DETAILED DESCRIPTION OF THE INVENTION

[0041] Plasma-enhanced chemical vapor deposition (PECVD) has been widely used in microelectronics fabrication to deposit films, such as a SiO ₂ , at low temperatures. In the PECVD process, a radio frequency (RF) glow discharge (plasma) supplies part of the energy to promote a desired chemical reaction on the surface of the substrate. FIG. 1A is a schematic illustration of an exemplary PECVD system 100 with parallel plate electrodes 110 , 115 . The system 100 includes a chamber 120 , a vacuum system 130 , an RF generator 140 for generating a source plasma 145 , and a gas or fluid delivery system 150 for introduction of reactive gases. A wafer 160 for film deposition is placed on the grounded electrode 110 . Reactive gases are introduced into a reaction chamber 120 through inlet 125 of the gas delivery system. In order to promote a uniform distribution, the reactive gases typically are introduced into the chamber at a source positioned opposite or a distance from the wafer. The wafer-containing electrode may be rotated for further uniformity of deposition, as indicated by arrow 165 . The gas delivery system may include heating and cooling means (not shown) for maintaining a constant gas and chamber temperature. Wafers are transferred into and out of chamber 120 by a robot blade (not shown) through an insertion/removal opening (not shown) in the side of chamber 120 . Two or more chambers may be connected. In at least some PECVD systems, the chambers share reactive gases, but have individual RF controls.

[0042] FIG. 1B is an enlarged view of the PECVD reaction chamber illustrating an exemplary delivery system for the reactive gases used in the PECVD process. The gases are introduced through inlet 125 into a heated gas distribution head (showerhead) 170 , which has outlets 180 at spaced intervals. As shown by arrows in FIG. 1 B, the reactive gases then flow into the plasma gas. The energy of the plasma is transferred into the gas mixture, transforming the gas into reactive radicals, ions and other highly excited species. The energetic species then flow over the wafer, where they are deposited as a thin film. Since the formation of the reactive species takes place in the gas phase, the wafer can be maintained at low temperatures.

[0043] The term “target output” represents the desired processing outcome of a plasma enhanced chemical vapor deposition process. Some tolerance is built into the profile, so that the profile includes the target value and acceptable standard deviations therefrom. Film thicknesses falling within the standard deviation would not require updating of the deposition recipe. Thus, use of the term “target output” includes the target value and the standard deviation therefrom.

[0044] The term “wafer” is used in a general sense to include any substantially planar object onto which a film is deposited. Wafers include monolith structures or structures having one or more layers, thin films or other features already deposited thereon. “Thin film” and “film” may be used interchangeably, unless otherwise indicated.

[0045] An exemplary PECVD deposition system includes two or more chambers in which deposition of material occurs. The chambers may carry out the same process or different processes; some CVD systems are based on a series of operations, while some use parallel processing schemes. The chambers may thus process wafers in parallel, that is, each deposition chamber deposits a film on a wafer at the same time. The deposition recipe for each chamber may be the same or different. In one or more processes, the chambers share some processing parameters while others are independently controlled. For example, gas flow of reactant gases is common to both (or all) chambers, but RF power, RF time and showerhead spacing are independently controlled in each chamber.

[0046] The present invention is described with reference to PECVD, however is it readily apparent that other plasma CVD processes, such as high density plasma processes, are also contemplated. The present invention also is applicable to CVD systems using either a batch process or inline process. An inline process refers to a process in which all wafers going through a system go through a sequence of steps and those steps may be carried out in different chambers, whereas a batch process refers to a process in which a wafer goes to any one of the chambers in the system, where the entire deposition is then carried out. The PECVD processes described above may be modeled to provide a format for improving the deposition process. The process model should accurately predict the thin film characteristics (output) for a given set of input conditions. The run-to-run film characteristics are improved or maintained by adjusting the deposition parameters during plasma enhanced deposition to correct for unmodeled effects or to correct for drift in the deposition process conditions. Run-to-run control can be defined as wafer-to-wafer or lot-to-lot, depending upon the processes being controlled and the available methods for monitoring output.

[0047] According to one or more embodiments of the present invention, an initial model is developed based upon knowledge of the film deposition process, as is shown in a flow diagram ( FIG. 2 ). An initial understanding of the system is acquired in step 200 , which is used to design and run a design of experiments (DOE) of step 210 . The DOE desirably is designed to establish the relationship between or among variables that have a strong and predictable impact on the processing output one wishes to control, e.g., film thickness or some other film property. The DOE provides data relating to process parameters and process outcome, which is then loaded to the advanced process control system in step 220 . The advanced process control system may be a controller or computer that uses the data to create and update the model. The model can be represented as raw data that reflects the system, or it can be represented by equations, for example multiple input-multiple output linear, quadratic and general non-linear equations, which describe the relationship among the variables of the system. Process requirements such as output targets and process specification are determined by the user in step 225 , which are combined with the DOE data to generate a working model in step 230 .

[0048] In developing the model, film properties of interest 302 are identified and outcome determinative processing variables 304 are selected for the model, as illustrated schematically in FIG. 3 . The specific film properties of interest may vary depending upon the type of film deposited, and thus the film properties of interest 302 and processing variables 304 of FIG. 3 are shown by way of example.

[0049] Regardless of the type of film substance for which a model is created, to obtain DOE data, an experiment is run which perturbs or varies values of the processing variables of interest about a center point (or median value). One or more processing variables can be varied. The film properties of interest in the resultant film are measured for each combination of inputs. Data can be acquired empirically, by carrying out a series of experiments over a range of parameter values of the processing variables. The data is fit to the appropriate curve (linear or non-linear) to define the model.

[0050] Undoped silica glass (USG) is commonly deposited by PECVD and a model development is discussed below with specific reference to USG, although it is readily apparent that the methodology can be used to develop models for any other PECVD film deposition process, as well as CVD (i.e., non-plasma enhanced) processes.

[0051] In one or more embodiments of the present invention, the film properties of interest for USG film include one or more of film thickness, film thickness uniformity, stress and refractive index (RI). In at least some embodiments of the present invention, the model accounts for two or more film properties, for example, the model describes the effect of process variables on film thickness (deposition rate) and film stress, or on film thickness and refractive index. Process variables for deposition of the USG film include one or more of ozone flow rate, reactive gas flow rate, carrier gas flow rate, chamber pressure and showerhead spacing (distance) from the substrate, as well as total deposition time. Deposition time is controlled by RF power and RF time. For the deposition of USG films, reactive gases typically include ozone (O ₃ ), oxygen (O ₂ ), and tetraethylorthosilicate (TEOS) or, alternatively, silane (SiH ₄ ).

[0052] Models for other film deposition systems can be similarly developed using the processing variables and film properties specific to those films. For example, the deposition of dielectric anti-reflective coating (DARC) films may be modeled and controlled in a manner similar to that described for the USG-based films. The precursors in the case of DARC films are silane (SiH ₄ ) and N ₂ O, which are reacted within a plasma to deposit the resultant film. The flow rates of these gases along with the spacing and deposition time are used to control the average film thickness, thickness uniformity, refractive index, extinction coefficient and stress.

[0053] The deposition of fluorosilicate glass (FSG) films can also be modeled. In this case, the average film thickness, film thickness uniformity, refractive index, and fluorine concentration can be modeled and controlled using spacing, deposition time, SiF ₄ flow rate, and high and low frequency RF power. In one or more embodiments, a component of the control of this film type is that both thickness and dopant concentration are controlled simultaneously.

[0054] For Black Diamond™ (BD) films, which are deposited using trimethylsilane (TMS) as a reactant gas, the effects of RF power, spacing, O ₂ flow rate, TMS flow rate, chamber pressure and deposition time are variables that can be included in a model used to control average film thickness, film thickness uniformity and refractive index. Furthermore, the carbon content of BD films can be provided as feed-forward data to the etching tool, as this factor is relevant in control of etch rate. Using a relationship between the etch uniformity and carbon content of the BD film, the etch recipe can be modified to moderate etch rate nonuniformity.

[0055] On the Producer™ system from Applied Materials of Santa Clara, Calif., the RF time, and hence deposition time, of each chamber is individually controllable, and the gas flow rates are common to all chambers of the PECVD system. In one or more embodiments of the present invention, the model can distinguish between the two types of processing variables (individual and common) and account for them accordingly. As discussed herein below, the model permits simultaneous optimization of more than one variable.

[0056] In one or more embodiments of the present invention, the model defines two or more different film property, e.g., film thickness, regions of the wafer. As is shown in FIG. 4, a wafer is divided into radial regions 401 through 405 of varying width and area. The number, size and location of the regions also can vary and may be selected based upon any number of factors, including the variability or uniformity of the film property in a given region of the wafer. In one or more embodiments of the present invention, it is desirable that the film property in any given region be substantially uniform, particularly in those cases where, for example, a number of wafer thickness measurements within a region are averaged to define the region-averaged thickness profile. Thus, at the edges of the wafer where edge effects can be dramatic, narrow regions encompassing only the outer portions of the wafer may be selected. Near the center of the wafer where deposition effects may be subtler, a larger region may be defined. In one or more embodiments of the present invention, the regions are defined such that all azimuthal variation is averaged out. In one or more embodiments of the present invention, the use of an input value is contemplated to correct and account for azimuthal variation. Film property measurements taken within a region of the wafer are averaged to give the average thickness for that region.

[0057] By way of example (with particular regard to film thickness), the five wafer regions of FIG. 4 can be defined as shown in Table 1 for a wafer that is 95 mm in diameter. 1

[0058] With the regions defined as in Table 1, all thin film thickness measurement points with a radial distance from the wafer center greater than 5 mm, but less than 40 mm, are averaged together to give the thickness of region 401 . The thicknesses for all other regions are similarly calculated, but region 405 includes all points greater than 92 mm and up to and including 95 mm. Thus, a film is deposited by PECVD and, based upon post-measurements and deposition time and/or other processing variables, a film thickness and deposition rate can be determined for each region.

[0059] In one or more embodiments of the present invention, a film property of interest is film thickness and one of the processing parameters is deposition time. In one or more embodiments of the present invention, a processing parameter is spacing of the showerhead from the wafer. Because deposition only occurs when RF power is activated, RF power time is directly correlated to deposition time. Since there are separate RF sources for each chamber in the PECVD system, the RF power time for each chamber can be varied as a means for controlling final film thickness. This provides the ability to control thin film final thickness even in the presence of differences in the chamber performances. Varying the spacing between the showerhead and the wafer also can control the film thickness, in this case, independently for each chamber. Similarly, changes in the supplied RF power also effects deposition rate. While film thickness is the measured output, it is appreciated that the information can be represented as a film deposition rate (film thickness per unit time) or as a film thickness profile (film thickness per unit area).

[0060] While film thickness and thickness uniformity are typically the tool behaviors being modeled, models for other film properties, such as stress and RI, can be developed by manipulating RF power and showerhead spacing or other processing variables. Multiple models for different film properties can be developed and used to describe the deposition process.

[0061] Once data from DOE runs are obtained, regression methods (or any other suitable method) may be used to determine a model that obeys the behavior of the process within the range of inputs that were used in the experiments. In one or more embodiments, the model for an i-chamber system is defined as shown in eq. (1),

Film _— thickness _ij =DR _ij ·time _i (1)

[0062] where i is the ith chamber of deposition chamber, and DR _ij is the deposition rate for annular region j of the wafer in chamber i, where no Einstein summation has been used for the indices. The model is determined for each region of the wafer, and together the models define a film thickness profile across the wafer. Thus, the model can predict a film thickness profile by entering hypothetical variables into the model equation. In use, a measured film thickness profile is used to further refine the model in order to obtain updated parameters and thus an updated process recipe.

[0063] The processing variable for a basic model is typically process time; however, additional parameters can be included in the model. The relationship can be expressed generally as:

Q _ij =g ( x ₁ ,x ₂ . . . x _n ) (2a)

[0064] where Q is some film property in region j on a wafer in chamber i which is the result of a processing run; g( ) is some linear or nonlinear function of x ₁ , x ₂ , . . . , x _n on recipe parameters or tool state parameters which affect the resulting film property Q. If the film property of interest is thickness, the function g( ) represents the deposition rate as a function of recipe parameters or tool state parameters. The thickness for each region j of wafer in chamber i would then be derived by multiplying the deposition rate by the deposition time as shown below.

Film _— thickness _ij =g _ij ( x ₁ ,x ₂ , . . . ,x _n )· time _i (2b)

[0065] A model including additional processing parameters is shown in eqs. (2c)-(2e).

Film _— thickness _ij =( c _1ij ·spacing _i +c _2ij ·power _i +c _3ij ·TEOS _— flow+c ₄ )· time _i (2c)

Stress _i =( b _1i ·spacing _i +b _2i ·power _i +b _3i ·TEOS _— flow+b ₄ ) (2d)

RI _i =( a _1i ·spacing _i +a _2i ·power _i +a _3i ·TEOS _— flow _i +a ₄ ) (2e)

[0066] where c _1ij through c _4ij are the parameters which provide the contribution of the particular processing parameter to the deposition rate in region j for a wafer in the i ^th chamber; b _1i through b _4i are the parameters which provide the contribution of the particular processing parameter to the film stress for the wafer in the ith chamber, and a _1i through a _4i are the parameters which provide the contribution of the particular processing parameter to the refractive index of the film to the wafer in the ith chamber. In one or more embodiments of the present invention, the film property, e.g., film thickness, is modeled in defined annular regions on the wafer. In one or more embodiments of the present invention, film properties, e.g., stress and refractive index, are modeled for the entire film.

[0067] The model takes into account common processing parameters that affect all chambers and independent processing parameters that affect each chamber individually. The relationships can also be adapted to reflect models where one or more regions of the film correspond to different annular regions of the wafer. This allows the controller to perform controls on multiple film regions simultaneously. This multiple region control provides control of within wafer uniformity. Thus, the model can account for an unlimited number of processing variables and permits their optimization while taking into consideration whether they affect all or only individual deposition chambers, or whether they affect different regions of the film differently.

[0068] In one or more embodiments of the present invention, the model may be further augmented to include the effect of the tool state. The tool state takes into consideration the effect of wear and use on the tool, here, a PECVD apparatus. This function is typically expressed as a scaling factor that takes the tool state into consideration. Factors that can affect tool state include idle time (time since last film deposition) and frequency of cleaning (or number of wafers deposited between cleaning or other shut down operation, such as preventative maintenance).

[0069] The first wafers coated after the deposition system has been idle typically have a different deposition rate than subsequently coated wafers, a situation known as the “first wafer effect”. In one or more embodiments of the present invention, the model is further modified to account for the effect of tool idle time on deposition rate. The model accounts for such variations on deposition by monitoring the idle time of the system and adjusting the deposition rate accordingly. Thus, a statement is placed within the model, which reflects the effect of idle time on processing, such as:

[0070] If (idle time)>5 min

Deposition time=x; (3)

[0071] Else

[0072] Deposition time=y.

[0073] This captures idle time dependence within the model. In one or more embodiments of the present invention, the model has a more gradual change from one deposition rate to another and is given by the following equation:

DR _idle =DR _{no idle} ·( d ₁ ·tan ⁻¹ ( d ₂ ·idle _— time+d ₃ )+ d ₄ ) (4)

[0074] where DR _idle is the deposition rate with the effect of idle time, DR _{no{overscore ( )}idle} is the deposition rate when there is no idle time, d ₁ and d ₄ determines the maximum change in deposition rate which is caused by idle time, d ₂ determines the rate at which this change occurs, and d ₃ determines at what idle time the change in deposition rate begins to be significant. In the general case, the effect of deposition rate or idle time can be given by the following equation:

DR ^idle =ƒ( DR _{no idle} ,idle _— time,x ₁ ,x ₂ , . . . x _n ) (5)

[0075] where ƒ( ) is some function which describes how the deposition rate is a function of: the deposition rate when there is no idle time, the idle time, and other past or current processing parameters related to the controller, tool state, or wafer state, here denoted by x ₁ , x ₂ , . . . , x _n .

[0076] The “first wafer effect” is a member of a broader class of events, in which a single wafer measurement differs significantly from previous and subsequent measurements run on a specific tool or resource and, as such, does not represent an accurate representation of the process tool during normal operation. Accordingly, when these measurements are used in a feedback control system, this erroneous information may cause the system performance to deteriorate. These sudden changes can be the result of abrupt changes in the processing equipment, such as starting up the process after the system has been idle for a time, or it can be due to processing errors, such as an error in the metrology system. Since these sudden changes do not accurately reflect the subsequent behavior of the process tool, a methodology is used to evaluate the reliability of the measurement.

[0077] In one or more embodiments of the present invention, a methodology is provided within the model for assessing the reliability of the measurement. The methodology (i) estimates the intrinsic variation in the process, (ii) determines when a recent measurement is outside normal operating variation and, if so, marks the data as suspicious, and (iii) ignores the data until a trend is determined from subsequent data. This methodology allows the system to be sensitive to changes that occur over more than one wafer, but also provides the system with robustness over metrology failures or situations similar to the first wafer effect.

[0078] One or more of the embodiments of the present invention are also capable of learning the extent of differences in the film property relationships between the first wafer and the next wafer on a resource (chamber or tool) specific basis, referred to generally as the “x-wafer effect.” This is done by either maintaining a separate feedback loop for the first wafer versus the other wafers or by capturing the difference in the film property between first wafer and subsequent wafer on a resource specific basis. Resource specific refers to a particular tool or process being used.

[0079] Another case of an x-wafer effect in a CVD process arises out of the frequency of a “clean function” that is run in a chamber. If a clean recipe is run in between every wafer the film property tends to be identical; however, the clean recipe is time consuming. In this situation the chamber clean recipe is run every x wafers where x typically varies from 2 to 6. There is a change in the film property related to the nth wafer after the clean was run. The film property as a function of clean can be stated as (1+n*a)*K, where n is the wafer number after clean and a and K are variables relating the variation. Such variations can be captured by the model. The value of ‘a’ can also be learned on a resource specific basis.

[0080] It has been observed that, for some films, the film deposition rate increases with each wafer processed without an intervening chamber cleaning. In particular, high dielectric BLOk™ films show increasing film deposition rate with time. While not being bound to any mode or theory of operation, it is believed that material that builds up on the chamber during deposition affects deposition rate; and in the case of BLOk™ films, increases deposition time. Cleaning the chamber removes the built up residues and restores the chamber to its initial deposition conditions. Cleaning typically involves use of a plasma during film deposition.

[0081] If one can model the change in deposition rate as a function of chamber cleaning frequency, the number (or frequency) of chamber cleans can be reduced while concurrently reducing the film thickness variation. This results in an improvement in the overall efficiency of the process.

[0082] In a DOE-based model, a DOE identifies a correlation between number of wafers processed and deposition rate. One assumes that at least most film thicknesses are close to the target and k is a constant determined using DOE that defines how a particular run effects the deposition rate. Each wafer will have a different effect on the deposition rate because deposition of each wafer affects the deposition rate of the next wafer.

[0083] The model can be expressed as: 1 $\begin{array}{cc}\mathrm{FT}=\left(1+k\sum _{i=1}^{N_w}T_i\right)·\mathrm{DR}·t& \left(6\right)\end{array}$

[0084] where FT is the film thickness, T _i is the target which corresponds to the i ^th wafer since the last clean, N ^w is the number of wafers processed since the last chamber clean, DR is the scaled deposition rate that corresponds to a chamber that has just been cleaned and t is the deposition time. The model thus predicts the wafer thickness (and thus the change in affect in wafer deposition rate) based upon the number of wafers run and depends upon an understanding of how k reflects the effect of wafer number on film deposition rate.

[0085] Another way of modeling the system is to use an on-line estimation approach. The approach assumes a linear relationship between the i ^th wafer and the i ^th+1 wafer. The model is shown in the following eqs.:

{circumflex over (F)}T=D{circumflex over (R)}·t (7a)

D{circumflex over (R)}=DR _i−1 +( DR _i−1 −DR ) (7b)

[0086] 2 $\begin{array}{cc}{\mathrm{DR}}_j=\frac{{\mathrm{FT}}_j}{t_j}& \text{(7c)}\end{array}$

[0087] where F{circumflex over (T)} is the predicted film thickness for wafer i, D{circumflex over (R)} is the predicted deposition rate, which corresponds to the process to be performed on the wafer i, and t _i is the deposition time for wafer i. DR _j is the deposition rate for all j and FT _j is the film thickness for all j, where j represents all the wafers used in the process between cleanings. The model employs a reset mechanism such that each time the chamber is cleaned, D{circumflex over (R)} _i−1 and D{circumflex over (R)} _i are reset to some nominal values that are representative of the deposition rate of a clean chamber.

[0088] Application of the model provides variation in the deposition recipe with each successive wafer so that the film thickness remains constant even though the deposition rate is varying. Furthermore, application of the model would permit extension of the time (or number of wafers) between cleanings. Since the cleaning operation is relatively long, this improves the overall efficiency of the tool and an improvement of the yield of the resultant product.

[0089] Once a process model is available, the model can be used to calculate an optimal set of recipe parameters in order to deposit a uniform film to a desired thickness. Conversely, using models such as those just described, a prediction for region-averaged film thickness can be calculated given the deposition time and any other parameters that are measured or varied. By individually optimizing for the regions j of the wafer, greater control over the total surface is attainable. Thus, greater within wafer film uniformity is achieved.

[0090] An exemplary optimization method, which can be used in determining an updated model (based upon the differences between measured and predicted values for a target output), solves the equation: 3 $\begin{array}{cc}\underset{x}{\min }f\left(y^{\mathrm{sp}},g\left(x\right)\right)& \left(8\right)\end{array}$

[0091] where x is a vector of recipe parameters and other processing parameters corresponding to the deposition recipe; g(x) is the model for the PECVD process which predicts the film properties based on a recipe and measurements related to tool state; y ^sp is a vector of the desired average region film thicknesses and/or other controlled film properties; and f (y ^sp , g(x)) is some function which is meant to compensate for the deviation between the model predictions g(x) and the desired values y ^sp . The updated model then is used to determine an updated deposition recipe.

[0092] Thus, the optimization method suggests that the model need not correct for 100% of the deviation from predicted values. A function may also be used, as contemplated by one or more embodiments of the present invention, to reflect uncertainty in the measured or calculated parameters, or to “damp” the effect of changing recipe parameters too quickly or to too great an extent. It is possible, for example, that without this “damping” effect the controller overcompensates for the measured deviations thereby necessitating another adjustment to react to the overcompensation. This leads to oscillations that may take several runs before the final, optimized conditions are realized.

[0093] Based upon this control method, the post-deposition film thickness is measured and the difference between the predicted thickness and the final (i.e., actual) thickness is determined. Other controlled film properties are measured, as needed by the model. In one or more embodiments of the present invention, the film property is measured on a lot-to-lot basis. In one or more embodiments of the present invention, the reliability of the data is assessed before the data is used in updating the model.

[0094] The error in prediction, also known as a bias, can then be linearly added into the model such that the actual final thickness more closely matches the predicted (and typically targeted) thickness. This bias is added to each region j, of wafer in chamber i, which is modeled as is shown in the following equation:

Film _— thickness _ij =g ( x ₁ ,x ₂ , . . . x _n ) _ij ·time _i +e _ij (9)

[0095] where e _ij is the bias term, which arises due to the difference between the predicted and actual amount, deposited for region j of wafer in chamber i. The process of linearly updating a model with bias terms based upon the difference between a model prediction and an actual measurement is part of at least some feedback control in one or more embodiments of the present invention.

[0096] Instead of (and/or, in addition to) use of the aforementioned bias, one or more embodiments of the present invention contemplate that an updated recipe can be calculated to optimize the available recipe parameters and to drive the predictions to a target value. The recipe parameters are changed such that the film thickness is made constant even though the deposition rate may be varying. A methodology that automatically changes the recipe to achieve consistent film thickness not only improves the consistency of the resultant film thickness, but also improves the productivity of the tool, since the system is subject to less frequent down time for reconditioning. This consistent film thickness then improves the yield of the resultant product.

[0097] Process model development and optimization are carried out with reference to a specific deposition system. That is, conditions that effect the thin film characteristics are specific to the type of thin film being deposited and the tool used for deposition. It is recognized that many other films are and can be deposited using PECVD, and that models for their deposition can be similarly developed using the methodology and guidelines set forth herein. In one or more embodiments of the present invention, it is contemplated that a separate model (or at least a supplement to a composite model) is created for each thin film that is deposited. Alternatively, a model may be developed in reference to a previously developed model. This model may be product specific and take the original model and scale it based upon the differences between the products.

[0098] An example the use of an initial model developed as described herein above to control the run-to-run average thickness and the thickness uniformity of the deposition process and to provide a feedback loop for updating the deposition recipe is shown schematically in FIG. 5 . Briefly, one or more wafers is processed according to a first deposition recipe. The actual number of wafers depends on the complexity of the model and can be about 10, or as many as 20-30 or more. A thickness measurement is taken across the deposited film to obtain a film thickness profile, which is compared to the predicted film thickness profile calculated by the model. If the measured film thickness profile indicates deviation from the predicted results, those deviations are used to update the model to better reflect the behavior of the processing tool. The updated model is used to determine an updated recipe, which is then used in a feedback loop to progressively match the behavior of the processing tool and to optimize the recipe so as to improve or maintain within wafer film thickness uniformity.

[0099] According to the processing flow diagram in FIG. 5 , initial processing conditions (e.g., an initial tool state and initial wafer state) are identified that will provide a desired film deposition profile in step 500 . The initial conditions may be determined empirically or by using the processing model of one or more embodiments of the present invention. If a processing model is used, a controller can use this model to calculate step times and processing parameters (i.e., to set the recipe for one or more incoming wafers) to deposit a film having a target (in some cases, a flat) profile on an incoming profile with a desired thickness, as shown in step 510 . Thin films are deposited according to the initial deposition recipe in the PECVD tool at step 520 . The thickness of the deposited film is measured and deviation from the predicted thickness is determined in step 530 . In step 540 it is determined whether the deviation between the predicted and observed behavior exceeds an established tolerance. If the deviation is within acceptable ranges, no changes are made to the model and the recipe is unchanged (step 550 ). If the deviation is outside acceptable limits, then this information is marked to trigger a change in the model as described in step 560 and this information is fed back to the model in step 570 and thus into the controller where the deposition recipe is optimized according to an updated model that takes the deviation from the predicted value into consideration. The deposition step can be repeated and further updates of the deposition recipe are possible.

[0100] As is the case in most feedback systems, the process variables that are measured on-line (in this case with an integrated metrology unit on the tool) are updated in the model based upon the error between the prediction and the actual measurement. In the case of PECVD-processed films, one or more embodiments of the present invention contemplate that both uniformity and thickness are measured on-line and are used for updating the process model. Other controlled film properties can be measured on-line or off-line. In some cases these measurements would be performed on a lot-to-lot basis. That is, upon completion of the lot (usually 25 wafers) the wafers are brought to an external metrology tool where several wafers of the lot are measured.

[0101] In one or more embodiments of the present invention, film properties, e.g., stress and refractive index, are not measured and are handled in much the same way output constraints are handled in model predictive control. The use of output constraints in mode predictive control can be seen in the following optimization relationship: 4 $\begin{array}{cc}\underset{x}{\min }f\left(y^{\mathrm{sp}},g\left(x\right)\right)& \left(10\right)\end{array}$ s.t. h ( x )≦0,

[0102] where h(x) is some constraint that is placed on the prediction of an unmeasured output. In one or more embodiments of the present invention, output constraints for the PECVD tool are applied to control the prediction of stress and refractive index. This optimization formulation constrains the prediction of the model to be within some limit, or set of limits, while still finding recipe parameters which yield the desired thickness and uniformity. Thus, as long as the recipe parameters are within stated maximum and minimum values, it is assumed that constrained output values are within allowable maximum and minimum values.

[0103] In one or more embodiments of the present invention, a feedback control methodology combines the chambers into a single model using the average of the tool states for each of the chambers. The single model would use the feedback approach described above to apportion the bias adjustment across the different chambers in some predetermined way.

[0104] When multiple process tools perform in series, also known as being run within a module, the performance of one tool can have a strong effect on the performance of subsequent tools. Accordingly, the performance of subsequent tools may be optimized by adjusting the performance of previous tools. For the specific case of ILD CVD, the standard way of performing the task is to deposit a film that has the most uniform film possible. Then, the ILD CPM is tasked with removing a certain amount of this film with as uniform a removal rate as possible. Unfortunately, the CMP removal profile is not as uniform as the deposition profile from the CVD tool. However, by manipulating the profile which results from the CVD tool, the shortcomings of the CMP tool can be addressed by providing an incoming profile which alleviates the resulting non-uniformities caused by the CMP tool.

[0105] Also, in one or more embodiments of the present invention, a feedback control scheme uses the final thickness measurements to distribute feedback individually to all of the chambers. Because each chamber can be can be treated individually, the tool state, i.e., cleaning frequency and idle time, can be included in the model and feedback can be specific to the chamber and deposition recipe. This feedback control scheme is particularly useful when different deposition recipes are being carried out in each chamber or when drift varies between chambers. The ability to separately model each chamber provides a greater of degree processing flexibility, since it allows one to change the processing recipe in one chamber (perhaps because film properties are drifting) while keeping the processing recipe at the remaining chamber unchanged (perhaps where film properties are within target ranges). When changes to the processing recipe are made to only one chamber, chamber-specific processing parameters are adjusted.

[0106] Feedback and feedforward control algorithms are constructed for use in the above control process based on the above models using various methods. The algorithms may be used to optimize parameters using various methods, such as recursive parameter estimation. Recursive parameter estimation is used in situations such as these, where it is desirable to update the model on line at the same time as the input-output data is received. Recursive parameter estimation is well suited for making decisions on line, such as adaptive control or adaptive predictions. For more details about the algorithms and theories of identification, see Ljung L., System Identification—Theory for the User , Prentice Hall, Upper Saddle River, N.J. 2nd edition, 1999.

[0107] In one or more embodiments of the present invention, the deposition recipe may be updated in discrete increments or steps defined in the algorithms of the model. Also, in one or more embodiments of the present invention, the updated recipes may be determined by interpolation to the appropriate parameters.

[0108] Additional apparatus utilized to implement the feedforward and feedback loop include tools for measuring a film property, e.g., a film thickness measurement (metrology) tool to provide thickness data needed to calculate film deposition rate. The tool may be positioned relative to the PECVD apparatus so as to provide in-line measurements, or it may be located remote from the apparatus. The tool may use optical, electrical, acoustic or mechanical measurement methods. A suitable thickness measurement device is available from Nanometrics (Milpitas, Calif.) or Nova Measuring Instruments (Phoenix, Ariz.). Other tools may be integrated into the system to provide measurement of film properties, such as trench depth, dopant concentration, refractive index or any other measurable film property that is modeled and controlled. The measurement is made wafer-to-wafer or lot-to-lot and may provide in-line or off-line measurements.

[0109] A computer may be utilized to calculate the optimal film deposition recipe based upon the measured film thickness and calculated deposition rate, employing the models and algorithms provided herein. A suitable integrated controller iAPC (integrated advanced process control) is available from Applied Materials (Santa Clara, Calif.).

[0110] Various aspects of the present invention that can be controlled by a computer can be (and/or be controlled by) any number of control/computer entities, including the one shown in FIG. 6 . Referring to FIG. 6 a bus 656 serves as the main information highway interconnecting the other components of system 611 . CPU 658 is the central processing unit of the system, performing calculations and logic operations required to execute the processes of embodiments of the present invention as well as other programs. Read only memory (ROM) 660 and random access memory (RAM) 662 constitute the main memory of the system. Disk controller 664 interfaces one or more disk drives to the system bus 656 . These disk drives are, for example, floppy disk drives 670 , or CD ROM or DVD (digital video disks) drives 666 , or internal or external hard drives 668 . These various disk drives and disk controllers are optional devices.

[0111] A display interface 672 interfaces display 648 and permits information from the bus 656 to be displayed on display 648 . Display 648 can be used in displaying a graphical user interface. Communications with external devices such as the other components of the system described above can occur utilizing, for example, communication port 674 . Optical fibers and/or electrical cables and/or conductors and/or optical communication (e.g., infrared, and the like) and/or wireless communication (e.g., radio frequency (RF), and the like) can be used as the transport medium between the external devices and communication port 674 . Peripheral interface 654 interfaces the keyboard 650 and mouse 652 , permitting input data to be transmitted to bus 656 . In addition to these components, system 611 also optionally includes an infrared transmitter and/or infrared receiver. Infrared transmitters are optionally utilized when the computer system is used in conjunction with one or more of the processing components/stations that transmits/receives data via infrared signal transmission. Instead of utilizing an infrared transmitter or infrared receiver, the computer system may also optionally use a low power radio transmitter 680 and/or a low power radio receiver 682 . The low power radio transmitter transmits the signal for reception by components of the production process, and receives signals from the components via the low power radio receiver. The low power radio transmitter and/or receiver are standard devices in industry.

[0112] Although system 611 in FIG. 6 is illustrated having a single processor, a single hard disk drive and a single local memory, system 611 is optionally suitably equipped with any multitude or combination of processors or storage devices. For example, system 611 may be replaced by, or combined with, any suitable processing system operative in accordance with the principles of embodiments of the present invention, including sophisticated calculators, and hand-held, laptop/notebook, mini, mainframe and super computers, as well as processing system network combinations of the same.

[0113] FIG. 7 is an illustration of an exemplary computer readable memory medium 784 utilizable for storing computer readable code or instructions. As one example, medium 784 may be used with disk drives illustrated in FIG. 6 . Typically, memory media such as floppy disks, or a CD ROM, or a digital video disk will contain, for example, a multi-byte locale for a single byte language and the program information for controlling the above system to enable the computer to perform the functions described herein. Alternatively, ROM 660 and/or RAM 662 illustrated in FIG. 7 can also be used to store the program information that is used to instruct the central processing unit 658 to perform the operations associated with the instant processes. Other examples of suitable computer readable media for storing information include magnetic, electronic, or optical (including holographic) storage, some combination thereof, etc. In addition, at least some embodiments of the present invention contemplate that the medium can be in the form of a transmission (e.g., digital or propagated signals).

[0114] In general, it should be emphasized that various components of embodiments of the present invention can be implemented in hardware, software or a combination thereof. In such embodiments, the various components and steps are implemented in hardware and/or software to perform the functions of the present invention. Any presently available or future developed computer software language and/or hardware components can be employed in such embodiments of the present invention. For example, at least some of the functionality mentioned above could be implemented using the C, C++, or any assembly language appropriate in view of the processor(s) being used. It could also be written in an interpretive environment such as Java and transported to multiple destinations to various users.

[0115] Although various embodiments that incorporate the teachings of the present invention have been shown and described in detail herein, those skilled in the art can readily devise many other varied embodiments that incorporate these teachings. All references mentioned herein are incorporated by reference.