Issue



Making the move to fab-wide APC


09/01/2004







Major efforts are underway to move semiconductor manufacturing to fab-wide solutions for advanced process control after initial deployment, which was primarily aimed at greater control over single process steps and certain tool sets. It is now generally believed that fab-wide APC will be required in all wafer-processing plants within just a few years, but a number of hurdles must be cleared to fully realize the benefits. Key enablers and a flexible integration approach must be utilized collectively to realize fab-wide APC solutions.

Over the past 10 years, advanced process control (APC) —which includes run-to-run (R2R) control and fault detection and classification (FDC) — has evolved rapidly from an area of research applied primarily to chemical mechanical planarization (CMP) processes, to a critical component of fab "best practices" with significant proliferation in CMP, diffusion, lithography (CD and overlay), and etch processes [1].

This evolution has been driven by the need to maintain or improve yield and to increase throughput in the face of smaller geometries, larger wafers, and greater product mixes. In addressing this need, APC has been used successfully, primarily to increase accuracy and reduce variability in a process-centric fashion for fab applications such as lithography CD, etch CD, and CMP thickness [2, 3]. While this approach improves individual process performance, it does not guarantee improvement in line metrics such as yield and throughput [4].

In the past two years, the APC evolution has matured with the focus shifting to integrate APC fab-wide, targeting broader objectives such as overall fab yields, throughput rates, and electrical characteristics in addition to the traditional process-centric goals [2, 5, 6]. Unfortunately, due in large part to the process-centric origins of APC, a number of challenges must be addressed before a robust fab-wide APC solution can be realized. This state of affairs is reflected in the 2003 International Technology Roadmap for Semiconductors (ITRS), which states, "APC has demonstrated major value to the industry, and has been adopted by most manufacturers, to some extent, ...but a truly comprehensive APC manufacturing strategy is not yet reality..." [7].

The challenges to fab-wide APC deployment are being met through the development of key enablers for APC, which address both integration and technology issues associated with fab-wide deployment. These key enablers include:

  • a hierarchical approach to process control that allows control solution elements to be coordinated to achieve fab-wide targets;
  • a new process control system (PCS) Semi standard that will be the basis for APC component integration;
  • technology enhancements that will greatly expand APC capabilities in the future; and
  • R2R control and FDC algorithm enhancements, which are required in fab-wide solutions.

Hierarchical approach to APC

As the industry moves from process-centric "islands of control" to fab-wide solutions, a control strategy must be developed that supports control and optimization for fab-wide goals — such as yield — through the coordinated control of individual process parameters. The strategy that is generally being adopted breaks apart the control problem into layers, resulting in a hierarchical control structure as depicted in Fig. 1. In this illustration, the factory is divided into three layers: process, measurement, and control. The control layer is further subdivided into process, interprocess, and factory-level sublayers. At the lowest control layer, standard R2R control solutions use feedforward and feedback data to provide control of process quality parameters, such as CMP film thickness, to specified targets. FDC is used to determine if the tool is an acceptable region where control should be applied, and further serves to filter data into the controllers. At the intermediate control layer, interprocess control solutions coordinate groups of process controllers to achieve multiprocess targets. An example here is CD control coordinated between lithography and etch process steps.


Figure 1. A logical view of a hierarchical APC strategy.
Click here to enlarge image

At the top of the APC hierarchy, a factory control scheme coordinates all control solutions toward fab-wide objectives such as yield-to-throughput balancing and electrical characteristics. Yield management systems can coordinate information with FDC systems to better identify sources of yield loss. Note that Fig. 1 focuses exclusively on the R2R control and FDC components of the solution; in actual implementation, the solution could be extended to incorporate other components such as fab-wide productivity monitoring, fault prediction systems, maintenance management, and dynamic scheduling and dispatch.

Enter the PCS standard

Constant challenges in deployment of APC are the cost and reliability of integration. As we move to fab-wide solutions, the integration problem is magnified and includes not only tool integration for data collection, but also application-to-application integration to support interoperability among APC components and between APC and non-APC components. It is clear that there is a need for APC component interface standards to support interoperability as well as interchangeability or "plug-n-play" of APC components. The semiconductor industry recognized this need and recently developed the Process Control Systems (PCS) format (Semi standard E133) to address it [8, 5].

The PCS standard will facilitate the integration of process control systems into current and future fabs. This standard is focused on defining capabilities of the PCS functional groups of R2R control, fault detection (FD), fault classification (FC), fault prediction (FP), and statistical process control (SPC). This standard also specifies PCS interfaces that will enable these functional groups to interact effectively and share data among themselves and with the other interdependent factory systems, including systems within the process equipment. The PCS standard thus is expected to play a critical role in fab-wide APC proliferation by being a key element for cost-effective and reliable APC solution integration.

APC enhancements

APC technology has been enhanced greatly over the past 10 years. Advances in R2R control include multivariate control approaches and solutions demonstrated for CMP, diffusion, lithography, and etch; interprocess control solutions (such as lithography-to-etch CD control); solutions for rejecting maintenance and product disturbances; and methods for dealing with metrology issues (such as missing data). FDC advances include the proliferation of multivariate analysis and associated approaches for data reduction; application of techniques such as principle component analysis (PCA) and project-on-latent structures (PLS) to support the classification component of FDC; methods for pre-qualifying and continuously monitoring, and assessing tool data quality to support FD; and virtual metrology [2, 3, 5, 9, 10].

Many of these R2R control and FDC enhancements have been driven by requirements distilled from earlier APC implementations and have resulted in more robust solutions suitable for fab-wide deployment.

One example of a key APC enhancement is virtual metrology, which is an estimation process of predicting values of quality parameters of interest [10]. Traditionally, R2R control has focused on the control and improvement of quality product parameters that are directly measured, such as film thickness. In some cases, however, the cost of measuring all wafers in production is prohibitively high or the quality parameters cannot be directly measured.


Figure 2. "Type 2" virtual metrology solution for copper CMP control.
Click here to enlarge image

Virtual metrology techniques can be used to predict in-line measurable parameters, such as thickness, thereby reducing the frequency of measurements required to support control. Virtual metrology can also be used to predict parameters that are not directly measurable in-line, such as sheet resistance, thereby enabling control of these parameters. These two flavors of virtual metrology can be referred to as "type 1" and "type 2," respectively. Note that a virtual metrology approach will only work if a predictable relationship can be discerned between the metrology quality parameter of interest and the virtual metrology parameters. Furthermore, the process will always require regular tuning with techniques such as off-line metrology. Figure 2 shows a type 2 virtual metrology solution used to provide control of sheet resistance, Rs, by relating it to upstream metrology parameters and post CMP-process thickness [10].

Enhancements for fab-wide deployment

As R2R control and FDC deployment have matured, so has the technology to support them. For example, both R2R control and FDC have evolved from univariate to "univariate + multivariate" solutions. Multivariate solutions enable simultaneous control of film thickness and uniformity in a CMP process [3, 4]. As the evolution continues to fab-wide multicomponent solutions, technology enhancements found in only a small percentage of today's APC solutions will be required.

R2R control solutions maintain dynamic or adjustable models of the processes they are controlling. The modeling process determines which parameters on the tool should be adjusted, and by how much, to correct for process drift and shift. Once the models are developed, they must be continuously tuned to account for changing process conditions. This requires some form of downstream (post) metrology to evaluate the controller performance. The controllers can also use incoming (pre) metrology to compensate for incoming wafer variations. It is the method of tuning in response to metrology data and process conditions that, in large part, differentiates R2R control algorithms [3, 9].

A number of other capabilities that may be more important to realizing fab-wide R2R control are required, because they allow the solutions to provide effective automated control in the face of common practical issues such as tool and metrology errors, maintenance events, and product shifts (see table).

FDC solutions generally monitor various parameters on a tool, detect out-of-norm conditions, and relate them to problems with the tool [5, 11]. Today these solutions can include univariate and/or multivariate techniques, where fault conditions are linked to individual or multiple variables, respectively. While multivariate techniques are now a general requirement of FDC solutions, it is important to note that it has been reported that over 70% of faults can be determined through univariate techniques [5]. This information, combined with the fact that univariate analysis is much easier to understand, believe, and relate to a cause, clearly indicates the univariate analysis capabilities should also be a requirement of FDC solutions.

Other capabilities that should be required of all FDC process control solutions in a fab-wide environment are summarized in the table. Perhaps the most important of these capabilities is ease of use. FDC systems collect large amounts of data and often use complex algorithms. These systems can only be effective if they are used on a regular basis, however; thus, one challenge is to provide users with the right information so that they can develop and deploy effective FDC without unnecessary exposure to the complexities of the solution. Approaches that are addressing this challenge include distributed (e.g., web) user interfaces (UIs) with web-style drill-down approaches, model development wizards, and model templates and libraries.

Flexibility is the key

The ITRS states that fab-wide APC will be integrated at the data storage, user interface, and logic interaction levels [7]. This means that a fab-wide APC approach, represented in Fig. 3, that provides the following should be used: a common data management scheme for applications, access to APC data in a dashboard fashion where UIs are configured to user requirements, and APC and non-APC components to be integrated and utilized collectively in a configurable fashion to support fab-wide goals.


Figure 3. A flexible fab-wide APC infrastructure.
Click here to enlarge image

The primary components of this approach are briefly described as follows.

Control rules strategy editor. Perhaps the most important component of a fab-wide APC solution is a mechanism that supports the configuration and maintenance of control strategies. Fab-wide APC solutions must be highly configurable because they will involve multiple components to collectively provide APC solutions that cross process boundaries. APC system rules or "strategies" must be configured and reconfigured to support specific factory objectives.

PCS-compliant application interface. The PCS standard defines basic capability and interface requirements for all APC components, establishing a level of interoperability in APC systems. The cost of APC integration at a fab-wide level could easily be prohibitive if a standard integration approach is not used, so fab-wide APC systems should support the integration of PCS-compliant components.

Common data repository. A significant portion of the benefits arising from fab-wide APC results from the ability to target multiprocess to fab-wide goals, such as chamber-to-chamber comparisons and optimization, multiprocess control, and tying control to yield and throughput objectives. These capabilities require a fab-wide data warehousing strategy that provides these services and meets the performance requirements to support the individual applications and integration with non-APC capabilities.

"Dashboard" approach to UI configuration. A key impediment to widespread acceptance of R2R control and especially FDC on the factory floor has been the complexity of these systems as presented through their UIs. R2R control and FDC solutions are generally provided with their own configuration and operation UIs that can be very difficult to understand and cluttered with information that is of little interest to the operator.

This problem is not unique to semiconductor manufacturing and is being addressed with the "dashboard" UI approach [12]. With the dashboard approach, the user is presented in one or a few screens with information from multiple applications that are of the most interest (much like the dashboard in an automobile). Dashboard solutions for APC in semiconductor fabs should be:

Graphical and distributed. Graphical and web-distributed UIs allow users to quickly grasp APC concepts and support drill-down analysis, while allowing the UIs to port from the fab floor to the office environment.

Multicomponent and configurable. As with a traditional dashboard, the APC UI approach must allow data from multiple components to be provided on one UI, and allow the user to configure UIs to display the data that is most important.

Vendor-independent. Key components displayed in APC dashboard solutions should be provided by an APC application type, regardless of supplier. For example, if a R2R module from one supplier is replaced with another, a large portion of the R2R control dashboard components should remain unchanged. With this approach, a common dashboard can be used fab-wide with minimal re-training required whenever an APC application is replaced.

Linking to database or database middleware. An approach to provide this vendor-independent common dashboard has been to tie the UI to the common database rather than directly to the individual applications. While this approach has significant business and standardization implications, it is being considered in many industries, driven by the high cost of (re)training [12].

Looking ahead

APC has already demonstrated significant return-on-investment [13]; it is now considered a required component of many processes in the fab. Even so, the industry has only begun to tap the true power of APC. The move to fab-wide solutions will not only allow companies to tie APC to fab-wide yield and throughput objectives, but also to eventually integrate the logic of APC with other components such as scheduling and dispatch, maintenance management and prediction, and even supply chain management.

While this powerful vision requires a fully logically integrated system, it is important to note that device manufacturers can realize this vision in well-defined steps, with each step providing its own significant ROI. The key is following an integration strategy that supports the migration. The strategy for the next important step, namely the move to fab-wide APC, reinforces this long-term vision.

Acknowledgments

The author wishes to acknowledge the software solutions group (SSG) at Brooks Automation, especially Jianping Zhou, Mark Yelverton, Todd Nowak, Brad Schulze, and Kevin Dimond, for their contributions to and review of this article; and also International Sematech, especially Dr. Brad Van Eck, for support and promotion of APC.

References

  1. L. Peters, "APC is Gaining Widespread Adoption," Semiconductor International, Dec. 2001.
  2. M. Liu, "APC from the Foundry Perspective" (keynote address), Proc. AEC/APC Symposium XIV, Sept. 2003.
  3. J. Moyne, E. del Castillo, A. Hurwitz, Run-to-run Control in Semiconductor Manufacturing, CRC Press, 2000.
  4. C. El Chemali, J. Moyne, K. Khan, J. Colt, J. Chapple-Sokol, et al., "Multizone Uniformity Control of a CMP Process Utilizing a Pre and Post-Measurement Strategy," J. American Vacuum Soc., June/July 2000.
  5. Proc. AEC/APC Symposium XIV, Sept. 2003.
  6. C.A. Bode, T.J. Sonderman, "Controlling the Margins in 300mm Manufacturing," Solid State Technology, pp. 49–52, Feb. 2004.
  7. International Technology Roadmap for Semiconductors, 2003 edition, Semiconductor Industry Association, available at http://public.itrs.net.
  8. Semi E133: Provisional Standard for Automated Process Control Systems Interface, Semiconductor Equipment and Materials International, April 2004.
  9. J. Mullins, J. Zou, "Frequency Domain Stability and Performance Analysis of Moving Average Run-to-Run Controllers," Proc. AEC/APC Symposium XIV, Sept. 2003.
  10. J.S. Lin, P.H. Chen, S. Wu, F. Ko, M.S. Zhou, et al., "Model-based Wafer-to-Wafer Control for Cu CMP," VLSI Multilevel Interconnect Conf., Sept. 2003.
  11. G. Cherry, J. Qin, "Multivariate Analysis for Semiconductor Fault Detection and Classification and Tool Performance Assessment," Texas-Wisconsin Modeling and Control Consortium, Feb. 2004.
  12. J. Moyne, J. Korsakas, D. Tilbury, "Reconfigurable Factory Testbed (RFT): A Distributed Testbed for Reconfigurable Manufacturing Systems," Japan-USA Symposium on Flexible Automation, July 2004.
  13. T. Stanley, et al. "Cost and Revenue Impact of Advanced Process Control (APC) with and Emphasis on Run-to-Run Control (R2R)," Proc. Sematech AEC/APC Symposium XIV, Sept. 2002.

James Moyne is director of advanced process control (APC) technology at Brooks Automation Inc.'s Software Solutions Group, 2340 West Shangri La, Phoenix, AZ 85029; ph 734/516-5572, fax 602/861-1442, e-mail [email protected].

Extended capability requirements of fab-wide APC solutions

Run-to-run control

  • Provide control only in a region where control models are valid, e.g., when the tool is healthy (possibly as determined by an FD capability)
  • Accommodate missing, delayed, out-of-order, or "bad" metrology data
  • Provide robust control in the face of model error
  • Support model adjustments to accommodate shift disturbances such as PM events at a tool
  • Support model adjustments to accommodate product changes and automatically calculate model adjustments for newly introduced products
  • Provide mechanisms to evaluate the health and effectiveness of the R2R controller
  • Utilize intelligent metrology strategies, balancing metrology sampling frequency and accuracy against yield and throughput objectives
  • Provide PCS-standard compliant interface specifications for integration.

Fault detection and classification

  • Provide an easy-to-use solution
  • Operate over a fab-wide data framework to enable capabilities such as tool-to-tool comparative analysis
  • Provide PCS-standard compliant interface specifications for integration