CN100354776C - Semiconductor run-to-run control system with state and model parameter estimation - Google Patents

Abstract

A method for a run-to-run (R2R) control system includes processing materials using a process input and producing a process output, storing the process input in a database, the storing including using a timestamp, and storing at least one measurement of the process output in the database aligned with each process input using the timestamp. The method further includes iterating over the data in the database to estimate one or more coefficients for a model, and, if one or more measurements is missing, replacing the missing measurements based on a prediction from said model. The model is updated with said coefficient estimates. The method additionally includes iterating over the data from the database to estimate a process state, and, if one or more of the measurements is missing from the database, replacing the missing measurements based on prediction from the model. The model is updated with said process state estimate. A controller may receive the updated model and utilize the model to produce the next process input. The updated model may also be utilized to generate an estimate for a measurable process variable, wherein the estimate can be compared to an actual measurement to determine if the estimate is within confidence limits. If the estimate is not within confidence limits, a fault is indicated.

Description

Semiconductor run-to-run control system with state and model parameter estimation

Technical field

The present invention relates to the semiconductor processes field, relate in particular to batch (the Runto Run control) control in the semiconductor processes.

Background technology

Batch (R2R) control in the semiconductor processes and fault detect (FDC) refer to come the more practice of new equipment prescription setting based on the product measurement that takes place after finishing dealing with.This practice is applied in the semi-conductor industry because of measurement products characteristic at the scene.For example, in chemical-mechanical planarization (CMP) process, the liner rotational speed, compelling force and other parameters are measurable, and film thickness, the actual product variable is immeasurablel.Similarly, in photoetching technique, exposure power, time shutter and focus can determine that still actual overlay errors cannot be measured after resist develops, detachment process for a long time just takes place after photolithographic exposure.Other example exists, but main application is present in photoetching technique (covering and critical dimension (CD)), among plasma etching and the CMP.

The best method that realizes effective R2R control is the long method that postpones to use reliable process model of measuring because of often being associated with these realizations.But the common issue with among the R2R of these processes control and the FDC is, model coefficient be not well-known and the experiment of refining value too arduous.Come the demand of the method for while estimation procedure state and refined model coefficient to exist for using the data of collecting in the closed loop controller operating process.What complicate the issue is, some in some in the state variable and the model coefficient are with the fact of message reflection to the different component of problem.For example, some coefficients may be that product is relevant, but are not that instrument is relevant.In the state variable some may reflect the state in a certain respect of instrument, and other may reflect consumables or helper component, for example contribution of graticule.The method that adapts to these demands of effective R2R control aspect will solve many manufacturing difficulties of making in the semi-conductive process.

In the prior art, the state and the parameter estimation method of combination also were not available during batch control was used.The whole bag of tricks that relates to the problem part is known.In the prior art, state estimation is used for estimating tool state from the data of measuring, and suppose that product contribution is known, constant and is included in the model.But if model comprises some coefficient errors, inevitably accident when product changes, uses control system control production to cause that more than the process of a product tangible interference changes so.Product in the typical integrated circuit production line changes basically every batch of wafer and takes place.Therefore, the Interference Estimation transmission of constant model highly depends on the performance of model coefficient accuracy.

In the prior art, other realize a kind of estimator of adaptive form, and it estimates the model coefficient of characterization tool current state, but the additional interference that is provided by process tool (process tool) and consumables in the process is provided.This method is introduced the problem identical with independent Interference Estimation.Controller with all errors owing to model coefficient.Because different product has different model coefficients, and instrument disturbs with the product contribution in the coefficient and mixes, and when product was changed aborning, error propagation appeared in the output as stochastic error.But error is not at random, but instead owing to system's specification error reason correctly.

What need is that a kind of ability in conjunction with implementation additivity Interference Estimation and model parameter estimation makes source of error correctly designated, the method that allows product to change with the minimum system disturbance.What further need is the method for this problem, and its (i) is used in combination parameter and the state/parameter estimation of making data, with while refinement and submodel contribution; And (ii) estimate tool state.

Summary of the invention

Having reliable process model postpones for long measurement of realizing that effective batch of (R2R) control system is absolutely necessary, especially considering being associated with this system.Come the demand of the method for estimation procedure state and refined model coefficient to exist for using the data of in the closed loop controller operating process, collecting.

According to the present invention, provide a kind of state and parameter estimation method of combination in R2R control is used.Especially, the method that is used to control manufacture process comprises: material and production process output are handled in the use input, the process input is stored in the database, this storage comprises mark service time, and at least one measurement mark service time and the input of each process of process output is stored in the database alignedly.This method also is included in iteration on the data in the database, with one or more coefficients of estimation model, and if one or more measurement lose, replace based on prediction and lose measurement from described model.Model upgrades with described coefficient estimation.This method also is included in addition from iteration on the data of database, with the estimation procedure state, and if one or morely from database, losing of measuring, replace the measurement of losing of database based on the prediction that comes self model.Model is estimated to upgrade with described process status.Losing the replacement of measurement in parameter or state estimation procedure can impliedly or clearly finish.Controller can receive the model of renewal and utilize model to produce next process input.But whether the estimation that the model that upgrades also can be used for producing measured process variable is wherein estimated to compare with actual measurement to determine to estimate in fiducial limit.If estimate that fault is pointed out not in fiducial limit.

According to a kind of embodiment, this method comprises the one or more modules that are connected to database, these one or more modules comprise sort module at least, it is configured to will be from the measurement classification of the asynchronous reception of process according to time mark, the measurement that is classified comprises the measurement that arrives later, measures the error of calculating to allow next process to input to small part based on using the back to use.Sort module is also with the data qualification in the database, so that use in parameter estimation or state estimation procedure.

If actual measurement becomes available one period after model coefficient estimation or process status estimation, for one or more measurements of losing, the prediction that previously lost is measured is measured with actual measurement and is replaced, and actual measurement is stored in the database.In model coefficient estimation and process status estimation subsequently, actual measurement replaces the prediction measurement and uses.Wait for that may be variable the period that actual measurement becomes available in the present invention.

In one embodiment, state estimator wave filter and parameter estimator wave filter comprise the independent rolling time domain filtering (receding horizonfilter) that uses Constraint least square algorithm calculating to estimate.Least square optimization need be found the solution the quadratic equation program.

The invention provides a kind of method that is used to control manufacture process, comprising: the use input is handled and production process output; The process input is stored in the database; Explicitly be stored in database with each process input at least one measurement of process output; Iteration is with the estimation procedure state on from data of database; If one or more measurements are lost, then lose measurement based on the database replacement that is predicted as that comes self model from database; And with the described model of described process status estimation renewal.

According to said method of the present invention, if wherein actual measurement becomes available one period after process status is estimated, then for described one or more measurements of losing, replacing prediction with described actual measurement measures, described actual measurement is stored in the described database, and in process status is subsequently estimated, utilizes described actual measurement.

According to said method of the present invention, its middle controller receives the model that upgrades, and utilizes the model of described renewal to produce next process input.

According to said method of the present invention, its middle controller comprises Model Predictive Control (MPC) type controllers, direct modeling inverter type controller, and one or more in the simple integral controller.

According to said method of the present invention, the estimation of its middle controller use state makes the error minimize in each batch of process.

According to said method of the present invention, wherein this process is operated at least one semiconductor devices.

According to said method of the present invention, wherein database is coupled to one or more modules of operating on database.

According to said method of the present invention, wherein module comprises estimation module, computing module, sort module, matching module and data memory module.

According to said method of the present invention, wherein manufacture process is one or more batch (R2R) process in the following process: chemical-mechanical planarization (CMP) process, photoetching process and plasma etch processes.

According to said method of the present invention, be variable wherein said period.

According to said method of the present invention, wherein the described storage of the process of database input comprises provides time mark, and described at least one measurement mark service time of wherein process output is associated with each process input.

The present invention also provides a kind of control system, comprising: controller can provide the process input; Process is coupled to controller and imports and provide process output with receiving course; Database is configured to the measurement of receiving course input and process output, and database can be associated each process input and the output of each process; Model, output can supply a model; And state estimator, be coupled to and receive database output and model output, state estimator is connected to controller and can the production process state estimation, and be configured to based on described process status estimate will renewal model be provided to described controller.

According to above-mentioned control system of the present invention, wherein this process is operated at least one semiconductor devices.

According to above-mentioned control system of the present invention, its middle controller comprises following Model Predictive Control (MPC) type controllers, direct modeling inverter type controller, and one or more in the simple integral controller.

The invention provides a kind of device, comprise being used for the instrument that the use input is handled and production process is exported; Be used for the process input is stored in the instrument of database, this instrument that is used to store comprises provides time mark; At least one measurement mark service time that is used for process is exported and the input of each process are stored in the instrument of database alignedly; Be used on iteration with the instrument of estimation procedure state from data of database; Be used for replacing the instrument of the measurement of losing from database based on the prediction that comes self model; And be used to use described process status to estimate to upgrade the instrument of described model.

The invention provides a kind of method that is used to control manufacture process, comprising: the use input is handled and production process output; The process input is stored in the database; Explicitly be stored in database with each process input at least one measurement of process output; Iteration is wherein found the solution the constraint quadratic programming with the estimation procedure state when each iteration of data on from data of database; If one or more measurements are lost, lose measurement based on the database replacement that is predicted as that comes self model from database; And with the described model of described process status estimation renewal.

According to said method of the present invention, if wherein actual measurement becomes available one period after process status is estimated, then for described one or more measurements of losing, replacing prediction with described actual measurement measures, described actual measurement is stored in the described database, and in process status is subsequently estimated, utilizes described actual measurement.

According to said method of the present invention, its middle controller receives the model that upgrades, and utilizes the model of described renewal to produce next process input.

According to said method of the present invention, be variable wherein said period.

According to said method of the present invention, wherein the described storage of the process of database input comprises provides time mark, and described at least one measurement mark service time of wherein process output is associated with each process input.

The invention provides a kind of method that is used to control manufacture process, comprising: the database with process input data and relevant process output data is provided; Iteration is with the estimation procedure state on from data of database; If one or more measurements are lost, then lose measurement based on the database replacement that is predicted as that comes self model from database; And with the described model of described process status estimation renewal.

According to said method of the present invention, if wherein actual measurement becomes available one period after process status is estimated, then for described one or more measurements of losing, replacing prediction with described actual measurement measures, described actual measurement is stored in the described database, and in process status is subsequently estimated, utilizes described actual measurement.

According to said method of the present invention, its middle controller receives the model that upgrades, and utilizes the model of described renewal to produce next process input.

According to said method of the present invention, be variable wherein said period.

According to said method of the present invention, wherein said process output data mark service time is associated with described process input data.

According to said method of the present invention, find the solution the constraint quadratic programming when also being included in the each iteration of described data.

The invention provides a kind of controller that is used for manufacture process, comprising: be used for the device that the use input is handled and production process is exported; Be used for the process input is stored in the device of database; At least one that is used for process output measured the device that is stored in database with each process input explicitly; Be used on iteration with the device of estimation procedure state from data of database; Lose from database if be used for one or more measurements, then lose the device of measurement based on the database replacement that is predicted as that comes self model; And the device that is used for estimating the described model of renewal with described process status.

The invention provides a kind of controller that is used for manufacture process, comprising: the device that is used for the process input is stored in database; At least one that is used for process output measured the device that is stored in database with each process input explicitly; Be used on iteration and wherein when each iteration of data, find the solution the constraint quadratic programming with the device of estimation procedure state from data of database; Lose from database if be used for one or more measurements, then lose the device of measurement based on the database replacement that is predicted as that comes self model; And the device that is used for estimating the described model of renewal with described process status.

The invention provides a kind of controller that is used for manufacture process, comprising: the device that is used to provide database with process input data and relevant process output data; Be used on iteration with the device of estimation procedure state from data of database; Lose from database if be used for one or more measurements, then lose the device of measurement based on the database replacement that is predicted as that comes self model; And the device that is used for estimating the described model of renewal with described process status.

The invention provides a kind of method that is used to control manufacture process, comprising: the use input is handled and production process output; The process input is stored in the database; Explicitly be stored in database with each process input at least one measurement of process output; Iteration is to estimate to be used for one or more coefficients of model on from data of database; If one or more measurements are lost, then lose measurement based on the database replacement that is predicted as from described model from database during model coefficient is estimated; Upgrade described model with described coefficient estimation; Iteration is with the estimation procedure state on from data of database; If one or more measurements are lost, then lose measurement based on the database replacement that is predicted as from described model from database during process status is estimated; And estimate to upgrade described model with described process status.

According to said method of the present invention, if wherein actual measurement becomes available one period after model coefficient estimation or process status estimation, then for described one or more measurements of losing, replace the prediction measurement that previously lost is measured with described actual measurement, described actual measurement is stored in the described database, and subsequently model coefficient estimate and the process status estimation in utilize described actual measurement.

According to said method of the present invention, its middle controller receives the model that upgrades, and utilizes the model of described renewal to produce next process input.

According to said method of the present invention, its middle controller comprises Model Predictive Control (MPC) type controllers, direct modeling inverter type controller, and one or more in the simple integral controller.

According to said method of the present invention, the estimation of its middle controller use state makes the error minimize in each batch of process.

According to said method of the present invention, wherein this process is operated at least one semiconductor devices.

According to said method of the present invention, wherein database is coupled to one or more modules of operating on database.

According to said method of the present invention, wherein module comprises estimation module, computing module, sort module, matching module and data memory module.

According to said method of the present invention, wherein said model coefficient estimation and described process status are estimated to comprise provides fiducial limit.

According to said method of the present invention, comprise that also but the model that utilizes described renewal produces the estimation to measured process variable, and actual measurement is compared with described estimation to determine that described estimation is whether in described fiducial limit, if wherein described estimation not in described fiducial limit, indication fault then.

According to said method of the present invention, be variable wherein said period.

According to said method of the present invention, wherein the replacement of losing measurement of database is finished clearly or impliedly.

According to said method of the present invention, find the solution the constraint quadratic programming when also being included in the each iteration of described data.

According to said method of the present invention, wherein the described classification of process input comprises provides time mark, and described at least one measurement mark service time of wherein process output is associated with each process input.

The invention provides a kind of method that is used to control manufacture process, comprising: the use input is handled and production process output; The process input is stored in the database; Explicitly be stored in database with each process input at least one measurement of process output; Iteration is wherein found the solution the constraint quadratic programming to estimate to be used for one or more coefficients of model when each iteration of described data on from data of database; If one or more measurements are lost, then lose measurement based on the database replacement that is predicted as from described model from database during model coefficient is estimated; Upgrade described model with described coefficient estimation; Iteration is wherein found the solution the constraint quadratic programming with the estimation procedure state when each iteration of described data on from data of database; If one or more measurements are lost, lose measurement based on the database replacement that is predicted as from described model from database during process status is estimated; And with the described model of described process status estimation renewal.

According to said method of the present invention, if wherein actual measurement becomes available one period after model coefficient estimation or process status estimation, then for described one or more measurements of losing, use described actual measurement to replace the prediction measurement that previously lost is measured, described actual measurement is stored in the described database, and subsequently model coefficient estimate and the process status estimation in utilize described actual measurement.

According to said method of the present invention, its middle controller receives the model that upgrades, and utilizes the model of described renewal to produce next process input.

According to said method of the present invention, its middle controller comprises Model Predictive Control (MPC) type controllers, direct modeling inverter type controller, and one or more in the simple integral controller.

According to said method of the present invention, the estimation of its middle controller use state makes the error minimize in each batch of process.

According to said method of the present invention, wherein this process is operated at least one semiconductor devices.

According to said method of the present invention, wherein database is coupled to one or more modules of operating on database.

According to said method of the present invention, wherein module comprises estimation module, computing module, sort module, matching module and data memory module.

According to said method of the present invention, wherein said model coefficient estimation and described process status are estimated to comprise provides fiducial limit.

According to said method of the present invention, comprise that also but the model that utilizes described renewal produces the estimation that is used for measured process variable, and actual measurement is compared with described estimation to determine that described estimation is whether in described fiducial limit, if wherein described estimation not in described fiducial limit, indication fault then.

According to said method of the present invention, be variable wherein said period.

According to said method of the present invention, wherein the replacement of measuring for database lost is finished clearly or impliedly.

According to said method of the present invention, wherein the described storage of the process of database input comprises provides time mark, and described at least one measurement mark service time of wherein process output is associated with each process input.

The invention provides a kind of method that is used to control manufacture process, comprising: the database with process input data and relevant process output data is provided; Iteration is with one or more coefficients of estimation model on from data of database; If one or more measurements are lost, then lose measurement based on the database replacement that is predicted as from described model from database during model coefficient is estimated; Upgrade described model with described coefficient estimation; Iteration is with the estimation procedure state on from data of database; If one or more measurements are lost, then lose measurement based on the database replacement that is predicted as from described model from database during process status is estimated; And with the described model of described process status estimation renewal.

According to said method of the present invention, if wherein actual measurement becomes available one period after model coefficient estimation or process status estimation, then for described one or more measurements of losing, use described actual measurement to replace the prediction measurement that previously lost is measured, described actual measurement is stored in the described database, and subsequently model coefficient estimate and the process status estimation in utilize described actual measurement.

According to said method of the present invention, wherein said model coefficient estimation and described process status are estimated to comprise provides fiducial limit.

According to said method of the present invention, comprise that also but the model that utilizes described renewal produces the estimation to measured process variable, and actual measurement is compared with described estimation to determine that described estimation is whether in described fiducial limit, if wherein described estimation not in described fiducial limit, indication fault then.

According to said method of the present invention, be variable wherein said period.

According to said method of the present invention, wherein the replacement that database lost is measured is finished clearly or impliedly.

According to said method of the present invention, find the solution the constraint quadratic programming when also being included in the each iteration of described data.

The invention provides a kind of controller that is used for manufacture process, comprising: be used for the device that the use input is handled and production process is exported; Be used for the process input is stored in the device of database; At least one that is used for process output measured the device that is stored in database with each process input explicitly; Be used on from data of database iteration device with one or more coefficients of estimating to be used for model; Lose from database during model coefficient is estimated if be used for one or more measurements, then lose the device of measurement based on the database replacement that is predicted as from described model; Be used for upgrading the device of described model with described coefficient estimation; Be used on iteration with the device of estimation procedure state from data of database; Lose from database during process status is estimated if be used for one or more measurements, then lose the device of measurement based on the database replacement that is predicted as from described model; And the device that is used for estimating to upgrade described model with described process status.

The invention provides a kind of controller that is used for manufacture process, comprising: be used for the device that the use input is handled and production process is exported; Be used for the process input is stored in the device of database; At least one that is used for process output measured the device that is stored in database with each process input explicitly; Be used on from data of database iteration device, wherein when each iteration of described data, find the solution the constraint quadratic programming with one or more coefficients of estimating to be used for model; Lose from database during model coefficient is estimated if be used for one or more measurements, then lose the device of measurement based on the database replacement that is predicted as from described model; Be used for upgrading the device of described model with described coefficient estimation; Be used on iteration and wherein when each iteration of described data, find the solution the constraint quadratic programming with the device of estimation procedure state from data of database; Lose from database during process status is estimated if be used for one or more measurements, then lose the device of measurement based on the database replacement that is predicted as from described model; And the device that is used for estimating the described model of renewal with described process status.

The invention provides a kind of controller that is used for manufacture process, comprising: the device that is used to provide database with process input data and relevant process output data; Be used on iteration with the device of one or more coefficients of estimation model from data of database; Lose from database during model coefficient is estimated if be used for one or more measurements, then lose the device of measurement based on the database replacement that is predicted as from described model; Be used for upgrading the device of described model with described coefficient estimation; Be used on iteration with the device of estimation procedure state from data of database; Lose from database during process status is estimated if be used for one or more measurements, then lose the device of measurement based on the database replacement that is predicted as from described model; And the device that is used for estimating the described model of renewal with described process status.

Description of drawings

The present invention can be by understanding with reference to appended drawings better, and its many purposes, feature and advantage become obvious to those skilled in the art.Spread all over the use of the identical parameters numeral of a few width of cloth figure and represent same or analogous element.

Fig. 1 is labeled as " prior art ", and the exemplary process system of the semiconductor wafer of being carried out by semiconductor tools is described.

Fig. 2 is the block diagram of control system according to embodiments of the present invention.

Fig. 3 illustrates the sequential chart of the operation of state/parameter estimator according to embodiments of the present invention.

Fig. 4 is the process flow diagram of method according to embodiments of the present invention.

Fig. 5 explanation confidence region during fault detect according to embodiments of the present invention.

Embodiment

I. summarize

The embodiment of the present invention that describes below provides may be according to being implemented to the feature that concrete exploitation purpose, commercial interest and system requirements that another kind of realization may have nothing in common with each other change or develop from a kind of.Any this change that should be appreciated that feature of the present invention will be a daily task of benefiting from those skilled in the art disclosed by the invention.

The invention provides a kind of method and system that is used for effectively and accurately handling material, it is handled and measures from material processed, for example collects asynchronously in semiconductor processes.The present invention also provides a kind of state and parameter estimation method of combination in R2R control is used.

With reference to figure 1, the embodiment procedure for displaying of prior art control system input 106, it may be starting material, the incomplete material of handling, other processes input, for example semiconductor materials of perhaps handling on process tool 104.Controller 102 is connected to process tool 104 so that one or more control input signals to be provided.The material of process tool 104 processing procedures input 106 and output procedure output 108.

With reference now to Fig. 2,, shows according to a kind of R2R control system of the present invention.Control system 200 comprises the process 210 that is used to handle the material that is connected to controller 202.Controller 202 receiving course set-point r212 and output calculation process input u214.Process 210 receives actual process input v218, and it comprises the calculation process input 214 that the input that is caused by the operator disturbs Δ u216 to change.Actual process input v218 is further received by I/O thesaurus 208, and this I/O thesaurus 208 may be embodied as database.

Process 210, the product after for example process tool 104 is handled material and exported the processing that needs to measure.Process 210 is transmitted the parameter of product and product.The output of process tool 210 is measured.This measurement is shown as piece 224 and comprises: (i) measurement of carrying out at the scene; Soon the measurement of carrying out after (ii) handling; And the measurement of (iii) finishing later.Operation is with via transport function on measuring for control system 200, and for example disturbance transfer function matrix 220 increases and disturbs d222.Process measurement is shown as process output y226 and further is provided to I/O thesaurus 208.Importantly, measuring y226 collects in the processing of the real material from process tool 210 usually asynchronously.Therefore, this input measurement to process instrument 210 can be shown as dotted line.

I/O thesaurus 208 is provided to model 262 and state/parameter estimator 206 with data 228.I/O thesaurus 208 can be implemented as the input in accumulation past and the database of measurement.I/O thesaurus 208 also comprises the sorter that is used for according to process time labeled bracketing input and measurement, and handles and lose the rolling time domain filtering (receding horizonfilter) of measurement or suitably replace wave filter.Sorter also grouped data so that in state estimation or parameter estimation procedure, use.State/parameter estimator 206 also receives data and the model that upgrades is provided to controller 202 from model 262.Controller 202 is unlike controller 102, based on the Model Calculation process input of upgrading.Model is estimated to upgrade based on coefficient and the process status calculated by state/parameter estimator 206.

As described in more detail below, state/parameter estimator 206 uses more than the estimator of a wave filter as model 262 different components.State/parameter estimator 206 uses a wave filter to be used for parameter estimation and another wave filter is used for state estimation.

In the operation, between process batch (run), can not carry out by survey instrument 224 from having on the product of the process in past batch, one or a plurality of measurement.

That be connected to I/O thesaurus 208 is several module 250-258.Module comprises the data memory module 250 of catching and store the process output vector of measuring from the real process vector sum of I/O thesaurus 208.Matching module 252 is matched process input and output vector according to the process time mark of process time mark or a plurality of predictions.Sort module 254 is right according to process time labeled bracketing I/O vector.Sort module 254 also grouped data so that in state estimation or parameter estimation procedure, use.The process output of computing module 256 use Model Calculation predictions or the output of a plurality of process, and estimation module 258 is from the process input, the prediction of output and actual output vector are come the estimation procedure state vector.In addition, estimation module 258 is from process output, and the model prediction of output and actual output vector are estimated the model coefficient that upgrades.

Especially, data memory module 250 and matching module 252 comprise catches real process input v218, the time mark of additional representation process 210 beginnings, v218 stores in the right sorted table of process I/O to be kept in the I/O thesaurus 208 with input, catch product and measure y226, and with they with I/O thesaurus 208 in sorted table in the respective process input v218 routine of uniting.

I/O thesaurus 208 can be implemented as the right database of coordination I/O that is used to store the per unit product.In semiconductor processing environment, product unit can be implemented as a large amount of wafers or single wafer, depends on the particular scene that is used for state/parameter estimator 206 and the control granularity (granularity of control) of configuration parameter.In addition, length of field when configuration parameter may comprise (horizon length), the cut-off date of measurement and weight matrix.The cut-off date of measuring is that the time mark that all processes are imported and product is measured is ignored by system 200 before this.More configuration parameter is described below.

II. process model

Process model 262 can be as the sign of getting off:

x _k+1＝A ^Jx _k+B ^Ju _k+G ^Jd _k

d _k+1＝αd _k+w _k (1)

$y_k^j=C^j{x}_k+v_k$

The explanation that appears at the symbol in the equation (1) appears in the following form 1.The selection of α and the model of w allow those skilled in the art to mould the potential interference model of system.Form 2 provides the various selections for α and w, consequent interference model, and it is a white-noise process.In addition, interference may comprise several subvectors, and each subvector comprises about instrument, durable (for example graticule) or the gas producing formation information for the additional contribution of state model.Requiring the modification of these equatioies of contributing separately (1) of adaptation will be known for those skilled in the art.Matrix A, B and G are known from process is considered usually.But the coefficient in the Matrix C often shows the contribution of different product for the process result who measures in output variable y.Therefore, depend on particular procedure, the unknown disturbances that in variable d, embodies, and the unknown model coefficient that embodies in C may exist.

Form 1: the symbol of process model equation

Symbol	Definition
		k	Criticize or wafer index (counter)
x	The process status vector
		u	The process input vector
d	Process status additional interference vector
		y	Process output vector
w	The process status noise vector is assumed to be the white noise of normal distribution usually
		v	Process output noise vector is assumed to be the white noise of normal distribution usually
j	Product index (identifier)
		A	The process dynamic matrix
B	The input gain matrix
		G	State obstacle gain matrix
α	The interfere with dynamic matrix
		C	The output coefficient matrix

Form 2: use interfere with dynamic matrix design interference model

α	Consequent d
		0＜|α|＜1	Steady single order with white noise is dynamic
1	The white noise of integration
		0	White noise

III. controller

With reference to controller 202 discussed above, the realization of this controller may comprise the controller of several types, comprises Model Predictive Control (MPC) type controllers, direct modeling inverter type controller and simple pi controller.

IV. state estimation

In one embodiment, the state estimator wave filter of state/parameter estimator 206 is to use the rolling time domain filtering (receding horizon filter) that Constraint least square algorithm computing mode and output error are estimated.This least square optimization need be found the solution quadratic equation program (QP).Equation (2) provides the fundamental formular of the least square problem of estimator and represents.

$\min J=\underset{i=-1}{\overset{N-1}{\mathrm{\Σ}}}{w^{\′}}_i{Q}_i{w}_i+\underset{i=0}{\overset{N}{\mathrm{\Σ}}}{v^{\′}}_i{R}_i{v}_i$

{w _i，v _i}

Condition: $x_0={\stackrel{\‾}{x}}_0+w_{-1}$

x _i+1＝A _i+1x _i+B _i+1u _i+w _i i＝0，N-1 (2)

y _i＝C _ix _i+v _i i＝0，N

w _min≤w _i≤w _max

f _min≤Fx _i≤f _max

Variable J is a target function value, w _iBe that the middle i+1 of time domain (horizon) criticizes the state error of (lot), v _iBe the model prediction that i criticizes in the time domain (horizon) and measure between error, Q _i, R _iBe the weighting matrix that i criticizes in the time domain (horizon),

Be in the time domain (horizon) first, or the starting condition of the state of wafer, x _iBe that i criticizes in the time domain (horizon), or the state estimation of wafer, y _iBe that i criticizes in the time domain (horizon), or the measurement of wafer, A _I+1, B _I+1, C _I+1Be the state space matrices that i+1 criticizes in the time domain (horizon), and w _Min, w _Max, f _Min, F, f _MaxConstraint on the definition status sum of errors state estimation.Variable w _i, w _Min, w _Max, f _Min, x _iAnd f _MaxBe n _sThe vector of * 1 dimension, wherein n _sIt is state number.Variable u _iBe n _uThe vector of * 1 dimension, wherein n _uIt is the number of process input.Variable v _iAnd y _iBe n _yThe vector of * 1 dimension, wherein n _yIt is the number of process output.By algebraic operation, formulate is as follows again for equation (2):

min J＝w′(Q+M′ _ARM _A)w-2Y′RM _Aw+Y′RY (3)

{w}

Condition: $F_{\min }\≤F_M(F_{A}w+F_{B}u+F_C{\stackrel{\‾}{x}}_0)\≤F_{\max }$

Wherein, the following definition in (4) of the parameter in the objective function of equation (3):

$w=\left[\begin{array}{c}{w}_{-1}\\ w_0\\ w_1\\ w_2\end{array}\right]$ $u=\left[\begin{array}{c}0\\ u_0\\ u_1\\ u_2\end{array}\right]$ $y=\left[\begin{array}{c}{y}_0\\ y_1\\ y_2\\ y_3\end{array}\right]$ $Y=y-M_{B}u-M_C{\stackrel{\‾}{x}}_0$

$M_C=\left[\begin{array}{c}{C}_0\\ C_1{A}_1\\ C_2{A}_2{\mathrm{<figure-callout\; id="1"\; label="A"\; filenames="C0282339300391.png"\; state="\{\{state\}\}">A</figure-callout>}}_1\\ \&CenterDot;\\ \&CenterDot;\\ \&CenterDot;\\ C_N\underset{i=N}{\overset{1}{\mathrm{\&Pi;}}}{A}_i\end{array}\right]$

Wherein w has dimension (N+1) n _s* 1, u has dimension (N+1) n _u* 1, y and Y have dimension (N+1) n _y* 1, matrix M _AHas dimension (N+1) n _y* (N+1) n _s, M _BHas dimension (N+1) n _y* (N+1) n _u, M _CHas dimension (N+1) n _y* n _s, Q has dimension (N+1) n _s* (N+1) n _s, and R has dimension (N+1) n _y* (N+1) n _yThe intrafascicular approximately parameter of equation (3) is defined as in (5):

$F_{\min }=\left[\begin{array}{c}{f}_{\min }\\ \&CenterDot;\\ \&CenterDot;\\ \&CenterDot;\\ f_{\min }\end{array}\right]$ $F_{\max }=\left[\begin{array}{c}{f}_{\max }\\ \&CenterDot;\\ \&CenterDot;\\ \&CenterDot;\\ f_{\max }\end{array}\right]$

$F_C=\left[\begin{array}{c}I\\ A_1\\ A_2{\mathrm{<figure-callout\; id="1"\; label="A"\; filenames="C0282339300391.png"\; state="\{\{state\}\}">A</figure-callout>}}_1\\ \&CenterDot;\\ \&CenterDot;\\ \&CenterDot;\\ \underset{i=N}{\overset{1}{\mathrm{\&Pi;}}}{A}_i\end{array}\right]---\left(5\right)$

F wherein _Min, F _MaxHas dimension (N+1) n _s* 1, F _m, F _AAnd F _CHas dimension (N+1) n _s* (N+1) n _s, and F _BHas dimension (N+1) n _s* (N+1) n _u

When measurement is lost in existence in the I/O thesaurus 208, lose measurement and replace with model prediction.This replacement can be by calculating the measured value of prediction and using their replacements to lose to measure and finish clearly from model 262.Replace also and can pass through M in equation (3) and (4) _AWith the measuring corresponding row and be made as null value and impliedly finish with losing of Y.Any method is all impelled v in the equation (2) among both _jIn with lose that to measure corresponding element be 0.

State/parameter estimator 206 can be stored than the more past value of only estimating in the time domain (estimate horizon) of number N.Its storage up to and be included in all data that N-1 that batch (or wafer) with the oldest drop-out handle before immediately criticizes (or wafer).When new measurement reached, state/parameter estimator 206 served as an iteration on the right sorted table of the IO of group in being stored in I/O thesaurus 208 with N data point, uses each iteration to delete the oldest and next one point is increased in the time domain that N orders.Appear in any one group of N point if having the data point of losing value, state/parameter estimator 206 replaces model 262 predictions of that.The error of not supposing that is zero.State/parameter estimator 206 is still estimated the error of that in it finds the solution the process of QP.

When but the measured value of the value of before having lost becomes the time spent, the new measurement that I/O thesaurus 208 replaces in the sorted table that is kept in the database.When request process input next time, state/parameter estimator 206 uses 262 predictions of actual measurement substitution model for that, and before being measured, the renewal of current point, proofreaies and correct previous estimation based on the use of model prediction in the iteration of having a few of ordering from N.Though the time limit can force, do not need to lose value and in fixing period, become available, and can be variable the period that measurement becomes available.

With reference now to Fig. 3,, shows the order progress of the estimation interval of losing state of value/parameter estimator.In Fig. 3, the time domain (horizon) that is used for configuration status/parameter estimator 206 comprises time domain (N) 3.But, it will be appreciated by those skilled in the art that selected N is thought of as condition with system requirements and other.In addition, Fig. 3 illustrates iteration 302,304,306 and 308 four times, and each iteration is ended boundary 310,312,314 and 316 with time index identification, each iteration explanation.When each wave filter iteration (k), 302,304,306 and 308 o'clock, criticize arbitrarily or the measurement of wafer (j) is received.Insert new measurement in position in the tabulation of data memory module 250 in being stored in I/O thesaurus 208.Suitable position is determined based on rolling time domain filtering (receding horizon filter) based on the correlated process time mark.Fig. 3 illustrates as the new point of putting 322,324,326 and 328.State/parameter estimator 206 is fetched I/O classification of Data table from I/O thesaurus 208, and on continuous group of N data point iteration, with the current estimation of arrival process state.In the example that in Fig. 3, shows, lose for one of the value in the tabulation, as showing by point 332 in the iteration 302.According to a kind of embodiment of the present invention, state/parameter estimator 206 can be to that application model 262 prediction.Similarly, for the iteration j+1 that is designated reference number 304, model 262 prediction can be used for putting 334, except input thesaurus 208 must continue to keep more than N-1 data point, because lose value in the given data centre.At the iteration j+2 that is designated reference number 306, the data of previously lost obtain.Data memory module 250 is inserted into data 326 in the placeholder position of obliterated data in the I/O thesaurus 208.State/parameter estimator 206 iteration in whole tabulation is once more used real data to the data point of previously lost specifically then, rather than model 262 predictions.After this iteration is finished, a required only N-1 data point in the prediction of N value when I/O thesaurus 208 is trimmed to next iteration with data list.At the iteration j+3 that is designated reference number 308, system 200 is as operating for iteration j, because there is not the data point of losing in tabulation.

V. parameter estimation

Equation (6) is provided at that state/parameter estimator 206 is used for upgrading Matrix C in the parameter estimation procedure ^jThe equation of model coefficient.

$\min J=\underset{i=1}{\overset{N}{\mathrm{\&Sigma;}}}{({\stackrel{\&OverBar;}{y}}_i-{\stackrel{\widehat{\&OverBar;}}{y}}_i)}^{\&prime;}Q({\stackrel{\&OverBar;}{y}}_i-{\stackrel{\widehat{\&OverBar;}}{y}}_i)$ (for all batches i=1 of product j, N)

{C ^J}

Condition: ${\hat{x}}_0={\stackrel{\&OverBar;}{x}}_0$

${\hat{x}}_{i+1}=A_{i+1}{\hat{x}}_i+B_{i+1}{u}_{i}i=0,N-1$

${\hat{y}}_i={C^j}_i{\hat{x}}_{i}i=0,N---\left(6\right)$

${\stackrel{\&OverBar;}{y}}_i=y_i-y_{i-1}i=0,N$

${\stackrel{\widehat{\&OverBar;}}{y}}_i={\hat{y}}_i-{\hat{y}}_{i-1}i=0,N$

The parameter estimator filter class is similar to the state estimator wave filter, except optimizing the data of using from the like products/layer of a plurality of process tools.As discussed above, the data in sort module 254 classification I/O thesauruss 208 databases are so that used by the parameter estimator wave filter.The parameter estimator wave filter relies on data manager module and is updated in the data sequence so that will lose value when in place.Like this, parameter estimation is benefited from the useful processing of handling with non-ordered data of losing that data manager module provides in state estimation procedure.

With reference to figure 4, the above-described method of flowchart text.Model coefficient in the piece 402 regulation models 262 and interference Δ u 216 are initialized as initial value.Next, in piece 404, the process of recommendation input u uses current model 262 to calculate.Export target y and controller algorithm are also calculated.In piece 406, the operator is provided to process with actual input value v=u+ Δ u then.The material of making is handled in piece 408 then, and real process output y measures in piece 410.In piece 412, the new coefficient of state/parameter estimator 206 operation parameter algorithm for estimating computation models 262.Piece 414 regulation models 262 use the new model coefficient that calculates in piece 412 to upgrade.In piece 416, state/parameter estimator 206 user mode algorithm for estimating calculate new Interference Estimation, and this new Interference Estimation of model 262 usefulness is upgraded.Piece 418 relates to the optional process of the fault detect that describes below.When new recommendation process input u calculated in piece 404, the current model 262 that upgrades in piece 414 and 416 was used for carrying out and calculates.As reflecting in Fig. 4, parameter estimation and state estimation take place in independent step.But the order of these steps is inessential.

With reference to the controller among the figure 2 202, state/parameter estimator 206 guarantees that controllers 202 always visit up-to-date model in conjunction with Fig. 3 and Fig. 4.In addition, by exporting when measuring the prediction that replaces the model 262 in I/O (I/O) sequence when also not receiving, estimator 206 makes full use of its all available information, comprises process model 262, and does not need to handle the trouble restriction that wait is finished up to measurement.Common art methods ignore have lose measurement batch, suppose that basically that a collection of error is zero.Another kind of art methods is to postpone to estimate must provide new when being input in the process up to controller.The prior art method causes that too slow controller calculates, and does not satisfy the value problem of losing.On an average, it reduces really and loses the value number in the sequence, because by waiting for that controller allows more time to be used for finishing measurement in undeterminate batch or the wafer.

Those skilled in the art will recognize that, except above-described those, other filtering methods can use in parameter and state estimation.For example, firstorder filter or Kalman filter can be used.But above-described filtering method allows to use the Constraints Processing of these other filtering methods processing.

VI. fault detect

The a small amount of additional treatments of fault detect application need.In many fault detects are used, follow the starting condition of the state of maintenance event, for example the wet cleaning in the plasma etching device makes and when having fault really to exist fault detection algorithm is not obscured mutually with the identification fault.State/parameter estimator 206 estimates that by with above-mentioned same procedure in the state estimation paragraph original state compensates these skews from initial wafer is handled.In addition, fault detection algorithm is based on from the overlayer wafer, rather than the model that makes up of product wafer, stands the problem identical with the R2R controller.Matrix C ^jRelation between overlayer wafer model and the product wafer model is provided for each product j.These two kinds of abilities allow to reduce the cost of model development based on the initial modeling of overlayer wafer, and the offset of performance of the process tool of maintenance event is followed in compensation.

The required additional treatments of fault detect is that after state and parameter estimation, model 262 surpluses must be compared with measurement data.Because parameter estimation and state estimation all provide the letter of putting that they export separately to estimate that system can compare pattern allowance with the change level of expectation, and when falling within outside the fiducial interval of prediction, measurement data discerns fault.Before fault detect became reliably, measurement data must accumulate fully.It is believed that, measure about two to three times measurement data of state number and determine reliably enough whether fault takes place.In alternative embodiment, when fiducial limit was reduced to predefined value, fault detect may begin.

Discuss as following, two steps are involved in fiducial interval (CI) calculating: CI is calculated in (1) whiteness test and (2).Whiteness test provides test the character of the surplus (perhaps Yu Ce output " error ") of " degree of confidence " that CI calculates.

A. the CI of parameter estimation calculates

Below step be used for the CI that calculating parameter estimates,

1. test the whiteness of surplus

2. calculate the estimate covariance matrix of surplus

3. the result of use step (2) calculates the estimate covariance matrix P of estimated parameter _N

4. indicate P _N ^IiBe P _NI diagonal element, so

$Z=\frac{{\widehat{\mathrm{\&theta;}}}_N^i-{\mathrm{\&theta;}}^i}{\sqrt{P_N^{\mathrm{ii}}}}~N\left(\mathrm{0,1}\right)---\left(7\right)$

Wherein N (0,1) is the standardized normal distribution with zero mean and variance one.For given level of significance α (for example, 2.17 value is represented 3% level of significance),

${\widehat{\mathrm{\&theta;}}}_N^i-\mathrm{\&alpha;}\sqrt{P_N^{\mathrm{ii}}}<{{\mathrm{\&theta;}}^i<\widehat{\mathrm{\&theta;}}}_N^i+\mathrm{\&alpha;}\sqrt{P_N^{\mathrm{ii}}}---\left(8\right)$

Wherein

Be i estimated parameter, and θ ⁱBe i actual parameter.Referring to S.Weisberg, Applied Linear Regression (Second Edition), John Wiley﹠amp; (S.Weisberg uses linear regression (second edition), John Wiley﹠amp to Sons (1985); Sons (1985)), be hereby incorporated by.

1. whiteness test

In order to test the whiteness of surplus, get

${\mathrm{\&epsiv;}}_N=(Y_N^{\mathrm{\&gamma;}}-{\hat{Y}}_N^{\mathrm{\&gamma;}}){=\left[\begin{array}{ccccccccccc}{\mathrm{\&epsiv;}}_{11}& {\mathrm{\&epsiv;}}_{21}& ...& {\mathrm{\&epsiv;}}_{n_{y}1}& {\mathrm{\&epsiv;}}_{12}& ...& {\mathrm{\&epsiv;}}_{n_{y}2}& ...& {\mathrm{\&epsiv;}}_{1N}& ...& {\mathrm{\&epsiv;}}_{n_{y}N}\end{array}\right]}^{\&prime;}---\left(9\right)$

Y wherein _N ^γBe the vector of measuring output, and

It is the vector of prediction output.For losing measurement, the value of measurement replaces the predicted value of origin self model 262.Then, calculated below:

$R_i\left(0\right)=\frac{1}{N}\underset{m=1}{\overset{N}{\mathrm{\&Sigma;}}}{\mathrm{\&epsiv;}}_{i,\mathrm{<figure-callout\; id="2"\; label="m"\; filenames="C0282339300391.png"\; state="\{\{state\}\}">m</figure-callout>}}^2,$ ${\mathrm{NR}}_i\left(0\right)=\frac{R_i\left(0\right)}{R_i\left(0\right)}=1,$ i＝1，2，...，n _y (10)

$R_i\left(k\right)=\frac{1}{N}\underset{m=1}{\overset{N}{\mathrm{\&Sigma;}}}{\mathrm{\&epsiv;}}_{i,m}{\mathrm{\&epsiv;}}_{i,m-k},$ ${\mathrm{NR}}_i\left(k\right)=\frac{R_i\left(0\right)}{R_i\left(k\right)},$ i＝1，2，...，n _y (11)

Whiteness test may be presented below: the refusal as the surplus predicated error of white noise has usually,

$\left|{\mathrm{RN}}_i\left(k\right)\right|>\frac{\mathrm{\&alpha;}}{\sqrt{N}},$ For any k＞1, i=1,2 ..., n _y(12)

Wherein α represents level of significance, and can find (for example, 2.17 value is represented 3% level of significance) from the t statistical table.

2.CI calculate

After whiteness test, the following calculating of estimate covariance matrix of surplus:

${\mathrm{\&lambda;}}_N=\frac{1}{N-d}{(Y_N^{\mathrm{\&gamma;}}-{\hat{Y}}_N^{\mathrm{\&gamma;}})}^{\&prime;}(Y_N^{\mathrm{\&gamma;}}-{\hat{Y}}_N^{\mathrm{\&gamma;}})---\left(13\right)$

R _N＝λ _NI (14)

Wherein,

Definition in equation (9).

Estimated parameter P _NThe estimate covariance matrix followingly determine:

P _N＝(Φ _N′QΦ _N) ^-1Φ _N′QR _NQΦ _N(Φ _N′QΦ _N) ^-1 (15)

Next, indication P _N ^IiBe P _NI diagonal element, so

$Z=\frac{{\widehat{\mathrm{\&theta;}}}_N^i-{\mathrm{\&theta;}}^i}{\sqrt{P_N^{\mathrm{ii}}}}~N\left(\mathrm{0,1}\right)---\left(16\right)$

For given level of significance α,

${\widehat{\mathrm{\&theta;}}}_N^i-\mathrm{\&alpha;}\sqrt{P_N^{\mathrm{ii}}}<{\mathrm{\&theta;}}^i<{\widehat{\mathrm{\&theta;}}}_N^i+\mathrm{\&alpha;}\sqrt{P_N^{\mathrm{ii}}}---\left(17\right)$

B. the CI of state estimation calculates

Below step be used for the CI that computing mode estimates,

1. the whiteness disturbed of test evaluation state

2. calculate the estimate covariance matrix P of estimated state _N

3. indicate P _N ^IiBe P _NI diagonal element, so

$Z=\frac{{\hat{x}}_N^i-x^i}{\sqrt{P_N^{\mathrm{ii}}}}~N\left(\mathrm{0,1}\right)---\left(18\right)$

Wherein N (0,1) is the standardized normal distribution with zero mean and variance one.For given level of significance α (for example, 2.17 value is represented 3% level of significance),

${\hat{x}}_N^i-\mathrm{\&alpha;}\sqrt{P_N^{\mathrm{ii}}}<x^i<{\hat{x}}_N^i+\mathrm{\&alpha;}\sqrt{P_N^{\mathrm{ii}}}---\left(19\right)$

Wherein Be i estimated state, and x ⁱBe i virtual condition.

1. whiteness test

For the whiteness that the test evaluation state disturbs, get

${\mathrm{\&epsiv;}}_N={\left[\begin{array}{ccccccccccc}{\widehat{\mathrm{\&epsiv;}}}_{11}& {\widehat{\mathrm{\&epsiv;}}}_{21}& ...& {\widehat{\mathrm{\&epsiv;}}}_{n_{s}1}& {\widehat{\mathrm{\&epsiv;}}}_{12}& ...& {\widehat{\mathrm{\&epsiv;}}}_{n_{s}2}& ...& {\widehat{\mathrm{\&epsiv;}}}_{1N}& ...& {\widehat{\mathrm{\&epsiv;}}}_{n_{s}N}\end{array}\right]}^{\&prime;}---\left(20\right)$

ε wherein _NBe the vector that estimated state disturbs, n _sBe state number, and N estimate time domain (estimate horizon).Then, calculated below:

$R_i\left(0\right)=\frac{1}{N}\underset{m=1}{\overset{N}{\mathrm{\&Sigma;}}}{\widehat{\mathrm{\&epsiv;}}}_{i,\mathrm{<figure-callout\; id="2"\; label="m"\; filenames="C0282339300391.png"\; state="\{\{state\}\}">m</figure-callout>}}^2,$ ${\mathrm{NR}}_i\left(0\right)=\frac{R_i\left(0\right)}{R_i\left(0\right)}=1,$ i＝1，2，...，n _s (21)

$R_i\left(k\right)=\frac{1}{N}\underset{m=1}{\overset{N}{\mathrm{\&Sigma;}}}{\widehat{\mathrm{\&epsiv;}}}_{i,m}{\widehat{\mathrm{\&epsiv;}}}_{i,m-k},$ ${\mathrm{NR}}_i\left(k\right)=\frac{R_i\left(0\right)}{R_i\left(k\right)},$ i＝1，2，...，n _s (22)

Whiteness test may be presented below: the refusal that disturbs as the state of white noise has usually,

$\left|{\mathrm{NR}}_i\left(k\right)\right|>\frac{\mathrm{\&alpha;}}{\sqrt{N}},$ For any k＞1, i=1,2 ..., n _s(23)

Wherein α represents level of significance, and can find (for example, 2.17 value is represented 3% level of significance) from the t statistical table.

2.CI calculate

When state disturbs when being white, the covariance matrix of current estimated state is provided by following,

So

$Z=\frac{{\hat{x}}_N^i-x^i}{\sqrt{R_i\left(0\right)}}~N\left(\mathrm{0,1}\right)$ i＝1，2，...，n _s (25)

X wherein ⁱBe i state of process.For given level of significance α,

${\hat{x}}_N^i-\mathrm{\&alpha;}\sqrt{R_i\left(0\right)}<x^i<{\hat{x}}_N^i+\mathrm{\&alpha;}\sqrt{R_i\left(0\right)}---\left(26\right)$

Principal component analysis (PCA) (PCA) is used for detection failure.The minimum vector that loads is determined by percentage variance method, and surplus designs from the loading factor.

As illustrated in fig. 5, thresholding (from fiducial interval, for example 95%) definition hyperelliptic confidence region 502.Any surplus of these 502 outsides, zone will be thought fault.The example of fault is by reference number 504 signs.

When fault was detected, control system can be provided with alarm or closing tool.

V. other embodiments

It will be appreciated by those skilled in the art that, embodiment disclosed herein can be implemented as can be as one or more program products, in a variety of forms, the software program instructions that comprises the computer program distribution, and no matter be used for the program recorded medium or the signal bearing medium of the particular type that actual enforcements issues, the present invention is similarly suitable.But the example of program recorded medium and signal bearing medium comprises for example floppy disk of record type medium, and CD-ROM and tape transfer type media be numeral and analog communication links for example, and other medium memory and publishing system.

In addition, by block diagram, making of process flow diagram and/or example is used for stating various embodiments of the present invention to preceding detailed description.It will be appreciated by those skilled in the art that each block diagram part, flow chart step, and can be by the hardware of wide region by the operation and/or the composition of the operation instruction of example, software, firmware, and any combination realizes individually and/or jointly.As the skilled person will appreciate, the present invention can be completely or partially at the standard integrated circuit, in the special IC (ASIC), as the computer program that operates on for example one or more computing machines of the general-purpose machinery with suitable hardware, as firmware, perhaps realize as its any combination in fact, and design circuit system and/or write code for software or firmware and will fully in those skilled in the art's technical ability, consider the disclosure.

Though the present invention is about top embodiments set forth with change and to describe, these embodiments and variation are illustrative, and the present invention does not think be confined to these embodiments and variation on scope.Therefore, various other embodiments do not described here and modification and improvement can be in essence of the present invention and scope by following claims definition.

Claims (66)

1. method that is used to control manufacture process comprises:

The use input is handled and production process output;

The process input is stored in the database;

Explicitly be stored in database with each process input at least one measurement of process output;

Iteration is with the estimation procedure state on from data of database;

If one or more measurements are lost, then lose measurement based on the database replacement that is predicted as that comes self model from database; And

Estimate to upgrade described model with described process status.

2. according to the method for claim 1, if wherein actual measurement becomes available one period after process status is estimated, then for described one or more measurements of losing, replacing prediction with described actual measurement measures, described actual measurement is stored in the described database, and in process status is subsequently estimated, utilizes described actual measurement.

3. according to the model that the process of claim 1 wherein that controller receive to upgrade, and utilize the model of described renewal to produce next process input.

4. according to the method for claim 3, its middle controller comprises Model Predictive Control (MPC) type controllers, direct modeling inverter type controller, and one or more in the simple integral controller.

5. according to the method for claim 3, the estimation of its middle controller use state makes the error minimize in each batch of process.

6. according to the process of claim 1 wherein that this process operates at least one semiconductor devices.

7. according to the process of claim 1 wherein that database is coupled to one or more modules of operating on database.

8. according to the method for claim 7, wherein module comprises estimation module, computing module, sort module, matching module and data memory module.

9. according to the process of claim 1 wherein that manufacture process is one or more batch (R2R) process in the following process: chemical-mechanical planarization (CMP) process, photoetching process and plasma etch processes.

10. according to the method for claim 2, be variable wherein said period.

11. provide time mark according to the process of claim 1 wherein that the described storage of process input of database comprises, and wherein process output described at least one measure service time mark and be associated with the input of each process.

12. a control system comprises:

Controller can provide the process input;

Process is coupled to controller and imports and provide process output with receiving course;

Database is configured to the measurement of receiving course input and process output, and database can be associated each process input and the output of each process;

Model, output can supply a model; And

State estimator is coupled to and receives database output and model output, and state estimator is connected to controller and can the production process state estimation, and be configured to based on described process status estimate will renewal model be provided to described controller.

13. according to the control system of claim 12, wherein this process is operated at least one semiconductor devices.

14. according to the control system of claim 12, its middle controller comprises following Model Predictive Control (MPC) type controllers, direct modeling inverter type controller, and one or more in the simple integral controller.

15. a device comprises

Be used for the instrument that the use input is handled and production process is exported;

Be used for the process input is stored in the instrument of database, this instrument that is used to store comprises provides time mark;

At least one measurement mark service time that is used for process is exported and the input of each process are stored in the instrument of database alignedly;

Be used on iteration with the instrument of estimation procedure state from data of database;

Be used for replacing the instrument of the measurement of losing from database based on the prediction that comes self model; And

Be used to use described process status to estimate to upgrade the instrument of described model.

16. a method that is used to control manufacture process comprises:

The use input is handled and production process output;

The process input is stored in the database;

Explicitly be stored in database with each process input at least one measurement of process output;

Iteration is wherein found the solution the constraint quadratic programming with the estimation procedure state when each iteration of data on from data of database;

If one or more measurements are lost, lose measurement based on the database replacement that is predicted as that comes self model from database; And

Estimate to upgrade described model with described process status.

17. method according to claim 16, if wherein actual measurement becomes available one period after process status is estimated, then for described one or more measurements of losing, replacing prediction with described actual measurement measures, described actual measurement is stored in the described database, and in process status is subsequently estimated, utilizes described actual measurement.

18. according to the method for claim 16, its middle controller receives the model that upgrades, and utilizes the model of described renewal to produce next process input.

19. according to the method for claim 17, be variable wherein said period.

20. according to the method for claim 16, wherein the described storage of the process of database input comprises provides time mark, and described at least one measurement mark service time of wherein process output is associated with each process input.

21. a method that is used to control manufacture process comprises:

Database with process input data and relevant process output data is provided;

Iteration is with the estimation procedure state on from data of database;

If one or more measurements are lost, then lose measurement based on the database replacement that is predicted as that comes self model from database; And

Estimate to upgrade described model with described process status.

22. method according to claim 21, if wherein actual measurement becomes available one period after process status is estimated, then for described one or more measurements of losing, replacing prediction with described actual measurement measures, described actual measurement is stored in the described database, and in process status is subsequently estimated, utilizes described actual measurement.

23. according to the method for claim 21, its middle controller receives the model that upgrades, and utilizes the model of described renewal to produce next process input.

24. according to the method for claim 22, be variable wherein said period.

25. according to the method for claim 21, wherein said process output data mark service time is associated with described process input data.

26., find the solution the constraint quadratic programming when also being included in the each iteration of described data according to the method for claim 21.

27. a controller that is used for manufacture process comprises:

Be used for the device that the use input is handled and production process is exported;

Be used for the process input is stored in the device of database;

At least one that is used for process output measured the device that is stored in database with each process input explicitly;

Be used on iteration with the device of estimation procedure state from data of database;

Lose from database if be used for one or more measurements, then lose the device of measurement based on the database replacement that is predicted as that comes self model; And

Be used for estimating the device of the described model of renewal with described process status.

28. a controller that is used for manufacture process comprises:

Be used for the process input is stored in the device of database;

At least one that is used for process output measured the device that is stored in database with each process input explicitly;

Be used on iteration and wherein when each iteration of data, find the solution the constraint quadratic programming with the device of estimation procedure state from data of database;

Lose from database if be used for one or more measurements, then lose the device of measurement based on the database replacement that is predicted as that comes self model; And

Be used for estimating the device of the described model of renewal with described process status.

29. a controller that is used for manufacture process comprises:

Be used to provide the device of database with process input data and relevant process output data;

Be used on iteration with the device of estimation procedure state from data of database;

Lose from database if be used for one or more measurements, then lose the device of measurement based on the database replacement that is predicted as that comes self model; And

Be used for estimating the device of the described model of renewal with described process status.

30. a method that is used to control manufacture process comprises:

The use input is handled and production process output;

The process input is stored in the database;

Explicitly be stored in database with each process input at least one measurement of process output;

Iteration is to estimate to be used for one or more coefficients of model on from data of database;

If one or more measurements are lost, then lose measurement based on the database replacement that is predicted as from described model from database during model coefficient is estimated;

Upgrade described model with described coefficient estimation;

Iteration is with the estimation procedure state on from data of database;

If one or more measurements are lost, then lose measurement based on the database replacement that is predicted as from described model from database during process status is estimated; And

Estimate to upgrade described model with described process status.

31. method according to claim 30, if wherein actual measurement becomes available one period after model coefficient estimation or process status estimation, then for described one or more measurements of losing, replace the prediction measurement that previously lost is measured with described actual measurement, described actual measurement is stored in the described database, and subsequently model coefficient estimate and the process status estimation in utilize described actual measurement.

32. according to the method for claim 30, its middle controller receives the model that upgrades, and utilizes the model of described renewal to produce next process input.

33. according to the method for claim 32, its middle controller comprises Model Predictive Control (MPC) type controllers, direct modeling inverter type controller, and one or more in the simple integral controller.

34. according to the method for claim 32, the estimation of its middle controller use state makes the error minimize in each batch of process.

35. according to the method for claim 30, wherein this process is operated at least one semiconductor devices.

36. according to the method for claim 30, wherein database is coupled to one or more modules of operating on database.

37. according to the method for claim 36, wherein module comprises estimation module, computing module, sort module, matching module and data memory module.

38. according to the method for claim 30, wherein said model coefficient estimation and described process status are estimated to comprise provides fiducial limit.

39. method according to claim 38, comprise that also but the model that utilizes described renewal produces the estimation to measured process variable, and actual measurement is compared with described estimation to determine that described estimation is whether in described fiducial limit, if wherein described estimation not in described fiducial limit, indication fault then.

40. according to the method for claim 31, be variable wherein said period.

41., wherein the replacement of losing measurement of database is finished clearly or impliedly according to the method for claim 30.

42., find the solution the constraint quadratic programming when also being included in the each iteration of described data according to the method for claim 30.

43. according to the method for claim 30, wherein the described classification of process input comprises provides time mark, and described at least one measurement mark service time of wherein process output is associated with each process input.

44. a method that is used to control manufacture process comprises:

The use input is handled and production process output;

The process input is stored in the database;

Explicitly be stored in database with each process input at least one measurement of process output;

Iteration is wherein found the solution the constraint quadratic programming to estimate to be used for one or more coefficients of model when each iteration of described data on from data of database;

If one or more measurements are lost, then lose measurement based on the database replacement that is predicted as from described model from database during model coefficient is estimated;

Upgrade described model with described coefficient estimation;

Iteration is wherein found the solution the constraint quadratic programming with the estimation procedure state when each iteration of described data on from data of database;

If one or more measurements are lost, lose measurement based on the database replacement that is predicted as from described model from database during process status is estimated; And

Estimate to upgrade described model with described process status.

45. method according to claim 44, if wherein actual measurement becomes available one period after model coefficient estimation or process status estimation, then for described one or more measurements of losing, use described actual measurement to replace the prediction measurement that previously lost is measured, described actual measurement is stored in the described database, and subsequently model coefficient estimate and the process status estimation in utilize described actual measurement.

46. according to the method for claim 44, its middle controller receives the model that upgrades, and utilizes the model of described renewal to produce next process input.

47. according to the method for claim 46, its middle controller comprises Model Predictive Control (MPC) type controllers, direct modeling inverter type controller, and one or more in the simple integral controller.

48. according to the method for claim 46, the estimation of its middle controller use state makes the error minimize in each batch of process.

49. according to the method for claim 44, wherein this process is operated at least one semiconductor devices.

50. according to the method for claim 44, wherein database is coupled to one or more modules of operating on database.

51. according to the method for claim 50, wherein module comprises estimation module, computing module, sort module, matching module and data memory module.

52. according to the method for claim 44, wherein said model coefficient estimation and described process status are estimated to comprise provides fiducial limit.

53. method according to claim 52, comprise that also but the model that utilizes described renewal produces the estimation that is used for measured process variable, and actual measurement is compared with described estimation to determine that described estimation is whether in described fiducial limit, if wherein described estimation not in described fiducial limit, indication fault then.

54. according to the method for claim 45, be variable wherein said period.

55. according to the method for claim 44, wherein the replacement of measuring for database lost is finished clearly or impliedly.

56. according to the method for claim 44, wherein the described storage of the process of database input comprises provides time mark, and described at least one measurement mark service time of wherein process output is associated with each process input.

57. a method that is used to control manufacture process comprises:

Database with process input data and relevant process output data is provided;

Iteration is with one or more coefficients of estimation model on from data of database;

If one or more measurements are lost, then lose measurement based on the database replacement that is predicted as from described model from database during model coefficient is estimated;

Upgrade described model with described coefficient estimation;

Iteration is with the estimation procedure state on from data of database;

If one or more measurements are lost, then lose measurement based on the database replacement that is predicted as from described model from database during process status is estimated; And

Estimate to upgrade described model with described process status.

58. method according to claim 57, if wherein actual measurement becomes available one period after model coefficient estimation or process status estimation, then for described one or more measurements of losing, use described actual measurement to replace the prediction measurement that previously lost is measured, described actual measurement is stored in the described database, and subsequently model coefficient estimate and the process status estimation in utilize described actual measurement.

59. according to the method for claim 57, wherein said model coefficient estimation and described process status are estimated to comprise provides fiducial limit.

60. method according to claim 59, comprise that also but the model that utilizes described renewal produces the estimation to measured process variable, and actual measurement is compared with described estimation to determine that described estimation is whether in described fiducial limit, if wherein described estimation not in described fiducial limit, indication fault then.

61. according to the method for claim 58, be variable wherein said period.

62. according to the method for claim 57, wherein the replacement that database lost is measured is finished clearly or impliedly.

63., find the solution the constraint quadratic programming when also being included in the each iteration of described data according to the method for claim 57.

64. a controller that is used for manufacture process comprises:

Be used for the device that the use input is handled and production process is exported;

Be used for the process input is stored in the device of database;

At least one that is used for process output measured the device that is stored in database with each process input explicitly;

Be used on from data of database iteration device with one or more coefficients of estimating to be used for model;

Lose from database during model coefficient is estimated if be used for one or more measurements, then lose the device of measurement based on the database replacement that is predicted as from described model;

Be used for upgrading the device of described model with described coefficient estimation;

Be used on iteration with the device of estimation procedure state from data of database;

Lose from database during process status is estimated if be used for one or more measurements, then lose the device of measurement based on the database replacement that is predicted as from described model; And

Be used for estimating to upgrade the device of described model with described process status.

65. a controller that is used for manufacture process comprises:

Be used for the device that the use input is handled and production process is exported;

Be used for the process input is stored in the device of database;

At least one that is used for process output measured the device that is stored in database with each process input explicitly;

Be used on from data of database iteration device, wherein when each iteration of described data, find the solution the constraint quadratic programming with one or more coefficients of estimating to be used for model;

Lose from database during model coefficient is estimated if be used for one or more measurements, then lose the device of measurement based on the database replacement that is predicted as from described model;

Be used for upgrading the device of described model with described coefficient estimation;

Be used on iteration and wherein when each iteration of described data, find the solution the constraint quadratic programming with the device of estimation procedure state from data of database;

Lose from database during process status is estimated if be used for one or more measurements, then lose the device of measurement based on the database replacement that is predicted as from described model; And

Be used for estimating the device of the described model of renewal with described process status.

66. a controller that is used for manufacture process comprises:

Be used to provide the device of database with process input data and relevant process output data;

Be used on iteration with the device of one or more coefficients of estimation model from data of database;

Lose from database during model coefficient is estimated if be used for one or more measurements, then lose the device of measurement based on the database replacement that is predicted as from described model;

Be used for upgrading the device of described model with described coefficient estimation;

Be used on iteration with the device of estimation procedure state from data of database;

Lose from database during process status is estimated if be used for one or more measurements, then lose the device of measurement based on the database replacement that is predicted as from described model; And

Be used for estimating the device of the described model of renewal with described process status.

Images