JP2003209035A - Setting method, setting program and setting device for parameter control value in semiconductor manufacturing process - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To shorten a time until the yield of a semiconductor product is improved or stabilized by analytically setting a parameter control value in each process of a semiconductor manufacturing line. <P>SOLUTION: The setting method of the parameter control value in the semiconductor manufacturing process comprises an initial setting process 2 setting an arbitrary initial value as the control value of the parameter in the semiconductor manufacturing processes A<SB>1</SB>-AN; an estimating process 3 estimating the characteristic value of a semiconductor product based on the set parameter control value; a difference calculating process 4 obtaining a difference between the estimated characteristic value and a preset target value of the semiconductor product; and a correcting process 9 obtaining the correcting amount of the parameter control value in the semiconductor manufacturing process A<SB>1</SB>-AN based on a difference between the estimated characteristic value and the target value to correct the set parameter control value when the difference between the estimated characteristic value and the target value is not within a predetermined range. <P>COPYRIGHT: (C)2003,JPO

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method, a program and an apparatus for setting a parameter control value in a semiconductor manufacturing process.

[0002]

2. Description of the Related Art A semiconductor manufacturing line has many semiconductor manufacturing processes such as "oxidation", "opening", and "diffusion". A parameter control value (for example, a parameter aim value, a parameter upper limit value or a lower limit value) is set for each process.

Here, the parameter means a parameter between each operation of the semiconductor manufacturing process and the characteristic value of the semiconductor product, and means a characteristic of the semiconductor determined as a result of each operation of the semiconductor manufacturing. Further, the parameter target value is a target value of semiconductor characteristics determined as a result of each process operation of semiconductor manufacturing.
The parameter upper limit value or lower limit value refers to the upper limit value or the lower limit value of the semiconductor characteristics determined as a result of each process operation of semiconductor manufacturing.

For example, in the "oxidation" step, the thickness of the oxide film serves as a parameter, and the target value of the thickness of the oxide film serves as a parameter target value.

In the following, setting of the parameter aim value will be described, but the same applies to other parameter management values.

[0006] In each process, work is carried out so as to realize this parameter target value, and a semiconductor product is completed.
The characteristics (properties) of the final semiconductor product are determined by the parameter values of each process. Therefore, it is necessary to set a parameter aim value in each process such that the final semiconductor product characteristic has a predetermined target value.

The parameter target value of each process is set when the semiconductor manufacturing line is started up. In general, the parameter aim value of each process is set based on the experience of an engineer by repeating simulation and trial production.

[0008]

In a semiconductor manufacturing line, a method of setting a parameter aim value so that a characteristic of a semiconductor product is a target value by repeating simulation and trial production or based on the experience of an engineer is used. There is a problem that it takes a long time before the product yield is improved or stabilized.

Further, even in one process of the semiconductor manufacturing line, if the parameter value (the value of the parameter actually realized in the process or the measured value of the parameter value realized in the process) deviates from the parameter target value, Even if the process is performed so that the parameter value is close to the target value of the parameter in other processes, the characteristics of the completed semiconductor product deviate from the target value, and the performance as a semiconductor product,
Value may be reduced.

The present invention has been made in view of the above circumstances. Firstly, the yield of semiconductor products is improved by analytically setting parameter control values in each process of a semiconductor manufacturing line. The purpose is to shorten the time until stabilization.

Secondly, when the parameter value deviates from the parameter control value in any step of the semiconductor manufacturing line, the parameter control value in the subsequent steps is set so that the characteristic of the semiconductor product becomes the target value. The purpose is to adjust and prevent deterioration of the performance and value of semiconductor products.

[0012]

[Means for Solving the Problems] Specific means taken to realize the present invention will be described below.

A first invention is an initial setting step of setting an arbitrary initial value as a parameter control value in a semiconductor manufacturing process, and a predicting step of predicting a characteristic value of a semiconductor product based on the set parameter control value. A difference calculation step of obtaining a difference between the predicted characteristic value and a preset target value of the semiconductor product, and when the difference between the predicted characteristic value and the target value is not included in a predetermined range. A semiconductor manufacturing method, which includes: a correction step of obtaining a correction amount of a parameter control value in the semiconductor manufacturing process based on a difference between the predicted characteristic value and the target value, and correcting the set parameter control value. This is a method of setting parameter control values for a process.

As a result, the characteristic value of the semiconductor product is analytically determined based on the relationship between the parameter value of each process of semiconductor manufacturing and the characteristic value of the semiconductor product, instead of relying on the intuition and experience of an engineer. It is possible to set the parameter control value of each process of semiconductor manufacturing so as to match with, and it is possible to stabilize the yield of semiconductor products at a high level and in a short period of time.

A second invention is a prediction for predicting a characteristic value of a semiconductor product based on a parameter value realized in a semiconductor manufacturing process that has been carried out and a parameter management value set in a semiconductor manufacturing process that has not been carried out. A step, a difference calculation step of obtaining a difference between a predicted characteristic value and a preset target value of the semiconductor product, and a difference between the predicted characteristic value and the target value is included in a predetermined range. If not, based on the difference between the predicted characteristic value and the target value, the correction amount of the parameter control value set in the unimplemented semiconductor manufacturing process is obtained, A parameter management value setting method for a semiconductor manufacturing process, comprising a correction step of correcting a set parameter management value.

As a result, even if the parameter value deviates from the target value in a certain process of the semiconductor manufacturing line, the characteristic value of the semiconductor product is set to the target value by correcting the parameter target value in the subsequent process. Can be matched.

In the above first or second invention,
The predicting step, the difference calculating step, and the correcting step may be repeatedly executed until the difference between the predicted characteristic value and the target value falls within a predetermined range.

Thus, in the first aspect of the present invention, if the work of each process is performed so as to realize the parameter control value, the parameter control value such that the characteristic of the semiconductor product becomes the target value can be obtained.

Further, in the second aspect of the present invention, if the work of each step to be performed from now on is carried out so as to realize the parameter control value, the parameter control value such that the characteristic of the semiconductor product becomes the target value is obtained. You can

Further, in the correction process of each of the above inventions, the correction amount may be obtained so that the parameter control value after correction is equal to or higher than a predetermined lower limit value and equal to or lower than a predetermined upper limit value.

The parameter control value has, for example, a physically or technically feasible range. By defining the upper limit value or lower limit value to specify this feasible range, by modifying the parameter control value to be included in the range,
A feasible parameter control value can be obtained.

Further, in the correction process of each of the above inventions, an abnormality may be notified when the parameter control value after correction does not exceed a predetermined lower limit value or less than a predetermined upper limit value.

As a result, in the first aspect of the present invention, it is possible to notify that the appropriate parameter control value could not be obtained.

Further, in the second invention, since the parameter control value of the process to be carried out is not included in the feasible range specified by the upper limit value and the lower limit value, the final semiconductor product characteristics are A warning can be sent that the target value will no longer be met. Then, by discarding the product in the process of manufacture at the stage of receiving this warning, it is not necessary to carry out the subsequent steps on the product in the process of manufacture, and the efficiency of semiconductor manufacturing can be improved.

Each of the above inventions can be implemented by a computer. That is, each step of each of the above inventions may be implemented by a function realized by a computer that reads a computer program. By using the program for setting the parameter management value of the semiconductor manufacturing process or the recording medium recording the program, the above inventions can be easily realized by the computer for the computer, the computer system, the server, and the client. .

The means for implementing the steps of the above inventions may be provided in the parameter control value setting device in the semiconductor manufacturing process.

[0027]

BEST MODE FOR CARRYING OUT THE INVENTION Embodiments of the present invention will be described below with reference to the drawings. In addition, in the embodiment of the present invention, the setting of the parameter aim value will be described, but the same applies to the setting of other parameter management values.

(First Embodiment) In the present embodiment, a characteristic of a semiconductor product is made to match a target value by a predictive model obtained from a relation between a parameter value of a semiconductor manufacturing process and a characteristic of a semiconductor product. A method, a program, and a system for setting a parameter control value in a semiconductor manufacturing process will be described below.

FIG. 1 is a block diagram illustrating the configuration of a parameter management value setting system (hereinafter referred to as “setting system”) in the semiconductor manufacturing process according to the present embodiment.

The setting system 1 has an initial setting function 2, a semiconductor characteristic prediction function 3, and a difference calculation function 4. The setting system 1 also includes a correction function 9 including a difference determination function 5, a parameter correction amount calculation function 6, a correction amount addition function 7, and a correction amount determination function 8.

Each of the functions 2 to 9 is realized by reading a parameter management value setting program (hereinafter referred to as “setting program”) 12 for the semiconductor manufacturing process, which is recorded on the recording medium 11, into a computer.

The initial setting function 2 includes a semiconductor manufacturing line L.
Initial values of the parameter aim values in the respective steps A _{1 to} A _N are set.

The initial value setting function 2 outputs the set initial value to the semiconductor characteristic prediction function 3.

The semiconductor characteristics prediction function 3 inputs the parameter target value in each step A ₁ to A _N of the semiconductor manufacturing line L, and the characteristic value of the semiconductor product based on the parameter target value in each step A ₁ to A _N It predicts and outputs the characteristic predicted value to the difference calculation function 4.

The difference calculation function 4 obtains an error by subtracting the characteristic predicted value from the characteristic target value of the semiconductor product, and outputs this error to the difference determination function 5.

The difference judgment function 5 judges whether or not the error is within the allowable range. If the error is within the allowable range, the setting system 1 finishes the parameter aim value setting operation. Output to the quantity calculation function 6.

The parameter correction amount calculation function 6 calculates the correction amount of the parameter aim value of each process A _{1 to} A _N of the semiconductor manufacturing line L from the error between the characteristic target value and the characteristic predicted value of the semiconductor product, and calculates each process. Correction amount addition function 7 corresponding to A _{1 to} A _N
Output to.

The correction amount adding function 7 adds the correction amount to the set parameter aim value to correct it, and outputs the corrected parameter aim value to the semiconductor characteristic predicting function 3.

The correction amount judgment function 8 judges whether the correction amount is less than or equal to a threshold value, and corrects the parameter aim value of each process A _{1 to} A _N of the semiconductor manufacturing line L until the correction amount becomes less than or equal to the threshold value. repeat.

That is, the correction amount determination function 8 is in a situation where the correction amount is less than or equal to a predetermined value and the parameter target value hardly changes from the current state, but the error between the characteristic target value and the characteristic predicted value of the semiconductor product is allowed. Prevents an infinite loop that occurs when the range is not reached.

A parameter management value setting method (hereinafter, referred to as "setting method") of the semiconductor manufacturing process executed by the setting system 1 will be described below.

FIG. 2 is a flow chart showing the procedure of the setting method according to the present embodiment.

First, when the initial value of the parameter aim value of each process A _{1 to} A _N of the semiconductor manufacturing line L is set by the initial setting function 2 (STEP 201), each initial value is input to the semiconductor characteristic prediction function 3. , Predictive characteristic values of semiconductor products are obtained (STEP203).

Next, the difference calculation function 4 calculates the error between the characteristic target value and the characteristic predicted value of the semiconductor product (STEP20
Four).

Next, the difference judgment function 5 judges whether the error between the characteristic target value and the characteristic predicted value of the semiconductor product is less than a threshold value (whether it is within an allowable range) (STEP 205), and the threshold value is determined. If the following is satisfied, the characteristic of the semiconductor product becomes a value close to the target value depending on the parameter aim values of the respective steps A _{1 to} A _N of the semiconductor manufacturing line L at this point, so the operation ends.

When the error between the characteristic target value and the characteristic predicted value of the semiconductor product exceeds the threshold value, the error between the characteristic target value and the characteristic predicted value of the semiconductor product is input to the parameter correction amount adding function 6, The correction amount of the parameter aim value of each process A _{1 to} A _N is calculated (STEP 206).

Here, the correction amount judgment function 8 judges whether or not the correction amount of the parameter aim value of each process A _{1 to} A _N is equal to or more than the threshold value (STEP 207), and the correction amount of the parameter aim value is threshold. If the above value is output to the correction amount summing function 7 the modification of the parameter target value for each step a ₁ to a _N correspond to the respective steps a ₁ to a _N.

On the other hand, if the correction amount of the parameter aim value is not more than the threshold value, it is judged that further correction of the parameter aim value has no effect, and the operation ends.

The correction amount of the parameter aim value is set to A _{1 in} each step.
When input to the correction amount addition function 7 corresponding to ~ A _N , the correction amount addition function 7 adds the correction amount of each process A _{1 to} A _{N to} the parameter target value of each process A _{1 to} A _N , The parameter aim values of each process A _{1 to} A _N are corrected (STEP 20
8).

With respect to the corrected parameter aim values of the respective processes A _{1 to} A _N , the predicted characteristic value of the semiconductor product is obtained again by the semiconductor characteristic prediction function 3, and the characteristic target value of the semiconductor product and the semiconductor product are calculated in the same procedure. until the correction amount parameter target value of the error until below the threshold or the steps a ₁ to a _N of the estimated characteristic value is equal to or lower than the threshold, steps a ₁ to a semiconductor manufacturing line L The modification of the parameters in _N is repeated (STEP 202).

A case where a three-layer neural network is used for the semiconductor characteristic prediction function 3 and the parameter correction amount calculation function 6 will be described below.

FIG. 3 is a block diagram showing a schematic configuration of the setting system 1 when a three-layer type neural network is used for the semiconductor characteristic prediction function 3 and the parameter correction amount calculation function 6. In FIG. 3, functions 2 to 4, 6, and 7 are shown.
, And other functions are omitted.

In the initial setting function 2, an initial value p ⁰ _i (i = 1, 2, ... N) of the parameter aim value of each process A _{1 to} A _N is set (N is the number of processes in the semiconductor manufacturing line L). , This initial value p
⁰ _i (i = 1,2, ... N) is each process A _{1 to} A of the semiconductor manufacturing line L
Parameter in _{the N} target value _{p i (i = 1,2, ...} N) as 3
It is input to the semiconductor characteristic prediction function 3 formed of a layered neural network.

[0054] In the semiconductor estimated characteristic feature 3, the parameter target value in each step A ₁ to A _N of the semiconductor manufacturing line L is input, characteristic estimation value s ^pred _k of semiconductor products is output. This calculation is based on the forward calculation of the three-layer neural network. That is, it is calculated as in the following equations (1) to (6).

[0055]

[Equation 1]

W1 _{j, i} is a coupling load value between the first layer i-th element and the second layer j-th element, and w2 _{k, j} is a coupling load value between the second-layer j-th element and the third layer k-th element. , .theta.2 _j is the threshold of the second layer j-th element, .theta.3 _k is the threshold of the third layer k-th element, p _i is the i th signal from the outside, the p _i as it is the first layer i th It is the output signal y1 _{i of the} device.

X2 _{j and} y2 _j are input signals and output signals of the j-th element in the second layer, and x3 _k and y3 _k are input signals and output signals of the k-th element in the third layer.

Although f2 and f3 are sigmoid functions shown in equation (7), the output of the sigmoid function is (0,1), so
The function f3 of the third layer element may be an identity function.

[0059]

[Equation 2]

N1 is the number of first layer elements, N2 is the number of second layer elements, N3
Is the number of elements in the third layer.

In this way, in the output calculation of the three-layer type neural network, when an input signal is given, the signal is transmitted in order from the first layer to the subsequent layer, and the output signal is output from the third layer element.

The characteristic predicted values of the semiconductor products are output to the difference calculation function 4, respectively.

In the difference calculation function 4, the characteristic target value of the semiconductor product is set.
s ^des _k (k = 1,2, ..., N3) predictive value s of semiconductor product characteristics
^pred _k (k = 1, 2, ..., N3) is subtracted and input to the parameter correction amount calculation function 5.

In the parameter correction amount calculation function 5, the difference e _k = s between the characteristic target value of the semiconductor product and the characteristic predicted value of the semiconductor product.
^{From des} _k -s ^pred _k (k = 1,2, ..., N3), the semiconductor manufacturing line L
The correction amount Δp _i (i = 1, 2, ..., N) of the parameter aim value in each of the steps A _{1 to} A _N is calculated. This calculation is based on the backward calculation of the three-layer neural network of the semiconductor characteristic prediction function 3. That is, it is calculated as in the following equations (8) to (12).

[0065]

[Equation 3]

Ε is a correction coefficient and is a small positive value. f2 '
(x) and f3 '(x) are the derivatives of the functions f2 (x) and f3 (x), respectively, but in the case of the identity function, 1, the sigmoid function f (x) = {1+
If exp (-x)} ^-1 , then f '(x) = f (x) {1-f (x)}.

The correction amount Δp _i (i = 1, 2, ..., N) of the parameter aim value in each process A _{1 to} A _N of the semiconductor manufacturing line L is calculated by the correction amount adding function 7 in each process A _{1 to} A _N. Is added to the parameter aim value p _i (i = 1,2, ..., N) of
The parameter aim values of _{1 to} A _N are modified.

The correction of the parameter aim value is repeated until the difference between the predicted characteristic value of the semiconductor product and the target value becomes equal to or less than the threshold value, and the difference between the predicted characteristic value of the semiconductor product and the target value becomes the threshold value. When it becomes below, the parameter aim value at that time is set, and the correction of the parameter aim value is repeated until the correction amount of the parameter aim value of each process A _{1 to} A _N becomes equal to or less than the threshold value, and each process A ₁ The operation is stopped when the correction amount of the parameter aim value of ~ A _N becomes equal to or less than the threshold value.

As described above, the semiconductor characteristic prediction function 3
When a three-layer type neural network is used for the parameter correction amount calculation function 6, the neural network needs to learn semiconductor characteristics in advance.

In the following, as a learning method of the neural network in the semiconductor characteristic prediction function 3 and the parameter correction amount calculation function 6, the case of using a generally known error backpropagation learning method will be described.

The learning of the neural network requires learning data composed of a pair of an input signal and a teacher signal which is a desired output signal when the input signal is input to the neural network.

Here, a plurality of prototypes are flown to the semiconductor manufacturing line L by changing the parameter aim values of the respective processes A _{1 to} A _N to produce semiconductor products, and at that time, each process A ₁
~ A _{N using} the parameter value measured by the measuring device as an input signal, obtain learning data in which the characteristic value of the completed semiconductor product is used as a teacher signal, and learn the neural network based on the input signal and the teacher signal. To do.

Each step A of the semiconductor manufacturing line L₁~ A_NIn total
The measured parameter value is p_n= (Pⁿ ₁, p ⁿ ₂, ..., pⁿ _N) (N = 1,
2, ..., M) (M is the number of prototypes, N is the semiconductor manufacturing line L)
Number of steps).

The characteristic value of the completed semiconductor product is T _n =
(T ⁿ ₁ , T ⁿ ₂ , ..., T ⁿ _N3 ) (n = 1, 2, ..., M). N3 is the number of types of characteristic values of semiconductor products.

Each process T of the semiconductor manufacturing line L for each prototype₁
~ T_nParameter value p_n= (Pⁿ ₁, pⁿ ₂, ..., pⁿ _N) (N = 1,2,
…, M) is a half composed of three-layer neural network
It is input to the conductor characteristic prediction function 3. Semiconductor characteristic prediction function
At 3, each process T of the semiconductor manufacturing line L₁~ T_nParame
When a data value is entered, the semiconductor product characteristic prediction value s_n= (S
ⁿ ₁, sⁿ ₂,…, Sⁿ _N3) (N = 1,2, ..., M) is output.

This calculation is based on the forward calculation of the three-layer neural network. This calculation is similar to the case of calculating the predictive characteristic value of the semiconductor product described above, but the formulas (13) to (18) are shown below in order to make the learning calculation formula easy to understand. Note that (13) to (18), the parameter value of n-th prototype ^{_{p n = (p n 1,}} p n 2, ..., p n N) shows a calculation formula in the case where the inputted.

[0077]

[Equation 4]

P ⁿ _i is a parameter value of the i-th process of the n-th trial manufacture, which is the output signal y1 ⁿ _i of the first-layer i-th element as it is.

X2 ⁿ _{j and} y2 ⁿ are the input and output signals of the second layer j-th element when the n-th prototype parameter value is input,
x3 ⁿ _j and y3 ⁿ _j are input signals and output signals of the k-th element of the third layer when the parameter values of the n-th prototype are input, and the output signals y3 ⁿ _j become the characteristic predicted values of the semiconductor product.

W1 _{j, i} is the coupling load value between the first layer i-th element and the second layer j-th element, and w2 _{k, j} is the coupling load value between the second-layer j-th element and the third layer k-th element. , Θ2 _j is the threshold value of the j-th element of the second layer, θ3 _k is the threshold value of the k-th element of the third layer, and f2 and f3 are sigmoid functions f (x) = {1 + exp (-x)} ^{- Although} it is ¹ , the output of the sigmoid function is (0,1), so the function f3 of the third layer element may be an identity function.

N1 is the number of elements in the first layer and corresponds to the number N of steps in the semiconductor manufacturing line L. N2 is the number of second layer elements, and N3 is the number of third layer elements.

In learning, the coupling weight value and the threshold value are adjusted so that the characteristic predicted value of the semiconductor product, which is the output value of the neural network, matches the measured characteristic value of the semiconductor product, which is the teacher signal.

First, the error function E is expressed by the following (19), (2
It is defined as in equation (0).

[0084]

[Equation 5]

The coupling weight value and the threshold correction amount are calculated so as to reduce the error function E. This calculation method is shown in the following equations (21) to (26).

[0086]

[Equation 6]

Η is a learning coefficient, which is a small positive value. f2 '
(x) and f3 '(x) are derivatives of the functions f2 (x) and f3 (x), respectively, but if the sigmoid function f (x) = {1 + exp (-x)} ^-1 ,
f '(x) = f (x) {1-f (x)}, which is 1 in the case of the identity function.

Correction amount Δw1 _{j, i} (i = 1,2, ..., N) of the combined weight value
1; j = 1,2, ..., N2), Δw2 _{k, j} (j = 1,2, ..., N2; k = 1,2, ..., N
3), threshold correction amount Δθ2 _j (j = 1,2, ..., N2), Δθ3 _k
(K = 1,2, ..., N3) is a coupling load value w1 _{j, i} (i = 1,2, ..., N1; j =) as shown in the following equations (27) to (30).
1,2, ..., N2), w2 _{k, j} (j = 1,2, ..., N2; k = 1,2, ..., N3), threshold θ2 _j (j = 1,2, ..., N2) ), Θ3 _k (k = 1,2, ..., N3), and correction is performed.

[0089]

[Equation 7]

Then, the correction of the coupling weight value and the threshold value is repeated until the error function value becomes equal to or less than the threshold value, and when the error function value becomes equal to or less than the threshold value, the learning ends, and the neural network becomes a semiconductor characteristic. It is used as the prediction function 3.

The parameter correction amount calculation function 6 may perform the same learning as the three-layer type neural network of the semiconductor characteristic prediction function 3 described above, and may perform the backward calculation at the time of calculation.

In the setting method, the setting program, and the setting system according to the present embodiment described above, the relationship between the parameter value of each process A _{1 to} A _N of the semiconductor manufacturing line L and the characteristic value of the semiconductor product is used. The parameter aim value of each process A _{1 to} A _N of the semiconductor manufacturing line L can be analytically set so that the characteristic value of the semiconductor product matches the target value.

Therefore, the yield of semiconductor products is stabilized at a high level in a short period of time as compared with the case where the parameter aim value of each process A _{1 to} A _N of the semiconductor manufacturing line L is set based on the intuition and experience of an engineer. be able to.

(Second Embodiment) In the present embodiment, the characteristics of a semiconductor product being manufactured are predicted by a prediction model obtained from the relationship between the parameter values of the semiconductor manufacturing process and the characteristics of the semiconductor product. A method, a program, and a system for correcting the parameter aim value of the subsequent process so that the predicted characteristic matches the target value will be described.

FIG. 4 is a block diagram illustrating the configuration of the setting system according to this embodiment. In FIG. 4, the same parts as those in FIG. 1 are designated by the same reference numerals, and the description thereof will be omitted or briefly described. Here, only different parts will be described in detail.

The setting system 13 has a semiconductor characteristic prediction function 14 and a difference calculation function 4. Also, setting system 1
3 has a correction function 16 including a difference determination function 5, a parameter correction amount calculation function 15, a correction amount addition function 7, and a correction amount determination function 8.

Each function 4, 5, 7, 8, 14 to 16 is realized by the setting program 18 recorded in the recording medium 17 being read by the computer.

In the present embodiment, it is assumed that the semiconductor flows through the processes A _{1 to} A _N on the semiconductor manufacturing line L and the processes up to the process A _n are completed.

In the already completed steps A _{1 to} A _n , the parameter values (parameter measurement values) of the steps A _{1 to} A _n are measured by the measuring devices B _{1 to} B _n .

The semiconductor characteristic predicting function 14 performs the steps A _{1 to} A _n.
And the parameter aim values of the steps A _{n + 1 to} A _N (N is the number of steps in the semiconductor manufacturing line L) are input, the characteristic values of the semiconductor product are predicted and output to the difference calculation function 4.

The difference calculation function 4 subtracts the characteristic predicted value of the semiconductor product from the characteristic target value of the semiconductor product and inputs it to the parameter correction amount calculation function 15.

The parameter correction amount calculation function 15 corrects the parameter aim values of the steps A _{n + 1 to} A _N that have not been completed based on the error between the characteristic target value of the semiconductor product and the characteristic predicted value of the semiconductor product. The amount is output to the correction amount addition function 7.

[0103] correction amount summing function 7 adds the correction amount of the parameter target value step A _{n + 1} ~A _N and the parameter target value of each process A _{n + 1} ~A _N, Step A _{n + 1} ~ Correct the parameter aim value of A _N.

The correction function 16 compares the parameter aim value of each process A _{n + 1 to} A _N with the upper and lower limit values of the possible values of the parameter, and the parameter aim value is out of the upper and lower limit range. In this case, a notice to the effect that this semiconductor product being manufactured will be destroyed will be given.

The setting method executed by the setting system 13 will be described below.

FIG. 5 is a flow chart showing the procedure of the setting method according to the present embodiment.

First, a variable n representing the progress of the semiconductor manufacturing process.
Is set to 1 (STEP401).

Then, the work of the process A _n is performed (STEP40
3), the parameter value of the process A _n is measured by the measuring device B _n (STEP 404).

Parameter values of the steps A _{1 to} A _n and the step A
_{n + 1} to A parameter target value of _N is a semiconductor characteristic estimation function 1
4 is input and the predicted characteristic value of the semiconductor product is obtained (STEP 406).

Next, the difference calculation function 4 subtracts the characteristic predicted value of the semiconductor product from the characteristic target value of the semiconductor product, and calculates the difference between the characteristic target value of the semiconductor product and the characteristic predicted value (STEP 407).

Next, the difference judgment function 5 compares the difference between the characteristic target value and the characteristic predicted value of the semiconductor product with the allowable amount (ST
EP408), if the difference is less than the allowable amount, jump to STEP412.

If the difference between the characteristic target value and the characteristic predicted value of the semiconductor product is larger than the allowable amount, the difference between the characteristic target value and the characteristic predicted value of the semiconductor product is input to the parameter correction amount calculation function 15, and the unexecuted process A is performed. The correction amount of the parameter aim value in _{n + 1 to} A _N is calculated (STEP 409).

Here, the correction amount judgment function 8 compares the correction amount of the parameter aim value with the threshold value (STEP 41
0), if the correction amount of the parameter target value is less than or equal to the threshold value, jump to STEP412.

On the other hand, if the correction amount of the parameter aim value is larger than the threshold value, the correction amount adding function 7 adds the parameter aim value and the correction amount of the parameter aim value to correct the parameter aim value (STEP411). .

The characteristic values of the semiconductor product are predicted again using the corrected parameter aim values of the respective processes A _{n + 1 to} A _N , and the characteristic target value of the semiconductor product and the predicted characteristic value of the semiconductor product are calculated in the same procedure. each step up or incomplete error is less than the threshold value of the a _{n + 1} to a correction amount of the parameter target value of _n is decreased below the threshold, the steps of the semiconductor manufacturing line L a _{n +} Fixed parameter target value of ₁ to a _N are repeated (STEP 405).

In STEP 412, the correction function 16 compares the parameter aim value of each process A _{n + 1 to} A _N with the upper and lower limit values of the parameter, and the parameter aim value is out of the upper and lower limit range. If so, the semiconductor manufacturing operation is terminated and this semiconductor is discarded. If all the target values of the parameters are within the upper and lower limit values, 1 is added to n (STEP 413), and the work of the step A _n is performed again (STEP 403). The corrected value is used as the parameter aim value of the process A _n at this time. In this way ST
The steps from EP403 to STEP412 are repeated from n = 1 to n = N (STEP402), and the process ends when n = N + 1.

A case in which a three-layer neural network is used as the semiconductor characteristic prediction function 14 and the parameter correction amount calculation function 15 will be described below.

FIG. 6 is a block diagram showing a schematic configuration of the setting system 13 when a three-layer neural network is used for the semiconductor characteristic prediction function 14 and the parameter correction amount calculation function 15. In FIG. 6, the functions 4, 7,
The relationship between 14 and 15 is shown, and other functions are omitted.

It is assumed that the semiconductor flows through the processes A _{1 to} A _n on the semiconductor manufacturing line L and the processes up to the process A _n are completed.

In the steps A _{1 to} A _n which have already been completed, the measuring device B is used.
The parameter values of each process A _{1 to} A _n are measured by _{1 to} B _n .

In the semiconductor characteristic prediction function 14, steps A ₁ to
Parameter values ^{_{A n p obs i (i =}} 1,2, ..., n) in Step A _{n + 1} ~
When the parameter target value p _i (i = n + 1, ..., N) of A _N (N is the number of semiconductor manufacturing line processes) is input, the characteristic prediction value s ^pred _k (k = 1,2) of the semiconductor product is input. , ..., N3) is output.

This calculation is based on the forward calculation of the three-layer type neural network. That is, the following (31) to (3
It is calculated as in equation 7).

[0123]

[Equation 8]

P ^obs _i (i = 1,2, ..., n) is a parameter value of the semiconductor manufacturing process, and other symbols are as described above.

The characteristic predicted values of the semiconductor products are output to the difference calculation function 4, respectively.

In the difference calculation function 4, the characteristic target value of the semiconductor product is set.
s ^des _k (k = 1,2, ..., N3) predictive value s of semiconductor product characteristics
^pred _k (k = 1, 2, ..., N3) is subtracted, and the value is input to the parameter correction amount calculation function 15.

In the parameter correction amount calculation function 15, the difference e _k between the characteristic target value of the semiconductor product and the predicted characteristic value of the semiconductor product.
= s ^des _k -s ^pred _k (k = 1,2, ..., N3), the process A _{1 to} A _N of the semiconductor manufacturing line L that has not been completed yet.
_{n + 1 to} A _N parameter aim value correction amount Δp _i (i = n + 1,
…, N) is calculated. This calculation is based on the backward calculation of the three-layer neural network of the semiconductor characteristic prediction function 14. That is, it is calculated as in the following equations (38) to (42).

[0128]

[Equation 9]

The correction amount Δp _i (i = n + 1, ..., N) of the parameter target values of the steps A _{n + 1 to} A _N is calculated by the correction value adding function 7 in the parameters of the steps A _{n + 1 to} A _N. The target value p _i (i = n + 1,…,
N), and the parameter aim values of the steps A _{n + 1 to} A _N are corrected.

Then, until the difference between the characteristic predicted value of the semiconductor product and the target value becomes less than the threshold value or the correction amount of the parameter aim value of each process A _{n + 1 to} A _N becomes less than the threshold value. The correction of the parameter aim values of the steps A _{n + 1 to} A _N is repeated, and the difference between the characteristic predicted value and the target value becomes less than or equal to the threshold value, or the parameter aim values of the steps A _{n + 1 to} A _N are changed. when the correction amount is equal to or lower than the threshold, application of the process a _{n + 1} is performed according to the parameter target value at that time, the parameter values from step a _{n + 1} is measured by the measuring device B _{n + 1.}

Next, step A₁~ A_{n + 1}Parameter value p^obs
_i(I = 1,2, ..., n + 1) and process A_{n + 2}~ A _NParameter of
Value p_iBased on (i = n + 2, ..., N), the characteristics of semiconductor products are estimated.
Measured.

The above is repeated until the parameter aim value of the final process A _N is determined, and the parameter aim value of each process is corrected so that the characteristics of the semiconductor product always match the target value.

As described above, the semiconductor characteristic prediction function 1
4 and the parameter correction amount calculation function 15, when a three-layer type neural network is used, this neural network is learned in advance as in the first embodiment.

In the setting method, the setting program, and the setting system according to the present embodiment described above, when the parameter value deviates from the parameter target value in a certain process of the semiconductor manufacturing line L, the characteristics of the semiconductor product are the target values. The target values of the parameters of the subsequent steps are adjusted so as to match with, and it is possible to prevent deterioration of the performance and value of the completed semiconductor product.

In each of the above embodiments, the case where a neural network is used as a method of predicting the characteristic value of a semiconductor product is described, but other methods such as multivariate analysis may be used. . Multivariate analysis is a method of statistically analyzing and elucidating data when an event is caused by a number of interrelated factors. By this multivariate analysis, the relationship between the parameter value or the parameter aim value and the characteristic value of the semiconductor product can be analyzed and used as a prediction model.

Further, the respective constituent elements of the setting systems 1 and 13 according to the above-mentioned respective embodiments may be freely combined, may be divided into a plurality of elements, and the arrangement position may be changed.

The setting programs 12 and 18 may be distributed on a plurality of computers and operate in cooperation with each other.

Further, in each of the above embodiments, the case where one parameter is handled in one step has been described, but the same method can be used when a plurality of parameters are handled in one step.

In each of the above embodiments, learning data is acquired from the prototype. When using the prototype like this, cut the prototype semiconductor product,
Parameter values can be measured by disassembling and destroying.
However, when the target value of a parameter for a semiconductor product in the process of manufacturing is modified, the product cannot be destroyed because it is flowing through the line. If the parameter value cannot be measured without destruction, work is being done so that the parameter value approaches the parameter target value, and it is considered that the parameter value is close to the target value, and the parameter target value remains unchanged. The prediction may be performed as a parameter value.

[0140]

As described above in detail, in the present invention,
It is an object of the present invention to set a parameter control value in each process of a semiconductor manufacturing line analytically and shorten the time until the yield of semiconductor products is improved or stabilized.

Further, in the present invention, even if the parameter value deviates from the parameter control value in any step of the semiconductor manufacturing line, the parameter control value in the subsequent steps is set so that the characteristic of the semiconductor product becomes the target value. Can be adjusted to prevent deterioration of the performance and value of semiconductor products.

[Brief description of drawings]

FIG. 1 is a block diagram illustrating the configuration of a parameter management value setting system in a semiconductor manufacturing process according to a first embodiment of the present invention.

FIG. 2 is a flowchart showing a procedure of a parameter management value setting method in a semiconductor manufacturing process according to the same embodiment.

FIG. 3 is a block diagram showing a system configuration when a three-layer neural network is used for a semiconductor characteristic prediction function and a parameter correction amount calculation function in the same embodiment.

FIG. 4 is a block diagram illustrating the configuration of a parameter management value setting system in a semiconductor manufacturing process according to a second embodiment of the present invention.

FIG. 5 is a flowchart showing a procedure of a parameter management value setting method in a semiconductor manufacturing process according to the same embodiment.

FIG. 6 is a block diagram showing a system configuration when a three-layer neural network is used for a semiconductor characteristic prediction function and a parameter correction amount calculation function in the same embodiment.

[Explanation of symbols]

1, 13 ... Parameter management value setting system for semiconductor manufacturing process 2 ... Initial setting function 3, 14 ... Semiconductor characteristic prediction function 4 ... Difference calculation function 5 ... Difference determination function 6, 15 ... Parameter correction amount calculation function 7 ... Correction amount addition function 8 ... correction amount determining function 9,16 ... correction function 11, 17 ... recording medium 12, 18 ... semiconductor parameter control values of the manufacturing process set program L ... semiconductor manufacturing line a ₁ to a _n ... step B ₁ .about.B _n ... Measuring device

Claims (9)

[Claims] 1. An initial setting step of setting an arbitrary initial value as a parameter control value in a semiconductor manufacturing process, a prediction step of predicting a characteristic value of a semiconductor product based on the set parameter control value, and a predicted characteristic. A difference calculation step of obtaining a difference between a value and a preset target value of the semiconductor product, and if the difference between the predicted characteristic value and the target value is not included in a predetermined range, the predicted value is calculated. Based on the difference between the characteristic value and the target value, the correction amount of the parameter control value in the semiconductor manufacturing process is obtained, and the correction process for correcting the set parameter control value. Value setting method. 2. A predicting step of predicting a characteristic value of a semiconductor product based on a parameter value realized in a semiconductor manufacturing process that has been carried out and a parameter control value set in a semiconductor manufacturing process that has not been carried out, Difference calculation step for obtaining the difference between the characteristic value and the preset target value of the semiconductor product, when the difference between the predicted characteristic value and the target value is not included in a predetermined range, Based on the difference between the predicted characteristic value and the target value, the correction amount of the parameter control value set in the unfinished semiconductor manufacturing process is obtained, and the correction amount is set in the unfinished semiconductor manufacturing process. A method of setting a parameter control value in a semiconductor manufacturing process, comprising a correction step of correcting a parameter control value. 3. The parameter control value setting method for a semiconductor manufacturing process according to claim 1, wherein the predicting step, the difference calculating step, and the correcting step are performed with the predicted characteristic value and the target value. And a parameter management value setting method in a semiconductor manufacturing process, which is repeatedly executed until a difference between the parameter control value and the value is included in a predetermined range. 4. The parameter control value setting method for a semiconductor manufacturing process according to claim 1, wherein in the correction step, the parameter control value after correction has a predetermined lower limit value or more and a predetermined value. A parameter management value setting method for a semiconductor manufacturing process, characterized in that a correction amount is calculated so as to be equal to or less than an upper limit value. 5. The method for setting parameter control values in a semiconductor manufacturing process according to claim 1, wherein in the correction step, the parameter control value after correction is equal to or higher than a predetermined lower limit value and a predetermined upper limit value. A method for setting parameter control values in a semiconductor manufacturing process, which notifies an abnormality when the following conditions are not met. 6. An initial setting function for setting an arbitrary initial value as a parameter control value in a semiconductor manufacturing process in a computer, and a prediction function for predicting a characteristic value of a semiconductor product based on the set parameter control value, A difference calculation function for obtaining a difference between a characteristic value predicted by a prediction function and a preset target value of the semiconductor product, and a difference between the predicted characteristic value and the target value is included in a predetermined range. When there is not, based on the difference between the predicted characteristic value and the target value, to obtain a correction amount of the parameter control value in the semiconductor manufacturing process, to realize a correction function for correcting the set parameter control value A program for setting parameter control values for semiconductor manufacturing processes. 7. A predicting function for predicting a characteristic value of a semiconductor product on a computer based on a parameter value realized in a semiconductor manufacturing process that has been carried out and a parameter control value set in a semiconductor manufacturing process that has not been carried out. And a difference calculation function for obtaining a difference between a characteristic value predicted by the prediction function and a preset target value of the semiconductor product, and a difference between the predicted characteristic value and the target value within a predetermined range. When not included in the, based on the difference between the predicted characteristic value and the target value, the correction amount of the parameter control value set in the unimplemented semiconductor manufacturing process is determined, and the unimplemented semiconductor A parameter management value setting program for a semiconductor manufacturing process for realizing a correction function for correcting a parameter management value set in a manufacturing process. 8. An initial setting means for setting an arbitrary initial value as a parameter control value in a semiconductor manufacturing process, a predicting means for predicting a characteristic value of a semiconductor product based on the set parameter control value, and the predicting means. Difference calculation means for obtaining the difference between the predicted characteristic value and the preset target value of the semiconductor product, and when the difference between the predicted characteristic value and the target value is not included in a predetermined range And a correction unit that determines a correction amount of a parameter control value in the semiconductor manufacturing process based on a difference between the predicted characteristic value and the target value, and corrects the set parameter control value. Characteristic semiconductor manufacturing process parameter control value setting device. 9. A predicting means for predicting a characteristic value of a semiconductor product based on a parameter value realized in a semiconductor manufacturing process that has been carried out and a parameter management value set in a semiconductor manufacturing process that has not been carried out, Difference calculating means for obtaining the difference between the characteristic value predicted by the predicting means and the preset target value of the semiconductor product, and the difference between the predicted characteristic value and the target value is included in a predetermined range. If not, based on the difference between the predicted characteristic value and the target value, the correction amount of the parameter control value set in the unimplemented semiconductor manufacturing process is obtained, A parameter control value setting device in a semiconductor manufacturing process, comprising: a correction unit that corrects a set parameter control value.