after the change of the manufacturing conditions are changed. The number of man-hours for re-testing can be reduced. Further, in the experiment method, since many manufacturing factors can be examined with a small number of experiments and a function formula can be obtained with a minimum number of experiments, the number of work steps and costs can be reduced. In addition, the entire process is automated using a computer, and semi-automatic processes are provided for each statistical calculation process to improve work efficiency and reflect empirical judgments in addition to theoretical judgments based on statistical calculations. can do. As a result, the electrical or physical characteristics of the semiconductor device can be more accurately and efficiently controlled.

[Brief description of the drawings]

FIG. 1 is a view showing a flowchart in a semiconductor device manufacturing method according to an embodiment of the present invention;

FIG. 2 is a flowchart showing an experimental method in the embodiment of the method of manufacturing a semiconductor device according to the present invention;

FIG. 3 is a flowchart showing a conventional method for manufacturing a semiconductor device.

Claims (1)

[Claims] 1. A manufacturing method for setting optimum condition values for a plurality of manufacturing conditions that may affect the characteristics of a semiconductor device, the method comprising: performing an experiment for determining the optimum condition values; For an experiment, a step of preparing an initial condition value card which is data defining each condition value for the manufacturing conditions included in the entire manufacturing process of the semiconductor device, and selecting the manufacturing conditions to be subjected to the experiment A step of combining each of the two condition values of the selected manufacturing conditions; a step of creating each condition value card which is data recording the combination; and a step of forming each of the condition values in a part of the initial condition value card. Performing an initial experiment reflecting the data; selecting important conditions from the manufacturing conditions based on the results of the initial experiment; and condition values for the selected important manufacturing conditions. Increasing the number of conditions, changing or adding each condition value card for the increased manufacturing conditions, a step of performing an additional experiment by reflecting each condition value data on a part of the initial condition value card, A step of expressing a characteristic of the semiconductor device based on a result of an experiment and the additional experiment by a function formula including a condition value of the manufacturing condition; and a step of determining the optimum condition value based on the function formula. A method for manufacturing a semiconductor device, comprising: