Methods for providing run to run process control using a dynamic tuner

Abstract

Methods for providing run to run process control using a dynamic tuner are provided. Once such method includes receiving a data point for a process output parameter, determining whether the data point is within a desired range for the process output parameter, setting, when the data point is within the desired range, a dynamic lambda value equal to a preselected base lambda value, setting, when the data point is not within the desired range, the dynamic lambda value equal to a value based on the preselected base lambda value, a degree of difference between the data point and a target for the process output parameter, and a scale factor, calculating an exponentially weighted moving average using the dynamic lambda value, and adjusting the process control parameter in accordance with the exponentially weighted moving average.

Description

FIELD

The present invention relates to manufacturing processes, and more specifically to methods for providing run to run process control using a dynamic tuner.

BACKGROUND

Run to run process control has been widely developed and utilized in different manufacturing industries where the processing conditions of next run are adjusted based on prior run results. A simple exponentially weighted moving average (EWMA) tuner is often adopted in the run to run process control system to estimate the deviation of model parameters. The simple EWMA tuner can be expressed using the formula:P_n+1=λ×R_n+(1−λ)×P_n

where:

• • P=Process Control Parameter
  • R=Predicted Process Model Parameter based on Actual Data (e.g., new data point)
  • λ=EWMA weighing factor (0<λ<1)
  • n=nth run

The EWMA weighing factor, lambda, is usually carefully selected for a process in order to get adequate process tuning. One such approach involves determining an optimum/fixed lambda value based on process capability index values and is described in U.S. Pat. No. 7,809,459 to Morisawa et. al. The simple EWMA tuner is usually quite effective to bring a process with moderate drifts under control, but may not be able to tune the process quick enough to address severe process shifts. Using a large fixed lambda value with the EWMA tuner as might be used in the Morisawa system could accelerate process tuning, but it does not allow for adjustment of the lambda value on the fly and thus introduces a high risk of process over-tuning.

A predictor corrector control scheme, which can also be referred to as a double EWMA controller, is proposed in an article by S. Butler and J. Stefani entitled, “Supervisory Run-to-Run Control of Polysilicon Gate Etch Using In Situ Ellipsometry,” IEEE Trans. Semiconduct. Manufact., vol. 7, no. 2, pp. 193-201, 1994. In this predictor control scheme, a second EWMA tuner is used to compensate for the error incurred from the simple EWMA tuner, so extra tuning can be applied to the model control parameter when large process drifts occur. However, while this approach may be capable of adapting to small changes in process output, it is not well suited to adapting to high magnitude or short time interval type changes in process output.

Another approach that deals with both process drifts and process shifts at the same time has been proposed by R. Guo, A. Chen, and J. Chen, in a portion of a book entitled, “Chapter 19 An Enhanced EWMA Controller for Processes Subject to Random Disturbances,” Run-to-Run Control in Semiconductor Manufacturing, edited by J. Moyne, E. del Castillo, and A. M. Hurwitz, CRC Press LLC, 2001. In the enhanced EWMA controller proposed by Guo, a baseline lambda is utilized to compensate for smaller process drifts. Two EWMA control charts are used as the detection tools for large and medium process shifts. If there is an out-of-control signal on the charts, a dynamic tuning loop is triggered and the lambda value is reset to a higher value. The lambda value is then decreased gradually over the next few runs, and eventually back to the baseline lambda. However, while this approach may be capable of some dynamic tuning using the control charts to tune to particular shifts, it is not well suited for rapid fluctuations in process output as the approach does not tune particularly fast.

Accordingly, an improved run to run control system that can address these shortcomings is needed.

SUMMARY

Aspects of the invention relate to methods for providing run to run process control using a dynamic tuner. In one embodiment, the invention relates to a method for providing run to run process control of a process control parameter using a dynamic tuner, the method including receiving a data point for a process output parameter, determining whether the data point is within a desired range for the process output parameter, setting, when the data point is within the desired range, a dynamic lambda value equal to a preselected base lambda value, setting, when the data point is not within the desired range, the dynamic lambda value equal to a value based on the preselected base lambda value, a degree of difference between the data point and a target for the process output parameter, and a scale factor, calculating an exponentially weighted moving average using the dynamic lambda value, and adjusting the process control parameter in accordance with the exponentially weighted moving average.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG. 1 is a flowchart of a general method for providing run to run process control of a process control parameter in accordance with one embodiment of the invention.

FIG. 2 is an illustration of an exponentially weighted moving average formula with a dynamic lambda that can be used to provide dynamic control of a process model/control parameter in accordance with one embodiment of the invention.

FIG. 3 is a flowchart of a particular method for providing run to run process control of a process model/control parameter using the dynamic tuning formula of FIG. 2 in accordance with one embodiment of the invention.

FIG. 4 is a graph illustrating the comparative performance of a dynamic EWMA tuner versus a simple EWMA tuner using a relatively low fixed lambda value for an example process output parameter that experiences a relatively severe process shift in accordance with one embodiment of the invention.

FIG. 5 is a graph illustrating the comparative performance of a dynamic EWMA tuner versus a simple EWMA tuner using a relatively high fixed lambda value for an example process output parameter that experiences the relatively severe process shift in accordance with one embodiment of the invention.

FIG. 6 is a graph illustrating the comparative performance of a dynamic EWMA tuner versus a simple EWMA tuner using a relatively high fixed lambda value for an example process output parameter without the relatively severe process shift of FIGS. 4 and 5 in accordance with one embodiment of the invention.

FIG. 7 is a table illustrating the comparative performance of a dynamic EWMA tuner versus a simple EWMA tuner using various lambda values, where the performance results include the number of data points that fell outside of the desired process output range and the resulting wafer to wafer sigma in accordance with one embodiment of the invention.

DETAILED DESCRIPTION

Referring now to the drawings, embodiments of processes for providing run to run process control of a process control parameter are illustrated. The processes apply an exponentially weighted moving average (EWMA) formula with a dynamic lambda to control a process parameter (e.g., process control parameter or process model parameter). When the process output parameter extends beyond a preselected/desired range, the dynamic lambda can have an aggressive value determined in part by a preselected scale factor and a degree to which the process output parameter extends beyond a target range. On the other hand, when the process output parameter is within the preselected/desired range, the dynamic lambda can have a less aggressive value (e.g., base lambda value) such as a preselected constant lambda value. The EWMA is then calculated using the dynamic lambda value and a process control parameter is adjusted in accordance with the calculated EWMA. The process can then repeat with a new data value, thereby enabling dynamic tuning.

FIG. 1 is a flowchart of a general process 100 for providing run to run process control of a process control parameter in accordance with one embodiment of the invention. The process first receives (102) a data point for a process output parameter (e.g., “key process output parameter”, “just process output parameter” or “process output”). The process then determines (104) whether the data point is within a preselected or desired range for the process output parameter. The preselected/desired range can be defined by a preselected upper desired limit for the process output parameter and a preselected lower desired limit for the process output parameter.

The process then sets (106), when the data point is within the desired range, a dynamic lambda value equal to a preselected base lambda value. In several embodiments, when the data point is within the desired range, the process uses the base lambda value that corresponds to a less aggressive value than the value the process will use when the data point is beyond the desired range. The process then sets (108), when the data point is not within the desired range, the dynamic lambda value equal to a value based on the preselected base lambda value, a degree of difference between the data point and a target for the process output parameter, and a scale factor. The scale factor can be selected based on empirical data and a preselected degree of aggression that is desired for tuning the process control parameter to meet the target for the process output. In several embodiments, a higher scale factor corresponds to a higher degree of aggression. In several embodiments, the scale factor is selected to be greater than 0.

The target value of the process output is usually chosen based on process or device performance requirements. Different processes have different target values. After the target value is set for a process output, process control can be adjusted based on the target value, which is usually not changed unless the process or device performance requirements change.

The process then calculates (110) an exponentially weighted moving average (EWMA) using the dynamic lambda value. In several embodiments, the conventional formula for EWMA is used along with the dynamic lambda value to generate the EWMA output value. The process then adjusts (112) the process control parameter in accordance with the exponentially weighted moving average output value.

In one embodiment, the process control parameter can be a polishing rate, a deposition rate, an etch rate, an angle control for a material removal process, or a chemical mechanical planarization rate. In one embodiment, the process control parameter corresponds to a back pressure for a chemical mechanical planarization process. In other embodiments, the process control parameter can correspond to other control parameters of a manufacturing process. In several embodiments, the process control parameter can correspond to control parameters of a manufacturing process for magnetic transducers used in magnetic storage devices, including, for example sliders for hard disk drives.

In one embodiment, the process can perform the sequence of actions in a different order. In another embodiment, the process can skip one or more of the actions. In other embodiments, one or more of the actions are performed simultaneously. In some embodiments, additional actions can be performed.

FIG. 2 is an illustration 200 of an exponentially weighted moving average formula with a dynamic lambda 202 that can be used to provide dynamic control of a process control parameter in accordance with one embodiment of the invention. In several embodiments, the formula 202 can be used in conjunction with the control processes of FIG. 1 or FIG. 3. The illustration 200 further includes a variable definition block 204 that provides definitions for each of the variables in the formula 202. While the expression in the formula 202 for the estimated process parameter (e.g., P_n+1=λ_d×R_n+(1−λ_n)×P_n) is a known formula for computing an exponentially weighted moving average, the use of a dynamic lambda value and the corresponding formula (λ_d=|Y_n−Y_target|×SF×DF) is not believed to be known in the art. As indicated in the variable definition block 204, the dynamic lambda and base lambda each are values between 0 and 1. The deciding factor (DF) is either 0 or 1 effectively acting as a switch between using the base lambda value (e.g., when the DF is 0) and the dynamic lambda (e.g., when the DF is 1) depending on whether the current/new value for the key process output parameter (Y) is within a desired range defined by a preselected upper desired limit (UDL) and a preselected lower desired limit (LDL).

In a number of embodiments, the scale factor (SF) is greater than 0. In some embodiments, the scale factor (SF) is greater than 1. In one embodiment, the scale factor (SF) is greater than 1 and less than 10. In other embodiments, other suitable values for the scale factor (SF) can be used. In several embodiments, a higher scale factor (SF) corresponds to a higher degree of aggression for tuning the process control parameter to meet the target of the process output. In one such embodiment, the scale factor (SF) is selected such that the dynamic lambda remains between 0 and 1. In another such embodiment, the scale factor (SF) is increased to compensate for a relatively low data point (e.g., maintain lambda between 0 and 1).

FIG. 3 is a flowchart of a particular process 300 for providing run to run process control of a process control parameter using the dynamic tuning formula 202 of FIG. 2 in accordance with one embodiment of the invention. The process can first obtain (302) a new process output parameter data (Yn). The process calculates (304) the predicted process model/control parameter, Rn, based on the new data (Yn). The process then determines (306) whether the new process output parameter data (Yn) is outside of a desired range. If the new process output parameter data (Yn) is not outside of the desired range, then the process sets (308) the deciding factor (DF) equal to 0 and as a result the dynamic lambda formula simplifies and the process sets (310) the dynamic lambda (λ_d) equal to the base lambda (λ_b). From block 310, the process adjusts (312) the process model/control parameter in accordance with the EWMA formula (e.g., P_n+1=λ_d×R_n+(1−λ_d)×P_n) and the dynamic lambda value.

On the other hand, if the new process output parameter data (Yn) is outside of the desired range, then the process sets (314) the deciding factor (DF) equal to 1 and calculates (316) the dynamic lambda (λ_d) according to the formula equal to the base lambda (λ_b) plus a value equal to the scale factor (SF) multiplied by the absolute value of the difference between the new process output parameter data (Yn) and a preselected target value for the process output parameter (Y_target). In such case, the process continues from block 316 and adjusts (312) the process model/control parameter in accordance with the EWMA formula (e.g., P_n+1=λ_d×R_n+(1−λd)×P_n) using the calculated dynamic lambda value. From block 312, the process uses (318) the process model/control parameter to determine the process conditions for the next run/iteration of the process. The process then executes (320) the next run/iteration of the process by returning to block 302 to obtain new process output parameter data and thereby restarts the process.

In one embodiment, the process can perform the sequence of actions in a different order. In another embodiment, the process can skip one or more of the actions. In other embodiments, one or more of the actions are performed simultaneously. In some embodiments, additional actions can be performed.

FIG. 4 is a graph 400 illustrating the comparative performance of a dynamic EWMA tuner 402 versus a simple EWMA tuner 404 using a relatively low fixed lambda value 406 for an example process output parameter that experiences a relatively severe process shift in accordance with one embodiment of the invention. As can be seen in FIG. 4, the fixed lambda value 406 is set at 0.2 (a relatively low value for the fixed lambda) while the dynamic lambda value 408 varies during the process and ranges from about 0.2 to about 0.7. The desired range 410 for the key process output parameter is marked on the graph with a box formed with dashed lines while the target for the key process output parameter is about 0.

Note that the severe process shift occurred at about run number 6, where the key process output parameter became much higher than the target, which is 0. Under the simple EWMA tuner 404 with the fixed lambda value of 0.2, the key process output did not go back to the desired range until about run number 18. However, using the dynamic EWMA tuner 402, where the base lambda was set also to be 0.2 and the dynamic lambda could go up to 0.7, the key process output was tuned back into the desired range at about run number 10, which is much faster than the simple EWMA tuner 404. The dynamic lambda 408 from run number 6 to run number 9 is higher than the base lambda 406 since the key process output is out of the desired range. The dynamic lambda values 408 for those runs also changed on the fly, depending on how far the key process output was away from the target. Once the key process output gets into the desired range, the dynamic tuner 402 uses the base lambda 406 to continue tuning the process closer to the target.

FIG. 5 is a graph 500 illustrating the comparative performance of a dynamic EWMA tuner 502 versus a simple EWMA tuner 504 using a relatively high fixed lambda value 506 for an example process output parameter that experiences the relatively severe process shift in accordance with one embodiment of the invention. As can be seen in FIG. 5 and as is distinguished from the simple EWMA tuner of FIG. 4, the fixed lambda value 506 is set at 0.7 (a relatively high value for the fixed lambda) while the dynamic lambda value 508 varies during the process in a range from about 0.2 to about 0.7. The desired range 510 for the key process output parameter is marked on the graph with a box formed with dashed lines while the target for the key process output parameter is about 0.

As mentioned above, using a large lambda value in the simple EWMA tuner could accelerate process tuning, but it also introduces a high risk of process over-tuning. FIG. 5 illustrates that when using a large fixed lambda value of 0.7 for the simple EWMA tuner in place of the relatively low fixed lambda of FIG. 4, the simple EWMA tuner 504 brings the key process output closer to the target quickly, similar to the dynamic tuner 502. However, when there is no severe process shift, the same simple EWMA tuner 502 will also over-tune the process output parameter.

FIG. 6 is a graph 600 illustrating the comparative performance of a dynamic EWMA tuner 602 versus a simple EWMA tuner 604 using a relatively high fixed lambda value for an example process output parameter without the relatively severe process shift of FIGS. 4 and 5 in accordance with one embodiment of the invention. As can be seen in FIG. 6, the fixed lambda value 606 is set at 0.7 (a relatively high value for the fixed lambda) while the dynamic lambda value 608 varies during the process in a range from about 0.2 to about 0.7. The desired range 610 for the key process output parameter is marked on the graph with a box formed with dashed lines while the target for the key process output parameter is about 0.

As illustrated in FIG. 6, which is similar in regard to the graph of FIG. 5 except that the severe process shift has been removed, the simple EWMA tuner 604 using the relatively high fixed lambda value 606 can and does over-tune the process output parameter. More specifically, the key process output is outside of the desired range for a total of 7 runs when using the same simple EWMA tuner 604 with the fixed lambda value of 0.7 as was used in FIG. 5. On the other hand, when using the same dynamic tuner 602, only 2 runs were out of the desired range. The wafer-to-wafer sigma of the key process output using the dynamic tuner 602 was also lower than that of the simple EWMA tuner 604.

FIG. 7 is a table 700 illustrating the comparative performance of a dynamic EWMA tuner 702 versus a simple EWMA tuner 704 using various lambda values, where the performance results include the number of data points that fall outside of the desired process output range 706 and the resulting wafer to wafer sigma 708 in accordance with one embodiment of the invention. As can be seen from FIG. 7, the dynamic EWMA tuner 702 outperformed the simple EWMA tuner 704 despite the simple EWMA tuner 704 having used various fixed lambda values including 0.2, 0.45, and 0.7. More specifically, the dynamic EWMA tuner 702 had much fewer data points outside of the desired range than the simple EWMA tuner 704. Also, the dynamic EWMA tuner 702 had lower wafer to wafer sigma (e.g., wafer to wafer variation) than the simple EWMA tuner 704.

While the above description contains many specific embodiments of the invention, these should not be construed as limitations on the scope of the invention, but rather as examples of specific embodiments thereof. Accordingly, the scope of the invention should be determined not by the embodiments illustrated, but by the appended claims and their equivalents.

Claims

1. A method for providing run to run process control of a process control parameter using a dynamic tuner, the method comprising: receiving a data point for a process output parameter at a processor;determining, by the processor, whether the data point is within a desired range for the process output parameter;setting, by the processor, when the data point is within the desired range, a dynamic lambda value equal to a preselected base lambda value;setting, by the processor, when the data point is not within the desired range, the dynamic lambda value equal to a value based on the preselected base lambda value, a degree of difference between the data point and a target for the process output parameter, and a scale factor;calculating, by the processor, an exponentially weighted moving average using the dynamic lambda value;adjusting, by the processor, the process control parameter in accordance with the exponentially weighted moving average to produce an adjusted process control parameter; andcontrolling a manufacturing device using the adjusted process control parameter.

2. The method of claim 1, further comprising calculating a predicted control process parameter based on the data point, wherein the calculating the exponentially weighted moving average using the dynamic lambda value comprises using the predicted process control parameter.

3. The method of claim 1, wherein the desired range is determined based on empirical data.

4. The method of claim 1, wherein the scale factor represents a preselected degree of aggression for tuning the process control parameter to meet the target for the process output parameter.

5. The method of claim 1, wherein a higher value for the scale factor corresponds to a higher degree of aggression for tuning the process control parameter to meet the target for the process output parameter.

6. The method of claim 1, wherein the scale factor is a number greater than 0.

7. The method of claim 1, wherein the process control parameter is selected from the group consisting of a polishing rate, a deposition rate, an etch rate, and an angle control for a material removal process.

8. The method of claim 1, wherein the process control parameter comprises a back pressure for a chemical mechanical planarization process.

9. The method of claim 1, wherein the setting, when the data point is not within the desired range, the dynamic lambda value comprises setting the dynamic lambda value equal to a first value of the preselected base lambda value plus an absolute value of the data point minus the target for the process output parameter where the first value is multiplied by the scale factor.

10. The method of claim 1, wherein the setting, when the data point is not within the desired range, the dynamic lambda value comprises decreasing the dynamic lambda value when a value, equal to an absolute value of the data point minus the target for the process output parameter, decreases.

11. The method of claim 1, wherein the setting, when the data point is not within the desired range, the dynamic lambda value comprises increasing the dynamic lambda value when a value, equal to an absolute value of the data point minus the target for the process output parameter, increases.

12. The method of claim 1, wherein the scale factor is selected based on a degree of difference between the data point and the target for the process output parameter such that the dynamic lambda value is greater than 0 and less than 1.

13. The method of claim 1, wherein the preselected base lambda value is greater than 0 and less than 1.

14. The method of claim 1, wherein the dynamic lambda value is greater than 0 and less than 1.