APC Software for 300 mm Fabs
Anthony Toprac and W. Jarrett Campbell, Yield Dynamics Inc., Austin, Texas -- Semiconductor International, 8/1/2001
| At a Glance |  | ||
| | ||
| |||
The imperative for maximizing tool productivity and shrinking device tolerances has made typical approaches to manufacturing control much more expensive than they were previously. For example, engineers commonly develop process recipes that minimize process sensitivity to recipe variable perturbations. This approach minimizes the impact of drift or step change in recipe variable control. Although effective, the cost of tool time for the necessary experiments makes this passive control approach to solving process control problems cost-prohibitive.
Alternatively, active control methods allow the user to change recipe variables to compensate for drift and step changes in tool operation or the condition of incoming product. The automated version of this control methodology is termed run-to-run control. In contrast to passive control methods, a recipe developed for run-to-run control has an engineered amount of sensitivity to variation in one or more recipe variables. These variables become the control "knobs" for driving the process to target.
|
|
APC systems effectively multiply capital productivity by minimizing tool time devoted to qualification and testing. As a result, the return on investment of APC software systems is very high and can quickly cover the cost of purchase and implementation.4 Arguably, the competitiveness of a new 300 mm manufacturing plant rests on the quality of its automated control implementations as much as any other factor that contributes to its success.
Advanced control landscape
|
|
While successful in improving control of the given process step, a cell controller is unable to comprehend complex interactions among multiple process steps. Since the electrical performance of finished product represents the sum of all process step results and their interactions, cell control is not capable of driving some of the most critical electrical properties to target.
Framework systems
In recent years, discussions of advanced control software systems have focused on APC frameworks. A framework is a software infrastructure that facilitates efficient implementation of control applications by providing the basic functionality that is common across all control implementations. As such, an APC framework is not a useful product by itself, just as the Windows operating system is not useful without applications such as Word and Excel. While infrastructure software does have value, as a commercial product it holds little interest to fab engineers interested in addressing their control issues.
|
|
A comprehensive suite of run-to-run and fault detection control applications paves the way for global, fabwide control solutions. In fabwide control, a global model of the process comprehends information from all process steps and electrical tests. The model, which may be either physical5 or empirical,6 captures the sum of all process results as well as their interactions. Under fabwide control, the global model sets process targets for run-to-run controllers at multiple operations based on production process data and electrical measurements from product wafers.5
Software standards
Software standards specifications are necessary to employ APC applications and enable fabwide process control. In September 1999, SEMI first published its E93-0200 Provisional Specification for CIM Framework Advanced Process Control Component.7 This specification defines the role of the APC component in the larger CIM framework outlined by SEMI's E81-0600 Provisional Specification for CIM Framework Domain Architecture.8 As a companion to the CIM framework spec, the SEMI E96-0200 Guide for CIM Framework Technical Architecture9 serves as a technical map for the implementation of APC framework components in the context of the CIM framework.
The evolution of the E93 APC framework specification began in 1995, when Advanced Micro Devices (AMD, Sunnyvale, Calif.) and Honeywell (Morristown, N.J.) entered into a joint program called the Advanced Process Control Framework Initiative (APCFI), under support of a NIST Advanced Technology Program grant. The goal of the APCFI was to specify and build a proof of concept for a component-based distributed computing architecture that would allow rapid integration and deployment of APC applications into today's wafer fabs.
Simultaneously, Austin, Texas-based SEMATECH (now International SEMATECH) had an initiative underway to define a new component-based CIM framework architecture. This initiative leveraged the APCFI work to define an APC framework architecture. Reports on these two efforts can be found on International SEMATECH's Web site.10
|
|
This SEMATECH APC framework architecture was then taken to SEMI as a candidate for an APC specification. After the industry reviewed and modified the architecture, E93-0200 was published in September 1999. As shown in Figure 5, only the control execution component (analogous to the SEMATECH plan executor component) of the architecture survived the standards process. In addition, E93 defines a family of interfaces known as algorithm execution components, which is similar to the application interface component specified by SEMATECH.
|
|
One popular misconception is that compliance with E93 allows a semiconductor manufacturer to readily switch out an APC application or internal framework component from one software vendor's product to another. In reality, the E93 specification is far too general for plug-and-play interoperability. Key components such as databases and operator interfaces are conspicuously missing from the specification. Moreover, E93 does not address the biggest difficulty faced by the manufacturer — the integration of an entire APC application (e.g., overlay run-to-run control) into the existing CIM architecture in the fab. Although E93 does specify the interface between the process machine (or factory automation host) and the control execution component, this specification lacks sufficient detail. In any practical implementation of APC applications, vendors must build beyond the E93 spec in order to implement a working interface for the application.
The incomplete nature of the E93 spec was driven by a number of factors. First, the spec attempts to address all technologies labeled "APC," including run-to-run control, SPC, fault detection and sensor integration. The use cases and data requirements are very different for each of these technologies, requiring their own unique specification definitions. Another contributing factor to E93 specification vagueness is the relative immaturity of semiconductor APC software technology at the time the spec was written. It has been only in the past two to three years that there has been significant vendor activity in the APC market space. Prior to this, semiconductor manufacturers, regarding their systems as proprietary technology for competitive advantage, developed most APC software in-house.
One recent development that may become pivotal in establishing functional standards is the growth of open-source software. Increasingly, semiconductor manufacturers with in-house-developed APC software want to move their mature control system's support to the vendor community. Open sourcing this software could become an important means of enabling this change. By open sourcing reference implementations, solid standards could evolve based on the detailed understanding of the software's functional requirements and communication protocols. A manufacturer's mature control system technology would evolve into part of the APC software provider's product infrastructure. There are many precedents in the wider software industry of such open-source initiatives.11
Although E93 specification conformance is often referenced as a requirement in APC software purchase orders, compliance with this spec does little to enhance the value of the APC application unless the interfaces and custom enhancements to the spec are well-documented and open for interfacing by third parties. At the heart of both the CIM and APC framework specs is the requirement that vendors thoroughly document their product interfaces and allow other parties to develop interfaces with the vendor's software. This open interface concept has already been widely adopted by vendors of control software in other industries.12
Delivery of the original promise of interoperability between disparate software systems and vendor products requires strict conformance to an enhanced E93 specification, as well as detailed documentation, ideally provided by open sourcing of product interfaces.
Conclusions
In the past five years, advanced control methods have evolved from concept to implementations of proven value. The availability of commercially supported control applications, however, has lagged market demand. A full suite of control applications that interact to drive electrical parameters of finished product, although discussed in the literature, has yet to become available as a commercial product. The lag in availability of advanced control software systems has limited these implementations to manufacturers that have the ability to develop this capability on their own. Weak, poorly accepted software standards and the nonexistence of well-documented, published vendor-proprietary interfaces have not provided the manufacturer with choices for APC software that integrate easily and interoperate seamlessly with other software systems.
The competitive advantage that advanced control provides ensures that these systems will be required in newly built 300 mm wafer fabs. While the potential return on investment of these automated control systems is huge, the actual value added depends on the degree of process understanding used in developing the control methods. Even the most sophisticated control system requires a solid foundation of process understanding and accurate, meaningful quality metrics to provide value to the semiconductor manufacturer.
 |  |
Phone: 1-512-257-9500
Fax: 1-512-257-9503
e-mail: atoprac@ydyn.com
|  |
REFERENCES
- L.J. Currah, "U.S. Still Looks Good to Fab Builders," Semiconductor Business News, October 1997.
- D.N. Lauben, F.M. Traylor, "Activity Based Cost Planning— Modeling Fab Refurbishment and Greenfield Facilities," Semiconductor Fabtech, 9th ed., March 1999.
- S. Novak, "AMD Looking at 6 Sites Around Austin," Austin-American Statesman, July 3, 2001, p. D1.
- J. Baliga, "Advanced Process Control: Soon to Be a Must," Semiconductor International, July 1999, p. 76.
- C. May, "APC: A Recipe for Success in the Sub-Quarter Micron Era," Proc. of AEC/APC Workshop X, October 1998, p. 7.
- R.S. Guo, C.K. Tsai, J.H. Lee and S.C. Chang, "Intelligent Process Diagnosis Based on End-of-Line Electrical Test Data," Proc. Intl. Electronics Manufacturing Tech. Symp., October 1996, p. 347.
- SEMI E93-0200 Provisional Specification for CIM Framework Advanced Process Control Component, Semiconductor Equipment and Materials International (SEMI), www.semi.org.
- SEMI E81-0600 Provisional Specification for CIM Framework Domain Architecture, SEMI, www.semi.org.
- SEMI E96-0200 Guide for CIM Framework Technical Architecture, SEMI, www.semi.org.
- International SEMATECH, www.sematech.org.
- E.S. Raymond and B. Young, "The Cathedral and the Bazaar: Musings on Linux and Open Source by an Accidental Revolutionary," O'Reilly and Associates, 2001, www.tuxedo.org/~esr/writings/cathedral-bazaar.
- International Symposium on Open Control Systems, www.lassi.sci.fi/softsympo99.
- W.J. Campbell, Model Predictive Run-to-Run Control of Chemical Mechanical Planarization, University of Texas, August 1999.
- M. Miller, "Advanced Process Control Framework Initiative: Fab Deployment Projects," Proc. of SEMATECH AEC/APC Workshop IX, October 1997, p.151.
