A reliable control system for measurement on film thickness in copper chemical mechanical planarization system Hongkai Li, Zilian Qu, Qian Zhao, Fangxin Tian, Dewen Zhao, Yonggang Meng, and Xinchun Lu Citation: Review of Scientific Instruments 84, 125101 (2013); doi: 10.1063/1.4833396 View online: http://dx.doi.org/10.1063/1.4833396 View Table of Contents: http://scitation.aip.org/content/aip/journal/rsi/84/12?ver=pdfcov Published by the AIP Publishing Articles you may be interested in Chemical mechanical planarization of gold J. Vac. Sci. Technol. A 32, 021402 (2014); 10.1116/1.4863275 Effects of manganese oxide–mixed abrasive slurry on the tetraethyl orthosilicate oxide chemical mechanical polishing for planarization of interlayer dielectric film in the multilevel interconnection J. Vac. Sci. Technol. A 26, 996 (2008); 10.1116/1.2936225 Applications of Raman spectroscopy in copper chemical mechanical planarization: In situ detection of tantalum layer to dielectric layer transition J. Appl. Phys. 100, 014907 (2006); 10.1063/1.2212087 Investigation of the nonuniformities in polyurethane chemical mechanical planarization pads J. Vac. Sci. Technol. B 24, 25 (2006); 10.1116/1.2137339 Chemical mechanical planarization characteristics of W O 3 thin film for gas sensing J. Vac. Sci. Technol. A 23, 737 (2005); 10.1116/1.1868612
This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitationnew.aip.org/termsconditions. Downloaded to IP: 65.39.42.160 On: Mon, 08 Dec 2014 22:05:47
REVIEW OF SCIENTIFIC INSTRUMENTS 84, 125101 (2013)
A reliable control system for measurement on film thickness in copper chemical mechanical planarization system Hongkai Li, Zilian Qu, Qian Zhao, Fangxin Tian, Dewen Zhao, Yonggang Meng, and Xinchun Lua) State Key Laboratory of Tribology, Tsinghua University, Beijing 100084, China
(Received 3 June 2013; accepted 11 November 2013; published online 2 December 2013) In recent years, a variety of film thickness measurement techniques for copper chemical mechanical planarization (CMP) are subsequently proposed. In this paper, the eddy-current technique is used. In the control system of the CMP tool developed in the State Key Laboratory of Tribology, there are in situ module and off-line module for measurement subsystem. The in situ module can get the thickness of copper film on wafer surface in real time, and accurately judge when the CMP process should stop. This is called end-point detection. The off-line module is used for multi-points measurement after CMP process, in order to know the thickness of remained copper film. The whole control system is structured with two levels, and the physical connection between the upper and the lower is achieved by the industrial Ethernet. The process flow includes calibration and measurement, and there are different algorithms for two modules. In the process of software development, C++ is chosen as the programming language, in combination with Qt OpenSource to design two modules’ GUI and OPC technology to implement the communication between the two levels. In addition, the drawing function is developed relying on Matlab, enriching the software functions of the off-line module. The result shows that the control system is running stably after repeated tests and practical operations for a long time. © 2013 AIP Publishing LLC. [http://dx.doi.org/10.1063/1.4833396] I. INTRODUCTION
The progressively increasing number of multi-layer film required in ultra large scale integration (ULSI) technology and the decreasing feature size of the circuit components has tremendously increased the need for global surface planarization.1, 2 As one of the most widely used planarization techniques, chemical mechanical planarization (CMP) is presently the only technique that can offer excellent both local and global planarity on the wafer surface.3 As the name implies, the CMP process combines mechanical and chemical removal mechanisms in a synergistic effect,2 whereby a chemical reaction increases the mechanical removal rate of a material and bring about quicker planarization of the thin film. A typical CMP process includes a plate to which a pad is attached and a polishing head which holds the wafer and presses it against the pad. The head, in general, rotates about its axis at the same time the polishing plate rotates in the same direction, while the chemical slurry is continuously supplied to the pad surface through the slurry feeder.1, 4, 5 A schematic diagram of the CMP process is shown in Fig. 1. Since the electrical performance of the interconnection is determined by the resistive and capacitive elements that contribute to the gate delay and hence affect the transistor performance,6 in the past few years, copper has been widely used as interconnect metallization, replacing aluminum in integrated circuit interconnections.7 This switch was emerged and stimulated due to copper advantage characteristics, such as low resistivity and high immunity to electro-migration.8 The practical copper CMP process usually a) Author to whom correspondence should be addressed. Electronic mail:
[email protected] 0034-6748/2013/84(12)/125101/10/$30.00
involves two steps.9 The first step accomplishes the removal of copper overburden and stops on TaN/Ta liner. The second step involves the removal of TaN/Ta liner, and then about 300400 Å of the dielectric material as well. To determine the precise time at which the target material has been removed or the requisite layer thickness remains, each wafer’s polish profile must be monitored by the control system. So far, the in situ measurement technique for real-time monitoring of copper film thickness and detecting polishing termination is useful to solve underpolish and overpolish problems, which is significant for copper CMP process. After the copper CMP process (and post-CMP cleaning), the thickness of remaining copper film needs to be known, mainly for experiments, which provides the basis for subsequent optimization of CMP process parameters. Therefore, the off-line measurement technique is also necessary. CMP tool employs state-of-the-art mechanical control system and software to achieve exceptional performance and productivity.9 Developments in measurement and control system have been critical in establishing CMP as a viable manufacturing process. Based on the requirements mentioned above, a control subsystem for measuring copper film thickness on wafer surface is successfully realized in our CMP control system, integrating measurement technology into the CMP tool which is developed independently, and providing a reliable technical guarantee that the system runs efficiently. II. MEASUREMENT TECHNOLOGY A. Measurement method
The CMP system is designed for copper CMP process, which is used for removing the excess copper on wafer
84, 125101-1
© 2013 AIP Publishing LLC
This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitationnew.aip.org/termsconditions. Downloaded to IP: 65.39.42.160 On: Mon, 08 Dec 2014 22:05:47
125101-2
Li et al.
Rev. Sci. Instrum. 84, 125101 (2013)
FIG. 1. Schematic diagram of the CMP process.
surface. There are two modules, in situ and off-line, for measurement subsystem. Generally, it is difficult to accomplish in situ measurement, because the wafer is pressed face down against a rotating polishing pad during CMP, while a chemically and physically (abrasive) active slurry planarizes the wafer.2 So the condition of in situ measurement should be satisfied first when the method to measure the thickness of copper film is considered to choose. In recent years, a variety of in situ detection methods that make use of various principles, such as optical reflectance or frictional force,10 have been put forward. Generally, the spectra of reflected light from different layers on a pattern wafer demonstrates a clear transition from copper to the underlying dielectric layer. And the friction force between a pad and a wafer is varied when the wafer surface material changes. But they only aim at end-point detection, not real-time measuring. Recently, the eddy current method11–17 for detecting copper film thickness has been used in industrial CMP. And research on the eddy current method in CMP process has been appeared in industry quarters.18 As it is known, under the condition that a metal film, represented by a copper film, is exposed to a directional and intense magnetic field, a local eddy current is generated in the surface of copper film.19 Using this method, the eddy current generated in the copper film changes monotonically in proportion to film thickness.19 In addition, this method is a non-contact technique, avoiding to scratch the wafer surface. Meanwhile, it can achieve in situ measurement of the copper film thickness on wafer surface in real time. So it can be applied to CMP in situ measurement technology. What is more, it is completely suitable for off-line measurement. Therefore, the eddy current method is used in this study.
B. Measurement module
The measurement device contains an eddy-current probe, a signal generator, a front end circuit, a signal collection with a control processor, and a communication cable. The signal generator provides a basic operating signal for the sensor. The probe has a drive coil that induces a current in the copper film on the wafer surface, and the device first converts the change of collected signals into the output voltage.18 In a certain range, there exists a linear relationship between the output voltage and the thickness of copper film. So it can be used to measure the thickness of copper film on the wafer surface. When the frequency of the operating signal is fixed, the smaller the thickness of copper film is, the smaller
FIG. 2. Demonstration of probe motion trail during CMP process.
the output voltage amplitude is. Based on the calibration curve which has been set by the operator, the processor of the measurement device can give an absolute copper film thickness value. The resolution of the measurement module is about 15 nm, the repeatability error is less than 1.6%, and the sensitivity is about 2 mv/nm.
1. In situ module
The probe is embedded in the plate. The copper film thickness is detected by the probe through the pad. A slip ring is used to connect the probe and the signal processing circuit. The processor keeps communication with the control system through a coaxial cable. After a delay for polish process stabilization, the measurement device outputs data in every circle successively to the control system by DP bus. During the CMP process, the wafer rotates along with the polishing head and swing back and forth on the polishing pad radially. When the probe moves below the wafer, the measurement process is triggered. To minimize the measurement error and ensure that measurement data are valid, two Hall switches are used to indicate the accurate measurement range. When the probe enters and leaves the measurement range (shown in Fig. 2), two switching signals are given, respectively. The arc between a and b is the valid measurement path.
2. Off-line module
The probe is installed in a mechanical arm above the rotary table, on which the wafer is placed. The measurement principle is the same as the in situ module. According to the measurement mode set by the operator, the movement mechanism takes corresponding action, assisting the probe in accomplishing measurement of all points on wafer surface. After the measurement process, the processor calculates the thickness values. And then the calculation results are all sent to the control system by DP bus at once.
This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitationnew.aip.org/termsconditions. Downloaded to IP: 65.39.42.160 On: Mon, 08 Dec 2014 22:05:47
125101-3
Li et al.
Rev. Sci. Instrum. 84, 125101 (2013)
FIG. 3. Hardware structure of CMP control system.
Finally, it is demonstrated that the two modules can be integrated into a CMP system and run well based on the robust performance under actual conditions.
III. CONTROL SYSTEM A. In situ module
1. Architecture
At present, programmable logic controller (PLC) has been widely used in the industrial control area. The PLC is generally used as the lower computer to complete data collection in the control system,20 and control the output and input directly. The hardware of the CMP control system is composed of PLCs and industrial personal computer (IPC). A schematic diagram is shown in Fig. 3. Here, PLCs form the lower control system, and IPC forms the upper control system, which is responsible for monitoring every unit of CMP system including in situ module and off-line module through each PLC, assuring that the whole system runs stably, and managing local data read from the lower control system. The physical connection between IPC and PLCs is achieved by the industrial Ethernet. The OPC technology is used to transmit data between the upper and the lower.
As shown in Fig. 3, each in situ module which is integrated with the corresponding CMP unit is controlled by a single PLC. However, the off-line module is controlled by its own PLC4 according to the practical need. Using this structure, more units can be integrated as needed, and each unit has its own controller without interference on each other. The lower control system sets up a special storage area (DB), which is responsible for temporarily saving data from in situ module, including the uncalibrated values, the calibration table, and the thickness values. During the CMP process, the measurement module sends each thickness value to the PLC periodically, and the upper control system accesses data in PLC DB by OPC technology. Before the next period, the CMP control system analyzes the data for endpoint feature. If the pre-setting point of the copper film thickness is detected, the current polish step can be stopped and the process is ended by the control system subsequently. The operator can read the thickness values on the graphical user interface (GUI) of the upper control system in real time, knowing how copper film thickness changes. After the CMP process, if an operator wants to browse all the process data, the data can be displayed in a specified format on the interface. All the data, including the uncalibrated values, the thickness values, and the calibration table, can be stored in a local folder. The operator can set the file name and the save path, and does not need to care about the file format.
2. Process flow
The diagram of in situ module process flow is shown in Fig. 4. There are calibration mode and measurement mode in the control system. Before the measurement process, the calibration process is carried out under the actual polishing
FIG. 4. Diagram of process flow (calibration and measurement).
This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitationnew.aip.org/termsconditions. Downloaded to IP: 65.39.42.160 On: Mon, 08 Dec 2014 22:05:47
125101-4
Li et al.
condition, and the operator creates a complete calibration table according to the correlation between the uncalibrated values and the known thickness values, and then downloads the table to the measurement module. To eliminate the influence of different CMP recipes on measurement results, different calibration tables are stored as different groups. Before the calibration table is downloaded, a calibration group is chosen. In the measurement process, the operator first selects a calibration table and then sets the CMP end-point value. During the CMP process, the film thickness is calculated by the system using the selected calibration table, and the result is output periodically. As eddy current method can measure the different thickness of copper film, the operator can change the end-point set value as needed. 3. Software
As an important part of the upper control system, the software is running on the IPC and mainly contains GUI and communication module. Based on the requirements from the operator, the basic functions of the software include the establishment of the communication, sending the control commands, setting the parameters, and reading the data, as well as processing the data, etc. Here is a simple example in Fig. 5, where there is just one module, and simplifies the user interface properly to show the main functions. If the expansion is needed, the interface can manage more in situ modules in different CMP units by inserting new pages to tabwidget. a. GUI. Qt is a leading cross-platform application and UI development framework, using standard C++ and widely used for developing software applications with a GUI.21 Its powerful full-framework capabilities allow for the creation of highly performing native applications, and it provides the mechanism of Signal/Slot, which is the foundation of the Qt programming, and makes the signal transmission between different components safer and easier. With Qt, the developer can reuse code efficiently to target multiple platforms with one code base and create applications rapidly for one platform and easily build and run to deploy on another platform. The GUI of the software is the only way to interact with the control system for the operator. It helps the operator to send the control commands and parameters to the lower con-
Rev. Sci. Instrum. 84, 125101 (2013)
trol system promptly, and at the same time to get feedback about the in situ module, including measurement data, from the lower control system to the operator, and then to finish the subsequent processing. In the process of the software development, C++ is chosen as the programming language, in combination with the Qt framework to create a cross-platform GUI. The development is done in Qt OpenSource 4.6.0. On Windows 7 the program is compiled using the Microsoft Visual Studio compiler, as provided with Microsoft C++ 2008. The main functions of GUI are as below, and the implementations are introduced. (1) Command: Depending on the operated object, the commands mainly include: (1) access OPC serve in the sync or async way; (2) process the data in temporary array which is responsible for temporarily storage data received from PLC DB; (3) manage the local data file, especially use QFile and QTextStream to fulfill reading or writing a local file. To realize the different commands, different command buttons (QPushButton widgets) are first generated in the GUI, such as Open, Save, Download, and Thickness shown in Fig. 5. By means of Signal/Slot, the clicked signal of a button can be connected to a corresponding slot. In the slot, we code the corresponding function. (2) Input: To allow the operator to input information, we generate the spin box widget used for choosing a calibration group or a measurement group, and the one-line text editor used for setting the end-point value. Every time the value (or text) changes, the widget emits the valueChanged signal or textChanged signal. Then we write the value to specified variable in lower control system by OPC data access in the slot. To reduce the data error, a regular expression to determine whether the input is acceptable or not is defined. (3) Display: Many labels are used to display all kinds of information. On the GUI, the labels are divided into two categories, static labels and dynamic labels. The static labels are used for explaining other widgets, such as a button and a spin box. The dynamic labels are used for showing the system time, the running condition of the module, and the communication status. Since the dynamic labels have to be refreshed in real time, a timer is needed. The QTimer class provides repetitive and single-shot timers and a high-level programming interface for timer. So a QTimer is created to connect its timeout signal to the appropriate slot, and to call the start() (a member function). From then on it gives the timeout signal at constant intervals. In the slot, the contents of the dynamic labels are refreshed. When a variable in the lower control system changes, it can be shown on the corresponding label immediately and the operator can read it at once. An example for a 500-ms timer is shown below. The timerupdate slot is called every 500 ms to refresh the content of all dynamic labels QTimer ∗ timer = new QTimer; connect(timer, SIGNAL(timeout()),this, SLOT(timerupdate())); timer → start(500).
FIG. 5. GUI of in situ module.
(4) Table: The table is used for displaying the uncalibrated values and the calibration table. On the GUI, the table is constructed with the required numbers of rows and
This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitationnew.aip.org/termsconditions. Downloaded to IP: 65.39.42.160 On: Mon, 08 Dec 2014 22:05:47
125101-5
Li et al.
columns. When an operator inputs data in the cell of table, the table automatically creates an item variable which is used to store the data. When the operator wants to display the data, the data can be inserted into the table easily. b. Communication module. The OPC technology, which is widely used in the process industry and has become an industrial standard for accessing data,22, 23 is used to realize the data communication in the control system. It enables the connectivity and interoperability of plant floor information,24 and defines a set of standard COM objects, methods and properties that specifically address requirements for real-time factory automation, and process control applications. In OPC’s client/server model, the server application acquires, contains, and provides data to the client application. The OPC server provides standard interfaces, allowing clients to access to the server25 and exchange data in a common format. Because the OPC standards are based upon the computer industry standards, the technical reliability is assured. In the control system, the OPC server is installed in the IPC, and C++ is used to develop OPC client by the custom interface. In the control system, we organize all the variables in one group, although each unit has its own controller with a different address. When the operator logs in the GUI of the upper control system, the client of communication module connects to the server automatically. When the operator logs off the GUI, the client disconnects to the server. The first standard of OPC specifications is OPC Data Access (DA), used to move real-time data from the PLCs, DCSs, and other control devices to HMIs and other display clients. In the OPC DA specification, there are three types of reading request to the server: sync read (polled), async read, and exception (subscription), and two types of writing requests to the server: sync write and async write. The sync and async work with specific subsets of items from the group which are provided by the caller. The subscription sends back any item in the group that changes.26 In the communication module, the client uses the sync way to write the important variables, such as end-point value, in order to ensure that each parameter is set correctly. The async way is chosen to write other variables, so the consumption of CPU and network resources can be reduced. Compared to the sync and async, the subscription can effectively reduce the number of access to server, and handle large amounts of data, avoiding the network congestion, so the subscription is chosen for the read operation. B. Off-line module
The off-line module is used for the multi-points measurement after the CMP process, mainly consists of the following parts (shown in Fig. 6): eddy-current measurement device, rotary table, mechanical arm, wafer bracket, and related drivers. The rotary table and the mechanical arm make up the motion subsystem. According to the measurement mode set by the operator, the movement subsystem assists probe to complete the measurement of each point. The control system structure of the off-line module is shown in Fig. 3. The lower control system also needs to set up a special storage area (DB), which is responsible for temporarily saving data from the off-line module, including the
Rev. Sci. Instrum. 84, 125101 (2013)
FIG. 6. Schematic diagram of off-line module.
uncalibrated values, the calibration table, and the thickness values. Since the off-line module is more complicated, especially it has its own motion subsystem, a single PLC is assigned to it. The lower control system defines many variables for control commands and process parameters. Before each measurement, the upper control system sets all these variables. As for other sides, both of the in situ module and the off-line module have a lot of functions in common, this section mainly describes its technical characteristics in detail.
1. Process flow
There are also calibration mode and measurement mode in the off-line module control system, but the algorithm is more complicated. a. Calibration. Generally, the rotary table cannot keep horizontal absolutely, which causes the change of lift-off height between the wafer and the probe at different points and can affect the accuracy of measurement. Therefore, the method of multi-points calibration is adopted. The multipoints calibration is to fit the calibration curve, respectively, at each point, on the basis of the point number and distribution. When a wafer is being measured, the calculation of thickness at each point is done by its own calibration curve. During the calibration process, we first get an uncalibrated matrix, here name it X ⎛
x1,1 x2,1 .. .
⎜ ⎜ ⎜ X=⎜ ⎜ ⎝ xm−1,1 xm,1
x1,2 x2,2 .. .
··· ··· .. .
x1,n−1 x2,n−1 .. .
xm−1,2 xm,2
··· ···
xm−1,n−1 xm,n−1
x1,n x2,n .. .
⎞
⎟ ⎟ ⎟ ⎟ . (1) ⎟ xm−1,n ⎠ xm,n
The xr,c is the output value, when the probe is at point c of the standard sample r (0 ≤ r ≤ m, 1 ≤ c ≤ n). Each output value is the difference between the values at the measurement point and the initial position, eliminating the zero drift of the sensor. According to the thickness values of m standard samples (each sample has n points), we get a thickness matrix,
This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitationnew.aip.org/termsconditions. Downloaded to IP: 65.39.42.160 On: Mon, 08 Dec 2014 22:05:47
125101-6
Li et al.
Rev. Sci. Instrum. 84, 125101 (2013) TABLE I. 8-series points-distribution.
here name it Y ⎛
y1,1
⎜ ⎜ y2,1 ⎜ . Y=⎜ ⎜ .. ⎜ ⎝ ym−1,1 ym,1
y1,2
···
y1,n−1
y2,2 .. .
··· .. . ··· ···
y2,n−1 .. .
ym−1,2 ym,2
ym−1,n−1 ym,n−1
⎞
n
Num
m1
m2
m3
m4
m5
m6
m7
m8
m9
m10
⎟ ⎟ ⎟ ⎟ . (2) ⎟ ⎟ ym−1,n ⎠ ym,n
6 7 8 9 10
121 169 225 289 361
1 1 1 1 1
8 8 8 8 8
16 16 16 16 16
24 24 24 24 24
32 32 32 32 32
40 40 40 40 40
48 48 48 48
56 56 56
64 64
72
y1,n y2,n .. .
The yr,c is the thickness value at point c of the standard sample r (0 ≤ r ≤ m, 1 ≤ c ≤ n). Based on the correlation between matrix X and Y, we carry out linear fitting piecewise between the output values in each column of matrix X and the corresponding thickness values in each column of matrix Y, and obtain the calibration curve at each point. By this method, the calibration curves at total n points are obtained. Finally, stored all calibration curves. b. Measurement. When the copper film thickness of a prepared wafer is measured, the control system calculates each thickness value at each point by its own calibration curve automatically. Before each calibration or measurement, the rotary table should rotate back to the initial position, avoiding the change of position of each calibration or measurement point. In the calibration or measurement process, the film thickness value at one point is the average of all points in the local segment.
2. Points-distribution
There are two measurement modes, the XY model and the global mode, designed for the off-line module. In the XY model, all the points on two vertical diameters are measured. In the global model, the thickness values of all points distributed evenly on the whole surface are obtained. Considering the technique demands, 8-series points-distribution in the global model is used (Fig. 7). According to the total number of measurement points and the radial distance d from the outermost circle to the edge of the wafer (the measurement points on the outermost circle should be effective), all coordinates of measurement points on the wafer can be completely obtained. a. Number of circles and points-distribution on each circle. Here, n is the number of circles. num, which is the total number of measurement points, can be calculated by . According to the 8-series the formula, num =1+8 × n(n−1) 2 points-distribution, we can look up the table to get the number of measurement points on each circle, as shown in Table I.
FIG. 7. Points-distributions of different number: (a) 121 points; (b) 169 points; (c) 225 points; (d) 289 points; (e) 361 points.
This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitationnew.aip.org/termsconditions. Downloaded to IP: 65.39.42.160 On: Mon, 08 Dec 2014 22:05:47
125101-7
Li et al.
Rev. Sci. Instrum. 84, 125101 (2013)
In Table I, mi is the number of measurement points on circle i, and mi = 8 × (i − 1), (i = 2, 3, . . . , n). Whatever the number of total measurement points is, the center of wafer is always the first circle, and there is only one point there. The points on each circle are evenly distributed, and the central . angle between the two adjacent points is 360 mi b. Radial distance between two adjacent concentric circles. To ensure the principle of uniform distribution, the radial dis. R is tance r between two adjacent concentric circles is R−d n−1 the radius of wafer. 3. Software
Based on the requirements from operator, the basic functions of software also include the establishment of the communication, sending the control instructions, setting the parameters, and reading the data as well as processing the data, etc. The access to data in lower control system is the same as the in situ module. In the software development, Qt is still chosen to design GUI and the OPC technology is also used for the communication module. Since the drawing is a new and advanced characteristic in the control system software, which is necessary and useful to the off-line module, the implementation of the drawing is introduced below. a. Development environment. Matlab is a numerical language software based on the matrix programming unit with the function of efficient computation and potential function of graph processing.27 It is widely used in the engineering field, mainly for signal and image processing.28 Although Qt in Visual Studio platform has flexibility and high efficiency of C++ language, it is not good at high-performance drawing, cannot meet the needs of the drawing function development in the control system software. Therefore, Matlab is used here. b. Calling Matlab. There is no doubt that the combination between Qt and Matlab can achieve complementary advantages of each other in the practical application, so that we are able to find an optimal balance between the code execution speed and the programming efficiency, and reduce the development difficulty. Generally, there are some ways to achieve cooperation with Matlab:28 to apply Matlab’s engine, to call Matlab function by means of Matlab C/C++ function library, and to apply Matlab compiler. The Matlab engine provides a way of accessing and calling Matlab for an external language. With Matlab engine, the Matlab library can be used as a powerful programmable function library by the external language. This method can use almost all Matlab functions, but cannot be separated from Matlab environment. It means the user’s computer needs the Matlab software, and implementation efficiency is low. In the control system, the upper communicates with the lower in real time, and there is a large amount of data interaction. This requires the system has a good performance of real time. If we call Matlab function by means of Matlab C/C++ function library in the main program, namely, the drawing function is directly added into the main program, the real time of system can be affected undoubtedly. The Matlab compiler is a software tool for the improvement of Matlab development envi-
FIG. 8. Algorithm flow chart of drawing subprogram.
ronment by MathWorks. It can transform M language documents into different types of source code to generate the required MEX file (.dll), standalone application file (.exe), and COM component, and to enhance the implementation efficiency of the code greatly. So, considering the reliability and flexibility of the control system, we generate a standalone application file by the Matlab compiler. It is mainly used for drawing and can be called by the main program at any time. c. Implementation. Below is the algorithm of drawing subprogram, when the subprogram is called by main program (Fig. 8): First, reads all measurement data stored in the specified folder. Second, according to the total number of data, determines which type of measurement mode, and then generates the corresponding coordinate distribution. Third, corresponds the data to the coordinate of corresponding point one by one, and completes the drawing. Because the M language document needs to be compiled into a standalone application file, the Matlab compiler needs to be configured. Here, Lcc compiler is chosen. After the configuration, we compile the M language document and generate the standalone application file. The Matlab Component Runtime (MCR) is a set of standard dynamic link library to provide the necessary basic environment for running program. In order to run the program on the IPC, the MCR is installed.28 After each measurement process, the upper control system obtains a set of discrete measurement data. The main program first pre-processes all data, such as sorting, and then saves them in a specified file with specified format and file name, namely, the main program does all the preparations, as a data preprocessor. The output of image and complex numerical calculation of data are done by the drawing subprogram (Fig. 9). When the drawing subprogram receives the calling
FIG. 9. Diagram of relationship between main program and subprogram.
This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitationnew.aip.org/termsconditions. Downloaded to IP: 65.39.42.160 On: Mon, 08 Dec 2014 22:05:47
125101-8
Li et al.
Rev. Sci. Instrum. 84, 125101 (2013)
FIG. 10. Demonstrations of different functions: (a) Data Cursor; (b) Box Rotate; (c) Go to X-Y view.
instruction from the main program, the subprogram searches and reads the data automatically, then plots the figure, and then provides an intuitive analysis model for the operator (Fig. 10). Based on the idea above, only a slot for calling the draw subprogram is kept in the main program. When the subprogram starts to run, there is no influence on the running of the main program. IV. RESULTS
Fig. 11 shows the example of the calibration curve of in situ module, and Fig. 12 shows the experiment result from an actual CMP process. There are more experiment results in our previous paper.18 Before each test, the module is calibrated (for repeatedly continuous process, only once calibration is enough). If the sensor is changed, a new calibra-
FIG. 11. Example of one calibration curve of in situ module.
tion table also needs to be created. Here, the uncalibrated value, which is unitless, is attained from the output voltage of measurement device through high-speed A/D conversion. In Fig. 12, we can see that the thickness of copper film on wafer surface decreases as CMP process goes on. Figure 13 shows the experiment result of a multi-points measurement (off-line module), including the thickness values attained from several measurements, and Fig. 14 show the standard deviations on each measurement point. In the experiment, 5 measurements on the same wafer for 361points are taken. The maximum standard deviation is less than 4 nm. Based on a large number of experiment results, the repeatability of the measurement system are confirmed. At
FIG. 12. Measured data from an actual CMP process in the time period between 0 and 60 s.
This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitationnew.aip.org/termsconditions. Downloaded to IP: 65.39.42.160 On: Mon, 08 Dec 2014 22:05:47
125101-9
Li et al.
Rev. Sci. Instrum. 84, 125101 (2013) 1 Y.
FIG. 13. Example of one calibration curve of in situ module.
FIG. 14. Example of one calibration curve of in situ module.
the same time, these examples demonstrate that our control system performs well and stably, satisfying all needs of measurement. V. CONCLUSIONS
In this paper, we present a reliable control system for measuring copper film thickness on wafer surface, including the in situ module and the off-line module. A control system software of measurement modules is successfully realized and integrated into the whole CMP system. The system reaches to nano-scale measurement and has highly practical value in semiconductor manufacturing. Under the actual process conditions, the control system performances well and can implement all control functions effectively. ACKNOWLEDGMENTS
The work is supported by the Important National Science and Technology Major Projects of China (2008ZX02104001), and National Basic Research Program of China (2009CB724207).
Moon, Mechanical Aspects of the Material Removal Mechanism in Chemical Mechanical Polishing (CMP) (Department of Mechanical Engineering, University of California, Berkeley, 1999). 2 P. B. Zantye, A. Kumar, and A. K. Sikder, “Chemical mechanical planarization for microelectronics applications,” Mater. Sci. Eng. R 45, 89– 220 (2004). 3 P. B. Zantye, Process, Reliability and Integration Issues in Chemical Mechanical Planarization (Department of Mechanical Engineering, University of South Florida, 2005). 4 T. Wang and X. Lu, “Numerical and experimental investigation on multizone chemical mechanical planarization,” Microelectron. Eng. 88, 3327– 3332 (2011). 5 M. R. Oliver, Chemical Mechanical Planarization of Semiconductor Materials (Springer, 2004). 6 D. C. Edelstein, G. A. Sai-Halasz, and Y.-J. Mii, IBM J. Res. Dev. 39, 383 (1995). 7 S. P. Murarka, I. V. Verner, and R. J. Gutman, Copper-Fundamental Mechanisms for Microelectronic Applications (John Wiley and Sons, Inc., N.Y., 2000), p. 337. 8 Y. Ein-Eli and D. Starosvetsky, “Review on copper chemical-mechanical polishing (CMP) and post-CMP cleaning in ultra large system integrated (ULSI): An electrochemical perspective,” Electrochim. Acta 52, 1825– 1838 (2007). 9 M. Krishnan and J. W. Nalaskowski, “Chemical mechanical planarization: Slurry chemistry, materials, and mechanisms,” Chem. Rev. 110, 178–204 (2010). 10 V. R. Kakireddy, S. Mudhivarthi, and A. Kumar, J. Vac. Sci. Technol. B 26, 141 (2008). 11 R. Allen, C. Chen, T. Trikas, K. Lehman, R. Shinagawa, V. Bhaskaran, B. Stephenson, and D. Watts, “In-situ CMP copper endpoint control system,” in Proceedings of the IEEE International Symposium on Semiconductor Manufacturing, 2001 (IEEE, New York, 2001), pp. 391–394. 12 L. Li, S. G. Barbee, A. Halperin, and T. F. Heinz, U.S. patent US6072313 (6 June 2000). 13 K. R. Lehman, S. M. Lee III, W. H. Johnson, and J. Fielden, U.S. patent US6433541B1 (13 August 2002). 14 O. Nagano, Y. Yamazaki, M. Miyoshi, H. Kaneko, and T. Matsuda, U.S. patent US6563308B2 (13 May 2003). 15 W. L. Yin and A. J. Peyton, “Thickness measurement of metallic plates with an electromagnetic sensor using phase signature analysis,” IEEE Trans. Instrum. Meas. 57(8), 1803–1807 (2008). 16 W. L. Yin and A. J. Peyton, “Thickness measurement of non-magnetic plates using multi-frequency eddy current sensors,” NDT & E Int. 40(1), 43–48 (2007). 17 W. L. Yin, A. J. Peyton, and S. J. Dickinson, “Simultaneous measurement of distance and thickness of a thin metal plate with an electromagnetic sensor using a simplified model,” IEEE Trans. Instrum. Meas. 53, 1335–1338 (2004). 18 Z. Qu, Q. Zhao, Y. Meng, T. Wang, D. Zhao, Y. Men, and X. Lu, “In-situ measurement of Cu film thickness during the CMP process by using eddy current method alone,” Microelectron. Eng. 108, 66–70 (2013). 19 T. Fujita, K. Kitade, and T. Yokoyama, “Development of original end point detection system utilizing eddy current variation due to skin effect in chemical mechanical polishing,” Jpn. J. Appl. Phys. 50, 05EC09 (2011). 20 Y. Yang, G. Ren, and X. Yin, “Measurement and control system for traction motor based on OPC technology,” Adv. Mater. Res. 542–543, 208–211 (2012). 21 J. Blanchette and M. Summerfield, C++ GUI Programming with Qt4 [M] (Prentice Hall PTR, 2006). 22 J. Liu, K. W. Lim, W. K. Ho, K. C. Tan, A. Tay, and R. Srinivasan, “Using the OPC standard for real-time process monitoring and control,” IEEE Software 22, 54–59 (2005). 23 OPC Task Force, OLE for Process Control-OPC Overview, v. 1.0 (OPC Foundation, 1998), see www.opcfoundation.org/DownloadFile.aspx/ General/OPC%20Overview%201.00.pdf?RI=1. 24 D. W. Holley, “Understanding and using OPC for maintenance and reliability applications,” IEE Comput. Control Eng. 15(1), 28–31 (2004). 25 Y. Shimanuki, “OLE for process control (OPC) for new industrial automation systems,” in Proceedings of the IEEE International Conference on Systems, Man, and Cybernetics, 1999 (IEEE Press, 1999), Vol. 6, pp. 1048– 1050. 26 A. Chisholm, “A technical overview of the OPC data access interfaces,” in Proceedings of the ISA EXPO, 1998.
This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitationnew.aip.org/termsconditions. Downloaded to IP: 65.39.42.160 On: Mon, 08 Dec 2014 22:05:47
125101-10
Li et al.
27 F. W. Hong, “Hybrid programming between Matlab and VC++ in the inte-
grated environment,” in Proceedings of the 3rd IEEE Conference on Communication Software and Networks (ICCSN), 2011 (IEEE, Piscataway, NJ, 2011), pp. 377–378.
Rev. Sci. Instrum. 84, 125101 (2013) 28 L.
Hong, “The application guide of mixed programming between Matlab and other programming languages,” in Proceedings of the 2nd International Conference on Computer and Automation Engineering (ICCAE), 2010 (IEEE, Piscataway, NJ, 2010), pp. 185–189.
This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitationnew.aip.org/termsconditions. Downloaded to IP: 65.39.42.160 On: Mon, 08 Dec 2014 22:05:47
This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitationnew.aip.org/termsconditions. Downloaded to IP: 65.39.42.160 On: Mon, 08 Dec 2014 22:05:47
REVIEW OF SCIENTIFIC INSTRUMENTS 84, 125101 (2013)
A reliable control system for measurement on film thickness in copper chemical mechanical planarization system Hongkai Li, Zilian Qu, Qian Zhao, Fangxin Tian, Dewen Zhao, Yonggang Meng, and Xinchun Lua) State Key Laboratory of Tribology, Tsinghua University, Beijing 100084, China
(Received 3 June 2013; accepted 11 November 2013; published online 2 December 2013) In recent years, a variety of film thickness measurement techniques for copper chemical mechanical planarization (CMP) are subsequently proposed. In this paper, the eddy-current technique is used. In the control system of the CMP tool developed in the State Key Laboratory of Tribology, there are in situ module and off-line module for measurement subsystem. The in situ module can get the thickness of copper film on wafer surface in real time, and accurately judge when the CMP process should stop. This is called end-point detection. The off-line module is used for multi-points measurement after CMP process, in order to know the thickness of remained copper film. The whole control system is structured with two levels, and the physical connection between the upper and the lower is achieved by the industrial Ethernet. The process flow includes calibration and measurement, and there are different algorithms for two modules. In the process of software development, C++ is chosen as the programming language, in combination with Qt OpenSource to design two modules’ GUI and OPC technology to implement the communication between the two levels. In addition, the drawing function is developed relying on Matlab, enriching the software functions of the off-line module. The result shows that the control system is running stably after repeated tests and practical operations for a long time. © 2013 AIP Publishing LLC. [http://dx.doi.org/10.1063/1.4833396] I. INTRODUCTION
The progressively increasing number of multi-layer film required in ultra large scale integration (ULSI) technology and the decreasing feature size of the circuit components has tremendously increased the need for global surface planarization.1, 2 As one of the most widely used planarization techniques, chemical mechanical planarization (CMP) is presently the only technique that can offer excellent both local and global planarity on the wafer surface.3 As the name implies, the CMP process combines mechanical and chemical removal mechanisms in a synergistic effect,2 whereby a chemical reaction increases the mechanical removal rate of a material and bring about quicker planarization of the thin film. A typical CMP process includes a plate to which a pad is attached and a polishing head which holds the wafer and presses it against the pad. The head, in general, rotates about its axis at the same time the polishing plate rotates in the same direction, while the chemical slurry is continuously supplied to the pad surface through the slurry feeder.1, 4, 5 A schematic diagram of the CMP process is shown in Fig. 1. Since the electrical performance of the interconnection is determined by the resistive and capacitive elements that contribute to the gate delay and hence affect the transistor performance,6 in the past few years, copper has been widely used as interconnect metallization, replacing aluminum in integrated circuit interconnections.7 This switch was emerged and stimulated due to copper advantage characteristics, such as low resistivity and high immunity to electro-migration.8 The practical copper CMP process usually a) Author to whom correspondence should be addressed. Electronic mail:
[email protected] 0034-6748/2013/84(12)/125101/10/$30.00
involves two steps.9 The first step accomplishes the removal of copper overburden and stops on TaN/Ta liner. The second step involves the removal of TaN/Ta liner, and then about 300400 Å of the dielectric material as well. To determine the precise time at which the target material has been removed or the requisite layer thickness remains, each wafer’s polish profile must be monitored by the control system. So far, the in situ measurement technique for real-time monitoring of copper film thickness and detecting polishing termination is useful to solve underpolish and overpolish problems, which is significant for copper CMP process. After the copper CMP process (and post-CMP cleaning), the thickness of remaining copper film needs to be known, mainly for experiments, which provides the basis for subsequent optimization of CMP process parameters. Therefore, the off-line measurement technique is also necessary. CMP tool employs state-of-the-art mechanical control system and software to achieve exceptional performance and productivity.9 Developments in measurement and control system have been critical in establishing CMP as a viable manufacturing process. Based on the requirements mentioned above, a control subsystem for measuring copper film thickness on wafer surface is successfully realized in our CMP control system, integrating measurement technology into the CMP tool which is developed independently, and providing a reliable technical guarantee that the system runs efficiently. II. MEASUREMENT TECHNOLOGY A. Measurement method
The CMP system is designed for copper CMP process, which is used for removing the excess copper on wafer
84, 125101-1
© 2013 AIP Publishing LLC
This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitationnew.aip.org/termsconditions. Downloaded to IP: 65.39.42.160 On: Mon, 08 Dec 2014 22:05:47
125101-2
Li et al.
Rev. Sci. Instrum. 84, 125101 (2013)
FIG. 1. Schematic diagram of the CMP process.
surface. There are two modules, in situ and off-line, for measurement subsystem. Generally, it is difficult to accomplish in situ measurement, because the wafer is pressed face down against a rotating polishing pad during CMP, while a chemically and physically (abrasive) active slurry planarizes the wafer.2 So the condition of in situ measurement should be satisfied first when the method to measure the thickness of copper film is considered to choose. In recent years, a variety of in situ detection methods that make use of various principles, such as optical reflectance or frictional force,10 have been put forward. Generally, the spectra of reflected light from different layers on a pattern wafer demonstrates a clear transition from copper to the underlying dielectric layer. And the friction force between a pad and a wafer is varied when the wafer surface material changes. But they only aim at end-point detection, not real-time measuring. Recently, the eddy current method11–17 for detecting copper film thickness has been used in industrial CMP. And research on the eddy current method in CMP process has been appeared in industry quarters.18 As it is known, under the condition that a metal film, represented by a copper film, is exposed to a directional and intense magnetic field, a local eddy current is generated in the surface of copper film.19 Using this method, the eddy current generated in the copper film changes monotonically in proportion to film thickness.19 In addition, this method is a non-contact technique, avoiding to scratch the wafer surface. Meanwhile, it can achieve in situ measurement of the copper film thickness on wafer surface in real time. So it can be applied to CMP in situ measurement technology. What is more, it is completely suitable for off-line measurement. Therefore, the eddy current method is used in this study.
B. Measurement module
The measurement device contains an eddy-current probe, a signal generator, a front end circuit, a signal collection with a control processor, and a communication cable. The signal generator provides a basic operating signal for the sensor. The probe has a drive coil that induces a current in the copper film on the wafer surface, and the device first converts the change of collected signals into the output voltage.18 In a certain range, there exists a linear relationship between the output voltage and the thickness of copper film. So it can be used to measure the thickness of copper film on the wafer surface. When the frequency of the operating signal is fixed, the smaller the thickness of copper film is, the smaller
FIG. 2. Demonstration of probe motion trail during CMP process.
the output voltage amplitude is. Based on the calibration curve which has been set by the operator, the processor of the measurement device can give an absolute copper film thickness value. The resolution of the measurement module is about 15 nm, the repeatability error is less than 1.6%, and the sensitivity is about 2 mv/nm.
1. In situ module
The probe is embedded in the plate. The copper film thickness is detected by the probe through the pad. A slip ring is used to connect the probe and the signal processing circuit. The processor keeps communication with the control system through a coaxial cable. After a delay for polish process stabilization, the measurement device outputs data in every circle successively to the control system by DP bus. During the CMP process, the wafer rotates along with the polishing head and swing back and forth on the polishing pad radially. When the probe moves below the wafer, the measurement process is triggered. To minimize the measurement error and ensure that measurement data are valid, two Hall switches are used to indicate the accurate measurement range. When the probe enters and leaves the measurement range (shown in Fig. 2), two switching signals are given, respectively. The arc between a and b is the valid measurement path.
2. Off-line module
The probe is installed in a mechanical arm above the rotary table, on which the wafer is placed. The measurement principle is the same as the in situ module. According to the measurement mode set by the operator, the movement mechanism takes corresponding action, assisting the probe in accomplishing measurement of all points on wafer surface. After the measurement process, the processor calculates the thickness values. And then the calculation results are all sent to the control system by DP bus at once.
This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitationnew.aip.org/termsconditions. Downloaded to IP: 65.39.42.160 On: Mon, 08 Dec 2014 22:05:47
125101-3
Li et al.
Rev. Sci. Instrum. 84, 125101 (2013)
FIG. 3. Hardware structure of CMP control system.
Finally, it is demonstrated that the two modules can be integrated into a CMP system and run well based on the robust performance under actual conditions.
III. CONTROL SYSTEM A. In situ module
1. Architecture
At present, programmable logic controller (PLC) has been widely used in the industrial control area. The PLC is generally used as the lower computer to complete data collection in the control system,20 and control the output and input directly. The hardware of the CMP control system is composed of PLCs and industrial personal computer (IPC). A schematic diagram is shown in Fig. 3. Here, PLCs form the lower control system, and IPC forms the upper control system, which is responsible for monitoring every unit of CMP system including in situ module and off-line module through each PLC, assuring that the whole system runs stably, and managing local data read from the lower control system. The physical connection between IPC and PLCs is achieved by the industrial Ethernet. The OPC technology is used to transmit data between the upper and the lower.
As shown in Fig. 3, each in situ module which is integrated with the corresponding CMP unit is controlled by a single PLC. However, the off-line module is controlled by its own PLC4 according to the practical need. Using this structure, more units can be integrated as needed, and each unit has its own controller without interference on each other. The lower control system sets up a special storage area (DB), which is responsible for temporarily saving data from in situ module, including the uncalibrated values, the calibration table, and the thickness values. During the CMP process, the measurement module sends each thickness value to the PLC periodically, and the upper control system accesses data in PLC DB by OPC technology. Before the next period, the CMP control system analyzes the data for endpoint feature. If the pre-setting point of the copper film thickness is detected, the current polish step can be stopped and the process is ended by the control system subsequently. The operator can read the thickness values on the graphical user interface (GUI) of the upper control system in real time, knowing how copper film thickness changes. After the CMP process, if an operator wants to browse all the process data, the data can be displayed in a specified format on the interface. All the data, including the uncalibrated values, the thickness values, and the calibration table, can be stored in a local folder. The operator can set the file name and the save path, and does not need to care about the file format.
2. Process flow
The diagram of in situ module process flow is shown in Fig. 4. There are calibration mode and measurement mode in the control system. Before the measurement process, the calibration process is carried out under the actual polishing
FIG. 4. Diagram of process flow (calibration and measurement).
This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitationnew.aip.org/termsconditions. Downloaded to IP: 65.39.42.160 On: Mon, 08 Dec 2014 22:05:47
125101-4
Li et al.
condition, and the operator creates a complete calibration table according to the correlation between the uncalibrated values and the known thickness values, and then downloads the table to the measurement module. To eliminate the influence of different CMP recipes on measurement results, different calibration tables are stored as different groups. Before the calibration table is downloaded, a calibration group is chosen. In the measurement process, the operator first selects a calibration table and then sets the CMP end-point value. During the CMP process, the film thickness is calculated by the system using the selected calibration table, and the result is output periodically. As eddy current method can measure the different thickness of copper film, the operator can change the end-point set value as needed. 3. Software
As an important part of the upper control system, the software is running on the IPC and mainly contains GUI and communication module. Based on the requirements from the operator, the basic functions of the software include the establishment of the communication, sending the control commands, setting the parameters, and reading the data, as well as processing the data, etc. Here is a simple example in Fig. 5, where there is just one module, and simplifies the user interface properly to show the main functions. If the expansion is needed, the interface can manage more in situ modules in different CMP units by inserting new pages to tabwidget. a. GUI. Qt is a leading cross-platform application and UI development framework, using standard C++ and widely used for developing software applications with a GUI.21 Its powerful full-framework capabilities allow for the creation of highly performing native applications, and it provides the mechanism of Signal/Slot, which is the foundation of the Qt programming, and makes the signal transmission between different components safer and easier. With Qt, the developer can reuse code efficiently to target multiple platforms with one code base and create applications rapidly for one platform and easily build and run to deploy on another platform. The GUI of the software is the only way to interact with the control system for the operator. It helps the operator to send the control commands and parameters to the lower con-
Rev. Sci. Instrum. 84, 125101 (2013)
trol system promptly, and at the same time to get feedback about the in situ module, including measurement data, from the lower control system to the operator, and then to finish the subsequent processing. In the process of the software development, C++ is chosen as the programming language, in combination with the Qt framework to create a cross-platform GUI. The development is done in Qt OpenSource 4.6.0. On Windows 7 the program is compiled using the Microsoft Visual Studio compiler, as provided with Microsoft C++ 2008. The main functions of GUI are as below, and the implementations are introduced. (1) Command: Depending on the operated object, the commands mainly include: (1) access OPC serve in the sync or async way; (2) process the data in temporary array which is responsible for temporarily storage data received from PLC DB; (3) manage the local data file, especially use QFile and QTextStream to fulfill reading or writing a local file. To realize the different commands, different command buttons (QPushButton widgets) are first generated in the GUI, such as Open, Save, Download, and Thickness shown in Fig. 5. By means of Signal/Slot, the clicked signal of a button can be connected to a corresponding slot. In the slot, we code the corresponding function. (2) Input: To allow the operator to input information, we generate the spin box widget used for choosing a calibration group or a measurement group, and the one-line text editor used for setting the end-point value. Every time the value (or text) changes, the widget emits the valueChanged signal or textChanged signal. Then we write the value to specified variable in lower control system by OPC data access in the slot. To reduce the data error, a regular expression to determine whether the input is acceptable or not is defined. (3) Display: Many labels are used to display all kinds of information. On the GUI, the labels are divided into two categories, static labels and dynamic labels. The static labels are used for explaining other widgets, such as a button and a spin box. The dynamic labels are used for showing the system time, the running condition of the module, and the communication status. Since the dynamic labels have to be refreshed in real time, a timer is needed. The QTimer class provides repetitive and single-shot timers and a high-level programming interface for timer. So a QTimer is created to connect its timeout signal to the appropriate slot, and to call the start() (a member function). From then on it gives the timeout signal at constant intervals. In the slot, the contents of the dynamic labels are refreshed. When a variable in the lower control system changes, it can be shown on the corresponding label immediately and the operator can read it at once. An example for a 500-ms timer is shown below. The timerupdate slot is called every 500 ms to refresh the content of all dynamic labels QTimer ∗ timer = new QTimer; connect(timer, SIGNAL(timeout()),this, SLOT(timerupdate())); timer → start(500).
FIG. 5. GUI of in situ module.
(4) Table: The table is used for displaying the uncalibrated values and the calibration table. On the GUI, the table is constructed with the required numbers of rows and
This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitationnew.aip.org/termsconditions. Downloaded to IP: 65.39.42.160 On: Mon, 08 Dec 2014 22:05:47
125101-5
Li et al.
columns. When an operator inputs data in the cell of table, the table automatically creates an item variable which is used to store the data. When the operator wants to display the data, the data can be inserted into the table easily. b. Communication module. The OPC technology, which is widely used in the process industry and has become an industrial standard for accessing data,22, 23 is used to realize the data communication in the control system. It enables the connectivity and interoperability of plant floor information,24 and defines a set of standard COM objects, methods and properties that specifically address requirements for real-time factory automation, and process control applications. In OPC’s client/server model, the server application acquires, contains, and provides data to the client application. The OPC server provides standard interfaces, allowing clients to access to the server25 and exchange data in a common format. Because the OPC standards are based upon the computer industry standards, the technical reliability is assured. In the control system, the OPC server is installed in the IPC, and C++ is used to develop OPC client by the custom interface. In the control system, we organize all the variables in one group, although each unit has its own controller with a different address. When the operator logs in the GUI of the upper control system, the client of communication module connects to the server automatically. When the operator logs off the GUI, the client disconnects to the server. The first standard of OPC specifications is OPC Data Access (DA), used to move real-time data from the PLCs, DCSs, and other control devices to HMIs and other display clients. In the OPC DA specification, there are three types of reading request to the server: sync read (polled), async read, and exception (subscription), and two types of writing requests to the server: sync write and async write. The sync and async work with specific subsets of items from the group which are provided by the caller. The subscription sends back any item in the group that changes.26 In the communication module, the client uses the sync way to write the important variables, such as end-point value, in order to ensure that each parameter is set correctly. The async way is chosen to write other variables, so the consumption of CPU and network resources can be reduced. Compared to the sync and async, the subscription can effectively reduce the number of access to server, and handle large amounts of data, avoiding the network congestion, so the subscription is chosen for the read operation. B. Off-line module
The off-line module is used for the multi-points measurement after the CMP process, mainly consists of the following parts (shown in Fig. 6): eddy-current measurement device, rotary table, mechanical arm, wafer bracket, and related drivers. The rotary table and the mechanical arm make up the motion subsystem. According to the measurement mode set by the operator, the movement subsystem assists probe to complete the measurement of each point. The control system structure of the off-line module is shown in Fig. 3. The lower control system also needs to set up a special storage area (DB), which is responsible for temporarily saving data from the off-line module, including the
Rev. Sci. Instrum. 84, 125101 (2013)
FIG. 6. Schematic diagram of off-line module.
uncalibrated values, the calibration table, and the thickness values. Since the off-line module is more complicated, especially it has its own motion subsystem, a single PLC is assigned to it. The lower control system defines many variables for control commands and process parameters. Before each measurement, the upper control system sets all these variables. As for other sides, both of the in situ module and the off-line module have a lot of functions in common, this section mainly describes its technical characteristics in detail.
1. Process flow
There are also calibration mode and measurement mode in the off-line module control system, but the algorithm is more complicated. a. Calibration. Generally, the rotary table cannot keep horizontal absolutely, which causes the change of lift-off height between the wafer and the probe at different points and can affect the accuracy of measurement. Therefore, the method of multi-points calibration is adopted. The multipoints calibration is to fit the calibration curve, respectively, at each point, on the basis of the point number and distribution. When a wafer is being measured, the calculation of thickness at each point is done by its own calibration curve. During the calibration process, we first get an uncalibrated matrix, here name it X ⎛
x1,1 x2,1 .. .
⎜ ⎜ ⎜ X=⎜ ⎜ ⎝ xm−1,1 xm,1
x1,2 x2,2 .. .
··· ··· .. .
x1,n−1 x2,n−1 .. .
xm−1,2 xm,2
··· ···
xm−1,n−1 xm,n−1
x1,n x2,n .. .
⎞
⎟ ⎟ ⎟ ⎟ . (1) ⎟ xm−1,n ⎠ xm,n
The xr,c is the output value, when the probe is at point c of the standard sample r (0 ≤ r ≤ m, 1 ≤ c ≤ n). Each output value is the difference between the values at the measurement point and the initial position, eliminating the zero drift of the sensor. According to the thickness values of m standard samples (each sample has n points), we get a thickness matrix,
This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitationnew.aip.org/termsconditions. Downloaded to IP: 65.39.42.160 On: Mon, 08 Dec 2014 22:05:47
125101-6
Li et al.
Rev. Sci. Instrum. 84, 125101 (2013) TABLE I. 8-series points-distribution.
here name it Y ⎛
y1,1
⎜ ⎜ y2,1 ⎜ . Y=⎜ ⎜ .. ⎜ ⎝ ym−1,1 ym,1
y1,2
···
y1,n−1
y2,2 .. .
··· .. . ··· ···
y2,n−1 .. .
ym−1,2 ym,2
ym−1,n−1 ym,n−1
⎞
n
Num
m1
m2
m3
m4
m5
m6
m7
m8
m9
m10
⎟ ⎟ ⎟ ⎟ . (2) ⎟ ⎟ ym−1,n ⎠ ym,n
6 7 8 9 10
121 169 225 289 361
1 1 1 1 1
8 8 8 8 8
16 16 16 16 16
24 24 24 24 24
32 32 32 32 32
40 40 40 40 40
48 48 48 48
56 56 56
64 64
72
y1,n y2,n .. .
The yr,c is the thickness value at point c of the standard sample r (0 ≤ r ≤ m, 1 ≤ c ≤ n). Based on the correlation between matrix X and Y, we carry out linear fitting piecewise between the output values in each column of matrix X and the corresponding thickness values in each column of matrix Y, and obtain the calibration curve at each point. By this method, the calibration curves at total n points are obtained. Finally, stored all calibration curves. b. Measurement. When the copper film thickness of a prepared wafer is measured, the control system calculates each thickness value at each point by its own calibration curve automatically. Before each calibration or measurement, the rotary table should rotate back to the initial position, avoiding the change of position of each calibration or measurement point. In the calibration or measurement process, the film thickness value at one point is the average of all points in the local segment.
2. Points-distribution
There are two measurement modes, the XY model and the global mode, designed for the off-line module. In the XY model, all the points on two vertical diameters are measured. In the global model, the thickness values of all points distributed evenly on the whole surface are obtained. Considering the technique demands, 8-series points-distribution in the global model is used (Fig. 7). According to the total number of measurement points and the radial distance d from the outermost circle to the edge of the wafer (the measurement points on the outermost circle should be effective), all coordinates of measurement points on the wafer can be completely obtained. a. Number of circles and points-distribution on each circle. Here, n is the number of circles. num, which is the total number of measurement points, can be calculated by . According to the 8-series the formula, num =1+8 × n(n−1) 2 points-distribution, we can look up the table to get the number of measurement points on each circle, as shown in Table I.
FIG. 7. Points-distributions of different number: (a) 121 points; (b) 169 points; (c) 225 points; (d) 289 points; (e) 361 points.
This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitationnew.aip.org/termsconditions. Downloaded to IP: 65.39.42.160 On: Mon, 08 Dec 2014 22:05:47
125101-7
Li et al.
Rev. Sci. Instrum. 84, 125101 (2013)
In Table I, mi is the number of measurement points on circle i, and mi = 8 × (i − 1), (i = 2, 3, . . . , n). Whatever the number of total measurement points is, the center of wafer is always the first circle, and there is only one point there. The points on each circle are evenly distributed, and the central . angle between the two adjacent points is 360 mi b. Radial distance between two adjacent concentric circles. To ensure the principle of uniform distribution, the radial dis. R is tance r between two adjacent concentric circles is R−d n−1 the radius of wafer. 3. Software
Based on the requirements from operator, the basic functions of software also include the establishment of the communication, sending the control instructions, setting the parameters, and reading the data as well as processing the data, etc. The access to data in lower control system is the same as the in situ module. In the software development, Qt is still chosen to design GUI and the OPC technology is also used for the communication module. Since the drawing is a new and advanced characteristic in the control system software, which is necessary and useful to the off-line module, the implementation of the drawing is introduced below. a. Development environment. Matlab is a numerical language software based on the matrix programming unit with the function of efficient computation and potential function of graph processing.27 It is widely used in the engineering field, mainly for signal and image processing.28 Although Qt in Visual Studio platform has flexibility and high efficiency of C++ language, it is not good at high-performance drawing, cannot meet the needs of the drawing function development in the control system software. Therefore, Matlab is used here. b. Calling Matlab. There is no doubt that the combination between Qt and Matlab can achieve complementary advantages of each other in the practical application, so that we are able to find an optimal balance between the code execution speed and the programming efficiency, and reduce the development difficulty. Generally, there are some ways to achieve cooperation with Matlab:28 to apply Matlab’s engine, to call Matlab function by means of Matlab C/C++ function library, and to apply Matlab compiler. The Matlab engine provides a way of accessing and calling Matlab for an external language. With Matlab engine, the Matlab library can be used as a powerful programmable function library by the external language. This method can use almost all Matlab functions, but cannot be separated from Matlab environment. It means the user’s computer needs the Matlab software, and implementation efficiency is low. In the control system, the upper communicates with the lower in real time, and there is a large amount of data interaction. This requires the system has a good performance of real time. If we call Matlab function by means of Matlab C/C++ function library in the main program, namely, the drawing function is directly added into the main program, the real time of system can be affected undoubtedly. The Matlab compiler is a software tool for the improvement of Matlab development envi-
FIG. 8. Algorithm flow chart of drawing subprogram.
ronment by MathWorks. It can transform M language documents into different types of source code to generate the required MEX file (.dll), standalone application file (.exe), and COM component, and to enhance the implementation efficiency of the code greatly. So, considering the reliability and flexibility of the control system, we generate a standalone application file by the Matlab compiler. It is mainly used for drawing and can be called by the main program at any time. c. Implementation. Below is the algorithm of drawing subprogram, when the subprogram is called by main program (Fig. 8): First, reads all measurement data stored in the specified folder. Second, according to the total number of data, determines which type of measurement mode, and then generates the corresponding coordinate distribution. Third, corresponds the data to the coordinate of corresponding point one by one, and completes the drawing. Because the M language document needs to be compiled into a standalone application file, the Matlab compiler needs to be configured. Here, Lcc compiler is chosen. After the configuration, we compile the M language document and generate the standalone application file. The Matlab Component Runtime (MCR) is a set of standard dynamic link library to provide the necessary basic environment for running program. In order to run the program on the IPC, the MCR is installed.28 After each measurement process, the upper control system obtains a set of discrete measurement data. The main program first pre-processes all data, such as sorting, and then saves them in a specified file with specified format and file name, namely, the main program does all the preparations, as a data preprocessor. The output of image and complex numerical calculation of data are done by the drawing subprogram (Fig. 9). When the drawing subprogram receives the calling
FIG. 9. Diagram of relationship between main program and subprogram.
This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitationnew.aip.org/termsconditions. Downloaded to IP: 65.39.42.160 On: Mon, 08 Dec 2014 22:05:47
125101-8
Li et al.
Rev. Sci. Instrum. 84, 125101 (2013)
FIG. 10. Demonstrations of different functions: (a) Data Cursor; (b) Box Rotate; (c) Go to X-Y view.
instruction from the main program, the subprogram searches and reads the data automatically, then plots the figure, and then provides an intuitive analysis model for the operator (Fig. 10). Based on the idea above, only a slot for calling the draw subprogram is kept in the main program. When the subprogram starts to run, there is no influence on the running of the main program. IV. RESULTS
Fig. 11 shows the example of the calibration curve of in situ module, and Fig. 12 shows the experiment result from an actual CMP process. There are more experiment results in our previous paper.18 Before each test, the module is calibrated (for repeatedly continuous process, only once calibration is enough). If the sensor is changed, a new calibra-
FIG. 11. Example of one calibration curve of in situ module.
tion table also needs to be created. Here, the uncalibrated value, which is unitless, is attained from the output voltage of measurement device through high-speed A/D conversion. In Fig. 12, we can see that the thickness of copper film on wafer surface decreases as CMP process goes on. Figure 13 shows the experiment result of a multi-points measurement (off-line module), including the thickness values attained from several measurements, and Fig. 14 show the standard deviations on each measurement point. In the experiment, 5 measurements on the same wafer for 361points are taken. The maximum standard deviation is less than 4 nm. Based on a large number of experiment results, the repeatability of the measurement system are confirmed. At
FIG. 12. Measured data from an actual CMP process in the time period between 0 and 60 s.
This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitationnew.aip.org/termsconditions. Downloaded to IP: 65.39.42.160 On: Mon, 08 Dec 2014 22:05:47
125101-9
Li et al.
Rev. Sci. Instrum. 84, 125101 (2013) 1 Y.
FIG. 13. Example of one calibration curve of in situ module.
FIG. 14. Example of one calibration curve of in situ module.
the same time, these examples demonstrate that our control system performs well and stably, satisfying all needs of measurement. V. CONCLUSIONS
In this paper, we present a reliable control system for measuring copper film thickness on wafer surface, including the in situ module and the off-line module. A control system software of measurement modules is successfully realized and integrated into the whole CMP system. The system reaches to nano-scale measurement and has highly practical value in semiconductor manufacturing. Under the actual process conditions, the control system performances well and can implement all control functions effectively. ACKNOWLEDGMENTS
The work is supported by the Important National Science and Technology Major Projects of China (2008ZX02104001), and National Basic Research Program of China (2009CB724207).
Moon, Mechanical Aspects of the Material Removal Mechanism in Chemical Mechanical Polishing (CMP) (Department of Mechanical Engineering, University of California, Berkeley, 1999). 2 P. B. Zantye, A. Kumar, and A. K. Sikder, “Chemical mechanical planarization for microelectronics applications,” Mater. Sci. Eng. R 45, 89– 220 (2004). 3 P. B. Zantye, Process, Reliability and Integration Issues in Chemical Mechanical Planarization (Department of Mechanical Engineering, University of South Florida, 2005). 4 T. Wang and X. Lu, “Numerical and experimental investigation on multizone chemical mechanical planarization,” Microelectron. Eng. 88, 3327– 3332 (2011). 5 M. R. Oliver, Chemical Mechanical Planarization of Semiconductor Materials (Springer, 2004). 6 D. C. Edelstein, G. A. Sai-Halasz, and Y.-J. Mii, IBM J. Res. Dev. 39, 383 (1995). 7 S. P. Murarka, I. V. Verner, and R. J. Gutman, Copper-Fundamental Mechanisms for Microelectronic Applications (John Wiley and Sons, Inc., N.Y., 2000), p. 337. 8 Y. Ein-Eli and D. Starosvetsky, “Review on copper chemical-mechanical polishing (CMP) and post-CMP cleaning in ultra large system integrated (ULSI): An electrochemical perspective,” Electrochim. Acta 52, 1825– 1838 (2007). 9 M. Krishnan and J. W. Nalaskowski, “Chemical mechanical planarization: Slurry chemistry, materials, and mechanisms,” Chem. Rev. 110, 178–204 (2010). 10 V. R. Kakireddy, S. Mudhivarthi, and A. Kumar, J. Vac. Sci. Technol. B 26, 141 (2008). 11 R. Allen, C. Chen, T. Trikas, K. Lehman, R. Shinagawa, V. Bhaskaran, B. Stephenson, and D. Watts, “In-situ CMP copper endpoint control system,” in Proceedings of the IEEE International Symposium on Semiconductor Manufacturing, 2001 (IEEE, New York, 2001), pp. 391–394. 12 L. Li, S. G. Barbee, A. Halperin, and T. F. Heinz, U.S. patent US6072313 (6 June 2000). 13 K. R. Lehman, S. M. Lee III, W. H. Johnson, and J. Fielden, U.S. patent US6433541B1 (13 August 2002). 14 O. Nagano, Y. Yamazaki, M. Miyoshi, H. Kaneko, and T. Matsuda, U.S. patent US6563308B2 (13 May 2003). 15 W. L. Yin and A. J. Peyton, “Thickness measurement of metallic plates with an electromagnetic sensor using phase signature analysis,” IEEE Trans. Instrum. Meas. 57(8), 1803–1807 (2008). 16 W. L. Yin and A. J. Peyton, “Thickness measurement of non-magnetic plates using multi-frequency eddy current sensors,” NDT & E Int. 40(1), 43–48 (2007). 17 W. L. Yin, A. J. Peyton, and S. J. Dickinson, “Simultaneous measurement of distance and thickness of a thin metal plate with an electromagnetic sensor using a simplified model,” IEEE Trans. Instrum. Meas. 53, 1335–1338 (2004). 18 Z. Qu, Q. Zhao, Y. Meng, T. Wang, D. Zhao, Y. Men, and X. Lu, “In-situ measurement of Cu film thickness during the CMP process by using eddy current method alone,” Microelectron. Eng. 108, 66–70 (2013). 19 T. Fujita, K. Kitade, and T. Yokoyama, “Development of original end point detection system utilizing eddy current variation due to skin effect in chemical mechanical polishing,” Jpn. J. Appl. Phys. 50, 05EC09 (2011). 20 Y. Yang, G. Ren, and X. Yin, “Measurement and control system for traction motor based on OPC technology,” Adv. Mater. Res. 542–543, 208–211 (2012). 21 J. Blanchette and M. Summerfield, C++ GUI Programming with Qt4 [M] (Prentice Hall PTR, 2006). 22 J. Liu, K. W. Lim, W. K. Ho, K. C. Tan, A. Tay, and R. Srinivasan, “Using the OPC standard for real-time process monitoring and control,” IEEE Software 22, 54–59 (2005). 23 OPC Task Force, OLE for Process Control-OPC Overview, v. 1.0 (OPC Foundation, 1998), see www.opcfoundation.org/DownloadFile.aspx/ General/OPC%20Overview%201.00.pdf?RI=1. 24 D. W. Holley, “Understanding and using OPC for maintenance and reliability applications,” IEE Comput. Control Eng. 15(1), 28–31 (2004). 25 Y. Shimanuki, “OLE for process control (OPC) for new industrial automation systems,” in Proceedings of the IEEE International Conference on Systems, Man, and Cybernetics, 1999 (IEEE Press, 1999), Vol. 6, pp. 1048– 1050. 26 A. Chisholm, “A technical overview of the OPC data access interfaces,” in Proceedings of the ISA EXPO, 1998.
This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitationnew.aip.org/termsconditions. Downloaded to IP: 65.39.42.160 On: Mon, 08 Dec 2014 22:05:47
125101-10
Li et al.
27 F. W. Hong, “Hybrid programming between Matlab and VC++ in the inte-
grated environment,” in Proceedings of the 3rd IEEE Conference on Communication Software and Networks (ICCSN), 2011 (IEEE, Piscataway, NJ, 2011), pp. 377–378.
Rev. Sci. Instrum. 84, 125101 (2013) 28 L.
Hong, “The application guide of mixed programming between Matlab and other programming languages,” in Proceedings of the 2nd International Conference on Computer and Automation Engineering (ICCAE), 2010 (IEEE, Piscataway, NJ, 2010), pp. 185–189.
This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitationnew.aip.org/termsconditions. Downloaded to IP: 65.39.42.160 On: Mon, 08 Dec 2014 22:05:47
