JP2013026503A - Method of manufacturing semiconductor device, and manufacturing apparatus for semiconductor device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To precisely monitor a manufacturing process of a semiconductor device.SOLUTION: An output waveform of a third sensor is formed from a waveform A1 which corresponds to a dynamic range of the sensor by a process according to a recipe. After that, a substrate is processed according to the recipe so that a waveform A2 caused by the process is formed. The waveform A2 is a signal having a noise level. By normalizing the output waveform of the third sensor, a signal at a noise level is prevented from being amplified. An operational state of a semiconductor manufacturing apparatus is monitored by performing multivariate analysis using data provided by normalizing output waveforms of a first sensor, a second sensor, and the third sensor, respectively.

Description

  The present invention relates to a semiconductor device manufacturing method and a semiconductor device manufacturing apparatus.

  In the manufacturing process of a semiconductor device, it is important to grasp and monitor the operating state of a semiconductor device manufacturing apparatus (hereinafter referred to as a semiconductor manufacturing apparatus) in order to stabilize and improve the process. For this purpose, many sensors for detecting the operating state of the semiconductor device are attached to the semiconductor manufacturing apparatus. Furthermore, a mechanism for monitoring the manufacturing process of the semiconductor device more comprehensively is obtained by acquiring information on the position and orientation of the transfer robot that carries the substrate into the semiconductor manufacturing apparatus.

  As a method for monitoring a process in a manufacturing process of a semiconductor device, for example, a diagnosis recipe is executed in the semiconductor manufacturing device, data from a sensor is acquired, and the necessity of maintenance of the semiconductor device is determined based on the data. Sometimes. With this method, it is possible to determine whether maintenance is necessary without stopping the semiconductor device. Furthermore, after executing the diagnostic recipe, the normal production process can be immediately resumed.

  In addition, when the semiconductor manufacturing apparatus is a plasma semiconductor manufacturing apparatus, it is known to detect abnormal discharge by removing high-frequency components caused by plasma. The high frequency component can be extracted by performing Fourier transform on a waveform cut out at a short time interval in which the baseline is considered to be constant.

  Here, when the number of sensors increases, it becomes difficult to determine a change in the operating state of the semiconductor device from the output values of the individual sensors. Furthermore, there is an event that can be detected for the first time by combining fluctuations in output values of a plurality of sensors. Therefore, it is difficult for the univariate monitoring method for determining the operation status for each sensor to accurately grasp the change in the operation state of the semiconductor device. For this reason, in a semiconductor device to which a large number of sensors are attached, it is desirable to detect statistically significant fluctuations by multivariate analysis. As a method of multivariate analysis, principal component analysis, MT method (Mahalanobis Taguchi method) and the like are mainly used. Further, in multivariate analysis that handles arbitrary unit system data, analysis processing is performed after normalizing fluctuations in data output from each sensor in order to simultaneously handle data from different sensors.

JP 2010-267711 A JP2005-144043 JP 2009-147183 A JP 2004-47885 A

  However, depending on the sensor, the output value may not change or the amount of change may be very small. When the sensor output is zero point fluctuation or fluctuation below the observation resolution, normalization processing of the sensor output waveform provides data in which fluctuation of noise level data is amplified. For this reason, data that was originally only noise is treated as data with significant fluctuations. As a result, it may be difficult to determine the operating status of the semiconductor manufacturing apparatus.

On the other hand, if the noise-only sensor is removed from the monitoring target and analyzed, the influence of the noise-only sensor can be excluded, but detection of an abnormal waveform in the sensor becomes difficult.
The present invention has been made in view of such circumstances, and an object thereof is to accurately monitor the manufacturing process of a semiconductor device.

  According to one aspect of the embodiment, a step of acquiring first data output when processing the substrate from a sensor attached to a processing unit that processes the substrate for manufacturing a semiconductor device, and the first Obtaining second data having a value greater than the noise level of the data as the output of the sensor, and normalizing the output value of the sensor using both the first data and the second data; A step of performing a multivariate analysis using values obtained by normalizing output values acquired from a plurality of sensors including the sensor attached to the processing unit, and creating a determination value for determining an abnormality of the processing unit A manufacturing method is provided for a semiconductor device.

  Moreover, according to another viewpoint of embodiment, the process part which processes the board | substrate which manufactures a semiconductor device, The some sensor attached to the said process part in order to monitor the process with respect to the said board | substrate, The said some sensor An output of the at least one sensor, a device control unit for acquiring first data and second data having a value greater than a noise level of the first data, the first data and the second data Both the data are used to normalize the output value of the sensor, and a multivariate analysis is performed using data obtained by normalizing the output values of a plurality of sensors, thereby creating a determination value for determining an abnormality of the processing unit. And a semiconductor device manufacturing apparatus including the analyzing unit.

  Since data without significant fluctuations is not amplified by the normalization process, the operating status of the semiconductor device manufacturing apparatus can be accurately determined.

FIG. 1 is a block diagram showing an example of a schematic configuration of a semiconductor device manufacturing apparatus according to the first embodiment of the present invention. FIG. 2 is a flowchart of a process performed by the semiconductor device manufacturing apparatus according to the first embodiment of the present invention. 3A and 3B are diagrams for explaining the principle of processing performed by a conventional semiconductor device manufacturing apparatus. FIG. 3A shows measured data, and FIG. 3B shows data obtained by normalizing the waveform of FIG. 4A and 4B are diagrams for explaining the principle of processing by the semiconductor device manufacturing apparatus according to the first embodiment of the present invention. FIG. 4A shows actual measurement data with a waveform added, and FIG. ) Shows data obtained by normalizing the waveform. FIG. 5 is a block diagram showing a schematic configuration of a CMP apparatus as an example of a semiconductor device manufacturing apparatus according to the first embodiment of the present invention. FIG. 6 is a diagram illustrating an example of an output waveform of a first sensor of a conventional CMP apparatus. FIG. 7 is a diagram showing an example of a trend chart of the output waveform of the first sensor of the conventional CMP apparatus. FIG. 8 is a diagram illustrating an example of an output waveform of the second sensor of the conventional CMP apparatus. FIG. 9 is a diagram showing an example of a trend chart of the output waveform of the second sensor of the conventional CMP apparatus. FIG. 10 is a diagram illustrating an example of a graph in which the principal component analysis results of a conventional CMP apparatus are plotted. FIG. 11 is a diagram showing an example of a trend chart of the first main component for a conventional CMP apparatus. FIG. 12 is a diagram showing an example of factor analysis for a conventional CMP apparatus. FIG. 13 is a diagram showing an example of an output waveform of the first sensor of the CMP apparatus according to the first embodiment of the present invention. FIG. 14 is a diagram showing an example of a trend chart of the output waveform of the first sensor of the CMP apparatus according to the first embodiment of the present invention. FIG. 15 is a diagram illustrating an example of a graph plotting the principal component analysis results of the CMP apparatus according to the first embodiment of the present invention. FIG. 16 is a diagram showing an example of a trend chart of the first main component in the CMP apparatus according to the first embodiment of the present invention. FIG. 17 is a diagram showing an example of factor analysis for the CMP apparatus according to the first embodiment of the present invention. FIG. 18 is a block diagram showing a schematic configuration of a sputtering apparatus as an example of a semiconductor device manufacturing apparatus according to the second embodiment of the present invention. FIG. 19 is a flowchart of a process performed by the semiconductor device manufacturing apparatus according to the second embodiment of the present invention. 20A and 20B are diagrams for explaining the principle of processing by the semiconductor device manufacturing apparatus according to the first embodiment of the present invention. FIG. 20A shows actual measurement data to which waveforms are added, and FIG. ) Shows data obtained by normalizing the waveform. FIG. 21 is a diagram illustrating an example of an output waveform of a first sensor of a conventional sputtering apparatus. FIG. 22 is a diagram showing an example of a trend chart of the output waveform of the first sensor of the conventional sputtering apparatus. FIG. 23 is a diagram illustrating an example of an output waveform of the second sensor of the conventional sputtering apparatus. FIG. 24 is a diagram showing an example of a trend chart of the output waveform of the second sensor of the conventional sputtering apparatus. FIG. 25 is a diagram showing an example of a graph plotting the principal component analysis results of a conventional sputtering apparatus. FIG. 26 is a diagram showing an example of a trend chart of the first main component for a conventional sputtering apparatus. FIG. 27 is a diagram showing an example of factor analysis for a conventional sputtering apparatus. FIG. 28 is a diagram showing an example of an output waveform of the first sensor of the sputtering apparatus according to the second embodiment of the present invention. FIG. 29 is a diagram showing an example of an output waveform of the second sensor of the sputtering apparatus according to the second embodiment of the present invention. FIG. 30 is a diagram illustrating an example of a graph plotting the principal component analysis results of the sputtering apparatus according to the second embodiment of the present invention. FIG. 31 is a diagram showing an example of a trend chart of the first main component in the sputtering apparatus according to the second embodiment of the present invention. FIG. 32 is a diagram showing an example of factor analysis for the sputtering apparatus according to the second embodiment of the present invention. FIG. 33 is a block diagram showing an example of a schematic configuration of a semiconductor device manufacturing system according to an embodiment of the present invention.

The objects and advantages of the invention will be realized and attained by means of the elements and combinations particularly pointed out in the appended claims.
The foregoing general description and the following detailed description are exemplary and explanatory only and are not intended to limit the invention.

(First embodiment)
The first embodiment will be described in detail with reference to the drawings.
FIG. 1 shows an example of a semiconductor device manufacturing apparatus. A semiconductor device manufacturing apparatus 1 (hereinafter referred to as a semiconductor manufacturing apparatus 1) includes a processing unit 2 that performs a predetermined treatment on a substrate on which the semiconductor device is manufactured, and a control unit 3 that controls the processing unit 2. The processing unit 2 is provided with a large number of sensors 10, 11, and 12 that are used to control processing. The number of sensors 10-12 is not limited to three.

  The control unit 3 includes a device control unit 21 having a CPU (Central Processing Unit), a memory, and the like, an input / output device 22 that controls input / output of data, a display unit 23 that displays data, and a storage device that stores data 24.

  The apparatus control unit 21 performs a multivariate analysis on the processing control unit 31 that controls the processing unit 2, performs multivariate analysis on the outputs of the sensors 10 to 12, and calculates a determination value that determines the operation status of the processing apparatus 1; The function is divided into a determination unit 33 that evaluates the state of the semiconductor manufacturing apparatus 1 based on the multivariate analysis result.

  Examples of the input / output device 22 include a keyboard, a mouse, a communication control device, and a printing device. The display unit 23 includes a display, for example. The storage unit 24 stores recipes for processing a substrate, data such as coefficients used when data processing is performed by the analysis unit 32, determination reference values used by the determination unit 33, and the like.

Next, a method for manufacturing the semiconductor device of this embodiment will be described below.
First, the outline of the operation of the semiconductor manufacturing apparatus 1 will be described with reference to the flowchart shown in FIG. In step S101, the process control unit 31 reads a recipe from the storage device 24, executes the recipe, and executes a predetermined process on the substrate. The recipe includes processing the substrate and causing at least one sensor 10-12 to output a pseudo waveform. The processed substrate is unloaded from the processing unit 2 for another process, for example. Further, an unprocessed substrate is newly carried into the processing unit 2.

  In step S102, the analysis unit 32 performs multivariate analysis using the outputs of the multiple sensors 10-12. In subsequent step S103, the determination unit 33 determines the operating state of the semiconductor manufacturing apparatus 1 using the result of the multivariate analysis. Here, the reference value used in the determination process in step S104 uses data registered in advance in the storage device 24 as a model of a normal operating state. The normal operating state model is created, for example, by accumulating normal sensor data, performing principal component analysis using these data, and calculating parameters and variation data during normal operation.

  If the result of the determination is that the next substrate can be processed (Yes in step S104), the processing here ends. In this case, the process proceeds to the next substrate processing. On the other hand, if the result of the determination indicates that maintenance or the like of the semiconductor manufacturing apparatus 1 is necessary (No in step S104), the process proceeds to step S105, and the determination unit 33 issues an alarm. The alarm is displayed on the display unit 23, for example. In addition, the alarm may be transmitted to an information terminal used by the worker using e-mail or the like.

Next, details of the multivariate analysis in step S102 of FIG. 2 will be described below.
First, as a comparison, analysis by general data processing will be described. When the processing control unit 31 of the control unit 3 executes a general polishing recipe, output waveforms as illustrated in FIG. 3 for the three sensors 10 to 12 are obtained. The horizontal axis in FIG. 3 indicates time, and the vertical axis indicates the output waveform of the first sensor, the output waveform of the second sensor, and the output waveform of the third sensor in order from the top.

  FIG. 3A shows measured data (raw data) of output waveforms of the three sensors 10 to 12. The first sensor 10 shows a curved output, and the second sensor 11 shows a pulsed output. That is, the output waveform of the first sensor 10 continuously varies, and the output waveform of the second sensor 11 varies between the low level and the high level. That is, the output waveforms of these two sensors 10 and 11 are significantly changed. On the other hand, the value of the third sensor 12 does not vary. Even if there is a fluctuation in the waveform output from the third sensor 12, the fluctuation is considered to be within a noise range such as a zero point fluctuation or an output fluctuation below the observation resolution.

  Subsequently, FIG. 3B shows data obtained by normalizing actually measured data of output waveforms of the three sensors 10 to 12. Here, in the multivariate analysis, since the absolute values of the sensors 10 to 12 have no meaning, the analysis unit 32 of the control unit 3 normalizes the data of the sensors 10 to 12. As a result, a plurality of physical quantities having completely different values can be evaluated on the same scale, and abnormality detection of the processing unit 2 and factor analysis at the time of abnormality detection can be performed. Normalization is obtained, for example, by calculating an average deviation and variance for each of the sensors 10 to 12, and dividing the average deviation by the standard deviation.

  As shown in FIG. 3B, the first sensor 10 obtains a curved output and the second sensor 11 obtains a pulse-like output by normalization processing. In the third sensor 12, a waveform with a large variation in waveform is obtained by the normalization process. This is because the third sensor 12 has no significant waveform fluctuation in the actual measurement data, so that the noise is expanded beyond the resolution of the sensor 12 by the normalization process, and a waveform reflecting the shape of the noise is obtained. .

  As described above, after normalizing the output waveform of the third sensor 12, the data is data with large fluctuations. For this reason, if multivariate analysis is performed using these three sensors 10 to 12, the result is influenced by the waveform of the noise level of the third sensor 12. For this reason, even if the semiconductor manufacturing apparatus 1 is normally operating normally, it may be determined to be abnormal.

  On the other hand, the output waveform and the data after normalization in the manufacturing method of the semiconductor device of this embodiment will be described with reference to FIG. The horizontal axis in FIG. 4 indicates time, and the vertical axis indicates the output waveform of the first sensor 10, the output waveform of the second sensor 11, and the output waveform of the third sensor 12 in order from the top. As shown in FIG. 4A, the first sensor 10 shows a curved output, and the second sensor 11 shows a pulsed output. The output waveforms of the two sensors 10 and 11 are considered to show fluctuations. The value of the third sensor 12 becomes a waveform A2 (first data) having a substantially constant value after the pulse-shaped waveform A1 (second data) is first detected.

  In the output waveform of the third sensor 12, the first pulse-shaped waveform A1 is intentionally changed by the recipe of the semiconductor manufacturing apparatus 1, and the generation timing thereof is before the processing for the substrate starts. . The changed waveform A1 is a waveform that exceeds the noise level of the third sensor 12, and has such a shape that the waveform generated during normal operation and the dynamic range of the third sensor 12 can be understood. That is, in the recipe of the semiconductor manufacturing apparatus 1 of this embodiment, a step is added in which the third sensor 12 outputs the waveform A1 before starting the processing. Here, the value of the third sensor 12 does not change in the waveform A2 during the substrate processing following the waveform A1. Even if there is a fluctuation in the waveform A2, it is considered that the fluctuation is within a noise range such as a zero point fluctuation or an output fluctuation below the observation resolution.

  Then, the output waveform of each sensor 10-12 after a normalization process is shown in FIG.4 (b). The data of the first and second sensors 10 and 11 can be output in the form of curves and pulses, respectively. As the data of the third sensor 12, pulsed data corresponding to the first waveform A1 and substantially linear data are obtained. Conventionally, the noise level is amplified and the waveform is disturbed, but the presence of the waveform A1 prevents the noise of the waveform A2 from being amplified. As a result, data in which the amount of change is almost zero is obtained in the region of the waveform A2.

  Thus, in this embodiment, the processing unit 2 is operated according to the recipe, and the waveform A1 different from the waveform generated during the substrate processing is output from the third sensor 12. The processing control unit 31 receives the output waveforms of the sensors 10 to 12, and the analysis unit 32 performs normalization processing using each output waveform including the waveform A1. Further, the analysis unit 32 performs principal component analysis using the normalized data. At the time of principal component analysis, data corresponding to the waveform A1 is not considered. As a result, noise level fluctuations are not amplified during multivariate analysis. As a result, the influence of noise level data is reduced, and erroneous determination of the operating state of the semiconductor manufacturing apparatus 1 is prevented.

  Next, a chemical mechanical polishing (CMP) apparatus will be described as an example of application of the manufacturing method to the semiconductor device of this embodiment. In the following, first, the schematic structure of the CMP apparatus and the outline of the polishing process by the CMP method will be described, and then the conventional multivariate analysis will be described. Further, multivariate analysis in the manufacturing method for the semiconductor device of this embodiment will be described later.

  FIG. 5 shows a chemical mechanical polishing (CMP) apparatus 41 as an example of a semiconductor device manufacturing apparatus. The CMP apparatus 41 includes a processing unit 2 and a control unit 3. The processing unit 2 includes a platen (surface plate) 50 that serves as a rotary table. The platen 50 is rotatably supported by a rotating shaft 51. A polishing cloth 52 is affixed to the upper surface of the platen 50. Above the platen 50, a polishing head 55 for holding a substrate 4 for manufacturing a semiconductor device is rotatably arranged. The rotating shaft 56 of the polishing head 55 is disposed in parallel with the rotating shaft 51 of the platen 50. The polishing head 55 is provided with a pressure mechanism (not shown) that presses the substrate 4 against the polishing pad 52 with a predetermined pressure.

  Further, a pad conditioner 58 provided with a file member for sharpening the polishing cloth 52 of the platen 50 is disposed above the platen 50. The pad conditioner 58 is attached downward to the rotary shaft 59 and can rotate on the platen 50 and reciprocate on the platen 50. Further, above the platen 50, two slurry pipes 60 and 61 for supplying slurry onto the polishing cloth 52, and pure water (deionized water) can be supplied onto the polishing cloth 52 for cleaning or the like. A liquid serving 22 is provided.

  Here, a large number of sensors 10 to 12 are attached to the processing unit 2 of the CMP apparatus 41. For example, a first flow sensor (first sensor 10) and a second flow sensor (second sensor 11) are attached to the two slurry pipes 60 and 61, respectively. Further, a third flow rate sensor (third sensor 12) is attached to the liquid pipe 62. In addition, a sensor (not shown), for example, a sensor that detects the rotation speed of the platen 50, a sensor that detects the rotation speed of the polishing head 55, and a sensor that measures the polishing amount of the substrate 4 are attached.

Further, analysis processing in the CMP apparatus 41 will be described.
First, the case where the output waveforms of the sensors 10 to 12 are normalized as they are will be described below.
First, FIG. 6 shows an example of an output waveform of the first sensor 10 within a predetermined period. First sensor 1
For example, 0 is a sensor that measures the flow rate of the first slurry pipe 60. The horizontal axis represents the processing time, and the vertical axis represents the output waveform of the first sensor 10. The maximum flow rate of the first sensor 10 is 500 sccm. From FIG. 6, there is a first group GP1 in which the output waveform of the first sensor 10 mainly changes at a value greater than zero, and a second group GP2 in which the output waveform mainly changes at a value less than zero. Recognize. Furthermore, FIG. 7 shows a graph in which the average value of the output of the first sensor 10 is plotted for each processing day for the same data. The horizontal axis indicates the processing date, and the vertical axis indicates the average value of the output waveform. The output value of the first half of the processing date belongs to the first group GP1, and the output value of the second half of the processing date belongs to the second group GP2.

  For comparison, FIG. 8 shows an example of waveform data when the output value fluctuates due to a recipe command. FIG. 8 shows the measurement result of the output waveform of the second sensor 11, for example, the flow rate of the second slurry pipe 61. A maximum of about 350 sccm of fluid flowed through the second slurry pipe 61. Furthermore, there is no significant difference between treatment dates. This can also be seen from the average output value shown in FIG.

  Here, the output waveform of the first sensor 10 shown in FIGS. 6 and 7 is within the allowable range in both the first group GP1 and the second group GP2. FIG. 10 shows the result of principal component analysis using the outputs of a large number of sensors including this data. The horizontal axis represents the first principal component, and the vertical axis represents the second principal component.

Analysis focusing on the principal component is a technique used to handle multiple factors comprehensively instead of considering them one by one when multiple factors are considered for one problem. . For example, after calculating the average and standard deviation of each sensor 10-12, a correlation coefficient matrix of the outputs and principal components of each sensor 10-12 is created, and the eigenvalues and eigenvectors of the correlation coefficient matrix are calculated. The correlation coefficient matrix is such that when the output values of the plurality of sensors s ₁ , s ₂ , s ₃ ,... Are x ₁ , x ₂ , x ₃ ,. z ₁ = a ₁₁ × x ₁ + a ₁₂ × x ₂ + a ₁₃ × x ₃ +... The second principal component is expressed as z ₂ = a ₂₁ × x ₁ + a ₂₂ × x ₂ + a ₂₃ × x ₃ +... The third and subsequent principal components are similarly expressed. Further, each coefficient a ₁₁ , a ₁₂ , a ₁₃ ,... Is an element in the first row, each coefficient a ₂₁ , a ₂₂ , a ₂₃ ,. A correlation coefficient matrix is formed by using each coefficient of the third and subsequent principal components as elements.

  Next, the value of the principal component is obtained from the eigenvector of the correlation coefficient matrix. Also, the contribution ratio of each principal component is obtained from the eigenvalues of the interphase coefficient matrix. The contribution ratio of the principal component is obtained by dividing the eigenvalue by the sum of the eigenvalues. After calculating the contribution ratio of each principal component, the first principal component and the second principal component are determined from the one with the larger eigenvalue.

FIG. 10 is a graph in which the score values of each sample are plotted in a coordinate system composed of a first principal component and a second principal component. The horizontal axis represents the first principal component, and the vertical axis represents the second principal component. The score value of each sample is obtained by using the value obtained by normal processing the outputs of the plurality of sensors 10 to 12 acquired simultaneously and the coefficients of the interphase coefficient matrix, and the value of z ₁ for the first principal component and the second principal component. Z ₂ is calculated for each. For example, each output waveform of each of the sensors 10 to 12 obtained at the same time on a specific day is treated as one set of data, that is, one sample, and these data are subjected to multivariate analysis to obtain a score value of the first principal component. z ₁ and the score value z ₂ of the second principal component are calculated. Thus plotting the score value z ₂ of the score value z ₁ and the second main component of the first principal component as the score value of the sample that is calculated. One plot is made according to the score value calculated from one set of data. This score value becomes a determination value for determining the operation status of the CMP apparatus 41. In FIG. 10, paying attention to the first principal component on the horizontal axis, it can be seen that the plotted data is bipolarized into a minus side set and a plus side set.

  Further, FIG. 11 shows the result of examining the trend by arranging the data of the first principal component in time series. The horizontal axis indicates the processing date, and the vertical axis indicates the first main component. As shown in FIG. 11, it can be seen that the trend changes midway. The processing day of this change point coincides with the day when the trend of the output value of the flow sensor shown in FIG. 7 changes.

  FIG. 12 shows the result of analyzing the factor of the sample in order to analyze the cause of the trend change. The horizontal axis lists the physical quantities detected by the sensors 10-12. The vertical axis represents the contribution rate of factors that affect the operating status of the CMP apparatus 41. The factors are listed on the horizontal axis. For example, from the left, the sweep torque of the pad conditioner 58, the rotational torque of the platen 50, the rotational torque of the polishing head 55, the flow rate of the second slurry pipe 61, the pad conditioner 58 Sweep position, pressure for pressing the polishing head 55 against the platen 50, pressing pressure for the pad conditioner 58, flow rate of the first slurry pipe 60, pressure in the third zone of the polishing head 55, retainer for holding the substrate 4 of the polishing head 55 There are the pressure of the ring, the pressure of the inner tube that presses the substrate 4 of the polishing head 55, the rotational speed of the polishing head 55, the rotational speed of the platen 50, and the rotational speed of the pad conditioner 58. These factors correspond to the number of sensors 10 to 12, that is, the type and number of parameters monitored through the sensors 10 to 12. The contribution ratio of the factor is calculated as a ratio of data fluctuation due to the factor to the total fluctuation.

  Since the contribution ratio of the first sensor 10 is prominent, it is presumed that the data has been polarized because the flow rate of the first slurry pipe 60 is the main cause. However, in practice, there is no significant variation in the flow rate of the first slurry pipe 60, and the first sensor 10 only outputs a noise level waveform. In other words, the fluctuation of the noise level of the first sensor 10 is amplified by the normalization process, so that an abnormality occurs in the CMP apparatus 41 that normally operates normally, and the cause is the first slurry pipe 60. It was judged. In this way, multivariate analysis of sensor data including noise as it is may cause false alarms due to noise. In addition, there is a higher possibility of overlooking fluctuations in other factors that should be captured. In addition, there is a possibility that the fluctuation of the factor to be noticed is buried in the noise of the first sensor 10 and overlooked.

In contrast, data processing according to this embodiment will be described.
First, as shown in step S <b> 101 of FIG. 2, the process control unit 31 reads a recipe from the storage unit 24 and executes it. The recipe is roughly divided into a first step of supplying slurry from the first slurry pipe 60 before polishing the substrate 4 and a subsequent second step of normal substrate processing. Here, the normal substrate processing executed by the recipe will be described below.

  In the second step of normal substrate processing, a polishing head 55 that first holds a substrate 4 as a workpiece is placed on a polishing cloth 52, and a platen 50 and a polishing are supplied while supplying slurry from slurry piping 60 and 61. The head 55 is rotated. The polishing cloth 52 and the substrate 4 are rubbed through the abrasive grains of the slurry to generate heat. As a result, the temperature at the interface between the substrate 4 and the polishing pad 52 rises and chemical mechanical polishing proceeds. When the chemical reaction starts, heat generated by the chemical reaction is also superimposed, and the temperature rise is promoted. Eventually, a steady state is reached, and the surface of the substrate 4 is polished at a constant speed. When polishing is completed, the polishing head 55 is first lifted upward, and high-pressure cleaning is performed with pure water to remove polishing residues on the surface of the substrate 4. Further, the pad conditioner 58 is pressed against the polishing pad 52 while supplying pure water from the liquid pipe 62, and the polishing pad 52 is sharpened by the pad conditioner 58 while rotating both.

By executing the recipe in this way, data as shown in FIG. 13 is obtained as the output waveform of the first sensor 10. The output waveform of the first sensor 10 includes an output waveform A3 (second data) output before substrate polishing in the first step, and an output waveform A4 of the second step started after the output of the output waveform A3. (First data), and these are output waveforms A3
, A4 are continuous data. The output waveform A4 of the first sensor 10 after the start of polishing is substantially zero. On the other hand, the waveform A3 is larger than the output waveform A4 of the noise level, and has a shape and a size that significantly changes with respect to the output waveform A4. As the generation timing of the waveform A3 and the shape of the waveform A3, for example, a flow rate of 1/10 or more of the maximum flow rate may be flowed in the polishing liquid in a short time of 1 to 5 seconds during pad conditioning. Similar waveform changes may be generated for the other sensors 11 to 12.

  Here, when the recipe was executed a plurality of times on different processing dates, the output waveforms A3 and A4 had substantially the same profile regardless of the processing date. In FIG. 14, the average value of the output of the first sensor 10 for each date is plotted. The horizontal axis indicates the date, and the vertical axis indicates the average value of the output. It can be seen that there is no significant difference depending on the treatment date.

  FIG. 15 shows the result of principal component analysis using the outputs of a large number of sensors including this data. The horizontal axis represents the first principal component, and the vertical axis represents the second principal component. In calculating the principal component, data corresponding to the output waveform during the substrate processing excluding the added waveform is used. It can be seen that the data varies greatly in the first principal component on the horizontal axis. In addition, many data that are out of the range of the score values of the main components obtained in the normal state indicated by an ellipse also appear. The determination unit 33 compares the score value of the sample with a reference value stored in advance in the storage unit 24 (corresponding to step S103 in FIG. 2). If the data obtained by the principal component analysis falls within the area indicated by the ellipse in FIG. 15, it is considered to be within the reference value. That is, the plot deviating from the ellipse in FIG. 15 deviates from the reference value, and it is determined that there is a possibility that the operation state of the CMP apparatus 41 is abnormal. When it deviates from the reference value (No in step S104), the determination unit 33 commands the issuing of an alarm (step S105).

  Here, FIG. 16 shows the result of examining the trend by arranging the data of the first principal component in time series in order to evaluate the variation of the score value of the sample and the deviation from the reference value. It can be seen that the trend changes in two places along the way. That is, it is considered that a significant change has occurred in the CMP apparatus 41 when the trend changes. Further, FIG. 17 shows the result of analyzing the cause of the trend change. On the horizontal axis, the same factors as in FIG. 12 are arranged. In the example shown in FIG. 17, the contribution ratio of the fluctuation of the rotational torque of the polishing head 55 is prominent. From this, it can be seen that in the CMP apparatus 41, the rotational torque of the polishing head 55 varies. The analysis results shown in FIGS. 16 and 17 may be performed only when the determination unit 33 determines that the operating state of the CMP apparatus 41 is abnormal, or may be always performed regardless of whether there is an abnormality.

  As described above, in this embodiment, the step of processing the substrate 4 and the step of intentionally changing the output of the sensors 10 to 12 are provided in the recipe. In addition to the output waveform, a waveform having a significant variation is obtained. The added waveform suppresses amplification of a noise level signal included in the output waveform during substrate processing. As a result, the influence of noise in multivariate analysis can be reduced. For this reason, erroneous determination of the operating state of the semiconductor manufacturing apparatus 1 can be prevented.

  In addition, the intentional output of the sensors 10 to 12 added by the recipe is performed under a condition that does not affect the actual process and the characteristics of the semiconductor device. For this reason, the operation state of the semiconductor manufacturing apparatus 1 can be inspected without stopping the manufacture of the semiconductor device. Here, the outputs of all the sensors 10 to 12 may be changed intentionally depending on the recipe. You may form a recipe so that only the output of the specific sensors 10-12 may fluctuate. Further, the step of outputting a waveform larger than the noise level may be performed after the normal substrate processing step. When a normal substrate processing step is divided into a plurality of steps, a step of outputting a waveform larger than the noise level may be provided between the plurality of steps.

  The determination processing in steps S104 and S105 in FIG. 2 may issue an alarm if one of the normal ranges shown in the ellipse in FIG. 15 is exceeded, or if the data exceeding the normal range exceeds a predetermined number. An alarm may be issued.

  Here, when the semiconductor manufacturing apparatus 1 is an etching apparatus and the output waveform of the gas flow rate sensor is changed according to the recipe, all the gases are short-time, for example, 1 to 5 seconds at the start of processing or evacuation, Add a step to the recipe to flow about 1/10 or more of the maximum flow rate. The gas may actually flow through a chamber in the etching apparatus, or may flow through a bypass line or a vent line.

  Moreover, when the semiconductor manufacturing apparatus 1 has a vacuum degree sensor and changes the output waveform of such a sensor according to the recipe, for example, a state where the pressure is controlled to be low so that the change in the vacuum degree becomes large, Add a step to the recipe that will result in a high vacuum.

  Furthermore, when the semiconductor manufacturing apparatus 1 is a CVD apparatus and the output waveform of the gas flow sensor is changed according to the recipe, all gases are supplied for a short period of time, for example, 1 second to 2 seconds before starting the process or before evacuating. Add a step to the recipe to flow about 1/10 or more of the maximum flow rate.

(Second Embodiment)
The second embodiment will be described in detail with reference to the drawings. The same components as those in the first embodiment are denoted by the same reference numerals. Moreover, the description which overlaps with 1st Embodiment is abbreviate | omitted. In this embodiment, the value of the sensor is not intentionally changed by the recipe, but a waveform prepared in advance is added. As a suitable example of this embodiment, for example, it may be difficult to change the value of the sensor greatly immediately before or after the process. An example of such a sensor is a cryopump temperature sensor.

The principle of this embodiment will be described below.
First, FIG. 18 shows a sputtering apparatus 81 as an example of a semiconductor manufacturing apparatus. The sputtering apparatus 81 includes the processing unit 2 and the control unit 3.
The processing unit 2 has a support 86 for the substrate 4 in the vacuum chamber 85, and a target 87 is disposed above the substrate 4. A power source 88 is connected to the target 87. A gas supply device 89 is connected to the vacuum chamber 85 by piping, and a cryopump 90 is connected as an exhaust device. The sputtering apparatus 81 also has another pump (not shown) connected to the cryopump 90.

  The apparatus control unit 21 of the control unit 3 is functionally divided into a processing control unit 31, an analysis unit 32, a determination unit 33, and a waveform forming unit 34. The waveform forming unit 34 adds virtual waveform data to the output waveform (measured data) of at least one of the sensors 10 to 12.

  Further, sensors 10 to 12 are attached to the processing unit 2. Examples of the sensors 10 to 12 include a flow sensor for gas supplied to the vacuum chamber 85, a sensor for measuring a voltage applied to the target, a temperature sensor for a cryopump, and the like. Furthermore, a plurality of other sensors (not shown) are also attached to the processing unit 2.

  Further, the sputtering apparatus 81 forms a thin film on the substrate 4 according to the flowchart of FIG. In step S <b> 101, the process control unit 31 reads a recipe from the storage device 24, executes the recipe, and forms a thin film on the surface of the substrate 4. The processed substrate 4 is unloaded from the processing unit 2 for another process, for example. Further, the next substrate 4 is carried into the processing unit 2.

  In subsequent step S101A, the waveform forming unit 34 adds the waveform stored in the storage unit 24 to the output waveform of the at least one sensor 10-12. The added waveform data corresponds to an example of a waveform generated when a significant numerical change occurs in each of the sensors 10 to 12, for example, a sweep waveform. The waveform forming unit 34 reads waveform data (second data) stored in advance in the storage device 24 for each sensor and adds it to the end of the actual output waveform (first data). The timing of adding the waveform data is a time region that does not affect the processing, and the waveform data may be added to the beginning of the actually measured data.

  In step S <b> 102, the analysis unit 32 performs multivariate analysis from the outputs of the multiple sensors 10 to 12. In subsequent step S103, the determination unit 33 performs a determination process. Here, the result of multivariate analysis of sensor data in a normal processing state is used as the reference value used in the determination processing in step S104.

  As a result of the determination, if the processing of the next substrate can be continued (Yes in step S104), the processing here ends. In this case, the process proceeds to the next substrate processing. On the other hand, if the result of the determination indicates that maintenance or the like of the semiconductor manufacturing apparatus 1 is necessary (No in step S104), the process proceeds to step S105, and the determination unit 33 issues an alarm.

Here, an example of waveform addition by the waveform forming unit 34 will be described with reference to FIG.
As shown in FIG. 20A, the first sensor 10 has a waveform in which the numerical value rises linearly at the end of the curved output. Similarly, the second sensor 11 has a waveform in which the numerical value rises linearly at the end of the pulse output. As the value of the third sensor 12, a waveform in which the numerical value rises linearly is added to the end of the waveform having a substantially constant value.

  As shown in FIG. 20 (b), after normalization processing, data obtained by adding data such that the numerical value rises linearly is obtained after the data corresponding to the actual measurement data of each of the sensors 10-12. It is done.

  As the waveform data to be added, there is a sweep waveform having a normal output range width of the sensors 10 to 12. Further, the minimum value and the maximum value of the long-term output values of the sensors 10 to 12 may be examined, and a waveform including the minimum value and the maximum value may be created.

  Next, a manufacturing method for the semiconductor device of this embodiment will be described. In the following, multivariate analysis is generally described as a comparative example first. Further, multivariate analysis in the manufacturing method for the semiconductor device of this embodiment will be described later.

  Here, the temperature of the cryopump does not deviate significantly from the set value. However, there is a possibility that the sensor output may contain noise or fluctuate due to temperature fluctuations. The temperature fluctuation is, for example, about plus / minus 1K at 10K setting, and about plus / minus K at 100K setting. For this reason, if the waveform data having a minute fluctuation is subjected to multivariate analysis as it is, this noise or fluctuation may be regarded as a significant fluctuation.

  FIG. 21 is a diagram in which the waveform data of the temperatures of the 10K and 100K cryopumps 90 of the actually measured data before virtual waveform data is added by the waveform forming unit 34 are superimposed for a plurality of days. The first sensor 10 varies between 10K and 11K. As shown in FIG. 22, the trend of the average value of the first sensor 10 is shown. Further, as shown in FIG. 23, the second sensor 11 varies between 99K and 100K. As shown in FIG. 24, the trend of the average value of the second sensor is shown. It can be seen that both the first and second sensors 10 and 11 generate 1K vibration that seems to be noise or fluctuation. That is, the output waveforms of the first and second sensors 10 and 11 do not show a significant fluctuation that indicates an abnormality of the sputtering apparatus 81.

  FIG. 25 shows the result of principal component analysis using waveform data of temperature without adding virtual data. The X axis represents the first principal component, and the Y axis represents the second principal component. There are many plots that are greatly out of the range of the ellipse obtained from the reference value. FIG. 26 shows a trend chart of the first main component. It can be seen that the score values of the first principal component vary in the negative direction.

  Furthermore, the result of analyzing the cause of the trend change is shown in FIG. The horizontal axis lists the physical quantities detected by the sensors 10-12. Factors include, for example, the third flow meter from the left, the first flow meter, DC current, voltage, power, first cooling stage temperature, second cooling stage temperature, chamber internal pressure (part 1), exhaust pipe pressure, There are chamber internal pressure (part 2), heater / susceptor temperature, and chamber internal pressure (part 3). The vertical axis represents the contribution ratio of the score value of the first principal component. From FIG. 27, it is considered that the abnormal state of the sputtering apparatus 81 is mainly caused by the fluctuation of the first cooling stage temperature, that is, the output waveform of the first output sensor 10.

  As described above, when the output of the first sensor 10 that does not show a significant fluctuation is normalized, the noise component is amplified, and an analysis result as if the normal first cooling stage temperature is abnormal is obtained. It will come out. In order to prevent such erroneous determination, in this embodiment, erroneous determination is prevented by adding a waveform as follows.

  That is, in this embodiment, a thin film is formed on the substrate 4 as normal substrate processing in step S101 of FIG. At this time, a waveform corresponding to the process is output from each sensor 10-12. In step S101A, a waveform different from the output by the recipe is added to the output waveforms of these sensors 10-12. Specifically, as shown in FIG. 28, a 15K waveform A6 (second data) is added to the end of the output waveform A5 (first data) of the first sensor 10 during execution of the recipe. Similarly, as shown in FIG. 29, a 110K waveform A8 (second data) is added to the end of the output waveform A7 (first data) of the second sensor during recipe execution.

  The waveforms A6 and A8 added in this way do not actually change the parameters of the sputtering apparatus 81, but add preregistered data to the measured data (A5, A7). The output waveforms A5 and A7 during the substrate processing are almost zero. On the other hand, the waveforms A6 and A8 are larger than the noise level output waveforms A5 and A7, and have shapes and sizes that change significantly with respect to the output waveforms A5 and A7.

  Further, in the analysis processing in step S102, normalization processing is performed using the waveforms of the respective sensors 10 to 11 including the added waveforms A6 and A8. The result of the principal component analysis using the normalized data is shown in FIG. The horizontal axis represents the first principal component, and the vertical axis represents the second principal component. By adding the output waveform to the measured data by the waveform forming unit 34 and performing multivariate analysis, it can be seen that the variation of the first principal component that has occurred due to the temperature factor of the cryopump 90 has disappeared. Here, data corresponding to the waveforms A5 and A7 in the process excluding the added waveforms A6 and A8 are used for calculating the principal component.

  Here, a large fluctuation was not seen in the trend chart shown in FIG. Further, as shown in the result of the factor analysis in FIG. 32, it is understood that the factor of the first principal component fluctuation is a factor other than the first cooling stage temperature, that is, the pressure fluctuation of the exhaust pipe. Since there is a plot (determination value) exceeding the ellipse based on the reference value, the determination unit 33 issues an alarm.

As described above, in this embodiment, when a recipe that intentionally changes the output waveforms of the sensors 10 to 12 cannot be executed, the normalization process is performed with a waveform prepared in advance added. Since the multivariate analysis is performed, the influence of the noise level waveform can be reduced. As a result, the same effect as in the first embodiment can be obtained.

  Here, in the sputtering apparatus 81, a sensor such as a gas flow rate can be dealt with by changing the recipe as in the first embodiment. That is, in one semiconductor manufacturing apparatus 1, there may be a sensor that adds a waveform according to a recipe step and a sensor that adds data prepared in advance without using a recipe. In this case, a waveform is added to the output waveform of at least one sensor by a recipe, and virtual data is added to the output waveform of at least one sensor.

Each embodiment also includes a program that causes a computer to execute steps S101 to S105. Furthermore, the recording medium storing the program and storing the program in a downloadable state in the computer are also included in the embodiments.
Moreover, in each embodiment, the control part 3 may perform apparatus control and multivariate analysis, and other apparatuses and operators may perform determination processing.

  Further, as shown in a schematic diagram of a semiconductor device manufacturing system in FIG. 33, an analysis unit 32, a determination unit 33, and a waveform forming unit are included in the device control unit 121 of the computer 101 connected to the semiconductor manufacturing device 1 via a network. 34 may be provided. The computer 101 includes a device control unit 21, a display unit 122, an input / output device 123, and a storage unit 24. The storage unit 24 stores data such as coefficients used when data processing is performed by the analysis unit 32, determination reference values used by the determination unit 33, and the like. In such a semiconductor device manufacturing system 100, the same effects as those of the second embodiment can be obtained.

  Here, the computer 101 of the semiconductor device manufacturing system 100 shown in FIG. 33 may not have the waveform forming section 34. In this case, the waveform is added to the output waveform of at least one of the sensors 10 to 12 according to the recipe executed in the semiconductor manufacturing apparatus 1. In this semiconductor device manufacturing system 100, the same effects as in the first embodiment can be obtained.

  All examples and conditional expressions given here are intended to help the reader understand the inventions and concepts that have contributed to the promotion of technology, and such examples and It is to be construed without being limited to the conditions, and the organization of such examples in the specification is not related to showing the superiority or inferiority of the present invention. While embodiments of the present invention have been described in detail, various changes, substitutions and variations can be made thereto without departing from the spirit and scope of the present invention.

The features of the embodiment will be additionally described below.
(Additional remark 1) The process of acquiring the 1st data output when processing the said board | substrate from the sensor attached to the process part which processes the board | substrate which manufactures a semiconductor device, The noise level of said 1st data Obtaining the second data having a larger value as the output of the sensor, normalizing the output value of the sensor using both the first data and the second data, and the processing unit A multivariate analysis using values obtained by normalizing output values acquired from a plurality of sensors including the sensor, and creating a determination value for determining an abnormality of the processing unit. A method of manufacturing a semiconductor device.
(Supplementary Note 2) The first step of the recipe registered in the storage unit is performed to acquire the second data, the second step of the recipe is performed to process the substrate, and the first step is performed. 1. The method for manufacturing a semiconductor device according to appendix 1, wherein data of 1 is acquired.
(Supplementary note 3) The semiconductor according to supplementary note 1 or supplementary note 2, wherein in the step of obtaining second data, the second data registered in the storage unit is added to the first data. Manufacturing method on the device.
(Supplementary Note 4) The method of manufacturing a semiconductor device according to any one of Supplementary notes 1 to 3, wherein the second data is a waveform or a sweep waveform corresponding to a dynamic range of the sensor.
(Supplementary note 5) Supplementary notes 1 to Supplementary notes including steps of previously registering a determination value when the processing unit is operating normally as a reference value and issuing an alarm when the generated determination value exceeds the reference value 5. The method for manufacturing a semiconductor device according to claim 4.
(Additional remark 6) As the output of the process part which processes the board | substrate which manufactures a semiconductor device, the several sensor attached to the said process part in order to monitor the process with respect to the said board | substrate, and at least 1 sensor of the said several sensor An apparatus control unit for acquiring first data and second data having a value greater than a noise level of the first data, and the sensor using both the first data and the second data. An analysis unit that normalizes the output values of the plurality of sensors, performs multivariate analysis using data obtained by normalizing the output values of the plurality of sensors, and creates a determination value for determining an abnormality of the processing unit. An apparatus for manufacturing a semiconductor device.
(Additional remark 7) It has a memory | storage part which memorize | stored the recipe containing the step which processes the said board | substrate, acquires said 1st data, and acquires the said 2nd data. The manufacturing apparatus of the semiconductor device of description.
(Additional remark 8) It has the memory | storage part which memorize | stored the recipe which processes the said board | substrate, and acquires said 1st data, and said 2nd data added to said 1st data 6. A manufacturing apparatus of a semiconductor device according to 6.
(Supplementary note 9) The semiconductor device manufacturing apparatus according to supplementary note 8, wherein the recipe includes a step of processing the substrate to acquire the first data, and a step of acquiring the second data.
(Supplementary Note 10) The supplementary notes 6 to 9 include a judgment unit that holds a judgment value when the processing unit is operating normally as a reference value and issues an alarm when the created judgment value exceeds the reference value. The manufacturing apparatus of the semiconductor device as described in any one.
(Additional remark 11) The 1st data output from the sensor attached to the process part which processes the board | substrate which manufactures a semiconductor device is acquired, The 2nd data which has a value larger than the noise level of said 1st data Acquired as the output of the sensor, normalizes the output value of the sensor using both the first data and the second data, and normalizes the data acquired from a plurality of sensors attached to the processing unit A manufacturing program for a semiconductor device, which causes a computer to perform a multivariate analysis process using the converted values.

1 Semiconductor manufacturing equipment (semiconductor equipment manufacturing equipment)
2 Processing Unit 3 Control Unit 4 Substrate 24 Storage Unit 31 Processing Control Unit 32 Analysis Unit 33 Determination Unit 41 CMP Device 81 Sputtering Device

Claims (6)

Obtaining a first data output when processing the substrate from a sensor attached to a processing unit for processing a substrate for manufacturing a semiconductor device, and having a value greater than a noise level of the first data Obtaining second data as an output of the sensor;
Normalizing the sensor output value using both the first data and the second data;
A multivariate analysis using each of the output values acquired from a plurality of sensors including the sensor attached to the processing unit, and a determination value for determining an abnormality of the processing unit is created; and ,
A method for manufacturing a semiconductor device, comprising:   The first step of the recipe registered in the storage unit is performed to acquire the second data, the second step of the recipe is performed to process the substrate, and the first data is The method for manufacturing a semiconductor device according to claim 1, wherein the method is obtained.   3. The semiconductor device according to claim 1, wherein in the step of acquiring second data, the second data registered in the storage unit is added to the first data. Production method. A processing unit for processing a substrate for manufacturing a semiconductor device;
A plurality of sensors attached to the processing unit to monitor processing on the substrate;
An apparatus control unit that acquires first data and second data having a value larger than a noise level of the first data as an output of at least one of the plurality of sensors;
The output value of the sensor is normalized using both the first data and the second data, and multivariate analysis is performed using data obtained by normalizing the output values of a plurality of sensors, and the processing unit An analysis unit for creating a determination value for determining an abnormality of
An apparatus for manufacturing a semiconductor device, comprising:   6. The semiconductor according to claim 5, further comprising a storage unit storing a recipe including a step of processing the substrate to acquire the first data and a step of acquiring the second data. Equipment manufacturing equipment.   The storage unit that stores a recipe for processing the substrate to acquire the first data and the second data added to the first data. Semiconductor device manufacturing equipment.

Images