Wafer Handling and Fab Automation

Improving productivity through automated software-driven manufacturing

Robert A. Pisa, PRI Automation

A new system relies on computer-integrated manufacturing to achieve automated material handling and complete fab management.

The semiconductor industry fuels its growth by continuing to bring more and more functions to increasingly complex chips while at the same time reducing the cost to the consumer. Productivity improvements realized by independent device manufacturers and foundries are measured in functions per dollar, numbers of transistors per unit area of silicon, or reduced manufacturing costs. Historically, three key factors have combined (albeit at different rates) to improve IC manufacturing output by 25­30% per year:

• Feature size reductions. It is believed that shrinking feature sizes will remain the key contributor to productivity in the near future. However, whether optical lithography technologies will keep pace with ever-smaller device features remains to be seen. To date they have outlived many predictions of an early demise.
• Wafer size increases. It is unlikely that efforts to grow 400- and 450-mm silicon ingots will bear fruit in the foreseeable future, indicating that productivity gains are not likely to come from further increases in wafer diameter. Indeed, the move to 300-mm manufacturing has experienced many false starts and is taking much longer than expected because of delays in developing appropriate technologies, process tools, and standards and as a result of unfavorable market and manufacturing economics. Manufacturing ramp-up in 300-mm fabs is scheduled to begin in earnest only in 2001, while full production in some factories is expected to commence a year later.
• Yield enhancements. As illustrated in Figure 1, productivity improvements attributable to yield enhancement strategies have been modest in the recent past and are predicted to decrease, despite the efforts of process and yield enhancement engineers who interpret fab data and apply remedial measures and despite the development of yield-measuring tools, data collection systems, and data-analyzing applications. The upcoming challenge will be to preserve and improve line and sort yields as 300-mm production ramps to high volume (20,000 wafer starts per month or more).

Figure 1: Graph of semiconductor advances, their contribution to present productivity increases, and future projections.

The IC industry faces the challenge of lifting 300-mm volume manufacturing up to current manufacturing standards and beyond. The new 300-mm fabs will have to match the performance of today's world-class 200-mm fabs, which operate with line yields of approximately 98­99% and defect densities of 0.15 defects per square inch. Moreover, 300-mm manufacturing will require the types of automated material-handling systems that are used optionally in 200-mm facilities today, as well as new methods of material tracking and information handling. This article describes a system developed by PRI Automation (Billerica, MA) that relies on computer-integrated manufacturing (CIM) to achieve automated material handling and complete fab management.

Productivity Enablers

A report by the Competitive Semiconductor Manufacturing program at the University of California (Berkeley) cites five key productivity enablers and their direct impact on IC manufacturing performance: automated equipment data entry, automated capture of engineering data, computerized yield analysis, recipe management, and reticle management.¹ These enablers and their impact on semiconductor manufacturing are summarized in Table I.²

Productivity Enabler Impact on Manufacturing Performance Line Yields Equipment Throughput Labor Productivity Defect Densities Cycle Time Automated equipment data entry x x x. . . Automated capture of engineering data x x x . . Computerized yield analysis . . . x . Recipe management . x . . x Reticle management . x . . x

Table I: Product enablers and their effects on semiconductor manufacturing.

A complete factory management system addresses these productivity enablers. PRI Automation's system, based on computer-integrated manufacturing, represents the merger of an automated material-handling system (AMHS) with a factory management software system (FMSS). The CIM system includes computers and computer system architecture, networks and network architecture, and software and software architecture. The FMSS is the software and software architecture part of the overall CIM system and includes many types of enabling and intermediate software and, most important, applications. Applications provide information to users in a variety of forms and perform myriad functions: material flow direction; flexible, configurable reporting; capacity and operational planning; shop-floor scheduling and dispatching; equipment maintenance management; management of durables (such as reticles); statistical process control; advanced process control (run-to-run and fault detection and classification); yield analysis; and design of experiments. Such applications are crucial to a factory management system because they help reduce cycle times, optimize resource utilization, improve throughput, increase device performance, and enhance device and product yield.

Software-Driven Manufacturing

The factory management software system is the brain of the factory management system. Integrating all of the software and hardware components of the FMSS into the overall manufacturing system is complex and can cost as much as 20­30% of the entire computer-integrated manufacturing budget. Aside from seeking to reduce costs and system complexity, an efficacious computer-integrated system must help reduce work in progress (WIP) inventory, perform rapid computer start-up, and improve material-handling capacity.

A high-level factory management software system must process and respond to various types of data: corporate data, materials inventory data, manufacturing data, factory activity and process data, materials movement data, and scheduling data.

Corporate Data. Every fab receives directives from an enterprise resource planning system about product demand (i.e., what to build) and delivery information (i.e., when to deliver it). Semiconductor companies often do not consist merely of a single production facility. When a diverse range of products is manufactured, a company must address which of its factories has the process capability to produce a given product, the due (delivery) date of the product, and the fab's current workload and future capacity.

Materials Inventory Data. The enterprise resource planning system feeds data to the manufacturing line, including data on incoming raw materials and durables, such as reticles and front-opening unified pods (FOUPs). Fabs need materials of all types to produce products. While usually relying on other corporate systems to maintain stock levels, place orders with vendors, and receive materials, the FMSS stores specifications for when and where materials are used in the process flow.

Manufacturing Data. The FMSS publishes or pushes data both up and down the chain to run the fab and report results such as standard direct costs, product routings, process step capacities, process duration, WIP status, equipment status, in-line data, WIP location, and the location of durables. The software system stores all of the data required to build a product. In order for corporate accounting systems to roll up the cost of products, process engineers calculate costs by determining production time, materials involved, and the equipment used at each process step. With this data, the cost basis of each product and the value of the line inventory can be determined at any time.

In addition to inventory information, the FMSS stores "current state" data. These data include the step in the process flow each product is at, the product's physical location, the product's status (whether it is in a queue, in process, on hold, or being reworked), the physical location and status of all process durables, and, finally, the status of the process tools (whether they are available, down, or undergoing preventive maintenance). With these data, the scheduling system can determine when to process each lot on all the tools in the fab and thus optimize productivity.

Factory Activity and Process Data. In a manual environment the FMSS directs the operator via user interfaces, while in a semiautomated or fully automated environment it directs fab operations via system-to-system and system-to-process communication. For example, it moves material from one point to another, establishes the next process step and what to do at that step, and directs the collection of manual or automatic data.

Going beyond the simple storage of factory manufacturing data, the FMSS plays an active, dynamic role in material processing. For example, if a plan to move material from point A to point B must change because of the need for inspection and subsequent rework, the physical location of the product and its next process step are updated automatically. Also, if metrology results (e.g., the oxide thickness of dielectric deposition) indicate that a change in a subsequent process is required (e.g., a time or pressure adjustment of chemical-mechanical polishing), the software can automatically direct the change.

Materials Movement Data. FMSS has a control software function that controls the physical delivery of such materials as product, engineering, monitor, and test wafers; FOUPs; and reticles. The automated material-handling system accomplishes all material movements via the software system and material control system, performing exactly the steps it has been programmed to perform. It relies on the specifications stored in the FMSS and reacts to changing needs in the fab's dynamic environment. This optimizes the use of system as well as fab resources.

Scheduling Data. The overall optimization of the fab is accomplished with the factorywide scheduling of every lot and the use of every tool (including durables such as reticles and FOUPs) over a fixed time horizon. Scheduling regulates all of the work in the entire fab, while dispatching is performed mostly on the local (area) level. Since local optimization depends on the immediate status and the immediate future status of the resources in other areas, dispatching may or may not be appropriate.

Organizing Fab Activities

One objective of the FMSS is to collapse and reduce the numbers of layers in the software architecture. Thus, as delineated in Figure 2, the system organizes fab activities on three levels that represent logical software functions and are not necessarily the levels often associated with software architecture. This simplified structure minimizes the need to integrate disparate data elements, reducing or perhaps eliminating duplication across multiple systems.

Figure 2: The factory management software system organizes fab activities on three levels.

Level 0: Get the Data. To realize improved productivity, the system must be able to make informed decisions. This requires that a sufficient amount of data be available to the FMSS and that some or all of that data be available to other components in the CIM environment. Data requirements include the following:

• Data timeliness--real-time data.
• Various data types, such as WIP status, equipment status, recipe status, reticle status, operator availability, metrology data, point-to-point transport time, and data on environmental conditions.
• Data availability--data must be accessible to any delegated person and any system, from anywhere.

Raw data, which are crucial because they undergird higher-level applications, are provided by equipment vendors via tools' semiconductor equipment communications standard (SECS) communication ports. Although standards exist for governing data communications between tools and host systems, recent tests have shown that there is much room for improvement in the types and quality of data streams and their ability to remotely control tools.³ While process tool automation is accomplished by writing equipment drivers and station controllers as intermediary layers in the software architecture, such station controllers are often designed with specific point-to-point communications capabilities and are difficult to change and upgrade once a fab is operating. Moreover, developing station controllers involves a good deal of business logic. Before the development of factory management software, station controllers were used to fill gaps in the functionality of the host system. While gap filling works, its implementation is costly. Consequently, a cost-effective, web-enabled integration system has been developed that makes all desired process tool data immediately available, including data from sensors that are added on or that do not enter the system via the SECS port. Thus, engineers and technicians can now observe (and even control) a process tool from any browser. In the future, this technology will be incorporated into the process tool controller.

Level 1: Run the Factory. Level 1 is composed of two sublevels, which partially overlap each other:

• Level 1A--intermediate applications, which actually run the fab and make real-time decisions that improve process control, yield, product performance, and resource utilization while reducing WIP.
• Level 1B--additional applications that are used in more of an off-line mode in contrast to the on-line or real-time functioning of Level 1A applications.

As depicted in Figure 3, level 1A applications combine the traditional functions of statistical process control and manufacturing execution systems with new active applications such as advanced process control, fault detection and classification, advanced equipment control, and WIP and resource dispatching.

Figure 3: Level 1A applications that control and improve fab operations.

The most common aspect of advanced process control is run-to-run control, which may render statistical process control obsolete in the future. After developing models for a given process on a given tool, the run-to-run procedure adjusts (within limits) the parameters and set points of process recipes to keep the process on target. This is accomplished by using process metrology data in a feed-forward and feed-back fashion. Advanced process control requires a homogeneous FMSS environment, because data must be moved up and down the entire chain from one process step or tool to the next while the requisite data systems interact when necessary.

Fault detection and classification signal changes in a tool's behavior before an out-of-control or out-of-spec condition arises. This function allows fab personnel to conduct preventive maintenance and repairs when needed and not necessarily at predetermined times, which may reduce tool use.

WIP productivity improvements result when an active dispatcher drives manufacturing to meet a published, optimized, factorywide schedule. The material control system--the software link to the AMHS system--receives a schedule well in advance of production so that material is prepared for the various process steps and the AMHS can optimize the use of its resources. This helps even out the peak demands on the transport system. The schedule also determines how the AMHS will deliver other resources, such as reticles, to the process tools. The result is that the right material is at the right place at the right time. Additionally, when the manufacturing execution system specifies that product must be rerouted for rework, a new schedule becomes available, thus reoptimizing the factory management system.

As shown in Figure 4, level 1B applications "work the data" and feed the results back to the level 1A applications, leading to an overall improvement in production.

Figure 4: Level 1B applications for working the data.

In off-line mode, applications such as design of experiments determine lot splits and process flows that minimize disruptions while squeezing the greatest amount of information out of the least amount of data. The integration of a design of experiment system with the manufacturing execution system can provide for the automatic generation of splits, rejoins, and rerouting in the execution system.

Other applications deal with specific portions of the factory data that are collected. One such application is a tool performance tracking platform, which was written by Sematech to improve overall equipment effectiveness. The FMSS equipment connectivity component spools every SECS message to a formatted file that may be postprocessed by the performance tracking platform. Tests on this platform indicated that a high percentage of the SECS messages from various process tools were inaccurate or not available, indicating the importance of Level 0 data gathering.³

Level 1B applications also use in-line metrology data from test, monitor, and product wafers to perform data correlation and data-mining functions. Thus, wafer sort yields can be analyzed for virtually any factory attribute, such as process tools, recipes, recipe parameters, recipe results, streaming tool data, or reticles.

Level 2: Corporate Strategic Goals. Level 2 applications involve a company's strategic goals and other corporate goals such as customer service issues. These applications, which are based on manufacturing data from the fab, include factory planning, simulation, cost of ownership models, capacity analyses, capital equipment planning, standard direct cost and margin models, and an available-to-promise function.

Conclusion

With the advent of 300-mm semiconductor manufacturing, the demand for intelligent factory automation will come increasingly to the fore. Automated, software-driven systems will improve productivity, optimize resource utilization, and respond quickly to changes on the fab floor involving process and product modifications, nonproduct wafers, hot lots, or process control issues. The economic and ergonomic issues of 300-mm manufacturing will create a strong demand for material-handling automation systems that can be integrated into a fabwide management software system.

Acknowledgment

Figure 1 and Table I are reproduced from "Productivity Enablers in Semiconductor Manufacturing," a presentation by Douglas Scott of PRI Automation (Billerica, MA) at Semicon Japan 1999.

References

1. RC Leachman, J Plummer, and N Sato-Misawa, "Understanding Fab Economics," in Competitive Semiconductor Manufacturing Program Report 47 (Berkeley, CA: University of California, 1999).
2. D Scott, "Productivity Enablers in Semiconductor Manufacturing" (paper presented at Semicon Japan, Tokyo, Japan, December 1­3, 1999).
3. C Deller and L Eng, "Automatic Data Collection," draft document (Austin, TX: International Sematech).

Robert A. Pisa is director of system solutions at PRI Automation in Billerica, MA, where he is responsible for defining complete system solutions for improving productivity at semiconductor wafer fabs. Before his present assignment he was the product manager for Promis Encore! at PRI's factory management software dvision (formerly Promis Systems). Pisa has been with PRI Automation/Promis Systems since 1996 during which time he has held a variety of product management positions. Before joining the company, he held various management positions in semiconductor process engineering, engineering management, and wafer fab management at Analog Devices and Polaroid. He received a BS in photographic science and instrumentation from the Rochester Institute of Technology in Rochester, NY. (Pisa can be reached at 978/670-4270 ext. 3358 or rpisa@pria.com.)

MicroHome | Search | Current Issue | MicroArchives
Buyers Guide | Media Kit