CN102467089B

CN102467089B - Process control method of semiconductor technology and system thereof

Info

Publication number: CN102467089B
Application number: CN 201010540104
Authority: CN
Inventors: 谢凯
Original assignee: Beijing North Microelectronics Co Ltd
Current assignee: Beijing North Microelectronics Co Ltd
Priority date: 2010-11-10
Filing date: 2010-11-10
Publication date: 2013-09-11
Anticipated expiration: 2030-11-10
Also published as: CN102467089A

Abstract

The invention provides a process control method of a semiconductor technology and a system thereof as well as a semiconductor control system containing the process control system, so that a problem that current univariate process control can not meet demands can be solved. The method comprises: a multivariate statistical process control model corresponded to a technological process that needs to be monitored is started; a monitoring sample is received, wherein the monitoring sample contains multiple technological data of all variables in the technological process that needs to be monitored; according to the multivariate statistical process control model and the multiple technological data of all the variables in the technological process that needs to be monitored, a fault value corresponded to the technological process that needs to be monitored is obtained; and it is determined whether the fault value exceeds the control limit of the multivariate statistical process control model; if so, a quantitative analysis is carried out on the fault for finding the fault cause. According to the invention, a fault cause can be obtained by a quantitative analysis, so that technological personnel can get rational understanding on a fault alarm.

Description

Course control method for use in the semiconductor technology and system

Technical field

The present invention relates to process control technology, particularly relate to course control method for use and system in a kind of semiconductor technology, and the semiconductor control system that comprises this Process Control System.

Background technology

In the process industrial (as oil refining, petrochemical complex, metallurgy etc.) that with continuous production is feature, process control has entered into the control of production overall process from separate device control, be a kind of comprehensive automation control model that integrates optimization, dispatches, manages.

Univariate statistics process control technology (Univariate Statistical Process Control, be called for short USPC) be a kind of process control technology commonly used, main X-control chart (Xbar-X), average-standard deviation control chart (Xbar-S), mean-range chart (Xbar-R) and the monodrome-moving range figure control charts such as (X-MR) of adopting monitored variable, when having process data, monitored variable surpasses control in limited time, control chart can capture this unusual (or claiming fault), and sends warning message; Simultaneously, use indexs such as historical sample data computation process capability index Cp, Cpk, weigh also consistance, the extent of stability of monitoring industrial processes.

In the semiconductor production manufacture process, because complex procedures, the characteristics that the cost of raw material is high, require same batch product quality of each production link to reach consistance highly, therefore also use univariate statistics process control technology (USPC) that the variable in producing is carried out complete monitoring, thereby instruct the slip-stick artist to find and handle the unusual variable fluctuation situation of appearance, guarantee the stable of production run.

But, the all types of control charts of USPC can only be monitored a variable at every turn, raising along with the semiconductor manufacturing facility complexity, controllable parameter also constantly increases, need the variable of monitoring also just more and more, if monitor a plurality of variablees simultaneously, not only calculated amount is huge, and the switching between control chart shows also very loaded down with trivial details.

In addition, because each variable is separately statistics monitoring, can not reflect the incidence relation that exists between variable, therefore can cause the erroneous judgement of fault and fail to judge.For example, with reference to Fig. 1, be that two variablees sample in the USPC control chart shows normal synoptic diagram.Two variablees are presented in separately the control chart, although show two process datas in each control chart near the last lower control limit shown in the dotted line among the figure (being the process data in the circle among Fig. 1), do not surpass control limit triggering fault alarm.But with reference to Fig. 2, be the synoptic diagram of two variablees sample display abnormality in the associated diagram of variable.From the associated diagram of these two variablees, can see two unusual process datas being arranged away from other process data straight lines and trigger fault alarm (being the process data in the circle among Fig. 2).Though the associated diagram of variable can be judged abnormal conditions, when monitored variable quantity more for a long time, corresponding variable associated diagram is also more, calculates and checks that the variable associated diagram is loaded down with trivial details equally.And in the prior art, can only the qualitatively analyze fault alarm based on the associated diagram of two variablees of USPC technology, and can not make the technologist not have rational understanding to fault alarm to its quantitative test, can not fundamentally solve the relevant issues of fault alarm.

In sum, along with improving constantly of semiconductor manufacturing facility complexity, failure monitoring is proposed higher requirement, the USPC technology can not satisfy this demand.

Summary of the invention

The invention provides course control method for use and system in a kind of semiconductor technology, and the semiconductor control system that comprises this Process Control System, to solve the problem that existing single argument process control can't satisfy the demands.

In order to address the above problem, the invention discloses the course control method for use in a kind of semiconductor technology, comprising:

Start the Multivariable Statistical Process Control model that needs monitoring process step correspondence;

Receive the monitoring sample, described monitoring sample comprises the described a plurality of process datas that need each variable in the monitoring process step;

According to described Multivariable Statistical Process Control model and the described a plurality of process datas that need each variable in the monitoring process step, obtain the corresponding described fault score value that needs the monitoring process step;

Judge whether described fault score value exceeds the control limit of described Multivariable Statistical Process Control model, if, then quantitative test failure cause.

Described quantitative test failure cause comprises: according to each statistic of a plurality of process datas of fault signature database and described each variable, obtain the fault signature data statistic; Described fault signature database is to set up based on the univariate statistics process control technology; According to described fault signature data statistic and preset the fault alarm rule, obtain fault score contribution plot, described fault score contribution plot has recorded fault score and the corresponding fault alarm type of each variable; In conjunction with described fault score contribution plot and described fault signature data statistic, quantitative test failure cause.

Preferably, under off-line state, described method comprises: reception comprises a plurality of monitoring samples that need the monitoring process step; Each needs the fault score value of monitoring process step to obtain correspondence respectively; Judge respectively whether each fault score value exceeds the corresponding control limit that needs the Multivariable Statistical Process Control model of monitoring process step, if, then quantitative test failure cause.

Preferably, under presence, described reception monitoring sample is specially: reception in real time comprises a certain monitoring sample that needs the monitoring process step.

Preferably, by adding up the same fault score value that needs the monitoring process step in each silicon chip etching technology process, set up the multivariate statistics control chart.

Preferably, described startup needs before the Multivariable Statistical Process Control model of monitoring process step correspondence, also comprise: set up the Multivariable Statistical Process Control model, specifically comprise: set the configuration information of modeling, described configuration information comprises processing step, data screening parameter, modeling variable, degree of confidence, the pivot computing method of appointment; Screen according to data screening parameter and the modeling variable a plurality of process datas to each variable in the processing step specified in each training sample; To the training sample compute statistics after the screening; Utilize statistic, degree of confidence and the pivot computing method of training sample to set up the Multivariable Statistical Process Control model, and the control of definite model limit; Judge whether there is the training sample that surpasses the control limit in institute's established model, if there is no, then preserve the model of setting up; If exist, then remove out-of-limit sample, and described statistic calculating is carried out in the remaining training sample circulation of utilization, model is set up and determining step, does not have the training sample above the control limit in institute's established model.

Preferably, described screening comprises: utilize the data screening device automatically in the processing step of appointment, training sample is carried out the screening pre-service of named variable; Wherein, described data screening device with the processing step be unit pre-configured the screening parameter value of all optional variablees under each processing step.

The present invention also provides the Process Control System in a kind of semiconductor technology, comprising:

Model imports module, is used for starting the Multivariable Statistical Process Control model that needs monitoring process step correspondence;

Data reception module is used for receiving the monitoring sample, and described monitoring sample comprises the described a plurality of process datas that need each variable in the monitoring process step;

Model computation module is used for obtaining the corresponding described fault score value that needs the monitoring process step according to described Multivariable Statistical Process Control model and the described a plurality of process datas that need each variable of monitoring process step;

Failure analysis module is used for judging whether described fault score value exceeds the control limit of described Multivariable Statistical Process Control model, if, then quantitative test failure cause.

Described failure analysis module comprises: first analytic unit, be used for each statistic according to a plurality of process datas of fault signature database and described each variable, and obtain the fault signature data statistic; Described fault signature database is to set up based on the univariate statistics process control technology; Second analytic unit is used for according to described fault signature data statistic and presets the fault alarm rule, obtains fault score contribution plot, and described fault score contribution plot has recorded fault score and the corresponding fault alarm type of each variable; Integerated analytic unit is used in conjunction with described fault score contribution plot and described fault signature data statistic, quantitative test failure cause.

Preferably, described system also comprises: model building module is used for setting up the Multivariable Statistical Process Control model.

Preferably, described model building module comprises: the model dispensing unit, and for the configuration information of setting modeling, described configuration information comprises processing step, data screening parameter, modeling variable, degree of confidence, the pivot computing method of appointment; The data screening unit is used for screening according to data screening parameter and the modeling variable a plurality of process datas to each variable in the specified processing step of each training sample; The statistical computation unit is used for the training sample compute statistics after the screening; The pre-established unit of model is used for utilizing statistic, degree of confidence and the pivot computing method of training sample to set up the Multivariable Statistical Process Control model, and the control of definite model limit; The iteration modeling unit is used for judging whether institute's established model exists the training sample that surpasses the control limit, if there is no, then preserves the model of setting up; If exist, then remove out-of-limit sample, and utilize remaining training sample to carry out in conjunction with statistical computation unit and model pre-established unit circulation that described statistic is calculated, model is set up and judge, in institute's established model, there is not the training sample that surpasses the control limit.

Preferably, described system also comprises: the data screening device, and being used for the processing step is the screening parameter value of all optional variablees under pre-configured each processing step of unit; The data screening module, be used for according to model specified processing step and named variable, utilize the data screening device automatically at processing step and the named variable of model appointment, utilize the data screening device automatically in the processing step of model appointment, the monitoring sample is carried out the screening pre-service of named variable, and the monitoring sample after will screening passes to described model computation module, and the monitoring sample after described model computation module utilizes model to screening calculates.

The present invention also provides a kind of semiconductor control system, comprises apparatus control system and Process Control System, it is characterized in that: described Process Control System comprises the arbitrary described Process Control System of claim 8 to 12.

Preferably, described semiconductor control system also comprises: data acquisition system (DAS) is used for collecting device control system institute control equipment at the monitoring sample of art production process, and stores in the database; Database is used for storage monitoring sample.

It is data source with described database that semiconductor is controlled described Process Control System, directly receives the monitoring sample from this database.

Compared with prior art, the present invention has the following advantages:

First, the present invention proposes a kind of Multivariable Statistical Process Control (Multivariate Statistical Process Control that is applicable to semiconductor technology, be called for short MSPC) method, this method can be utilized the incidence relation between a plurality of variablees, the unusual of monitoring sample reflected by the fault score value that is obtained by a plurality of aggregation of variable, and according to described fault score value the quantitative test reason that is out of order, thereby make the technologist to the rational understanding of fault alarm, and then fundamentally solve the relevant issues of fault alarm.Based on described fault score value, the present invention can just can accurately analyze abnormal conditions by a multivariate statistics control chart of having drawn the fault score value to the monitoring of sample, the fault erroneous judgement the problem includes: of having avoided univariate statistics process control (USPC) problem and the problem of failing to judge; And the present invention just can judge guilty culprit easily and quickly by multivariate statistics control chart intuitively, need not to switch loaded down with trivial detailsly and check a large amount of univariate statistics process control charts, alleviated operational load greatly, improve the control level, further improved production efficiency and the product yield of semiconductor equipment.

Second, during quantitative test failure cause of the present invention, on the basis that the multivariate statistics control chart is provided, fault score contribution plot and the fault signature data statistic of reflection univariate statistics situation also are provided in conjunction with the USPC technology, the solid dot that the user is transfinited in the multivariate statistics control chart is double-clicked and can be opened simultaneously, by this figure and table can be intuitively, clearly analyze the variable that breaks down and the basic reason of fault.

The 3rd, during quantitative test failure cause of the present invention, also in conjunction with the USPC technology fault alarm information has been carried out grade classification, the mode that this classification is reported to the police quantizes fault type, can reflect the fault variable by the situation of change of quantitative change to qualitative change, both realize capturing major failure fast in qualitative mode, realize again can doing further quantitative test to major failure, make fault analysis more comprehensively, the result has more actual directive significance.Simultaneously, the alert levels of this quantification can provide more horn of plenty, warning accurately for machine control system, and the fault handling function of further enhancing and the current control system of refinement is for process control provides advanced decision-making and scheduling feature.

The 4th, the present invention has also designed the data screening device and has come monitoring sample or model training sample are carried out pre-service, this data screening device with the processing step be unit pre-configured the screening parameter value of all optional variablees under each processing step, therefore in pre-treatment step, can directly import, carry out automatically thereby the troublesome operation that the data screening parameter will manually be set is optimized for by system.

The 5th, the present invention is to provide at the MSPC of tool level (directly towards board equipment) based on the real-time process data and use, directly the database with data acquisition system (DAS) is data source, to satisfy the requirement of control in real time.

The 6th, the present invention can also realize failure monitoring and the diagnosis under off-line and the online two states.

Description of drawings

Fig. 1 is that two variablees sample in the USPC control chart shows normal synoptic diagram in the prior art;

Fig. 2 is the synoptic diagram of two variablees sample display abnormality in the associated diagram of variable in the prior art;

Fig. 3 is the synoptic diagram of two variablees sample display abnormality in the MSPC control chart among the present invention;

Fig. 4 is the process flow diagram of setting up the MSPC model in the embodiment of the invention;

Fig. 5 is the Multivariable Statistical Process Control figure that the MSPC model is set up institute's reference in the process in the embodiment of the invention;

Fig. 6 is the MSPC off-line procedure control method process flow diagram in the described a kind of semiconductor technology of the embodiment of the invention;

Fig. 7 is the multivariate statistics control chart of a certain processing step in the embodiment of the invention;

Fig. 8 is MSPC fault score contribution plot in the embodiment of the invention;

Fig. 9 is the online course control method for use process flow diagram of MSPC in the described a kind of semiconductor technology of the embodiment of the invention;

Figure 10 is the Process Control System structural drawing in the described a kind of semiconductor technology of the embodiment of the invention;

Figure 11 is the Organization Chart of the described a kind of semiconductor control system of the embodiment of the invention.

Embodiment

For above-mentioned purpose of the present invention, feature and advantage can be become apparent more, the present invention is further detailed explanation below in conjunction with the drawings and specific embodiments.

The present invention proposes a kind of Multivariable Statistical Process Control (MultivariateStatistical Process Control that is applicable to semiconductor technology, be called for short MSPC) method, MSPC is also referred to as fault diagnosis and classification (Fault Detection and Classification, FDC) technology, can utilize incidence relation between a plurality of variablees to come unusual (or fault) situation of variable in the monitoring industrial processes, compare the fault variable that can occur in the identification equipment production run easily and quickly by succinct control chart with the USPC technology.For example, at example shown in Figure 1, the sample of two variablees in the USPC control chart shows normal, but with reference to Fig. 3, be the synoptic diagram of these two variablees sample display abnormality in the multivariate statistics control chart, from the control chart of Fig. 3, can directly find out this two unusual process datas.

The present invention is applicable to the failure monitoring of complicated procedures of forming in the semiconductor fabrication, mainly comprise Multivariable Statistical Process Control (MSPC) model foundation, utilize the MSPC model to monitor and fault analysis two parts.To be that example is elaborated below with the semiconductor etching process.

1, the MSPC model is set up

Characteristics according to the conductor etching process, the etching production overall process of each wafer (silicon chip) needs through some processing steps, and the recipe of each step (technical recipe) arranges difference, be that the setting value (or claim desired value) of each variable in each step is not necessarily identical, therefore, need be that unit sets up the model that adapts respectively with the step.

With reference to Fig. 4, it is the process flow diagram of setting up the MSPC model in the embodiment of the invention.

Step 401 is that unit imports a plurality of training samples with wafer, and each training sample comprises a plurality of process datas of each variable in each processing step;

Described training sample is used for model training, and in conductor etching technology, the sample data of a training sample comprises a plurality of process datas of each variable in each processing step of etching wafer (silicon chip).In this step, generally be that unit imports training sample with wafer, the modeling here is specially adopts a plurality of training samples that specified processing step is carried out modeling.

Step 402, the given process step; Need the monitoring process step in the corresponding practical application of the processing step of appointment here;

Step 403 arranges the data screening parameter and chooses the variable that participates in modeling;

The processing step of above-mentioned appointment, data screening parameter, modeling variable and subsequent step 406 preset confidence, pivot computing method all belong to the main configuration information of modeling, in addition also comprise other configuration informations.These configuration informations are in the initial setting in a single day of modeling, and step 405 is generally no longer changed to 408 iteration modeling process in the back.Certainly, according to the characteristics of processing step, the different process step need be set up different models, and is therefore generally also inequality at the corresponding configuration information of different models.

Step 404 is screened according to data screening parameter and the modeling variable a plurality of process datas to each variable in the processing step specified in each training sample;

Described screening is preferred pre-treatment step, is used for filtering out the training sample data that meet modeling demand from the training sample that imports, and namely a plurality of process datas to each variable in the processing step specified in each training sample screen.Wherein, described modeling variable refers to participate in the variable of modeling, i.e. each variable in the specified processing step in the training sample of Dao Ruing, but may only use wherein Partial Variable during modeling, therefore need be configured into row filter according to model.In addition, described data screening parameter comprises a plurality of parameters, and the training sample that is used for initially importing carries out processing such as denoising, to guarantee modeling precision.For example, can be set by the user the number of the process data of removing beginning and finishing at different variablees, to determine whether to eliminate the labile factor in this stage; Simultaneously, can also set maximal value, the minimum value scope of process data at different variable permission users, to remove noise data.

In practical operation, generally the data screening parameter is set and chooses the modeling variable by manual type, and then screen importing training sample.Since comprise a lot of variablees in each processing step, and the different process step has nothing in common with each other to the data screening parameter of same variable and the requirement of modeling variable, if it is each modeling all arranges one by one by manually-operated, very loaded down with trivial details.Based on this, present embodiment proposes a kind of data preprocessing method of optimization, and the design data screening washer arranges with the reduced data screening.Described data screening device with the processing step be unit pre-configured the screening parameter value of all optional variablees under each processing step.The data screening device can be constructed by file layouts such as Excel, and the recipe with the typical etch processes of wafer shown in the table 1 is example below:

Step ID	1	2	3	4	5	6	7	8	9	10
											Step Name	BT_sta	BT	ME1_sta	ME1	ME2_sta	ME2	OE_sta	OE	purge	dechuck
PM.S1.Prc	10	10	10	15	10	75	10	26	5	3
											Pressure	7	7	10	10	30	30	30	30	0	15
SRFpower	0	350	0	500	0	250	350	350	0	300
											BRFpower	0	40	0	80	0	75	40	40	0	0
GasCl2	0	0	60	60	50	50	15	15	0	0
											GasHBr	0	0	0	0	150	150	170	170	0	0
GasCF4	50	50	0	0	0	0	0	0	0	0
											GasCHF3	0	0	0	0	0	0	0	0	0	0
GasCH2F2	0	0	0	0	0	0	0	0	0	0
											GasSF6	0	0	30	30	0	0	0	0	0	0
GasAr	0	0	0	0	0	0	0	0	0	0
											...

Table 1

In the table 1, the Step ID in first row represents the processing step sequence number, and the Step Name in second row represents the processing step title, and what represent except Step ID and Step Name in first example is the title of each variable.As can be seen from Table 1, same variable is different in the recipe of different process step value in the etching process.

The data screening device of corresponding tables 1 structure is as shown in table 2:

Figure DEST_PATH_RE-GSB00000429838000011

Table 2

In the Excel file shown in the table 2, the a certain processing step of the corresponding etching wafer of each worksheet (being each sheet), the ParameterName field is to all optional variablees in should step, Step-start represents that this processing step begins the process data of part, Step-Stop represents the process data of latter end, and Max-limit represents maximum control limit, and Min-limit represents minimum control limit, Include is used for setting the variable that participates in modeling, and " 1 " representative is selected.After data screening device structure finishes, can directly import the MSPC system and use, the troublesome operation that the data screening parameter manually is set is optimized for by system carry out automatically.

Step 405 is to the training sample compute statistics after the screening;

Namely the training sample that meets modeling demand after the screening is carried out statistical computation, the statistic that obtains comprises Min (minimum value), Max (maximal value), Mean (mean value), Stdev (standard deviation) etc.These statistics will be for Modeling Calculation.Particularly, carry out statistical computation based on a plurality of process datas of each variable in the specified processing step in each training sample after the screening, to obtain each statistic.

Step 406 is selected degree of confidence and pivot computing method;

Described degree of confidence be used for to be determined the control limit of model, and optional degree of confidence is as 95%, 99%, 99.9% etc.Described pivot computing method are used for calculating the pivot number, namely select representative variable again and be used for modeling from a plurality of variablees that participate in modeling, thereby reduce calculated amount and computation complexity.Optional pivot computing method such as 90%CPV (cumulative percent variance, accumulative total of variance and number percent), AE (Average Eigenvalue, mean eigenvalue) etc.

In addition, except can using the pivot computing method, also can use computing method such as neural network or decision tree.

Step 407 utilizes statistic, degree of confidence and the pivot computing method of training sample to set up the Multivariable Statistical Process Control model, and the control of definite model limit;

Modeling process is calculated the pivot number for utilizing the pivot computing method, and the utilization of preference pattern algorithm utilizes statistic, degree of confidence and the pivot number etc. of training sample to calculate, and then can obtain the control limit of model.Described control limits the use of in judging whether processing step has fault to produce, and model is in case determine, control limit that should model is also namely determined.In addition, in modeling process, also need the Multivariable Statistical Process Control figure of preference pattern use and the Multivariable Statistical Process Control figure of screening model data, the Multivariable Statistical Process Control figure of described screening model data is used for step 408 and judges whether to exist the training sample that surpasses the control limit.Wherein, optional model algorithm such as SPE, Hotelling T ², φ etc., corresponding, control chart also can be from SPE (Squared prediction error, square prediction error), Hotelling T ², φ is (a kind of in conjunction with SPE and Hotelling T ²Algorithm) in select.

The MSPC model of setting up can be with reference to shown in Figure 5.Fig. 5 is the Multivariable Statistical Process Control figure that the MSPC model is set up institute's reference in the process in the embodiment of the invention, what this figure showed is after model tentatively builds up, the failure condition of corresponding etching the 6th processing step of a plurality of training samples that imports, and this model has been selected SPE, T simultaneously ²With the calculating of φ and the Multivariable Statistical Process Control figure of screening model data.At SPE, T ²In the φ control chart, the 6th processing step of the corresponding wafer of each solid dot, and each solid dot produces by a monitoring sample; Solid line representative control limit among the figure, the solid dot that exceeds number line is specially the solid dot that exceeds the control limit, and this solid dot shows that this wafer fault occurred in this 6th processing step etching process.

Step 408 judges whether there is the training sample that surpasses the control limit in institute's established model;

Particularly, with reference to Fig. 5, if namely there is not the solid dot that exceeds the control limit in the above-mentioned training sample that surpasses the control limit that do not exist, then accepts the model set up, and continue step 409; If exist, then remove the training sample that produces the solid dot that surpasses the control limit, namely remove the training sample that produces the solid dot that surpasses the control limit in a plurality of training samples that import in the step 401, and return step 405, and utilize remaining training sample to recomputate statistic, and the continuation subsequent step, up to meeting the requirements.

Set up process at model, the user can remove the training sample that surpasses the control limit, does not have solid dot to exceed control in the figure of similar Fig. 5 and only is limited to.Preferably, the model that obtains by the iteration modeling is optimum model, so present embodiment adopts the mode of iteration modeling.The iteration modeling process is: the described statistic calculating of step 405, the model foundation of step 407 and the determining step of step 408 are carried out in circulation, do not have the training sample that surpasses the control limit in institute's established model.Because the number of training of each iteration is reducing, therefore the control extreme position that at every turn calculates can be different, when all sample points in the model are all within the control limit, namely determined final control extreme position.In this iterative process, select unified degree of confidence and pivot computing method.

Step 409 is preserved the model of setting up.

After the iteration modeling finishes, then fixed model is preserved, in the production monitoring process for use in reality.

In above-mentioned modeling process, preferred, can also be to the compute statistics of a plurality of training samples of original importing, and preserve, for the slip-stick artist with reference to use.

Above-mentioned flow process is the flow process of the processing step of some appointments being carried out modeling, for the modeling of other processing steps that need monitor, also can not repeat them here with reference to above-mentioned flow process.

2, MSPC monitoring and fault analysis

The MSPC model just can utilize the MSPC model that the etching production run of wafer is monitored after setting up.Described monitoring comprises monitored off-line and two kinds of patterns of on-line monitoring, describes by Fig. 6 and flow process shown in Figure 9 respectively below.

With reference to Fig. 6, it is the MSPC off-line procedure control method process flow diagram in the described a kind of semiconductor technology of the embodiment of the invention.

Under the monitored off-line pattern, the MSPC system can calculate and analyze all data that import, and catches the wafer sample that surpasses the control limit.Concrete steps are as follows:

Step 601 starts the Multivariable Statistical Process Control model that each needs monitoring process step correspondence;

Under the monitored off-line pattern, need pre-determine which processing step and need carry out off-line analysis, start the MSPC model of corresponding these steps then.

Step 602, reception comprise a plurality of monitoring samples that need the monitoring process step, and each monitoring sample comprises a plurality of process datas that need each variable in the monitoring process step;

Under the monitored off-line pattern, also need all to receive the monitoring sample in these processing steps that need monitor, for example receive the wafer monitoring sample that the 6th step of etching technics and the 9th goes on foot in the table 1, the monitoring sample in each step has comprised a plurality of process datas of each variable.

Step 603 is screened pre-service to the monitoring sample that receives;

This step is preferred process, can utilize above-mentioned data screening device to screen, and the content of screening comprises Step-start, Step-Stop, Max-limit, Min-limit, Include etc.Certainly, the employed MSPC model of different process step difference, the parameter of screening and relevant parameter value also can be different.

Step 604, according to each described Multivariable Statistical Process Control model and each described a plurality of process data that need each variable in the monitoring process step, each needs the fault score value of monitoring process step to obtain correspondence respectively;

Be unit with the processing step namely, utilize the MSPC model that the monitoring sample after screening is calculated, each that is comprised by the monitoring sample needs a plurality of process datas of each variable in the monitoring process step comprehensively to obtain the fault score value that each needs the monitoring process step;

At the different process step, will use different MSPC models to carry out statistical computation.Utilizing each MSPC model to carry out in the calculation process, the statistic of a plurality of process datas after the screening of all load earlier a plurality of process datas of needing each variable in the monitoring process step, calculating named variable, degree of confidence and the pivot computing method of utilizing statistic, model to determine then, last calculating again needs the fault score value of monitoring process step, and the control chart that uses by model shows.

In the MSPC model, finally represent the normal or abnormal conditions of a monitoring sample in technological process by the fault score value.Described fault score value is the result of calculation that a plurality of process datas according to each variable in the need monitoring process step comprehensively obtain, and has taken into full account the incidence relation between the variable.

Step 605 judges respectively whether each fault score value exceeds the corresponding control limit that needs the Multivariable Statistical Process Control model of monitoring process step, if, then quantitative test failure cause.

In actual applications, a kind of adoptable analytical approach is to utilize the various types of control charts of Multivariable Statistical Process Control modeling rendering, by the control chart intuitive analysis reason that is out of order.For example, utilize the MSPC model that the fault score value of the corresponding processing step of each monitoring sample is drawn the multivariate statistics control chart, and corresponding given process step is switched demonstration.Described multivariate statistics control there is shown the fault score value of same processing step in each wafer etching process, has also marked the position of control limit, and the solid dot that surpasses this control limit is the trouble spot.

Exhibition Fig. 7 is the MSPC off-line analysis figure of a certain processing step in the embodiment of the invention.MSPC model that should processing step SPE, T have been used simultaneously ²With φ multivariate statistics control chart, because fault score value and the normal range value of trouble spot differ too much, so the control of each multivariate statistics control chart is limit near 0 position that is worth.But from any one control chart, can judge the trouble spot of breaking down, namely surpass the solid dot of control limit, and judged result be consistent.

In addition, owing to can calculate and analyze all data that receive under the off-line mode, therefore for the multivariate statistics control chart of other processing steps, off-line mode provides similar operation interface shown in Figure 5, can import the sequence number of processing step in Step ID, demonstration will be switched to multivariate statistics control chart that should given step in the below at interface.

Preferably, at each solid dot (can be called for short out-of-limit solid dot) that exceeds model control limit in the described multivariate statistics control chart, can also draw fault score contribution plot and the fault signature data statistic of the monitoring sample that produces this out-of-limit solid dot respectively.At which processing step fault has taken place though can qualitatively analyze have which wafer from above-mentioned multivariate statistics control chart, can't further analyze quantitatively is which variable of this processing step has caused this fault actually.Therefore, this step has further been drawn fault score contribution plot at the corresponding abnormal monitoring sample of out-of-limit solid dot.Described fault score contribution plot provides univariate fault score contribution margin and the corresponding univariate fault alarm type that the monitoring sample that produces this out-of-limit solid dot comprises, and therefore can analyze out of order basic reason from this figure.Simultaneously, in order to be illustrated more clearly in failure cause, the single argument compute statistics that can also comprise the monitoring sample that produces this out-of-limit solid dot in conjunction with univariate statistics process control (USPC) model, and utilize statistic to draw the fault signature data statistic of abnormal monitoring sample.Described fault score contribution plot and fault signature data statistic can be opened in the out-of-limit solid dot in double-clicking the multivariate statistics control chart and show.

Based on above analysis, described quantitative test failure cause specifically can comprise following processing:

At first, according to each statistic of a plurality of process datas of fault signature database and described each variable, obtain the fault signature data statistic; Described fault signature database is to set up based on USPC;

Wherein, preserved the tables of data of record product yield detailed technology index in the described fault signature database, these technical indicators are to draw on the basis of summing up the low reason of yield, described technical indicator comprises the Max_Limit described in the following table 3 (maximum control limit), Min_Limit (minimum control limit) etc., but does not comprise Mean (mean value), Max (maximal value), Min (minimum value), Stdev (standard deviation) and Range (scope).Each statistic of a plurality of process datas of described each variable refers to each variable is carried out the result of statistical computation, as calculating Min (minimum value), Max (maximal value), Mean (mean value), Stdev (standard deviation) etc.These statistics in conjunction with the every technical indicator in the fault signature database, just can be obtained the fault signature data statistic shown in the table 3.

Secondly, according to described fault signature data statistic and the fault alarm rule that presets, obtain fault score contribution plot, described fault score contribution plot has recorded fault score and the corresponding fault alarm type of each variable;

Described fault alarm rule refers to failure cause is carried out carrying out the rule that the fault alarm type is divided according to the order of severity, can determine according to practical application, for example the division rule shown in Xia Mian the table 4.

Again, in conjunction with described fault score contribution plot and described fault signature data statistic, quantitative test failure cause.

With reference to Fig. 8, it is MSPC fault score contribution plot in the embodiment of the invention; With reference to table 3, it is the USPC analytical table of MSPC fault.For certain the out-of-limit solid dot among Fig. 7, the position of this out-of-limit solid dot double-click can open Fig. 8 and below table 3.Wherein, fault score contribution plot shown in Figure 8 comprises two parts, shows SPE, T in the left side of figure ²With the single argument fault score value figure of φ, then show each univariate fault score contribution margin (Score) and corresponding fault and warning descriptor (Alarm) according to from high to low order on the right side of figure.In table 3, analyze the statistics that has provided each variable in conjunction with USPC.

ID	Parameter	Mean	Mar	Min	Std	Ranee	Max Limit	Minlimit	MeanPlus3Std	MeanMinus3Std	MeanOfModel	MarOfModel	StdOfModel
														1	rocessPressur	7.0329	7.3381	6.8055	0.0814	0.533	7.7	6.3	7.3166	6.7454	7.031	7.3671	6.6819
2	eliumPressur	7.8569	7.8569	7.8569	0	0	8.8	7.2	7.914	7.8969	7.9055	7.9622	7.8492
														3	HeliumLeakage	5.8299	8.2864	4.3677	1.1272	3.899	4	0.1	1.3258	1.3105	1.3181	1.458	1.2383
4	VatPosition	155.7	163	146	4.7322	17	1000	0	162.9542	137.537	150.2456	158	140
														5	SRFPower	347.8	348.2	347.3	0.2182	0.882	367.5	332.5	348.5093	347.4	347.9726	346.6326	347.1051
6	BRFPower	39.23	39.567	38.77	0.2055	0.796	42	38	39.7037	38.589	39.1463	39.6956	38.3832
														7	SRFRefl	0.7029	1.4191	0.2747	0.2737	1.144	5	0	1.4391	-0.081	0.6791	1.9913	0
8	BRFRefl	2.347	33.12	0.2747	5.6912	32.85	30	0	11.8074	-8.9096	1.4489	25.0401	0
														9	DCBias	75.604	83	4	19.297	79	91	0	143.0255	18.9791	81.0023	89	0
10	TopC1	58.7	58.7	58.7	0	0	1000	0	58.8591	58.5161	58.6876	59.2	57.9
														11	TopC2	59.1	59.1	59.1	0	0	1000	0	59.2456	59.0641	59.1548	59.7	58.2
12	BottomC1	48.513	50.3	37.3	3.9547	13	1000	0	60.8957	37.0584	48.9771	51.2	35.2
														13	BottomC2	39.825	53.7	38.5	3.546	15.2	1000	0	51.0239	29.1528	40.0884	54.7	38.4
14	GasCF4	50.08	50.08	50.08	0	0	52	48	50.0157	49.9982	50.007	50.09	49.91
														15	UCTemp	60.242	60.3	60.1	0.0739	0.199	65	55	60.1492	59.9703	60.0597	60.3	59.8
16	MCTemp	59.985	60	59.9	0.0357	0.1	65	55	60.0348	59.8652	59.95	60.1	59.8
														17	LCTemp	59.965	60	59.9	0.0357	0.1	65	55	60.0728	59.9048	59.988	60.1	59.8
18	ESCVoltage	1206.1	1207.1	1205.1	0.4523	1.007	1320	1080	1207.3025	1204	1205.7919	1208.6519	1204.303
														19	ESCTemp	65.238	65.36	65.19	0.047	0.17	68	62	65.4742	65.0136	65.2439	66.46	64.69

Table 3

Table 4

Based on above-mentioned Fig. 8 and table 3, table 4, the fault alarm analysis process is as follows:

Analysis with certain out-of-limit solid dot among Fig. 7 is example, in the fault score contribution plot of Fig. 8, the fault score of HeliumLeakage variable is the highest and make number one, and shows the possibility maximum that this variable breaks down, its fault and warning be described as " MeanAboveMaxLimit﹠amp; MeanAboveUpper3Std ".Further question blank 3, the mean value of this variable (Mean) is 5.8299 as can be seen, has surpassed the Maximum tolerance (Max_Limit) 4 of model specification, and the UCL of USPC system X-control chart control limit

I.e. 1.3258 (MeanPlus3Std).So corresponding tables 4, these data show that this fault has triggered 3 grades and 4 grades of warnings.In addition, the analytic process of other out-of-limit samples is the same.

Similar with top monitored off-line, multivariate multivariate statistics control chart that also can be by the MSPC model under the on-line monitoring pattern, fault score contribution plot and in conjunction with the fault signature data statistic of USPC, process data and analyze out of order basic reason notes abnormalities.But line model and off-line mode are different again, need to receive in real time the wafer sample data in the etching production run under the line model, carry out statistical computation and analysis, and in time send fault alarm information to control system, and idiographic flow as shown in Figure 9.

With reference to Fig. 9, it is the online course control method for use process flow diagram of MSPC in the described a kind of semiconductor technology of the embodiment of the invention.

Step 901 starts the Multivariable Statistical Process Control model that needs monitoring process step correspondence;

Under the on-line monitoring pattern, equally also needing to have pre-determined which processing step and need carry out on-line analysis, is that unit starts corresponding MSPC model then with the processing step.

Step 902 starts monitoring in real time;

Step 903 receives certain in real time and comprises a certain monitoring sample that needs the monitoring process step, and each monitoring sample comprises the described a plurality of process datas that need each variable in the monitoring process step;

Because the execution of processing step is order, so the reception of data also is the monitoring sample that receives only current processing step at every turn; When carrying out next processing step, if this processing step need be analyzed, then receive a plurality of process datas of each variable in this processing step again, and by MSPC model that should processing step is carried out statistical study.

Step 904 is screened pre-service to the monitoring sample that receives;

The processing of described screening pre-service and off-line model is similar, can utilize above-mentioned data screening device to screen, but only is that the data of the current processing step that receives are screened.

Step 905 according to described Multivariable Statistical Process Control model and the described a plurality of process datas that need each variable in the monitoring process step, is obtained the corresponding described fault score value that needs the monitoring process step;

After this processing step finishes, utilize corresponding MSPC model that the monitoring sample after screening under this processing step is calculated, comprehensively obtained the fault score value of this monitoring process step by a plurality of process datas of each variable in the need monitoring process step;

It is identical with the model calculating of off-line that described online model calculates, at this slightly.

Step 906 judges whether described fault score value exceeds the control limit of described Multivariable Statistical Process Control model, if, then quantitative test failure cause;

Namely can utilize corresponding MSPC model that the fault score value of need monitoring process step is drawn multivariate statistics control chart and demonstration.Multivariate statistics control chart and Fig. 7 of obtaining under line model are similar, but owing to online a plurality of processing steps of carrying out are in turn monitored, and are the figure of a dynamic change therefore.

Preferably, the solid dot that exceeds model control limit at fault score value in the described multivariate statistics control chart, can also be according to the failure cause quantitative test mode under the off-line mode, draw fault score contribution plot and the fault signature data statistic of monitoring sample respectively, the concurrent warning message that is out of order.

The fault score contribution plot that obtains under line model and fault signature data statistic also can be referring to Fig. 8 and table 3, at this slightly.

In addition, different with off-line mode is, also needs during on-line monitoring the fault of finding is in time reported to the police, so that the slip-stick artist in time makes corresponding measure, reduces the loss that fault is brought as far as possible.

Step 907 is waited for the monitoring sample of next processing step that receives the model appointment, and is returned step 903.

After this processing step finishes, continue to wait for the next execution that needs the processing step of monitoring, and utilize corresponding model to carry out the fault analysis of step 903 to 906.For example, suppose that the etching process of wafer relates to 10 processing steps, but only step 3, step 5 and step 9 are carried out on-line monitoring, then when carrying out step 3, utilize the MSPC model of corresponding step 3 that the sample data of step 3 is carried out fault analysis; When waiting until execution in step 5 then, the MSPC model of the corresponding step 5 of recycling carries out fault analysis to the sample data of step 5; At last, wait for again step 9 is carried out failure monitoring.

In sum, the MSPC method for supervising that provides of the embodiment of the invention has the following advantages:

Need to prove that above content only is that example describes with the semiconductor etching process, certainly, the present invention also is applicable to other semiconductor processes except etching technics.

Based on above content, the present invention also provides corresponding system embodiment.

With reference to Figure 10, it is the Process Control System structural drawing in the described a kind of semiconductor technology of the embodiment of the invention.

Described Process Control System comprises that mainly model imports module 1, data reception module 2, model computation module 3 and failure analysis module 4, and wherein, described model imports module 1 and is used for starting the Multivariable Statistical Process Control model that needs monitoring process step correspondence; Data reception module 2 is used for receiving the monitoring sample, and described monitoring sample comprises the described a plurality of process datas that need each variable in the monitoring process step; Model computation module 3 is used for obtaining the corresponding described fault score value that needs the monitoring process step according to described Multivariable Statistical Process Control model and the described a plurality of process datas that need each variable of monitoring process step; Failure analysis module 4 is used for judging whether described fault score value exceeds the control limit of described Multivariable Statistical Process Control model, if, then quantitative test failure cause.

Wherein, described Process Control System provides two kinds of monitoring modes, is respectively monitored off-line pattern and on-line monitoring pattern.Under off-line mode, at a plurality of processing steps that need in the technological process to monitor, it will be that unit imports a plurality of MSPC models with the step that model imports module 1; Data reception module 2 comprises a plurality of monitoring samples that need the monitoring process step with reception; At each processing step, failure analysis module 4 judges respectively whether each fault score value exceeds the corresponding control limit that needs the Multivariable Statistical Process Control model of monitoring process step, if, then quantitative test failure cause.

Under the line model, at a plurality of processing steps that need in the technological process to monitor, it also will be that unit imports a plurality of MSPC models with the step that model imports module 1; Data reception module 2 is then according to the execution sequence of processing step, receive the monitoring sample of certain monitoring step in real time, model computation module 3 and failure analysis module 4 utilize corresponding model to calculate and draw out multivariate statistics control chart that should processing step is shown, data reception module 2 is waited for the monitoring sample of (namely needing to monitor) next processing step that receives the model appointments then.In addition, described Process Control System can also comprise under the presence: the fault alarm module exceeds model control for the fault score value when multivariate statistics control chart monitoring sample and sends warning message in limited time.

Above-mentioned semiconductor processes control system can be utilized the incidence relation between a plurality of variablees, the unusual of monitoring sample reflected by the fault score value that is obtained by a plurality of aggregation of variable, therefore the monitoring to sample just can accurately analyze abnormal conditions by a multivariate statistics control chart of having drawn the fault score value, the fault erroneous judgement the problem includes: of having avoided univariate statistics process control (USPC) problem and the problem of failing to judge; And just can judge guilty culprit easily and quickly by multivariate statistics control chart intuitively, need not to switch loaded down with trivial detailsly and check a large amount of univariate statistics process control charts, alleviated operational load greatly, improve the control level, further improved production efficiency and the product yield of semiconductor equipment.

Based on above-mentioned semiconductor processes control system, in order further to analyze out of order basic reason, preferably, described failure analysis module 4 further can comprise: first analytic unit, be used for each statistic according to a plurality of process datas of fault signature database and described each variable, obtain the fault signature data statistic; Described fault signature database is to set up based on the univariate statistics process control technology; Second analytic unit is used for according to described fault signature data statistic and presets the fault alarm rule, obtains fault score contribution plot, and described fault score contribution plot has recorded fault score and the corresponding fault alarm type of each variable; Integerated analytic unit is used in conjunction with described fault score contribution plot and described fault signature data statistic, quantitative test failure cause.

That is: described Process Control System can also exceed each abnormal monitoring sample of model control limit at fault score value in the described multivariate statistics control chart, draws the fault score contribution plot of abnormal monitoring sample respectively; Wherein, described fault score contribution plot provides univariate fault score value and corresponding univariate fault and the warning message that the abnormal monitoring sample comprises, specifically can be referring to Fig. 8.

Simultaneously, described Process Control System can also be in conjunction with the USPC technology, each the abnormal monitoring sample that exceeds model control limit at fault score value in the described multivariate statistics control chart, the single argument compute statistics that the abnormal monitoring sample is comprised in conjunction with the univariate statistics process control model, and utilize described statistic to draw the fault signature data statistic of abnormal monitoring sample, specifically can be referring to table 3.The user double-clicks unusual process data position in the multivariate statistics control chart and can open fault score contribution plot and fault signature data statistic simultaneously, intuitively, clearly analyzes the variable break down and the basic reason of fault by this figure and table.

Preferably, for the quantitative analysis failure cause, described Process Control System can also combine described fault score contribution plot with the fault signature data statistic, according to failure cause described fault alarm information is carried out type and divides.This classification can reflect that the fault variable is by the situation of change of quantitative change to qualitative change, both realized capturing major failure fast in qualitative mode, realized again can doing further quantitative test to major failure, made fault analysis more comprehensively, the result has more actual directive significance.

Preferably, in order to remove the noise data in the monitoring sample, described Process Control System can also comprise data screening module 5, for the operation of reduced data screening module 5, can also comprise data screening device 6 simultaneously.It is the screening parameter value of all optional variablees under pre-configured each processing step of unit that described data screening device 6 is used for the processing step; Described data screening module 5 is used for according to model specified processing step and named variable, utilize the data screening device automatically in the processing step of model appointment, the monitoring sample is carried out the screening pre-service of named variable, and the monitoring sample after will screening passes to described model computation module, and the monitoring sample after described model computation module utilizes model to screening calculates.

Preferably, described Process Control System can also comprise: model building module 7 is used for setting up the Multivariable Statistical Process Control model.Described model building module 7 further can comprise:

The model dispensing unit, for the configuration information of setting modeling, described configuration information comprises processing step, data screening parameter, modeling variable, degree of confidence, the pivot computing method of appointment;

The data screening unit is used for according to data screening parameter and modeling variable the training sample under the given process step being screened; Preferably, described data screening unit can utilize data screening device 6 automatic in the processing step of appointment, training sample is carried out the screening pre-service of named variable;

The statistical computation unit is used for the training sample compute statistics after the screening;

The pre-established unit of model is used for utilizing statistic, degree of confidence and the pivot computing method of training sample to set up the Multivariable Statistical Process Control model, and the control of definite model limit;

The iteration modeling unit is used for judging whether institute's established model exists the training sample that surpasses the control limit, if there is no, then preserves the model of setting up; If exist, then remove out-of-limit sample, and utilize remaining training sample to carry out in conjunction with statistical computation unit and model pre-established unit circulation that described statistic is calculated, model is set up and judge, in institute's established model, there is not the training sample that surpasses the control limit.

In sum, the above-mentioned semiconductor processes control system that has comprised a plurality of preferred module can be based on the Multivariable Statistical Process Control model, and in conjunction with the univariate statistics process control model, directly perceived, succinct fault analysis and diagnosis is provided, further improved the control ability of semiconductor equipment.

Based on the said process control system, the embodiment of the invention also provides a kind of semiconductor control system that comprises this Process Control System.

With reference to Figure 11, it is the Organization Chart of the described a kind of semiconductor control system of the embodiment of the invention.

Described semiconductor control system comprises apparatus control system and Process Control System, described apparatus control system is used for the various sensor devices on the control production line, and for Process Control System provides Data Source, described Process Control System namely refers to system shown in Figure 10.

Figure 11 shows a kind of system diagram based on distributed design architecture, specifically comprises with lower module:

Clustering equipment control system (Cluster Tool Controller, CTC);

The transport module control system (Transfer Module Controller, TMC);

Technological process module controls system (Process Module Controller, PMC);

The factory automation module (Factory Automation, FA);

Data acquisition system (DAS) (Data Acquisition, DA);

The univariate statistics Process Control System (Univariate Statistical Process Control, USPC);

The Multivariable Statistical Process Control system (Multivariate Statistical Process Control, MSPC);

Real-time feedback control system (Run-to-Run Control, R2R);

Database (Data Base, DB).

Wherein, the upper level of control system be factory automation module FA, communicate by letter with factory Manufacturing Executive System MES (Manufacturing Execution System) by plant network; Next is clustering equipment control system CTC, embedded real-time feedback control system R2R among the CTC, same level also have data acquisition system (DAS) DA, USPC and MSPC, database D B; CTC is connecting the technological process module controls PMC of system of subordinate, and PMC is directly controlling the various sensor devices on the production line, as EndPoint Detector, V/I Probe, Other Hardware etc.; PMC and TMC carry out direct communication by LAN (Local Area Network) and CTC, R2R, DA, USPC and MSPC, database D B.

Based on above-mentioned distributed design architecture, because DA and PMC store all the monitoring samples in the wafer production run among the database D B, the MSPC system can directly be data source with DB, realizes data monitoring and fault analysis based on DB, to satisfy the requirement of control in real time.Therefore, described MSPC system is that a kind of multivariable process control based on the tool level is used, and described tool level namely refers to MSPC directly towards board control, can obtain real-time process data based on DB and realize data monitoring and fault analysis.

Each embodiment in this instructions all adopts the mode of going forward one by one to describe, and what each embodiment stressed is and the difference of other embodiment that identical similar part is mutually referring to getting final product between each embodiment.For system embodiment, because it is similar substantially to method embodiment, so description is fairly simple, relevant part gets final product referring to the part explanation of method embodiment.

More than to course control method for use and system in a kind of semiconductor technology provided by the present invention, and the semiconductor control system that comprises this Process Control System, be described in detail, used specific case herein principle of the present invention and embodiment are set forth, the explanation of above embodiment just is used for helping to understand method of the present invention and core concept thereof; Simultaneously, for one of ordinary skill in the art, according to thought of the present invention, the part that all can change in specific embodiments and applications, in sum, this description should not be construed as limitation of the present invention.

Claims

1. the course control method for use in the semiconductor technology is characterized in that, comprising:

Judge whether described fault score value exceeds the control limit of described Multivariable Statistical Process Control model, if, then quantitative test failure cause;

Described quantitative test failure cause comprises:

According to each statistic of a plurality of process datas of fault signature database and described each variable, obtain the fault signature data statistic; Described fault signature database is to set up based on the univariate statistics process control technology;

According to described fault signature data statistic and preset the fault alarm rule, obtain fault score contribution plot, described fault score contribution plot has recorded fault score and the corresponding fault alarm type of each variable;

In conjunction with described fault score contribution plot and described fault signature data statistic, quantitative test failure cause.

2. method according to claim 1 is characterized in that, under off-line state, comprising:

Reception comprises a plurality of monitoring samples that need the monitoring process step;

Each needs the fault score value of monitoring process step to obtain correspondence respectively;

Judge respectively whether each fault score value exceeds the corresponding control limit that needs the Multivariable Statistical Process Control model of monitoring process step, if, then quantitative test failure cause.

3. method according to claim 1 is characterized in that, under presence, described reception monitoring sample is specially: reception in real time comprises a certain monitoring sample that needs the monitoring process step.

4. according to claim 2 or 3 described methods, it is characterized in that, by adding up the same fault score value that needs the monitoring process step in each silicon chip etching technology process, set up the multivariate statistics control chart.

5. according to the arbitrary described method of claim 1 to 3, it is characterized in that described startup needs also to comprise: set up the Multivariable Statistical Process Control model, specifically comprise before the Multivariable Statistical Process Control model of monitoring process step correspondence:

Set the configuration information of modeling, described configuration information comprises processing step, data screening parameter, modeling variable, degree of confidence, the pivot computing method of appointment;

Screen according to data screening parameter and the modeling variable a plurality of process datas to each variable in the processing step specified in each training sample;

To the training sample compute statistics after the screening;

Utilize statistic, degree of confidence and the pivot computing method of training sample to set up the Multivariable Statistical Process Control model, and the control of definite model limit;

Judge whether there is the training sample that surpasses the control limit in institute's established model, if there is no, then preserve the model of setting up; If exist, then remove out-of-limit sample, and described statistic calculating is carried out in the remaining training sample circulation of utilization, model is set up and determining step, does not have the training sample above the control limit in institute's established model.

6. method according to claim 5 is characterized in that, described screening comprises:

Utilize the data screening device automatically in the processing step of appointment, training sample is carried out the screening pre-service of named variable; Wherein, described data screening device with the processing step be unit pre-configured the screening parameter value of all optional variablees under each processing step.

7. the Process Control System in the semiconductor technology is characterized in that, comprising:

Failure analysis module is used for judging whether described fault score value exceeds the control limit of described Multivariable Statistical Process Control model, if, then quantitative test failure cause;

Described failure analysis module comprises:

First analytic unit is used for each statistic according to a plurality of process datas of fault signature database and described each variable, obtains the fault signature data statistic; Described fault signature database is to set up based on the univariate statistics process control technology;

Second analytic unit is used for according to described fault signature data statistic and presets the fault alarm rule, obtains fault score contribution plot, and described fault score contribution plot has recorded fault score and the corresponding fault alarm type of each variable;

Integerated analytic unit is used in conjunction with described fault score contribution plot and described fault signature data statistic, quantitative test failure cause.

8. system according to claim 7 is characterized in that, also comprises:

Model building module is used for setting up the Multivariable Statistical Process Control model.

9. system according to claim 8 is characterized in that, described model building module comprises:

The data screening unit is used for screening according to data screening parameter and the modeling variable a plurality of process datas to each variable in the specified processing step of each training sample;

10. according to claim 7 or 9 described systems, it is characterized in that, also comprise:

The data screening device, being used for the processing step is the screening parameter value of all optional variablees under pre-configured each processing step of unit;

The data screening module, be used for according to model specified processing step and named variable, utilize the data screening device automatically in the processing step of model appointment, the monitoring sample is carried out the screening pre-service of named variable, and the monitoring sample after will screening passes to described model computation module, and the monitoring sample after described model computation module utilizes model to screening calculates.

11. a semiconductor control system comprises apparatus control system and Process Control System, it is characterized in that: described Process Control System comprises the arbitrary described Process Control System of claim 7 to 10.

12. system according to claim 11 is characterized in that, also comprises:

Data acquisition system (DAS) is used for collecting device control system institute control equipment at the monitoring sample of art production process, and stores in the database;

Database is used for storage monitoring sample.

13. system according to claim 12 is characterized in that:

Described Process Control System is data source with described database, directly receives the monitoring sample from this database.