This application claims the priority benefit of Taiwan application serial no. 103126779, filed on Aug. 5, 2014. The entirety of the above-mentioned patent application is hereby incorporated by reference herein and made a part of this specification.
The invention relates to a virtual metrology (VM) system and a VM method. More particularly, the invention relates to a VM system and a VM method configured to cluster data according to a plurality of predetermined patterns and perform a corresponding step if the obtained data meet expectations.
In the manufacturing-related industries, such as semiconductor and display panel manufacturing industries, quality control of workpieces produced by process equipment is required for ensuring stability of the manufacturing performance of the equipment and improving production and yield. Virtual metrology (VM) is a method to conjecture quality of a process workpiece according to the process data of the process equipment if no physical metrology operation is performed or can be performed on the process workpiece. When the VM technology is applied, the physical conditions or properties of chambers in one type of equipment or in one equipment are unlikely the same; hence, to make sure the conjecture accuracy, the conjecture model need be established according to the physical conditions or properties of different chambers in the equipment. In order not to spent significant labor cost and other costs on individually establishing respective models for different chambers in each equipment, an automatic virtual metrology (AVM) server and a method therefor have been proposed.
According to the AVM technology, a VM value and a feedback correction value are applied to ensure promptness and accuracy, e.g., the AVM process may be performed in a dual-phase manner. In phase I, the conjecture step is carried out by calculating the first-stage VM value of a certain workpiece right after the process data collection of the workpiece is completed, so as to satisfy the requirement for promptness. The Phase-II algorithm starts to collect the metrology data of a randomly selected workpiece in a cassette (for re-training or modulation) and then recalculate the phase-II VM data of all workpieces in the cassette to which the randomly selected workpiece belongs, so as to satisfy the requirement for accuracy. The AVM technology is also applied to generate a global similarity index (GSI) and a reliance index (RI) of the phase-I VM value and the phase-II VM value to quantify the reliability of the conjectured VM value. The AVM method saves tremendous time for introducing the VM algorithm to the chambers in the same type of equipment or in the same equipment as well as maintaining accuracy of real-time VM.
Please refer to
However, the existing VM schemes generally adopt the data driven methodology in different ways to ensure the correctness of the conjectured VM value. Such VM schemes often encounter the same problem, i.e., only issues arising from actual measurement can be recognized, and the correctness of data need be ensured by making corrections through feedback paths. In case of any unrecognizable issue, the feedback paths may worsen the problems, and thus real-time correction becomes impossible. Although a significant amount of the process data of manufacturing equipment may be collected, such data collection may merely serve to monitor the quality of the produced workpieces.
The invention is directed to a virtual metrology (VM) system and a VM method. In the VM system, the obtained data of tool conditions are clustered according to a plurality of predetermined patterns. The obtained data are calculated according to the patterns, so as to obtain a result. If the result meets expectations, a corresponding step is performed.
According to an embodiment of the invention, the corresponding step is a normal sampling step if the result indicates the obtained data meet one of the predetermined patterns. If the obtained data do not meet any of the predetermined patterns, an alarm is generated thereby, and the corresponding equipment may be shut down or other measures may be taken.
According to an embodiment of the invention, the corresponding step is a maintenance, repair, and overhaul step if the result indicates the obtained data meet one of the predetermined patterns.
According to an embodiment of the invention, if the obtained data of tool conditions do not meet any of the predetermined patterns, the data are stored; after certain amount of the data is accumulated, a new pattern is established, and the newly established pattern may dynamically serve as a basis for clustering data.
Several exemplary embodiments accompanied with figures are described in detail below to further describe the invention in details.
The invention is directed to a virtual metrology (VM) system and a VM method. In the VM system, the obtained data of tool conditions are clustered according to a plurality of predetermined patterns. The obtained data are calculated according to the patterns, so as to obtain a result. If the result meets expectations, a corresponding step is performed.
According to an embodiment of the invention, the corresponding step is a normal sampling step; that is, if the result indicates the obtained data meet one of the predetermined patterns, metrology is performed through carrying out the normal sampling step. For instance, the actual metrology data of a randomly selected workpiece are obtained as the basis for re-training or modulation, so as to satisfy the requirement for accuracy. If the obtained data do not meet any of the predetermined patterns, an alarm is generated thereby, and the corresponding equipment may be shut down or other measures may be taken. In an embodiment of the invention, if the obtained data do not meet any of the predetermined patterns, the data are stored and accumulated to a certain amount, and a new pattern for comparison is dynamically added.
In another embodiment of the invention, the corresponding step may be a maintenance step or a parts-replacing step. For instance, if the result meets one of the predetermined patterns, the maintenance step or the parts-replacing step is performed. If the result does not meet any of the predetermined patterns, the result is disregarded. However, if the manufacturing equipment is crashed because the maintenance step is not performed, the corresponding equipment parameters or data are stored and accumulated to a certain amount, and then a new pattern for comparison is dynamically added. A process apparatus is production equipment having a mechanical structure and various parts. The mechanical structure requires maintenance because of the accumulation of bi-products during production, and aging consumable parts require periodic replacement. The patterns can be associated with the accumulation state of bi-products during production or the aging state of the consumable parts.
According to the drawings, a VM system and a method therefor are provided below according to an embodiment of the invention, while the descriptions should not be construed as limitations to the invention.
The on-site system 210 is configured to performing processes for manufacturing various workpieces, and the manufacturing processes may be online/offline manual/automated manufacturing processes. Here, the on-site system 210 may include at least one manufacturing equipment 212, metrology equipment 214, other computer hosts, and other equipment that may be applied to complete the manufacturing processes. Through an automated interface 216, the on-site system 210 may communicate with and exchange data with the equipment control system 250 and the data storage system 220.
The manufacturing equipment 212 is, for instance, an ion implanter, a thermal reactor, an etcher, a lithography machine, and similar devices used in the semiconductor manufacturing process. The metrology equipment 214 may include ellipsometer, interferometer, scanning electron microscopy (SEM), or the like.
The automated interface 216 is connected to the manufacturing equipment 212, the metrology equipment 214, and other computer hosts, so as to obtain the process data. According to an embodiment of the invention, the automated interface 216, the manufacturing equipment 212, and the metrology equipment 214 may be connected through a connection interface for data exchange. Here, the connection interface includes a semiconductor equipment communication standards (SECS) interface, a generic model for communication and control of manufacturing equipment (GEM) interface, a SECS/GEM interface, and an equipment data acquisition (EDA) interface (i.e., interface A).
The on-site manufacturing system 210, the equipment control system 250, and the data storage system 220 communicate with each other and exchange data. The data storage system 220 may include a database, a file system, or any other data stored in a non-volatile memory or a volatile memory; here, the non-volatile memory is a hard disc, a cassette player, an optical storage medium, etc; the volatile memory may be a random access memory, for instance. In an embodiment of the invention, the data storage system 220 can obtain data from a plurality of data storage sources, e.g., from an equipment maintenance data source, a metrology data source, a process data source, and so on.
The data storage system 220 can store process data (e.g., manufacturing recipes), such as temperature, pressure, the used chemical materials, manufacturing time, etc. In addition, the data storage system 220 stores historical maintenance data of the manufacturing equipment, data of stocks, and so forth. The data storage system 220 also stores the process data obtained from the manufacturing equipment 212 of the on-site manufacturing system 210 or the metrology data obtained from the metrology equipment 214.
The process data may be process characterized parameters (e.g., different physical conditions or properties) executed by the manufacturing equipment, and these parameters may be obtained by sensors of the equipment and/or by operation parameters of the equipment and are collectively referred to as the process data. Among the process data, the data obtained by the sensors of the equipment exemplarily include pressure of a chamber, temperature, RF power, or RF reflection power, for instance. By contrast, among the process data, the operation parameters of the equipment are predetermined flow rates (of chemical reaction solvent, for instance), throttle valve settings (e.g., chemical chamber exhaust vacuum pump settings), and so forth.
The metrology data may include a wafer thickness metrology value (e.g., measured by an ellipsometer), a particle quantity metrology value (e.g., measured by scanning electron microscopy, SEM), a wafer curvature metrology value (e.g., measured by an interferometer), and so forth, for instance.
The equipment control system 250 may manage all or parts of operations in the manufacturing factory; besides, the equipment control system 250 may include online/offline manual/automated processes, and the control computations and operations are done by calculators or host servers. The processes, for instance, include a tracking and monitoring process performed on equipment, a delivery process of materials, a labor scheduling and adjusting process, etc. The equipment control system 250 may include a fault detection and classification (FDC) component 230 and a VM component 240.
The FDC component 230 may timely obtain data from the automated interface 216 of the on-site manufacturing system 210 and obtain data from the data storage system 220. The VM component 240 timely obtains the tools conditions of the manufacturing equipment or other data required by the VM method described herein through the automated interface 216 and obtain data from the data storage system 220. The FDC component 230 is connected to the VM component 240, and VM models of the VM component 240 may be employed to conjecture or measure data according to other metrology values or process data.
In the VM system and the VM method provided herein, the obtained data of tool conditions are clustered according to a plurality of predetermined patterns. The obtained data are calculated according to the patterns, so as to obtain a result. If the result meets expectations, a corresponding step is performed.
In an embodiment of the invention, the manufacturing environment is provided in the present embodiment with reference to
The data processing unit 330 is connected to the metrology equipment 310 to obtain the metrology data, the process data including manufacturing recipes stored in the data storage system depicted in
The conjecture module 350 is a software module or a firmware module the may be executed by parts of the equipment of the equipment control system 250; hence, no relevant descriptions are provided hereinafter, and the invention should not be limited thereto. Besides, the conjecture module 350 serves to gather the process data and cluster the process data to obtain a plurality of data clusters. The data clusters are compared with a plurality of patterns and are calculated according to the patterns to obtain a result. If the result meets expectations, a corresponding step is performed.
In an embodiment of the invention, the conjecture module 350 of the VM system includes an auto cluster module 354 and a plurality of VM processors; in the present embodiment, the VM system includes N VM processors, and only VM processors 356a, 356b, and 356c are provided herein for illustrative purposes. The conjecture module 350 obtains the process data Xi to Xn (e.g., the data obtained from the FDC component) of the equipment, and the obtained data are automatically clustered by an automatic clustering module 354, e.g., by selecting all or parts of the process data Xi to Xn required by each corresponding pattern.
The automatic clustering module 354 sends different data collections to the corresponding N VM processors; for instance, the data Xa (e.g., process data X1, X3, X5, and X7), the data Xb (e.g., process data X2, X4, X6, and X8), and the data Xc (e.g., process data X1, X5, X6, and X8) of the corresponding VM processors 356a, 356b, and 356c are sent to obtain different VM outputs 358a, 358b, and 358c. The data may be transformed to the output model 353 generated by the conjecture module 350.
The conjecture module 350 may selectively include a physical model 352 for making conjectures according to the obtained process data and generating an output model 351 based on the conjectures. The output model 353 generated by the conjecture module 350 and/or the output model 351 selectively generated by the conjecture module 350 may serve to generate the reliance index (RI) of the VM value, so as to quantify the reliability of the conjectured VM value or send the data to the next-stage equipment 360 for other purposes.
With reference to
After the process data Xi to Xn are completely clustered, in step S508, it is determined whether there is an existing model or pattern which is met; if yes, the conjectured value VP corresponding to a certain pattern is obtained by calculation. If there is no corresponding pattern, step S512 is performed to make sure no conjectured value corresponding to the data Y is output; in step S514, a corresponding step (e.g., issuing a warning alarm or temporarily stopping the equipment) is performed, and the manufactured and produced workpiece is physically measured. The obtained data and the measured result are respectively stored into individual sources; for instance, the data storage system depicted in
The predetermined patterns are obtained by performing multiple inductions on a plurality of manufacturing recipes (including the process data obtained from the same or similar past manufacturing processes) and corresponding metrology data to find a plurality of models. For instance, in exemplary groups A, B, C, and D, different process data have different mean coefficients, and the input data Y are obtained by defining the significantly accumulated historical data as follows:
Y=A1X1+A2X3+A3X5+A4X6+A5X7+A6X8+A0
Here, the conjectured value Yp may be obtained by calculating the process data X1, X3, X5, X6, X7, and X8, and the corresponding mean coefficients are A1, A2, A3, A4, A5, A6, and A0, which represents a pattern.
The significantly accumulated historical data may be the obtained metrology data. For instance, the metrology data may include a wafer thickness metrology value (e.g., measured by an ellipsometer), a particle quantity metrology value (e.g., measured by scanning electron microscopy, SEM), a wafer curvature metrology value (e.g., measured by an interferometer), and so forth. The significantly accumulated process data and the correspondingly obtained metrology data may be applied to establish different models and thus generate different patterns.
As to the auto-classification technology, a chemical vapor deposition (CVD) process is performed for oxide thickness prediction in an embodiment of the invention, which will be explained hereinafter.
The process data WHX1, X2, X3, . . . , X10) are obtained, wherein Wj is the jth wafer sample, Xi is the FDC parameter corresponding to the jth wafer sample, and in total there are N wafers. In the VM system and the VM method provided in an embodiment of the invention, the process data (e.g., Wj=(X1, X2, X3, . . . , X10)) of equipment are obtained. Here, Wj is the jth wafer sample, Xi is the FDC parameter corresponding to the jth wafer sample, and in total there are N wafers, whereby the data collections and the predetermined patterns are clustered. The patterns specifically meeting the manufacturing process or the equipment are obtained by perfoiming inductions on the accumulated historical data; for instance, four data clusters with the mean coefficients Ai, Bi, Ci, and Di are adopted.
Group A:
Y=A1X1+A2X3+A3X5+A4X6+A5X7+A6X8+A0;
Group B:
Y=B1X2+B2X3+B3X4+B4X5+B5X6+B6X7+B7X8+B8X9+B0;
Group C:
Y=C1X1 +C2X3+C3X5+C4X6+B5X7+C0;
Group D:
Y=D1X+B2X2+B3X4+B4X7+B5X8+B6X9+B7X10+B0
Through said equations, the corresponding conjectured value Yp can be obtained. If the obtained data meet one of the predetermined patterns, e.g., if the obtained data is compared with a predetermined threshold value or a predetermined value, a normal sampling step is performed for measurement.
According to the VM method described in an embodiment of the inventions, the corresponding step is a maintenance, repair or overhaul step if the result meets one of the predetermined patterns. If the result does not meet any of the predetermined patterns, the result is disregarded. However, if the manufacturing equipment is crashed because the maintenance repair or overhaul step is not performed, the corresponding equipment parameters or data are stored and accumulated to a certain amount, and then a new pattern for comparison is dynamically added. The process apparatus is the production equipment having the mechanical structure and various parts. The mechanical structure requires maintenance because of the accumulation of bi-products during production, and aging consumable parts require periodic replacement. The patterns can be associated with the accumulation state of bi-products during production or the aging state of the consumable parts.
Although the invention has been described with reference to the above embodiments, it will be apparent to one of ordinary skill in the art that modifications to the described embodiments may be made without departing from the spirit of the invention. Accordingly, the scope of the invention will be defined by the attached claims and not by the above detailed descriptions.
Number | Date | Country | Kind |
---|---|---|---|
103126779 | Aug 2014 | TW | national |