JP2006173373A - Manufacturing system of semiconductor product - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a manufacturing system equipped with a device abnormality monitoring system, capable of automatically updating the registration release of determination items of device abnormality monitoring and a determination standard. <P>SOLUTION: A manufacturing system of a product targeting a wafer comprises a device abnormality monitoring system 22 of a processing device 21 located upstream of a line, and a product abnormality monitoring system 52 of a product inspector 51 located downstream of the line. The manufacturing system comprises, as a line control system 60 therefor, a wafer trace system 61 for collecting data of product quality and a device status starting from an abnormality detection event, and a line quality management system 62 for checking and analyzing the trace collected data, to review monitoring items conforming to an up-to-date production line status, and to update the recomputed determination standard. <P>COPYRIGHT: (C)2006,JPO&NCIPI

Description

  The present invention relates to a manufacturing system for controlling a line for manufacturing a semiconductor product by integrating and processing fine elements on a substrate, and more particularly, to a manufacturing system having a function of monitoring a state change of a processing apparatus that causes a quality change of a product. It is related to effective technology.

  For example, a typical example of a semiconductor product manufactured by integrating and processing fine elements on a substrate is a wafer product manufactured by integrating and processing MOS transistors on a wafer. As inventions of methods and apparatuses for analyzing the cause of product quality fluctuation in a wafer product production line, there are those described in Patent Document 1 and Patent Document 2. These are inventions of a method and apparatus for analyzing the causal relationship between product quality information and manufacturing information. The product quality information targeted here is electrical characteristic information and yield information of semiconductor wafers, and manufacturing information is manufacturing equipment history information, manufacturing condition information, in-line measurement information, and equipment flow information related to the process process. is there.

  As a method for analyzing the causal relationship, multistage multivariate analysis (Y = A · X) is used in which product quality information is an objective variable (Y) and manufacturing information is an explanatory variable (X). Specifically, first, in order to avoid an incomputable problem and a lack of accuracy due to a multiple collinearity (Multiple Co-linear) phenomenon caused by simultaneous fluctuation of a plurality of elements of the explanatory variable, the elements of the explanatory variable are changed. Divide into a fixed number. Then, multiple regression analysis (Yi = A · Xi) is applied to all the divided groups, and the elements of the explanatory variables are narrowed down within each divided group by the variable increase / decrease method. It is a technique of extracting abnormal elements by repeating multiple regression analysis in combination with the narrowed explanatory variables in multiple stages.

Further, Non-Patent Document 1 discloses a projection method (Projection Method) for avoiding an incomputable problem and insufficient accuracy due to a multiple collinearity phenomenon caused by a plurality of elements of explanatory variables changing simultaneously. A plurality of calculation methods of the correlation model (Y = A · X) of the objective variable (Y) and the explanatory variable (X) based on Method) are described.
JP 2002-110493 A JP 2000-252180 A “Chemometrics, Data Analysis for the Laboratory and Chemical Plant”, WILEY (2003)

  By the way, in recent years, industrial products collectively referred to as a product category of “digital home appliances” are set products in which key devices such as a system LSI (SoC: System on Chip), a display panel, and a hard disk drive (HDD) are assembled. Each key device that determines the performance of a product is produced through a multi-step manufacturing process as a semiconductor device substrate (wafer) in which fine elements are integrated and processed on a substrate. The manufacture of key devices requires multi-step complicated and precise processes, and the number of processes exceeds several hundreds if a system LSI manufactured as a wafer is taken as an example. If the processing apparatus that handles one of the processes fluctuates outside a predetermined range, an abnormality occurs in the final quality of the product. For this reason, it is important for quality control of the production line to constantly monitor whether the state of each processing apparatus does not deviate from a predetermined range.

  As a result of the pursuit of processing miniaturization, the allowable range of the state variation of the processing process apparatus is becoming narrower year by year. A wafer becomes defective when the apparatus state deviates from the allowable fluctuation range due to an input error of an operation procedure of the processing process apparatus, an input error of processing conditions, or deterioration of parts of the processing apparatus. Since product inspection of wafers immediately after processing, which is performed for product abnormality monitoring during manufacturing, is generally performed by sampling inspection, there is a possibility that processing abnormality may be missed even if the apparatus state deviates from the allowable fluctuation range. In addition, there are processes that are difficult to inspect immediately after the completion of the process, such as an ion implantation process. Eventually, defects are found in the final product inspection in which 100% inspection is performed, but it takes several weeks to several months from each manufacturing process to final product inspection. And could have suffered significant damage.

  There are a plurality of processes in a production line for semiconductor products such as wafers, and a plurality of processing apparatuses are introduced in each process. Each processing apparatus outputs a plurality of state measurement values as state information. Preliminary monitoring is performed by setting appropriate judgment criteria for all (or selected important multiple) state measurement values in advance, and if an abnormality is detected, the equipment is immediately stopped and repaired. However, it is necessary to minimize the build-up of defective products.

  However, it is rare that the normal range of state measurement values of machining process equipment is deduced by physical theory, and if an appropriate criterion is not set, an unknown abnormality with no track record is missed. Or, there was a problem that a normal machining state was falsely determined as abnormal. Wafer products, which are key components, tend to be designed exclusively for each set product to be incorporated, and the number of products tends to increase accordingly. Different varieties increase the possibility of an unknown abnormality that has not been achieved. In addition, the possibility of an unknown abnormality increases when a new variety that is completely different from the conventional one is processed due to the change in generation of design basic rules and processes.

  The normal range of the state measurement value of the high-precision machining process apparatus is often obtained from statistical experience by repeating manufacturing and product inspection. In addition, the introduction of a new device in some processes and the disassembly and maintenance of the device may affect the entire production line, and the normal range of the conventional device state may fluctuate. Therefore, there has been a demand for a manufacturing system equipped with a device abnormality monitoring system that can automatically update registration criteria and determination criteria for device status monitoring once set from the latest manufacturing and product inspection results.

  In the techniques described in Patent Documents 1 and 2, product quality history information is obtained by multiple regression analysis (Y = A · X) with product quality history information as an objective variable (Y) and manufacturing history information as an explanatory variable (X). This is a technique for extracting manufacturing history information that is strongly connected to information. Since the semiconductor production line is mainly composed of continuous serial chain processes and the number of variables in the process is large, a multiple collinearity phenomenon (Multiple Co-linear) with simultaneously changing variables occurs. It is generally difficult to solve the multiple regression equation directly because of the stability of numerical calculation. Therefore, the elements of the manufacturing history information are divided into a fixed number, and multiple regression analysis (Yi = A · Xi) is applied to all the divided groups to narrow down the elements of the manufacturing history information within each divided group by the variable increase / decrease method, A multi-stage multivariate analysis method is adopted in which multiple regression analysis is applied again in multiple stages by combining the refined manufacturing history information. Therefore, although it is one means that always enables the execution of the calculation of the multiple regression model formula, only the statistical correlation strength (Y = A · X) between the product quality history information (Y) and the manufacturing history information (X). This is an analysis apparatus for enumerating and displaying cause candidates of product quality fluctuations obtained by paying attention to.

  Non-Patent Document 1 describes another means that always enables execution of multiple regression model equations even when there is multicollinearity by using a projection method. Although these methods always enable the execution of multiple regression model formulas, they do not necessarily guarantee that the obtained multiple regression model formulas reflect the actual causal relationship. There is a need to. In the calculation of a multiple regression model formula that handles many variables, it is important to narrow down the variables in advance in order to obtain a highly accurate multiple regression model formula that reflects the causal relationship.

  That is, in the conventional technology, in the production line of the microelement integrated semiconductor product that requires a large number of manufacturing processes and equipment, the state change of the manufacturing equipment that affects the final product quality before the corresponding product reaches the inspection process. If it is possible to detect in advance and detect it accurately, it is possible to take device measures to avoid mass production of defective products. However, the normal range of the state of the manufacturing equipment is not uniquely determined from the design specifications of the product, but it is necessary to determine it statistically or experimentally from the relationship with the product quality, because it cannot respond immediately to changes in the production line situation There was a problem of frequent false alarms and oversights.

  Therefore, in a production line that wants to suppress the quality fluctuations of microelement integrated semiconductor products, it is not only a list of products that cause product quality fluctuations and candidates for state measurement values, but a quantitative causal relationship with product quality fluctuations. There has been a demand for a manufacturing system equipped with a device abnormality monitoring system capable of canceling registration of device abnormality monitoring judgment items and automatically updating judgment criteria.

  Therefore, the present invention solves the above-mentioned problems, identifies the processing device that caused the product quality fluctuation and the state measurement value from the past results collection of manufacturing and product inspection in response to the report event of product abnormality occurrence. Or, identify the processing equipment that caused the product quality fluctuation and the state measurement value from the tracking tracking collection of manufacturing and product inspection in response to the notification event of equipment abnormality occurrence based on the current judgment criteria, and its quantitative causal relationship Provides a manufacturing system equipped with a device abnormality monitoring system that can determine the most appropriate device abnormality judgment criteria for the current production line status from the formula, deregister the device abnormality monitoring judgment items, and automatically update the judgment criteria The purpose is to do.

  In addition, when different types of elements are integrated, the present invention uses the difference in the types of elements to narrow down the number of variables handled in the calculation of the multiple regression model formula in advance, thereby changing product quality variation and processing. It is an object of the present invention to provide a manufacturing system that automatically obtains a quantitative causal relation with high accuracy between apparatus states.

  Of the inventions disclosed in the present application, the outline of typical ones will be briefly described as follows.

  In order to achieve the above object, the present invention is applied to a semiconductor product manufacturing system in which fine elements are integrated and processed on a material substrate using at least one processing apparatus and at least one product inspection apparatus. It has the following characteristics.

  The processing apparatus is connected to an apparatus abnormality monitoring system, and the apparatus abnormality monitoring system monitors each state measurement value of the processing apparatus based on a determination criterion that is statistically determined from the correlation with the product quality. The inspection apparatus is connected to a product abnormality monitoring system, and this product abnormality monitoring system monitors each product quality measurement value of the inspection apparatus based on a determination criterion determined from the design specifications.

  The apparatus abnormality monitoring system and the product abnormality monitoring system are connected to a line control system that centrally controls the entire line. This line control system inputs, tracks, and manages each product with a unique number name, collects and stores machining device state measurement values from the device abnormality monitoring system, and collects and stores product quality measurement values from the product abnormality monitoring system A wafer trace system. In addition, the line control system uses a product quality measurement data and device state measurement data collation analysis to identify the cause of the product quality variation caused by the device, and the statistical criteria of the device abnormality monitoring system. A line quality control system that performs derivation and updating is provided.

  In addition, in the case of a product integrating a plurality of types of elements such as a wafer of a CMOS semiconductor product, the line control system has a configuration including a fluctuation component extraction system that separates and extracts a common fluctuation component and an independent fluctuation component between a plurality of types of elements. Then, there is an effect that the analysis accuracy can be improved by narrowing down the target of collation analysis in advance.

  The line quality management system of the line control system identifies the cause of the quality variation and updates the determination criteria by collating and analyzing the product quality measurement data and the apparatus state measurement data in accordance with two types of abnormality detection events. One is when the apparatus abnormality monitoring system detects an abnormality in the apparatus state, and the other is when the product abnormality monitoring system detects an abnormality in product quality.

  When the apparatus abnormality monitoring system detects an abnormality in the apparatus state, the individual number name of the target processed product is traced and collated with the product inspection result. Product inspection can be performed either immediately after processing or at the final product quality inspection. Product inspection immediately after processing is often not 100% inspection, and it is not always possible to inspect immediately after any processing step. The final product quality inspection is 100% inspection. If the verified product inspection result is normal from the product criteria based on the design specifications, the product abnormality monitoring system has issued a false alarm, so the status item that is the target of device abnormality detection can be canceled from the monitoring target, Recalculate the device criteria and increase the value. If the verified product inspection result is abnormal from the product judgment criteria based on the design specifications, the correct answer is issued. Therefore, the status item that is the target of device abnormality detection remains unchanged and the device judgment criteria is recalculated. And update the values as needed.

  When the product abnormality monitoring system detects a product quality abnormality, the individual number name of the target processed product is reverse-traced and collated with the apparatus state measurement result upstream of the manufacturing process. The fact that the abnormality detection of the product abnormality monitoring system is the starting point means that any of the apparatus abnormality monitoring systems upstream of the manufacturing process has missed. Deriving the device that caused the product inspection abnormality and its state measurement items and criteria from the verified device status measurement results, registering the status items that should be subject to device abnormality detection as monitoring targets, and deriving device criteria Set.

  By taking the above configuration, it is possible to realize a production line including an abnormality monitoring system that automatically and automatically reviews the balance between correct and false alarm reports and missed detections of device abnormality detection.

  Among the inventions disclosed in the present application, effects obtained by typical ones will be briefly described as follows.

  According to the present invention, in a manufacturing line for semiconductor products in which fine elements are integrated and processed, it is possible to measure the state of the processing apparatus in many cases where the judgment criteria for monitoring the abnormality of the processing apparatus cannot be directly derived from the product design specifications or physical laws. Machine error monitoring that preempts product quality inspection immediately after machining by statistically deriving and setting the status parameters and judgment criteria of processing equipment that has an impact on product quality from the history of actual measurement and actual measurement of product quality By performing the above, maintenance of the processing apparatus can be performed at an early stage to prevent the formation of defective products.

  When the abnormality monitoring system of the processing device detects an abnormality for any status measurement value that has been registered for monitoring, the individual identification number name of the corresponding product is registered and traced in the trace system, and the product inspection is completed Check whether the status parameter has an effect on the product quality from the history of the actual status measurement value of the processing equipment and the inspection measurement actual value of the product quality. If there is no effect, the monitoring registration can be canceled to prevent false information and avoid unnecessary equipment maintenance.

  Also, when the abnormality monitoring system of the product inspection device detects an abnormality for any inspection measurement value, the individual identification number name of the corresponding product is registered in the trace system and backtracked, and the status measurement actual value of the processing device Check whether or not this machine status parameter has an effect on product quality from the history of inspection and measurement results of product quality. If there is an effect, set machine fault monitoring registration for this machine status parameter and determine its criteria. Is then statistically derived and set, and thereafter, the apparatus abnormality monitoring is performed in advance of the product quality inspection immediately after the processing, so that the processing apparatus can be maintained at an early stage to prevent the formation of defective products.

  Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings. Note that components having the same function are denoted by the same reference symbols throughout the drawings for describing the embodiment, and the repetitive description thereof will be omitted.

  Hereinafter, a semiconductor product manufacturing system of the present invention will be described with reference to a plurality of embodiments.

(Embodiment 1)
A basic embodiment of a semiconductor product manufacturing system according to the present invention will be described below with reference to FIGS. 1, 2, 3, and 4. FIG. FIG. 1 is a diagram showing a manufacturing system in which a wafer according to an embodiment of the present invention is a target product. FIG. 2 is a diagram showing in detail the apparatus abnormality monitoring system. FIG. 3 is a diagram showing the product abnormality monitoring system in detail. FIG. 4 is a diagram showing a computer system that realizes the manufacturing system of FIG. These embodiments are not limited to wafers as target products, and can be applied to general semiconductor products in which fine elements are integrated on a material substrate.

  First, the flow of wafer processing will be described with reference to FIG. The material wafer 1a is sequentially processed by a plurality of processing units constituting the production line, and finally the product quality is inspected by the product inspection apparatus 41 of the final product inspection unit 4 to be completed as a final product wafer 1b. There are two types of processing units: a processing unit (type 1) 2 and a processing unit (type 2) 3. The processing unit (type 1) 2 includes a processing device 21 and a product inspection device 51 immediately after processing, whereas the processing unit (type 2) 3 includes only the processing device 31 and does not include a product inspection device immediately after processing.

  In wafer manufacturing equipment, photolithographic processes and etching processes that can inspect product processing pattern dimensions with a scattering electron microscope (SEM) inspection apparatus, or film forming processes and polishing processes that can inspect product film thickness with a film thickness inspection apparatus are processed. It belongs to unit (type 1) 2. On the other hand, an ion implantation process and an annealing process that change the electrical characteristics and film quality characteristics of the wafer belong to the processing unit (type 2) 3 because it is difficult to inspect immediately after processing. A product abnormality monitoring system 42 or 52 is connected to the product inspection apparatus 41 or 51 to monitor abnormality of the product quality inspection measurement value. In addition, an apparatus abnormality monitoring system 22 or 32 is connected to the machining apparatus 21 or 31 to monitor an abnormality of the machining apparatus state measurement value.

  Next, the operation of the apparatus abnormality monitoring system 32 will be described using FIG. 1 and FIG. The operation of the device abnormality monitoring system 22 is the same. A state measuring device 31 s is incorporated in the processing apparatus 31 to measure the state of the processing apparatus 31 during wafer processing. The number of measurement parameters depends on the processing apparatus, and usually there are tens to hundreds of parameters. In FIG. 2, three machining state measurement values 31s1, 31s2, and 31s3 are shown as representatives, but the present invention is not limited to this. These processing state measurement values are sent from the processing apparatus 31 to the wafer trace system 61 of the line control system 60, and stored in the processing apparatus processing state measurement value database 64a together with the wafer number 1n.

  The machining state measurement values 31s1, 31s2, and 31s3 are sent to the determination unit 34c via the parameter selection unit 33s of the apparatus abnormality monitoring system 32. In FIG. 2, since the parameter selection switch 33s2 of the parameter selection unit 33s is open (OFF) and the parameter selection switches 33s1 and 33s3 are closed (ON), only the machining state measured values 31s1 and 31s3 are sent to the determination unit 34c. The parameter selection unit 33m sets and changes the open / close states of the parameter selection switches 33s1, 33s2, and 33s3 based on an instruction from the line quality management system 62 of the line control system 60.

  The determination unit 34 c includes determination criteria 34 h 1, 34 h 2, 34 h 3 and processing state measurement value comparators 34 c 1, 34 c 2, 34 c 3, performs abnormality determination, and sends the determination result to the wafer trace system 61. In FIG. 2, only the selected machining state measurement values 31s1 and 31s3 are determined. Determination criteria 34h1, 34h2, and 34h3 are set and changed by the determination criterion management unit 34m based on an instruction from the line quality management system 62 of the line control system 60.

  Next, the operation of the product abnormality monitoring system 42 will be described with reference to FIGS. 1 and 3. The operation of the product abnormality monitoring system 52 is the same. In FIG. 3, two product inspection measurement values 41s1 and 41s2 are shown as representatives, but the present invention is not limited to this. These product inspection measurement values are sent from the product inspection device 41 to the wafer trace system 61 of the line control system 60, and are stored in the final product wafer inspection measurement value database 64c or the wafer inspection measurement value database 64b during manufacturing together with the wafer number 1n. .

  All of the product inspection measurement values 41 s 1 and 41 s 2 are sent to the determination unit 44 c of the product abnormality monitoring system 42.

  The determination unit 44 c includes determination criteria 44 h 1 and 44 h 2 and product inspection measurement value comparators 44 c 1 and 44 c 2, performs abnormality determination, and sends a determination result to the wafer trace system 61. The determination criteria 44h1 and 44h2 of the product abnormality monitoring system 42 are values determined from the design specifications of the product, and are values obtained and set by the determination criterion management unit 44m from the product design system 45 when a new product is manufactured. The line control system 60 is not subject to change at any time. This is a point that is greatly different from the apparatus abnormality monitoring system 32 described above. Since many of the determination criteria of the apparatus abnormality monitoring system 32 cannot be uniquely determined from design specifications or the like, the correction of the determination criteria by the line quality management system of the present invention is effective as needed.

  Next, a computer system for realizing the manufacturing system according to the embodiment of the present invention will be described with reference to FIG. Each processing device 21, 31 and each product inspection device 41, 51 are connected to a computer network 70n, and communicate with a device abnormality monitoring computer 70c1, a product abnormality monitoring computer 70c2, and a line control computer 70c3 that are also connected.

  The apparatus abnormality monitoring systems 22 and 32 of FIG. 1 are realized by the apparatus abnormality monitoring program of the apparatus abnormality monitoring computer 70c1, and operate as each job that starts this program. 1 is realized by a product abnormality monitoring program of the product abnormality monitoring computer 70c2, and operates as each job that starts this program. The wafer trace system 61 and the line quality management system 62 in FIG. 1 are realized by a program of the line control computer 70c3, and operate as each job that starts this program.

  The machining state measurement values 31s1, 31s1, and 31s3 in FIG. 2 are stored in the machining apparatus machining state measurement value database 70d1 managed by the line control computer 70c3 via the computer network 70n. The product inspection measurement values 41s1 and 41s1 in FIG. 3 are stored in the in-manufacturing wafer inspection measurement value database 70d2 or the final product wafer inspection measurement value database 70d3 managed by the line control computer 70c3 via the computer network 70n.

  Next, with reference to FIGS. 5, 6, and 7, a criterion optimization algorithm based on criterion recalculation processing triggered by detection of abnormality in both the apparatus and the product, which is a feature of the present invention, will be described. 5 and 6 are diagrams showing an algorithm for optimizing the determination criterion for the apparatus abnormality detection trigger. FIG. 7 is a diagram illustrating an algorithm for optimizing the determination criterion for the product abnormality detection trigger.

  An algorithm for optimizing the determination criterion for the apparatus abnormality detection trigger (FIGS. 5 and 6) is executed by the computer according to the following procedure.

  In response to a wafer product processing completion event from an arbitrary processing apparatus (21, 31 in FIG. 1) (100 in FIG. 5), an apparatus abnormality monitoring computer (70c1 in FIG. 4) targets the processing apparatus. Monitoring is started (101). As step 1 (102), all the state measurement values (31s1, 31s2, 31s3 in FIG. 2) of this processing apparatus are acquired via the computer network (70n in FIG. 4) and transferred to the line control computer (70c3 in FIG. 4). Store in the sending database (70d1 in FIG. 4). As Step 2 (103), the processing device state measurement values (31s1 and 31s3 in FIG. 2) to be determined are sequentially selected based on the presetting of the parameter selection unit (33s in FIG. 2). As step 3 (104), this selected processing device state measurement value (31s1, 31s3 in FIG. 2) is used as a criterion (34h1, 34h3 in FIG. 2) and a comparator (34h in FIG. 2). 34c1 and 34c2) are collated to determine abnormality.

  The abnormality determination result is sent from the apparatus abnormality monitoring computer (70c1 in FIG. 4) to the line control computer (70c3 in FIG. 4), and a determination criterion optimization process is performed. The following steps branch depending on whether an abnormality is detected (105).

  If no abnormality is detected, the apparatus abnormality monitoring is terminated as it is (108).

  When an abnormality is detected, as shown in step 4 (106), a display instruction for a maintenance instruction of this processing apparatus is arbitrarily performed. In step 5 (107), the trace registration of the product wafer number (1n) for product inspection completion tracking is performed. At this point in time, the state measurement value of the machining apparatus in which an abnormality has been detected and its history necessary for the optimization of the judgment criteria can be prepared, but the product inspection measurement value cannot be prepared, and thus the algorithm processing is temporarily terminated (108).

  Next, upon receiving the inspection completion event of the wafer product for which the above-mentioned trace registration has been performed (200 in FIG. 6), the line control computer (70c3 in FIG. 4) resumes the determination algorithm optimization algorithm processing. The following algorithm processing is divided into two parts: a quality variation cause identifying unit and a determination criterion calculation and registration change unit.

  First, the line control computer starts the quality variation cause identification unit (201), and in step 1 (202), the product inspection measurement value of the corresponding wafer product and the past product inspection history data regarding the product of the same specification as the corresponding wafer product. Is obtained from the database (70d2 or 70d3 in FIG. 4). The product inspection monitoring value (70c2 in FIG. 4) of the corresponding wafer product has been acquired (described later). At the same time, the processing apparatus state measurement value of the corresponding wafer product and the past processing apparatus state history data regarding the product of the same specification as the corresponding wafer product are acquired from the database (70d1 in FIG. 4).

As an example, the transition of product inspection measurement value data including the history is shown in the chart of FIG. It is a product number w ₁₁ is a corresponding wafer product inspection measurement value of the product, w ₁₀ from the product number w ₁ is in the past of product inspection history data about the products of the same specification. FIG. 8A shows the case where the product inspection measurement value fluctuates abnormally, and FIG. 8B shows the case where the product inspection measurement value is within the normal fluctuation. The determination criteria 221 and 222 for the product inspection measurement values are fixed values determined in advance from the product design specifications. Product inspection measurement values and product inspection history data

Is the average of historical data

Subtract the standard deviation of the historical data

Divide by

Has been normalized.

As an example, the processing device state measurement value data including the history is shown in the chart of FIG. A product number w ₁₁ machining apparatus state measurement value of the corresponding wafer products, w ₁₀ from the product numbers w ₁ is in the past processing device state history data on products of the same specifications. FIGS. 9A and 9B show the case where the machining apparatus state measurement value fluctuates abnormally, and FIG. 9C shows the case where the machining apparatus state measurement value is within the normal fluctuation. The processing apparatus state measurement value determination criteria 231, 232, and 233 are statistically 3σ values obtained in advance from history data as initial values, but can be changed. Processing device state measurement value and processing device state history data

Is the average of historical data

Subtract the standard deviation of the historical data

Divide by

Has been normalized.

  The line control computer performs product inspection history data as step 2 (203).

And processing device status history data

Multiple regression equation between

Create Here, the first term a0 becomes zero when the average value is subtracted in advance.

Considering the data shown in FIG. 8 and FIG. 9, the past history (wafers w _{1 to} w ₁₀ ) is stable with no fluctuations more than the steady fluctuations, and the product inspection parameter y ₁ and the state of the processing apparatus in the corresponding wafer w ₁₁ Parameters x ₁ and x ₂ indicate abnormal fluctuations. The product inspection parameter y ₁ and the processing apparatus state parameter x ₃ of the wafer w ₁₁ are stable with no fluctuations beyond the steady fluctuations. Therefore, the line control computer creates a multiple regression equation for the product inspection parameter y ₁ . However, since the two processing device state parameters x ₁ and x ₂ are fluctuating at the same time, the line control computer distributes the fluctuation ratio to the regression coefficient,

Create a multiple regression equation as follows. Here, A ₁ and A ₂ are unit scale values of the _two processing apparatus state parameters x ₁ and x ₂ , and B ₁ is a unit scale value of the product inspection parameter y ₁ . Since the processing apparatus state measurement value measures a plurality of related parameters at the same time, a phenomenon in which the plurality of parameters fluctuate simultaneously in synchronization generally occurs.

In the present embodiment, since a phenomenon that fluctuates at the same time occurs only in the product w ₁₁ , a multiple regression equation in which the fluctuation ratio is distributed to the regression coefficients can be easily created. In general, fluctuations that do not synchronize with each other also occur in the past history data portion of the measurement value, and therefore, a reliable single regression distribution cannot create a reliable multiple regression equation. In such a case, the line control computer uses a solution called a latent structure projection method (PLS) or a partial least square method (PLS) shown in Non-Patent Document 1. Thus, a multiple regression equation is created in which the variation ratio is statistically distributed to the regression coefficient so as to increase the covariance between the product inspection measurement value and the apparatus state measurement value.

In step 3 (204), the line control computer selects a statistically significant processing apparatus state parameter having a high contribution degree from the regression coefficient of (Equation 12) as a quality variation cause parameter. From the regression coefficient ratio of (Equation 12), the contribution of x ₁ , x ₂ , x ₃ is

Therefore, the line control computer selects x ₁ and x ₂ with high contribution as quality variation cause parameters, and ends the quality variation cause identification unit (205). It is possible to select the quality variation cause parameter without deficiency by distributing the variation ratio to the regression coefficient as described above.

The line control computer, followed by starting the calculation and registration change unit criterion (206), in step 1 (207), for x ₁ and x ₂ selected as the quality variation cause parameter, product inspection parameter independently Create a regression equation with y ₁ . The independent regression equation is self-evident from the charts of FIGS. 8 and 9 in the present embodiment, but in general, the line control computer converts the contribution ratio of (Equation 13) to the variation ratio distribution regression equation of (Equation 12). By substituting

I ask.

In step 2 (208), the line control computer derives a processing apparatus state determination criterion from a product inspection determination criterion determined from the design specification based on the manufacturing history. Specifically, when the product inspection criteria is Yth ₁ ,

I ask.

  The line control computer (70c3 in FIG. 4) determines that the processing apparatus state determination criteria of the above (Equation 16) and (Equation 17) are abnormal in step 3 (209), step 4 (210), and step 5 (211). It is set in the parameter selection unit 33s and the determination unit 34c via the parameter selection unit 33m and the determination criterion management unit 34m of the device abnormality monitoring system (32 in FIG. 2) realized by the monitoring computer (70c1 in FIG. 4). The calculation of the reference and the registration update are finished (212). The abnormality monitoring determination reference updating unit 64 updates the determination reference.

  For the processing device state measurement value 31s1, the determination criterion 34h1 is rewritten to the value calculated by the above (Expression 16) while the selection state of the selection switch 33s1 is left as it is. Regarding the processing device state measurement value 31s2, since the selection switch 33s2 is in the release state, the selection state is changed, and the determination reference 34h2 is rewritten to the value calculated in the above (Equation 17). Since the processing apparatus state measurement value 31s3 does not exceed the current determination criterion, the selection switch 33s3 remains selected and the determination criterion 34h2 is not rewritten. If the product inspection measurement value does not fluctuate even though the processing device state measurement value 31s3 exceeds the criterion, the line control computer cancels the selection switch 33s3.

  Next, in the judgment criterion optimization algorithm based on the judgment criterion recalculation process triggered by the abnormality detection of both the apparatus and the product, which is a feature of the present invention, using FIG. 7, the judgment criterion of the product abnormality detection trigger is appropriate. The algorithm is explained.

  The algorithm for optimizing the judgment criterion for the product abnormality detection trigger (FIG. 7) is executed by the computer according to the following procedure.

  Upon receiving a wafer product inspection completion event from an arbitrary product inspection apparatus (41, 51 in FIG. 1) (300 in FIG. 7), a product abnormality monitoring computer (70c2 in FIG. 4) targets the wafer product. Abnormality monitoring is started (301). As step 1 (302), all the inspection measurement values (41s1, 41s2 in FIG. 3) of this product inspection apparatus are acquired via the computer network (70n in FIG. 4) and sent to the line control computer (70c3 in FIG. 4). Store in the database (70d1 in FIG. 4). In step 2 (303), the product inspection measurement value is collated by the judgment standard (44h1, 44h2 in FIG. 3) of the judgment unit (44c in FIG. 3) and the comparator (44c1, 44c2 in FIG. 3) to determine the product abnormality. I do.

  The product abnormality determination result is sent from the product abnormality monitoring computer (70c2 in FIG. 4) to the line control computer (70c3 in FIG. 4), and product abnormality determination standard optimization processing is performed. The following steps branch depending on whether or not product abnormality is detected (304).

  If the product abnormality is not detected, the product abnormality monitoring is terminated as it is (312).

  When the product abnormality is detected, in step 3 (305), the product inspection measurement value of the corresponding wafer product and the past product inspection history data regarding the product of the same specification as the corresponding wafer product are stored from the database (70d2 or 70d3 in FIG. 4). get. At the same time, the processing apparatus state measurement value of the corresponding wafer product and the past processing apparatus state history data regarding the product of the same specification as the corresponding wafer product are acquired from the database (70d1 in FIG. 4).

  The line control computer creates a multiple regression equation (Equation 11) between the product inspection history data (Equation 9) and the processing device state history data (Equation 10) as in Step 4 (306). Here, the first term a0 becomes zero when the average value is subtracted in advance.

The product inspection history data is the same as in FIG. In other words, it is assumed that the product inspection parameter y ₁ shows an abnormal fluctuation in the corresponding wafer w ₁₁ and the abnormality is detected exceeding the product judgment standard determined from the product specification. Therefore, the line control computer creates a multiple regression equation for the product inspection parameter y ₁ . On the other hand, the processing apparatus state history data is as shown in the chart of FIG. That is, as in FIG. 9, the processing apparatus state parameters x ₁ and x ₂ show fluctuations in the corresponding wafer w ₁₁ . However, it is assumed that the fluctuation is within the apparatus state determination criteria 241 and 242.

Again, since the two processing device state parameters x ₁ and x ₂ are fluctuating at the same time, the line control computer distributes the fluctuation ratio to the regression coefficient so that the same as described above (Formula 12). Create a regression equation. Here, A ₁ and A ₂ are unit scale values of the _two processing apparatus state parameters x ₁ and x ₂ , and B ₁ is a unit scale value of the product inspection parameter y ₁ . The multiple regression equation (Equation 12), product inspection parameters for the correlation of y ₁ and the processing apparatus state parameters x _1, x ₂ is formulated, products aberrant capable preceding detected by processing stage machining apparatus state parameters x _1, it is possible to direct the proper criteria for x _2.

  As described above, the plurality of processing apparatus state parameters may vary simultaneously with each other, and may include variations that are not synchronized with each other. In such a case, the line control computer uses a solution called a latent structure projection method (PLS) or a partial least square method (PLS) shown in Non-Patent Document 1. Thus, a multiple regression equation is created in which the variation ratio is statistically distributed to the regression coefficient so as to increase the covariance between the product inspection measurement value and the apparatus state measurement value.

In step 5 (307), the line control computer selects a statistically significant processing apparatus state parameter having a high contribution degree from the regression coefficient of (Equation 12) as a quality variation cause parameter. Since the contribution of x ₁ , x ₂ , x ₃ is (Equation 13) as described above from the ratio of the regression coefficients of (Equation 12), the line control computer determines x ₁ and x ₂ having high contributions. The quality variation cause parameter is selected, and the quality variation cause identifying unit is terminated. It is possible to select the quality variation cause parameter without deficiency by distributing the variation ratio to the regression coefficient as described above.

In step 6 (308), the line control computer creates a regression equation with the product inspection parameter y ₁ independently for x ₁ and x ₂ selected as the quality variation cause parameters. The independent regression equation is self-evident from the charts of FIGS. 8 and 10 in the present embodiment, but generally, the line control computer converts the contribution ratio of (Equation 13) to the variation ratio distribution regression equation of (Equation 12). By substituting into, respectively, (Equation 14) and (Equation 15) are obtained in the same manner as described above.

In step 7 (309), the line control computer derives a processing apparatus state determination criterion from a product inspection determination criterion determined from the design specification based on the manufacturing history. Specifically, when the product inspection determination criterion is Yth ₁ , (Equation 16) and (Equation 17) are obtained in the same manner as described above.

  In step 8 (310) and step 9 (311), the line control computer (70c3 in FIG. 4) uses the above-described processing apparatus state determination criteria in (Expression 16) and (Expression 17) as the apparatus abnormality monitoring computer (in FIG. 4). 70c1) is set in the parameter selection unit 33s and the determination unit 34c via the parameter selection unit 33m and the determination criterion management unit 34m of the apparatus abnormality monitoring system (32 in FIG. 2), and the product abnormality monitoring is terminated.

  For the processing device state measurement value 31s1, the determination criterion 34h1 is rewritten to the appropriate value calculated by the above (Equation 16) while the selection state of the selection switch 33s1 is left as it is. For the processing device state measurement value 31s2, since the selection switch 33s2 is in the released state, the selected state is set, and the value calculated by the above (Equation 17) for the determination criterion 34h2 is set. Since the processing apparatus state measurement value 31s3 does not exceed the current determination criterion, the selection switch 33s3 remains selected and the determination criterion 34h2 is not rewritten.

(Embodiment 2)
Hereinafter, another embodiment of the apparatus abnormality monitoring system for a semiconductor product manufacturing system according to the present invention will be described with reference to FIG.

  In the determination unit 34c of the apparatus abnormality monitoring system 32 shown in FIG. 2 of the first embodiment, the processing state measurement values 31s1, 31s2, and 31s3 of the respective processing apparatuses are individually determined based on the determination criteria 34h1, 34h2, and 34h3 and the comparator 34c1, respectively. Comparison was made using 34c2 and 34c3.

  In contrast, in the present embodiment, the determination unit 35c of the apparatus abnormality monitoring system 32 shown in FIG. 11 adds the machining state measurement values 31s1, 31s2, and 31s3 by multiplying the coefficients by the product-sum calculator 35a. Then, the integrated value is compared using one determination criterion 35h1 and the comparator 35c1. The coefficient required by the product-sum calculator 35a is set to the product inspection estimation coefficient 35e from the line quality management system 62 via the estimation coefficient management unit 35m. This product-sum operation integrated value is the product inspection estimated value of the above (Equation 12), and the determination criterion is set as the product determination criterion determined from the product specification in the same manner as the product abnormality monitoring.

  According to this configuration, since a plurality of machining apparatus state parameters that fluctuate simultaneously are integrated and determined, even if noise is applied to one of the machining apparatus state parameters, there is an effect that it is difficult to be affected. Further, since the product inspection determination standard determined from the product specification is used as it is as the apparatus determination standard, there is an effect that it is not necessary to calculate the determination standard. However, if the error term is large, the device determination criterion is created by subtracting the error term from the product inspection determination criterion.

(Embodiment 3)
Hereinafter, another embodiment of a semiconductor product manufacturing system according to the present invention will be described with reference to FIGS. 12, 13, 14 and 15. FIG.

  FIG. 12 is a diagram showing a manufacturing system that uses a wafer according to an embodiment of the present invention as a target product, and extracts fluctuation components among a plurality of types of elements from the manufacturing system described in FIG. 1 of the first embodiment. A system 65 is added. FIG. 13 is a diagram showing the fluctuation component extraction system 65 between the plural types of elements and the line quality management system 62 corresponding thereto. FIG. 14 is a diagram showing a cross-sectional structure (cross-sectional not shown) of PMOS and NMOS which are two field effect transistor elements constituting a complementary metal oxide semiconductor (CMOS) as an example of a plurality of kinds of elements.

  First, it will be described with reference to FIG. 14 that different types of elements are manufactured by a combination of a common manufacturing process and an independent manufacturing process. The PMOS (70p) and the NMOS (70n) are manufactured as CMOS constituent elements in such a way that they are mixed in adjacent regions on the wafer. Therefore, instead of manufacturing them in a completely separate manufacturing process, a common process is a common manufacturing process, and processes having different characteristics are processed in mutually independent manufacturing processes.

  The PMOS (70p) and the NMOS (70n) have the same cross-sectional dimensions but different electrical characteristics. The charge transport of the channel between the source (72p) and the drain (73p) of the PMOS (70p) is carried by holes, and the charge transport of the channel between the source (72n) and the drain (73n) of the NMOS (70n) is carried by electrons. . A voltage is applied to the gates (75p, 75n) by the gate electrodes (76p, 76n), an electric field is formed on the lower surface of the oxide insulating thin film (74p, 74n), and holes or electrons are collected by the electric field effect to generate a charge path. A certain channel is formed. Therefore, in a state where no voltage is applied to the gate, the channel portion needs to be in a state of insufficient charge.

  When using a P-type substrate, that is, a silicon wafer in which holes are excessive and electrons are insufficient by introducing impurities, the channel portion of the NMOS (70n) is in an electron insufficient state. Since the channel portion of the PMOS (70p) has an excessive number of holes as it is, an N well (71p) in a state of insufficient holes and excessive electrons is formed by implanting impurity ions into the silicon wafer. That is, the ion implantation process for forming the N well (71p) is a process unique to the PMOS.

  Further, the source (72p) and the drain (73p) of the PMOS (70p) are implanted with impurity ions that lack electrons when combined with silicon in order to be in a state of excessive holes, which are responsible for channel charge. On the other hand, the source (72n) and the drain (73n) of the NMOS (70n) implant excessive impurity ions when they are combined with silicon in order to bring about an excessive electron state, which is responsible for channel charge. That is, since different ions need to be implanted to form the source and the drain, different devices or devices with different operating conditions are used for PMOS and NMOS.

  On the other hand, the element isolation insulating trench (77p1, 77p2, 77n1, 77n2) is formed by digging a trench by lithography and embedding an oxide insulating film. In order to form an insulating groove having the same cross-sectional dimension and the same film quality between the PMOS and the NMOS, the processing is performed in the same process. The insulating thin films (74p, 74n), the gates (75p, 75n), the gate electrodes (76p, 76n), the spacers (80p1, 80p2, 80n1, 80n2), the source electrodes (78p, 78n) and the drain electrodes ( 79p and 79n) are also processed in the same process because they are formed with the same film quality having the same cross-sectional dimension between the PMOS and the NMOS.

  Here, the influence of the state variation of the processing apparatus will be considered for the product, that is, the gate threshold voltage which is one of the quality characteristics of the PMOS and NMOS. This gate threshold voltage is an important quality characteristic that determines the ON / OFF transition of the transistor.

  Even when the same gate voltage is applied, the amount of charge induced in the channel between the source and drain differs depending on the impurity concentration distribution. Therefore, when the number of ion implantations and the implantation depth vary due to the state variation of the ion implantation apparatus, the threshold voltage also varies. As described above, the ion implantation apparatus to be used or the processing conditions of the ion implantation apparatus are different in the impurity introduction regions of the PMOS and NMOS, so the threshold voltages of the PMOS and NMOS do not fluctuate simultaneously.

  On the other hand, even when the same gate voltage is applied, the shape of the electric field varies depending on the film thickness dimension of the gate insulating film (74p, 74n) and the width dimension of the gate (75p, 75n). The amount of charge induced in the channel also varies. Therefore, if these dimensions change due to the state change of the lithography apparatus that determines the width dimension and the film forming apparatus that determines the film thickness dimension, the threshold voltage also changes. As described above, since the shape processing of the PMOS and NMOS is processed at the same time, the devices used or the processing conditions of the devices are the same for the PMOS and NMOS, and the threshold voltages of the PMOS and NMOS vary simultaneously. As an apparatus related to lithography, there are an exposure apparatus and an etching apparatus. As an apparatus related to film formation, there are a CVD apparatus, a sputtering apparatus, a plating apparatus, and a CMP polishing apparatus.

  Next, by using the fact that different types of elements are manufactured by a combination of a common manufacturing process and an independent manufacturing process using FIGS. A system for constructing a product quality prediction regression equation will be described.

  In the first embodiment, the system that updates the determination parameter and the determination criterion of the processing apparatus state from two abnormality detections of the abnormality detection of the processing apparatus state and the abnormality detection of the product inspection has been described. The system according to the present embodiment is related to the update of the apparatus determination parameter and the determination standard starting from the abnormality detection of the latter product inspection, and particularly related to the variation of the product inspection value among the candidates for the processing apparatus of a plurality of processes. It is a system suitable for narrowing down some processing equipment and building accurate product quality prediction regression equations. In the regression equation creation step of Step 4 (306) in the update algorithm of the apparatus abnormality determination parameter and the determination criterion starting from the abnormality detection of the product inspection in FIG. 7 described in the first embodiment, the apparatus is assigned for each variation component. Detailed steps for constructing a product quality prediction regression equation are shown in FIG.

  As a detailed step 4-10 (410), the line control computer (70c3 in FIG. 4) extracts PMOS and NMOS fluctuation components from the measurement value history database of the final inspection for examining the quality of the product wafer. The quality of the product wafer is an electrical characteristic such as a threshold voltage that determines the ON / OFF transition of the transistor. More specifically, the fluctuation component extraction system 65 between a plurality of types of elements realized by the line control computer 70c3 is based on the final product wafer inspection measurement value database (64c in FIG. 13) of the product wafer, and performs PMOS and CMOS processing for each wafer. The threshold voltage history is acquired, and the distribution components are separated into independent variation components and common variation components. FIG. 13 shows a threshold voltage distribution chart 66. The horizontal axis represents the PMOS threshold voltage, and the vertical axis represents the NMOS threshold voltage. When the cross-sectional dimensions of the transistors fabricated at the same time in a common manufacturing process change, both the PMOS threshold voltage and the NMOS threshold voltage change in common and are distributed in the vicinity of a straight line of 45 degrees diagonally upward. The projection of the data points for each wafer onto the 45 ° diagonal line is the common variation component 66v2.

  On the other hand, when the amount of impurity introduced into transistors separately manufactured in different manufacturing processes fluctuates, the threshold voltage of PMOS and the threshold voltage of NMOS fluctuate independently of each other, and are separated from a straight line of 45 degrees diagonally upward. Distributed. The projection of the data point for each wafer onto a straight line at 45 degrees obliquely below is the independent variation component 66v1.

  As the next detailed step (420), the line control computer 70c3 determines whether the independent component has changed or the common component has changed, and branches the subsequent processing.

  When the independent component fluctuates, the independent fluctuation component extraction unit 65p1 realized by the line control computer 70c3 extracts the independent fluctuation component 66v1, and the element independent process regression equation of the quality fluctuation cause identification unit 63 of the line quality management system 62 The information is transferred to the construction unit 63r1 and detailed step 4-21 (421) is executed.

  In detailed step 4-21 (421), a processing device to be used in an independent process of PMOS and NMOS is allocated from a previously created process table or the like (not shown), and only the state measurement value history data of the device is processed. It is acquired from the device machining state measured value database 64a to the element independent process regression equation construction unit 63r1 through the data selector 63s1. Typical processing apparatuses used in the independent process include an ion implantation apparatus and a diffusion annealing apparatus.

  Next, the element independent process regression equation construction unit 63r1 realized by the line control computer 70c3 creates a regression equation between the independent variation component 66v1 and the device state measurement value history data as detailed step 4-31 (431). .

  On the other hand, when the common component varies, the common variation component extraction unit 65p2 realized by the line control computer 70c3 extracts the common variation component 66v2, and returns the element common process regression of the quality variation cause identification unit 63 of the line quality management system. The result is passed to the expression construction unit 63r2, and detailed step 4-22 (422) is executed.

  In detailed step 4-22 (422), a processing device used in a common process of PMOS and NMOS is allocated from a previously created process table or the like (not shown), and only the state measurement value history data of the device is processed. It is acquired from the device machining state measured value database 64a to the element common process regression equation construction unit 63r2 through the data selector 63s1. Typical processing apparatuses used in the common process include an exposure apparatus and etching apparatus related to lithography, or a CVD apparatus and CMP apparatus related to film formation.

  Next, the element common process regression equation construction unit 63r2 realized by the line control computer 70c3 creates a regression equation between the common variation component 66v2 and the apparatus state measurement value history data as detailed step 4-32 (432). .

  Thereafter, the abnormality monitoring determination reference updating unit 64 updates the apparatus state monitoring determination reference, taking over from step 5 onward in FIG.

(Embodiment 4)
Hereinafter, another embodiment of a semiconductor product manufacturing system according to the present invention will be described with reference to FIG.

  FIG. 16 is a diagram illustrating a method for controlling a machining apparatus provided with an apparatus state control unit. In contrast to FIG. 2 of the first embodiment, the processing apparatus 31 includes a state controller 31c1 in addition to the state measuring device 31s.

  For example, in general, in a semiconductor manufacturing apparatus, a process condition file called a recipe is created in advance in a storage device of the apparatus, and when the process condition file is selected according to the product to be processed, the apparatus itself converts each state into a file. In general, a method of controlling the specified condition value is adopted. For example, a process condition such as a gas flow rate or a vacuum degree is automatically controlled by a valve. Also, in order to cope with the drift of the machining capability (machining rate) of equipment, a method that can replace part of the process conditions that act as process time or adjustment factors with values specified from an external system has become widespread. ing.

  However, the calculation model that defines the relationship between the product design specifications that are the object of product inspection and the adjustment factors in the process conditions of the processing equipment necessary for calculating the optimum process conditions is based on experiments and history surveys in advance. The obtained model is used in a fixed manner, and there is a problem that the state change and disturbance of the apparatus that causes the product quality fluctuation cannot be reflected at any time.

  Therefore, in the present invention, the device control system 36 combined with the line quality management system 62 is configured, and the line quality management system 62 is created to recalculate the determination criteria starting from product abnormality detection. A configuration for updating a calculation model of an adjustment factor in a process condition based on a multiple regression equation (the same as the above (Expression 11)) indicating the latest relationship between product inspection history data and processing device state history data; did.

  As an example, if there are three device states as shown in FIG. 16 and the product quality is one, the multiple regression equation is

It becomes. The number of device states and the number of product qualities are examples, and are not limited thereto. A controllable state variable in the state controller 31c1 and x _1, the uncontrollable state variables and x _2, x _3, fixes the set value of the controllable state variable x ₁ is the device status target value calculation model

Calculated from

The line control system 60 transmits the device state target value calculation model to the model management unit 36m of the device control system 36. Model management unit 36m sets the model to device state target value calculation model 36e of the target value calculation portion 36c of the state variable x _1. Thereafter, using the set model, the device state target value calculator 36c1 receives the product specification target value y ₁ (36h), the device state x ₂ measurement value (31s2), and the device state x ₃ measurement value (31s3) as inputs. Then, a target value of a new apparatus state x ₁ for absorbing the fluctuation of the production line causing the product abnormality is calculated and set in the state controller 31c1 of the processing apparatus 31.

  As described above, the target value of the apparatus state can be recalculated starting from the product abnormality detection. The regression equation that is the basis of the model can be accurately calculated when the quality fluctuates, and this method combined with the product abnormality monitoring system does not miss that opportunity, and the model can be automatically recalculated at any time.

(Embodiment 5)
Hereinafter, another embodiment of a semiconductor product manufacturing system according to the present invention will be described with reference to FIGS.

  FIG. 17 is a diagram showing a processing dimension and a distribution dimension of implanted impurities, taking as an example one of the semiconductor transistor elements described in FIG. 14 of the third embodiment. FIG. 18 is a diagram showing the present embodiment in which the manufacturing system of FIG. 1 is applied to semiconductor manufacturing, using representative processes that affect product characteristics. FIG. 19 is a diagram illustrating processing steps of the present embodiment performed by a computer that implements the line control system 60.

  The electrical characteristics that determine the quality of the CMOS semiconductor element, such as the ON / OFF threshold voltage, vary depending on the processing dimensions, impurity implantation amount, and impurity distribution dimensions. Important processing dimensions that determine the threshold voltage include gate dimension L (75L in FIG. 17) and insulating film thickness T (74T). Further, as an important impurity implantation amount, an impurity implantation amount into the source 72p and the drain 73p can be mentioned, and an impurity distribution length dimension (72L, 72T, 73T) that determines an impurity distribution depth dimension (72T, 73T) and an electric gate dimension 72e. 73L) is also an important factor.

  The processing dimension 74T of the insulating film thickness T in FIG. 17 can be measured by the film thickness inspection apparatus 51a after processing by the film forming apparatus 21a of the insulating film forming unit 2a in FIG. Moreover, the gate length dimension 75L of FIG. 17 can be measured by the dimension inspection apparatus 51c after processing by the etching apparatus 21c of the gate etching unit 2c after processing by the gate film forming unit 2b of FIG.

  However, there is generally no dimension inspection apparatus that can measure the impurity distribution dimensions 72T, 73T, 72L, and 73L in FIG. 17 after implantation by the implantation diffusion apparatus 31a of the impurity implantation diffusion unit 3a in FIG. 18 or after diffusion. Also, there is generally no inspection apparatus that can directly measure the amount of impurities implanted into the source 72p and the drain 73p after the implantation. Therefore, in order to investigate the variation of the electrical characteristics due to the variation of the impurity, it is necessary to directly check the comparison with the state measurement value of the injection device or the diffusion device. The state parameters of the implantation apparatus related to the fluctuation of impurities include the number of scans of the impurity beam and the degree of vacuum in the chamber that affects the collision between the beam and other atomic molecules as the state parameters that determine the amount of impurities implanted as ions. There are control positions for various machine elements that determine the angle and shape of the beam. Further, the state parameters for determining the impurity distribution dimension include the angle between the beam and the product wafer and the implantation energy. The state parameters of the diffusion device that determine the impurity distribution dimension include temperature, heating energy, and heating time.

  When a regression equation is created by comparing the electrical characteristics variation of CMOS semiconductor elements and manufacturing parameters, the state parameters of all processing devices of each unit of the manufacturing line shown in FIG. 18 are included in the manufacturing parameters to be verified. If it analyzes, it will contain many noise components, and there exists a problem that it is difficult to make a highly accurate regression equation. Therefore, in the present embodiment, the measurable processing dimension inspection measurement value is combined with the implantation diffusion apparatus state measurement value in place of the non-measurable implantation diffusion inspection measurement value, and collated with the electrical characteristic variation of the CMOS semiconductor element. A regression equation was made. If there is an abnormality in the machining dimension inspection measurement value, this machining dimension inspection measurement value and the state measurement value of the processing device machined to that dimension were partially compared again to create a regression equation and fluctuated. A method for specifying the device state measurement value and calculating the criterion was adopted. A specific procedure will be described below with reference to FIG.

  When the product electrical property inspection of the new device wafer by the product inspection device 41 of the product electrical property inspection unit 4 is completed (500), product abnormality monitoring is started (501). In step 1 (502), the product electrical characteristic inspection measurement value is stored in the database 64c. In step 2 (503), the product abnormality monitoring system 42 compares the product electrical characteristic inspection measurement value with the determination criterion to make an abnormality determination.

  The processing branches depending on whether there is an abnormality (504). If there is no abnormality, the product abnormality monitoring is terminated (518), and if there is an abnormality, the process proceeds to Step 3.

In step 3 (505), and acquires the history of the history of the product electrical testing Y measurement values and history of the processing dimension inspection X ₁ measurements injection diffuser state X ₂ measurements from the database by matching the wafer number. In step 4 (506), multiple regression equations between the product electrical property test Y, the processing dimension test X ₁ and the injection diffusion device state X ₂

Create Vth is a product electrical property inspection measurement value, X ₁ element T is an insulation film thickness, X ₁ element L is a gate length dimension, X ₂ elements x ₁ and x ₂ are arbitrary items of injection diffusion apparatus state measurement values is there. The numbers in parentheses indicate the time-series order of history data collated with individual numbers. In step 5 (507), selecting the item of item or injection diffuser state X ₂ feature size of high contribution statistical significance test X ₁ from the magnitude of the regression coefficient of the polymerization regression equation.

By the high contribution items to product electrical testing Y type branching processing (508), the process proceeds to step 6 in the case of items of injection diffuser state X _2, steps in the case of the item processing dimension inspection X ₁ Proceed to step 8.

In step 6 (509), multiple regression equation between the scores of products electrical testing Y measurement values and the selected infusion diffuser state X ₂

Create x _p is an item of the selected injection diffusion device state X ₂ , and E _p · _rest is an effect of items other than x _p . In the following step 7 (510), the judgment criterion of the item x _p of the selected injection diffusion device state X ₂ from the judgment criterion Yth of the product electrical property inspection Y determined from the product specification.

Is derived.

On the other hand, the process proceeds to step 8 if high fields of contribution to product electrical testing Y is the item processing dimension inspection X ₁ (511), the device status X ₃ Measurement of processed dimensions processing device dimensions X ₁ applicable The value history is collated with the wafer number and obtained from the database. For fields processing dimension inspection X ₁ is an insulator thickness (74T), the dimensions processing device is an insulating film deposition apparatus 21a, the state parameters of the device, the flow rate of the raw material gas used in the deposition, the chamber There are high-frequency electric power that turns the gas into plasma, the temperature of the wafer that affects the film formation reaction, and the temperature of the wafer. For fields processing dimension inspection X ₁ is a gate length (75L), the dimensions processing apparatus is an etching apparatus 21c, the state parameters of the device, the flow rate of the material gas used for etching, the vacuum in the chamber There is a high frequency power that turns a gas into a plasma and a wafer temperature that affects the etching reaction.

Following step 9 (512), the regression equation between the selected machining dimensional inspection X ₁ items and dimension processing apparatus state X ₃

Create In step 10 (513), selecting the item dimension processing apparatus state X ₃ with high contribution statistical significance from the magnitude of the regression coefficients of the regression equation. Step 11 (514) In the multiple regression equation between the items of the selected machining dimensional inspection X ₁ item and the selected dimension processing apparatus state X ₃

Create In the above formula, the gate length L is representative, but the same applies to the insulating film thickness T. In step 12 (515), the criterion of the item of the selected dimension processing device state X ₃ from the criterion of the processing dimension inspection X ₁ determined from the product specification.

Is derived. In the above formula, the gate length dimension L is representative, but the same applies to the insulating film thickness T.

  Through the above steps, the line control computer specified the state item of the injection diffusion device or the dimension processing device that caused the fluctuation of the product electrical property inspection Y and derived the determination criteria. In step 13 (516), these determination criteria are set in the determination unit of each apparatus abnormality monitoring system. In step 14 (517), the selected apparatus status item is selected and set in the parameter selection section of each apparatus abnormality monitoring system.

  As described above, the dimension processing device that can directly inspect the processing results and the infusion diffusion device that cannot be inspected are integrated step by step, and the state analysis of all devices is collated with the product electrical property inspection with high accuracy. And the determination items of the apparatus abnormality monitoring system can be set and the determination criteria can be updated.

  As mentioned above, the invention made by the present inventor has been specifically described based on the embodiment. However, the present invention is not limited to the embodiment, and various modifications can be made without departing from the scope of the invention. Needless to say.

  For example, in the above-described embodiment, the manufacturing system has been described by taking the case where the target product is a MOS semiconductor product as a representative example. However, the manufacturing system is not limited to a semiconductor product, for example, on a glass substrate. Needless to say, the present invention can be similarly applied to a manufacturing system such as a liquid crystal display panel or a self-luminous display panel on which a thin film transistor is mounted.

  INDUSTRIAL APPLICABILITY The present invention is suitable for a semiconductor product manufacturing line that requires a large number of manufacturing processes and manufacturing apparatuses and that integrates and processes fine elements on a substrate. In particular, it is suitable for the case where the process state of the manufacturing apparatus is likely to fluctuate depending on the overhaul and maintenance operation time, and a part of the process state causes a fatal fluctuation in product quality. This is suitable when the determination criteria for process state monitoring cannot be uniquely obtained from the product design specification and must be statistically derived by comparing with the product quality inspection result.

  This target product is suitable not only for wafers integrated with semiconductor elements, but also for wafers integrated with MEMS (Micro Electro Mechanical Systems), plasma panels with integrated discharge cells, and liquid crystal panels with integrated liquid crystal molecular direction control transistors. is there.

In Embodiment 1 of this invention, it is a figure which shows the manufacturing system which used the wafer as the object product. In Embodiment 1 of this invention, it is a figure which shows the apparatus abnormality monitoring system of a manufacturing system in detail. In Embodiment 1 of this invention, it is a figure which shows the product abnormality monitoring system of a manufacturing system in detail. In Embodiment 1 of this invention, it is a figure which shows the computer system which implement | achieves a manufacturing system. In Embodiment 1 of this invention, it is a figure which shows the optimization algorithm (apparatus abnormality monitoring step) of the determination criterion of an apparatus abnormality detection trigger. In Embodiment 1 of this invention, it is a figure which shows the optimization algorithm (quality variation cause specific step and apparatus abnormality determination reference | standard update step) of the determination criterion of an apparatus abnormality detection trigger. In Embodiment 1 of this invention, it is a figure which shows the optimization algorithm (product abnormality monitoring step) of the determination criterion of a product abnormality detection trigger. (A), (b) is a figure which shows transition of product test | inspection measurement value data in Embodiment 1 of this invention. (A)-(c) is a figure which shows transition of processing apparatus state measured value data in Embodiment 1 of this invention. (A)-(c) is a figure which shows transition of the processing apparatus state log | history data in Embodiment 1 of this invention. In Embodiment 2 of this invention, it is a figure which shows the apparatus abnormality monitoring system of a manufacturing system in detail. In Embodiment 3 of this invention, it is a figure which shows the manufacturing system which used the wafer as the object product. In Embodiment 3 of this invention, it is a figure which shows the fluctuation | variation component extraction system between multiple types of elements, and the line quality management system corresponding to it. In Embodiment 3 of this invention, it is a figure which shows the cross-section of the semiconductor element which comprises CMOS. In Embodiment 3 of this invention, it is a figure which shows the regression type preparation step according to the fluctuation | variation component of multiple types of element. In Embodiment 4 of this invention, it is a figure which shows the apparatus control system of a manufacturing system. In Embodiment 5 of this invention, it is a figure which shows the processing dimension of a transistor, and the distribution dimension of an implantation impurity. In Embodiment 5 of this invention, it is a figure which shows the manufacturing system applied to semiconductor manufacture. In Embodiment 5 of this invention, it is a figure which shows the regression equation preparation step of 2 steps | paragraphs which mixed the dimension inspection value and the apparatus state measured value.

Explanation of symbols

  DESCRIPTION OF SYMBOLS 1a ... Material wafer, 1b ... Final product wafer, 1n ... Wafer number, 2 ... Processing unit (type 1), 2a ... Insulating film formation unit, 2b ... Gate film formation unit, 2c ... Gate etching unit, 3 ... Processing Unit (Type 2), 3a ... Impurity implantation diffusion unit, 4 ... Final product inspection unit, 21 ... Processing device (Type 1), 21a ... Film forming device, 21c ... Etching device, 22, 22a, 22c ... Device abnormality monitoring System (type 1), 31 ... Processing device (type 2), 31a ... Infusion / diffusion device, 31s ... Status measuring device, 32, 32a ... Device abnormality monitoring system (type 2), 33s ... Parameter selection unit, 33m ... Parameter selection 34c ... determination unit 34m ... determination reference management unit 35c ... determination unit 35m ... estimated coefficient management unit 36 ... device control system 41 ... product inspection device 2 ... Product abnormality monitoring system, 44c ... Determination unit, 44m ... Judgment reference management unit, 45 ... Product design system, 51 ... Product inspection device, 51a ... Film thickness inspection device, 51c ... Dimension inspection device, 52 ... Product abnormality monitoring system 52a ... Film thickness abnormality monitoring system, 52c ... Dimensional abnormality monitoring system, 60 ... Line control system, 61 ... Wafer trace system, 62 ... Line quality management system, 63 ... Quality variation cause identification unit, 64 ... Update of abnormality monitoring criteria 64a ... Processing device machining state measurement value database, 64b ... In-process wafer inspection measurement value database, 64c ... Final product wafer inspection measurement value database, 65 ... Variation component extraction system between multiple types of elements, 65p1 ... Independent variation component extraction Part, 65p2 ... common variation component extraction part, 70n ... computer network, 70c ... device abnormality monitoring computer, 70C2 ... products abnormality monitoring computer, 70C3 ... line control computer, 70D1 ... machining device machining state measurement value database, 70D2 ... being manufactured wafer inspection measurements database, 70D3 ... final product wafer inspection measurements database.

Claims (8)

A semiconductor product manufacturing system in which fine elements are integrated on a material substrate using at least one processing device and at least one product inspection device,
The manufacturing system includes:
An apparatus abnormality monitoring system that measures the state of the processing apparatus and monitors an abnormality of the state measurement value based on a preset criterion.
A product abnormality monitoring system that measures the quality of a semiconductor product by the product inspection device and monitors an abnormality of the quality measurement value based on a predetermined criterion.
It comprises a line control system that performs semiconductor product input control to the processing device and the product inspection device, and collects the state measurement value and the quality measurement value,
The line control system includes:
An identification number is assigned to each individual material substrate to be processed into a semiconductor product to control the input of each product to the device and the measurement value collection control, and the device abnormality detection of the device abnormality monitoring system is used as the starting point. A semiconductor product trace system that tracks a product individual to a product inspection device downstream of the production line, or backtracks a state measurement value of a processing device that has processed the product individual upstream of the production line based on the product abnormality detection of the product abnormality monitoring system. When,
From the verification analysis of the state measurement value of the processing apparatus tracked or back-tracked by the semiconductor product trace system and the quality measurement value of the semiconductor product from the detection of the apparatus abnormality detection or the product abnormality detection, it causes quality fluctuation. The processing device and the state measurement value of this processing device are identified, and the determination criteria of the specified processing device are calculated, and the abnormal monitoring of the state measurement value of the specified processing device is registered or canceled, and the determination criterion is updated. And a line quality control system that implements an abnormality detection as a starting point. The semiconductor product manufacturing system according to claim 1,
The manufacturing system includes:
Use different independent processing equipment as part of the manufacturing process, or use independent processing equipment with different processing conditions set in the same processing equipment as part of the manufacturing process, and use a common processing equipment. A semiconductor product in which at least two or more types of microelements that are processed using other manufacturing processes are integrated on a material substrate is the target of manufacture.
The line control system includes:
Regarding the quality measurement value measured by the product inspection apparatus, the inter-element variation component extraction system that separates the common variation component and the independent variation component between the different microelements of different types,
The line quality control system is
From the verification analysis of the inter-element common variation component of the quality measurement value and the state measurement value of the common processing device, the processing device that caused the quality variation and the state measurement value of this processing device are identified, or It is characterized by identifying the processing device that caused the quality variation and the state measurement value of this processing device from the collation analysis of the inter-element independent variation component of the quality measurement value and the state measurement value of the independent processing device. Manufacturing system for semiconductor products. In the semiconductor product manufacturing system according to claim 1 or 2,
As the product inspection device, two types of products, that is, at least one processing quality inspection device that measures the processing quality of a product in the middle of manufacturing a semiconductor product, and a final quality inspection device that measures the final quality of the product after the manufacture is finished Equipped with inspection equipment,
The line quality control system is
As a first step, the product processing quality measurement value of the processing quality inspection device is compared with the product final quality measurement value of the final quality inspection device, the inter-element common variation component, or the inter-element independent variation component From the analysis, identify the processing equipment that caused the quality fluctuation,
As a second step, the state measurement value of the processing apparatus that has caused the quality variation is identified from a collation analysis of the state measurement value of the specified processing apparatus and the product processing quality measurement value of the processing quality inspection apparatus. A semiconductor product manufacturing system comprising a two-step quality variation cause identification step for calculating a criterion for identifying a specified processing apparatus. In the semiconductor product manufacturing system according to any one of claims 1 to 3,
The processing device includes a state controller that controls a part of the state of the processing device to a processing condition value set from the outside,
The manufacturing system includes an apparatus state control system for calculating a processing condition value set from the outside,
The device state control system includes:
Storing an apparatus state target value calculation model derived from a collation analysis of a state measurement value of a processing apparatus and a quality measurement value of a semiconductor product performed by the line quality management system;
Using the device state target value calculation model, the product processing target value or the final product quality target value and the current or past measurement value of the uncontrollable state of the processing device are input as the controllable state of the processing device. Calculate the target value,
The calculated state target value of the processing device is set in the state controller to realize a processing device state that achieves the product processing target value or the final product quality target value regardless of the state variation of the processing device. Semiconductor product manufacturing system. In the semiconductor product manufacturing system according to any one of claims 1 to 4,
The semiconductor product manufacturing system according to claim 1, wherein the semiconductor product is a wafer product obtained by integrating semiconductor transistor elements as fine elements. In the semiconductor product manufacturing system according to claim 2,
The semiconductor product uses two types of semiconductor transistor elements, P-type and N-type, on a wafer by using different types of ion implanters or ion implanters set with different processing conditions as part of the manufacturing process. Integrated wafer products,
The quality measurement value of the product inspection device is an electrical property measurement value of the P-type and N-type semiconductor transistor elements of the wafer product,
The line quality control system determines the cause of the electrical characteristic variation from the collation analysis of the state measurement value of the ion implantation apparatus and the inter-element independent variation component of the electrical characteristic measurement value of at least two types of transistor elements of the wafer product. A semiconductor product manufacturing system characterized by identifying an ion implantation apparatus and a state measurement value of the ion implantation apparatus, and updating a determination criterion for the state measurement value of the identified ion implantation apparatus. In the semiconductor product manufacturing system according to claim 2,
The semiconductor product is a wafer product obtained by integrating and processing a P-type semiconductor transistor and an N-type semiconductor transistor on a wafer using different types of ion implantation apparatuses as part of the manufacturing process.
The quality measurement value of the product inspection apparatus is a measurement value of electrical characteristics of the P-type semiconductor transistor and the N-type semiconductor transistor of the wafer product,
The line quality control system is configured to perform an electrical characteristic from a collation analysis between a state measurement value of a processing apparatus other than the ion implantation apparatus and an inter-element common variation component of an electrical characteristic measurement value of at least two types of transistor elements of the wafer product. A semiconductor product characterized by identifying a processing apparatus related to lithography or film formation that causes fluctuations and a state measurement value of the processing apparatus, and updating a determination criterion of the state measurement value of the specified processing apparatus Manufacturing system. In the semiconductor product manufacturing system according to any one of claims 5 to 7,
As the product inspection apparatus, two types of product inspection apparatuses, that is, at least one processing dimension inspection apparatus that measures a processing dimension in the middle of manufacture of a wafer product, and a final characteristic inspection apparatus that measures the electrical characteristics of the product after the manufacture is finished. With
The apparatus abnormality monitoring system measures and monitors the state of an exposure apparatus, an etching apparatus, or a film forming apparatus that performs shape processing of a semiconductor transistor element, and an ion implantation apparatus or a diffusion apparatus that performs impurity implantation of the semiconductor transistor element. Equipped with equipment abnormality monitoring system,
The line quality control system is
As a first step, the manufacturing history of the wafer product is reverse-traced starting from the collection of electrical characteristic measurement values after the wafer product is manufactured, and the processing dimension measurement value history during the manufacturing of the wafer product is collected. In addition, an implantation apparatus in which impurities are implanted and diffused into the wafer product or a state measurement value history of the diffusion apparatus is collected,
As a second step, the collected measured dimension measurement value and the apparatus state measurement value are combined, and a collation analysis is performed between the combination data and the electric characteristic measurement value data. Identify the processed dimensions or state measurements of the impurity implanter,
As a third step, when the state change of the specified impurity implantation apparatus is the cause of the product electrical characteristic fluctuation, the determination criterion of the apparatus state monitoring of the specified impurity implantation apparatus is calculated,
When the variation in the processing dimension is the cause of the variation in the electrical characteristics of the product, as a fourth step, the state measurement value history of the semiconductor transistor element shape processing apparatus is collected, and the collected shape processing apparatus state measurement value data and Perform collation analysis with the machining dimension measurement value data, identify the state measurement value of the processing machine that caused the product machining dimension fluctuation from the collation analysis result, and calculate the criteria for monitoring the machine state of the identified machining apparatus A semiconductor product manufacturing system.

Images