JPH0524657B2 - - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to a semiconductor process control system, and in particular, detects information regarding the carrier distribution shape formed on a wafer using an optical sensor, and uses the result to control the semiconductor formation process. Regarding the method.

Conventional semiconductor process control, especially the control of the carrier distribution formation process by ion implantation or diffusion into the wafer, lacks a means for high-speed detection of the carrier concentration distribution (which has a large effect on the characteristics of the formed semiconductor elements). This made it difficult to automate the process. However, as previously shown by the applicants of the present invention, it has become possible to extract information regarding the carrier distribution shape within a wafer using an optical sensor (J.
Appl.Phys.51, 4125 (1980), patent application 1986-17012,
(Special application 1986-200191).

An object of the present invention is to provide a high-speed and highly accurate semiconductor formation process control method using optical carrier distribution measurement means.

As shown in the earlier application mentioned above, by analyzing the polarization deviation of the incident light and reflected light (hereinafter also referred to as ellipso data) when the wafer surface is irradiated with infrared laser light, the carrier inside the wafer can be determined. Distribution can be measured. Conventionally, the carrier distribution was measured by etching the wafer surface, which took a long time, but with this method, it is now possible to measure it at high speed. The present invention proposes a process control method that utilizes this new measurement method.

An embodiment of the present invention will be described below. FIG. 1 is a block diagram of one embodiment of the present invention. The initial condition setting unit 100 inputs a distribution shape parameter 150 in the depth direction of carriers to be formed on the wafer given from the outside, and uses a separately prepared ellipso data theoretical value calculation program to determine the distribution shape. The theoretical value 155 of the ellipso data that should be observed if it is true is calculated.

The control signal generation unit 110 uses the theoretical ellipso data value 155 and the actual ellipso data value 170 obtained from the wafer processed by the process to generate process control parameters 160 that improve the degree of agreement between the two. is calculated and passed to the carrier distribution forming device 120. The carrier distribution forming device 120 outputs the processed wafer 165. The ellipso data detection unit 130 irradiates a predetermined semiconductor element portion on this wafer with a light beam,
Polarization deviation data, ie, ellipsodata, between the incident light and the reflected light is determined.

FIG. 2 is a specific example of the above parameters 150. When it is desired to generate a carrier distribution represented by a distribution function 200 in the direction of depth Z, constant P ₀ , peak value P ₁ , peak depth P ₂ , two standard deviations P ₃ ,
By joining two Gaussian distributions using P ₄ ,
An example expressing this distribution is shown.

3 and 4 are examples of the ellipso data detection section 130 shown in FIG.
The details are shown in. FIG. 3 shows an apparatus for acquiring ellipsoscopic data by keeping the angular frequency ω _k of incident light constant and varying the incident angle θ _k (k=1, 2, . . . , m). 300 is an infrared generator, 310 is a rotating polarizer, 320 is a rotary encoder, 330 is a turntable, 340 is a rotating analyzer, 350 is a detector, and the acquired data is converted into actually measured ellipso data ( Δ _k , Ψ _k ) K=1, 2, ..., m.

FIG. 4 shows an apparatus for changing the angular frequency ω _k while keeping the incident angle θ _k constant; 400 is an infrared light source, 410 is a spectrometer, 420 is a polarizer, and 43
0 is a rotary analyzer, 440 is a rotary encoder,
450 is a detector.

FIG. 5 is a measurement situation diagram. In order to measure the carrier distribution at a predetermined portion Q (510) on the wafer 165, infrared light 500 is incident at an angular frequency ω _k and an incident angle θ _k , and reflected. A situation in which light 510 is obtained is shown.

FIG. 6 shows the system configuration in this embodiment. The large computer 600 runs the program P _rg.・1
(610) and a data file 620, and communicates with users via a terminal 630. P _rg. 1 performs the function of FIG. 1 100. Large computer 60
0 is connected to a small computer 640. In the figure, 680 indicates a line to another small computer. The small computer 640 has a program _Prg. 2 (650) and a data file 660. The terminal 670 also allows communication with the outside world.
The small computer 640 is the carrier distribution forming device 12
0 and the ellipso data detection device 130. Small computer 360 in Figures 3 and 4
This may be different from or the same as the small computer shown in FIG. 640. In the former case,
360 must be built into the device 130. In the following, for convenience of explanation, the device 130 includes three
60 is built-in, and 120 is also assumed to have a built-in small dedicated computer for device control.
P _rg.2 executes the function 110 in FIG.

The operation of this system will be explained below. FIG. 7 is an example of file 620 in FIG.
This file has a plurality of records for each product name corresponding to the type of wafer to be manufactured. In the figure, records 700 to 720 correspond to product name AG1. 730, 740, etc. correspond to the product name BG1. Items as shown in the figure are entered on each record. These contents will be explained below along with the explanation of system operation.

FIG. 8 is an explanatory diagram of the program operation of FIG. 6 610, and therefore details the initial condition setting process of FIG. 1 100.

Processing 800 is initial setting, and the following processing is performed.

(a) Read the product name from the outside via the terminal 630 (temporarily assume it is AG1).

(b) Read the number of samples N for the above product name.

(c) From the results of (a) and (b) above, 74 in the file in Figure 7
8, create N records with the input product name and other parts cleared (700 to 700 in Figure 1).
720). If items with the same name already exist, clear them and then do the same.

(d) Load the carrier distribution parameters (corresponding to the target distribution, 150 in FIG. 1) regarding the product name from the outside and enter them in the predetermined area 750 of the record created in (c) above.

(e) Read the number of cases m under the ellipso data measurement conditions ((ω _k , θ _k ) k=1, 2, ..., m) and (ω _k , θ _k ) for the product name from the outside and save them on the file. Fill in area 752.

(f) Read the permissible variation range of the carrier distribution formation process control condition parameters for the product name from the outside in the form of upper and lower limit values of each parameter, and set it in area 762.
Fill in. In the figure, the lower limit is indicated by x ₁ and the upper limit is indicated by ₁ .

(g) Enter the perturbation constant (parameter for perturbing the control signals x ₁ to x _o to the carrier distribution forming process, details will be described later) for the product name in area 764.
Fill in.

(h) Input other information related to the product name other than the above and write it in area 768.

As a result of the above, 7 of the N records for the product name
48,750,752,762,764,768
The data was entered in the section. 754,756,75
8,760,766 copies are in a clear state.

In the next process 810, control variable alternatives are input. The user is provided with control variables (typically temperature x ₁ , processing time) to control the carrier distribution formation process.
x ₂ is the main variable)) input N sets of alternatives,
Fill out 760 copies.

In the next process 820, target ellipso data (Ψ ^* _k , Δ ^* _k ) k=1, 2, . . . , m is created and written in the area 754. The calculation method is as follows.

Ψ _k = f - (ω _k , P ₀ , P ₁ , P ₂ , P ₃ , P ₄ , θ _k ) Δ _k = f ₂ (ω _k , P ₀ ,, P ₁ , P ₂ , P ₃ , P ₄ , θ _k ) Here, (ω _k , θ _k ) k = 1, 2, ..., m uses the ellipso data measurement conditions written in area 752, and P ₀ to P ₄ are 750 copies. Use distribution parameters. The functions f ₁ and f ₂ are obtained by integrating Maxwell's equation for calculating the polarization deviation, and the details have already been provided by the applicant of the present invention. J.APPl.Phys.51 4125 (1980), T.
Motooka and K.Watanabe, Damage Profile
Determination of ion−implanted Si−layer by
ellipsometry).

In the next step 830, an inquiry is made to the outside as to whether or not repetition is required, and if repetition is specified, the process returns to step 800 and similar processing is performed for other product names.

FIG. 9 is an explanatory diagram of the program operation of FIG. 6 650, and therefore details of the control signal generation section of FIG. 1 110. Below, the program 650 includes
The explanation assumes that it will operate in online mode.

(a) (Processing 900) Simultaneously with the startup of the small computer 640, the program 650 is also started and constantly performs processing for accepting interrupt signals.

(b) (Processing 910) If no interrupt signal has been generated, the process returns to 900 (912). If the interrupt signal is, for example, a termination instruction from the terminal 670, execution is terminated (914). When an interrupt signal is generated from the carrier distribution forming device 120, the process branches to path 916. In the case of an interrupt from ellipso data detection device 130, path 918 is branched.

(c) (Process 920) The product name of the product to be processed is read from the carrier distribution forming device 120.

(d) (Processing 922) Transfer the product name record read in the above (c) in the file in Figure 7 to the small computer file in Figure 6 660 (give the product name to the large computer and (Please have the relevant record forwarded to you).

(e) (Process 924) Examine the ellipso data actual measurement value section 756 of the transferred record in the file 660. If there is a record that remains cleared in the measured value section, the process branches to 926; otherwise, the process branches to 928.

(f) (Processing 930) Since the control experiment with the number of samples N set for the product name has not been completed, the first one of the records in which 756 copies in the file 660 are cleared (one at random) is ).

(g) Set a flag in the control status flag part of the above code (processing 931).

(h) (Process 932) Since the control experiment with the number of samples N set for the product name has been completed, the record in the file 660 is used to generate a control signal to be passed to the carrier distribution forming device 120. An example of the method is shown below.

(h.1) The last of the N records is used for work, and records with the number of control variables (n+1) are randomly selected from the other N-1 records. (In this case, the premise is that N-1>
must be set to n+1).

(h.2) Among n+1 records, choose the one with the largest deviation sum of 758 copies. The control variable on this record is written as 〓 _w .

(h.3) Find the center of gravity of the control variable (G=〓〓 _i /n) for the n+1 records excluding the above records.

(h.4) Regarding G〓 Point of symmetry of _w〓 Find _E using the following formula.

〓 _E = 2G−〓 _w (h.5) 〓 Randomize _E using the following formula.

〓 _E ′=〓 _E + δ〓 _EHere , δ〓 _E is a random vector, and its i-th element δx _Ei is determined by the following formula.

δx _Ei = c (- xi ) r _i where C is the perturbation constant for the product name, xi are the upper and lower limits of the i-th control variable, respectively, and r _i is a uniform random number between -1 and 1. , one sample for each i. The reason for deliberately randomizing 〓 _E to 〓 _E ′ will be explained later.

(i) (Processing 934) Write the above _〓E ' in the 760th copy of the work record (final record) of the file 660, and set a flag in the control status flag section (766) of the record.

(j) (Processing 936) Among the records in the file 660, a record whose control status flag section is set is found, and the data of the control variable section 760 is applied to the carrier distribution forming device.

(k) (Process 938) All records of the product name in file 660 are transferred to file 62 on the large computer side.
Save it over the corresponding record on 0.

(l) Return to processing 900 (940).

(m) (Processing 950) Ellipso data detection device 130
The product name of the wafer to be inspected is read from .

(n) (Process 952) All records in the file 620 regarding the product name are transferred to the file 660 from the large computer.

(o) (Process 954) Detect the record with the control status flag set on the file 660, and operate the ellipso data detection device according to the data written in the ellipso data measurement condition section (752 in Fig. 7). let

(p) (Processing 956) The ellipso data obtained in (o) above is
Enter the ellipso data actual measurement value section (756 in Figure 7) in the record with the above control status flag.

(q) (Process 958) Calculate the deviation between the ellipso data target value (754) and the actual measured value (756) on the record with the control status flag for each measurement condition case, and calculate the deviation ^m _k=1 (｜Ψ _k −Ψ ^* _k ｜＋｜
Find Δ _k −Δ ^* _k |). Here || is an absolute value symbol. Write the results on copy 758.

(R) (Processing 960) In addition to the relevant record (the record with the control status flag set), file 66
If there is an unexecuted record within 0 (a record in which the ellipso data actual measurement value part is in a clear state), do nothing and proceed to the next process (S). Otherwise, calculate the deviation sum of the record and the record. Compare the deviation sum of the record with the largest deviation sum among all other records other than . If the former is smaller than the latter, the latter record is placed over the former record. Through this process, records with control variables that reduce the sum of deviations are
It will replace records that do not.

(S) (Process 962) Clear the control status flag of the record in the file 660.

(T) (Process 938) Return the record on file 660 to 620.

By repeating the above operations, file 6
For each product name above 20, the control variable 760 is gradually improved, and the sum of deviations (see part 758) becomes smaller. That is, control variables of the carrier distribution forming device that provide a desired intra-wafer carrier distribution are autonomously searched for. The reason for randomizing 〓 _E and making it 〓 _E ′ in the above (h.5) is, for example, records 700 to 72 of product name AG1 in file 620.
This is to keep the control variables 〓1~〓 _N always having a certain degree of variation. If this is not implemented, then 〓1=〓2=……=〓
Considering the case where it converges with _N , it is no longer possible to generate new points using procedure (h.4). Therefore, even if the device 120 deteriorates and the optimum value of the control variable gradually shifts, it will not be possible to autonomously adapt to the new optimum value. On the other hand, if we continue to randomize C by selecting an appropriate minute amount using the method described in (h.5), a new control variable 〓 _E will be added to the device 120 every time the operation in (h.4) is performed. Because it keeps being handed over,
It becomes possible to autonomously adapt to changes in the optimum point of variables due to changes in the characteristics of the device, and it becomes possible to omit the work of regularly inspecting the device and reviewing control conditions.

In (h) above, one method for generating the new point 〓 _E was shown, but a more efficient method is the simplex algorithm (Nelder, JA and Mead, R.: A
simplex method for function minimization，
Comput.J.5 (1965)) is possible. In this case, the method for generating the new point 〓 _E becomes complicated. If the control condition flag is made multi-valued and the method of generating the new point 〓 _E is switched according to the value of the flag, 〓 _E can be generated in a framework similar to the above.

Next, other embodiments of the present invention will be described. In the above example, the theoretical ellipso data value (155) corresponding to the desired carrier distribution shape (see Figure 2) is used.
The control variable 160 to the device 120 was adjusted so that the ellipsodata (170) actually measured from the process approximated.

However, as another method, the desired distribution function is directly given as the signal 155, and the control signal generating section 110 is provided with a function for estimating the intra-wafer carrier distribution from the actually measured ellipso data. It is also possible to adjust the control variable 160 in a direction that minimizes the difference between the distribution and this estimated distribution. In order to achieve this, it is necessary to apply the carrier distribution estimation method to the applicant's earlier application (Japanese Patent Application No. 56-17012,
-200191), and the processing method for minimizing the sum of deviations in the above embodiment of the present invention may be used to reduce the deviation between the two distributions.

Next, the present invention and a method applied to the recycling of semiconductor products will be described.

In the carrier distribution formation process, if a distribution that is less than the planned concentration distribution or conversely exceeds the planned concentration distribution is formed, conventionally, the product must be discarded as a defective product. Ta. As can be seen from its content, the present invention provides means for controlling the carrier distribution of semiconductor products without destroying them. Therefore, for semiconductor products that are defective because they do not reach the expected concentration distribution, the amount of added impurities, the processing temperature of the diffusion furnace, etc. should be adjusted so that the ellipso data approaches the target ellipso data.
All you have to do is control the time etc. For defective products that exceed the planned concentration distribution, the ellipso data may be brought closer to the target ellipso data by adding a new impurity that offsets the effect of the impurity and diffusing it.

In carrying out these operations, it is necessary to precisely change the carrier distribution that has already been formed, so the carrier distribution forming process is controlled using, for example, the following model.

Δ _i =a _0i +a _i Δ ^* ₁ +a _2i Δ ^* ₂ +…+a _ni Δ ^* _n +b _1i Ψ ^* ₁ +
…＋
b _ni Ψ ^* _n +c _1i x ₁ +C _2i x ₂ +…+c _oi x _o i=1, 2,…, m Ψ _i =a′ _0i +a′ _1i Δ ^* ₁ +a′ _2i Δ ^* ₂₂ ^* +…+a ′ _ni Δ ^* _n
+b′ _1i Ψ ^* ₁
+…+b′ _ni Ψ ^* _n +c′ _1i x ₁ +c′ _2i x ₂ +…+c′ _oi x _o i=1, 2,…, m Here, (Δ _i , Ψ _i ) i=1~m This is the target ellipso data, and (Δ ^* _i , ψ ^* _ii = 1~m are the actual measured ellipsodata values of defective products. _x1 ~ _xo are the control variables of the diffusion furnace and additional impurities. Coefficient a _0i ~
c _oi , a′ _0i ~ c′ _oi are (Δ _i , Ψ _i ), (Δ ^* _i , Ψ ^* _i )
, x ₁ ~ x _o
It is estimated by, for example, the method of least squares using actual values such as . When recycling defective products,
With (Δ _i , ψ _i ) known as the target values and (Δ ^* _i , ψ ^* _i ) known as the current values, the above simultaneous equations are solved to find the control variables x ₁ to x _o ( In order to solve the equation, select n≦m.If additional processing is performed using this control variable on a defective product, a semiconductor element that gives ellipsodata close to the target ellipsodata (Δ _i , ψ _i ) In other words, it is possible to reproduce defective products.

According to the present invention, the carrier distribution forming process in the semiconductor manufacturing process can be fully automated, which was previously impossible.
High-performance control becomes possible, making it possible to improve and shorten product yield.

[Brief explanation of the drawing]

Fig. 1 is a block diagram of an embodiment of the control of the present invention, Fig. 2 is an explanatory diagram of a parameter expression method for carrier distribution, Figs. 3 and 4 are an explanatory diagram of an ellipso data acquisition device, and Fig. 5 is Ellipso data acquisition status diagram,
FIG. 6 is a system configuration diagram of an embodiment of the control method of the present invention, and FIG. 7 is a diagram of the data file 6 in FIG.
20, FIG. 8 is an explanatory diagram of the processing contents of the program 610 in FIG. 6, and FIG. 9 is an explanatory diagram of the processing contents of the program 650 in FIG.

Claims (1)

[Claims] 1. Infrared light is incident on the surface of a semiconductor wafer while changing at least the angular frequency or incidence angle of the infrared light, and the change in polarization of the reflected light from the surface of the semiconductor wafer is measured. Calculate the ellipsodata of the reflected light, and calculate the measured value of the ellipsodata of the reflected infrared light and the theoretical value of the ellipsodata that should be observed for the distribution in the depth direction of carriers that you want to form on the semiconductor wafer. 1. A semiconductor formation process control method, characterized in that the carrier distribution forming process of the semiconductor wafer is controlled so that the deviation value is reduced.