Striving to realize productivity in truly optimized fabs

Steve Fulton and Harvey Wohlwend, International Sematech Manufacturing Initiative (ISMI)

The drive to improve productivity and reduce costs in the semiconductor industry is accelerating. While productivity gains have historically been achieved by implementing technology advances such as device feature-size reductions, yield improvement, and wafer-size increases, other methods include the construction of 300-mm factories, which were designed to provide a quantum leap in manufacturing. Productivity gains from the ubiquitous use of automation and unprecedented levels of integration, which firmly established the factory host as the single dominant point of control, have fundamentally altered the fab productivity paradigm. Moreover, expanded advanced process control (APC) and fault detection and classification (FDC) applications take advantage of these enhanced capabilities and may be the first step in a virtual data revolution to sustain Moore's productivity curve.

Future technology and productivity challenges, such as those that appear in the factory integration chapter of the International Technology Roadmap for Semiconductors, include access to more and better data. Clearly, relatively few expert resources should focus on problem solving rather than on the search for data on process or equipment performance. Perhaps most important is the ability to uncover new opportunities in the manufacturing environment for productivity improvements.

Data Access

The International Sematech Manufacturing Initiative (ISMI) has spent the last three years defining and developing the e-manufacturing concept. In short, e-manufacturing is the use of advanced and emerging information technologies to provide automated, data-driven productivity optimization. As illustrated in Figure 1, three standardized data interfaces in the fab have been defined: Interface A, also known as the equipment data acquisition (EDA) interface, for equipment process and status data; Interface B for data exchange between factory applications; and Interface C for data between the factory environment and the rest of the world. (Interface C includes security and context considerations.) These interfaces have led to the use of advanced communications protocols, such as XML/SOAP for EDA, instead of SECS/GEM.^1,2

To meet future technology challenges and improve manufacturing productivity, a higher data-transfer rate is needed, in addition to more and better data. But that is just the beginning. For example, the data rate supported by SECS/GEM communication has historically been limited to 50–100 scalar variables at approximately 3 Hz, or about 300 data points per second. Additionally, the need to pass this data through the SECS/GEM port without conflicting with command and control messages has caused delays between the generation of sensor data and their availability to fab systems such as APC. SECS/GEM's limited data rate and inherent delays have placed severe limitations on productivity improvement.

Richer data sets, variables, and values that reveal process and tool conditions comprehensively are necessary, but they have not previously been available outside the process equipment. The EDA port has a minimum goal of 10,000 data points per second, which is at least 50 variables per process chamber at approximately 10 Hz, as shown in Figure 2. However, data requirements are specific to the type of equipment and process. For example, rapid thermal processors may require few variables at very high sample rates, such as 50 total data points at 100 Hz, while more-sedate tools, such as wet processors, may require many variables (e.g., the status of chemical-supply subsystems and equipment health) at lower sample rates, such as 2500 total data points at 2 Hz. Most importantly, data availability should not be limited by the transfer rate.

At this early stage of implementation, it is hard to envision what data will prove to be valuable or critical for a tool or process, as the discovery process has just begun to go beyond the limitations of SECS/GEM data transfer rates. While challenging, identifying the sensitivities and control parameters of specific processes or product families also raises competitive or proprietary issues. Toolmakers have expressed great concern about making every sensor, setting, event, or condition on their equipment accessible externally. Considering the complexity of semiconductor manufacturing tools and the high standard of operational performance expected of them, that concern is not without justification. Hence, close collaboration is required between OEMs and end-users to balance between the need for significantly more data and increasing performance risk. The first steps in that direction will be driven by APC and FDC applications, where access to more process variables at increased rates will be used to achieve improved models, enhanced decision making, and productivity gains.

Supplying fresh data where and when they are needed is required for each fab application to perform optimally, regardless of factors such as messaging overhead, sensor clock cycles, or the needs of other applications. In addition, the data must be well formed; they must be precise and accurate, have the proper number of significant digits, and have the correct format and time stamp. In light of current fab conditions, the need for improved data quality is obvious: Sensor data exceed the sensitivity of the sensor, subsystems and components in a single tool have conflicting time stamps and time-date formats, and data are rounded based on data bus or operating system limitations. In short, a data revolution is needed on the same scale as the 300-mm factory integration and automation changes that have taken place in the last few years.

Role of Prototypes

ISMI piloted an innovative approach to EDA standards, working to implement accelerated cycle of learning (ACL) prototypes, which are intended to be discovery vehicles for standards development and for identifying areas in need of improvement, or technical challenges requiring prompt attention. Improving standards based on proof-of-concept prototypes identifies incompatibilities, complexity, and holes that can be corrected early in the standards process. The result is a more mature set of standard requirements that has been rationalized for commercialization.

Initially, the ACL prototypes were used on equipment simulators, a relatively low-cost approach that could validate the standards' underlying content. The first prototypes focused directly on the core elements of EDA, the collection and messaging of data, and an evaluation of XML/SOAP as the messaging protocol. Work with the prototypes that investigated data collection from an equipment simulator and messaging to a host simulator identified the need for several refinements and clarifications that were predominately incremental and accelerated the maturity of the standard.

As presented in Figure 3, the XML/SOAP prototype discovered sensitivities that had not previously been known or expected. For example, investigators found that even when the SOAP 1.1 standard requirements are met, care should be taken to ensure interoperability when selecting a SOAP package for implementation. In addition, interactions involving the computer hardware and the operating system that SOAP will run on can affect overall communication performance significantly. ISMI has made these general findings public and has presented them at industry forums, preparing suppliers to make informed design choices.

Other prototypes focused on equipment self-description as well as standards' administrative and security requirements, applying control mechanisms to the collection and transfer of data from the equipment to host simulators. The challenges of certificate management in the fab environment and the mechanisms to be used by equipment suppliers resulted in incremental standards improvements. These prototypes highlighted the challenges associated with creating equipment metadata, the equipment self-description that describes the hardware configuration, components breakdown, and the software structure and mechanism that define the data available from the equipment. While creating equipment metadata is a relatively straightforward process once the structure and requirements are understood, the investigators did not fully appreciate the scope of the task because of the equipment's complexity. Even a relatively simple tool—simple in the sense that it runs mature processes, is well understood, has few variables, and does not perform a critical process step—produced a metadata set with more than 28,000 lines and a file size in excess of 10 Mbyte. Equipment self-description was highlighted as a critical foundation element of EDA that must be well reasoned and thoroughly implemented for the equipment to meet customer requirements in the fab.

Improvements were made to reflect the learning gained at each successive stage of implementation, accelerating the completeness and maturity of the standard, reducing overall industry risk, and lowering supplier-specific development costs substantially. When EDA-capable prototypes were being demonstrated on production-level equipment, the quality of the standard had matured to such an extent that the prototypes were essentially preproduction versions with production capability. Using the prototyping procedure, the investigators were able to confirm that every functional element and requirement in the standards could be implemented correctly. In addition, they were able to provide performance data showing that aggressive chipmaker goals could be met and, in many cases, exceeded without special effort.

It is difficult to estimate how much ACL prototyping has accelerated the adoption of EDA standards. However, it appears that the adoption cycle will be two years less than that required to implement the 300-mm communications standards. Perhaps more notably, during the EDA implementation ramp, the equipment suppliers avoided incurring development and implementation costs, implementing immature and incomplete standards that were subject to revision costs, and bearing additional support burdens. In contrast, during the five-year 300-mm standards ramp, revision costs, increased field support, delayed acceptance, and custom functionality cost suppliers more than $1 billion, and some issues remain unresolved.

While the costs of production and efficiency delays in the adoption of 300-mm standards have not been quantified, they certainly amount to hundreds of millions of dollars per year. On an annual basis, the cost-avoidance value of EDA's first-pass success may be lower because of the phased adoption ramp rate, but overall EDA costs could well exceed 300-mm costs. Nevertheless, clear and mature requirements enable equipment suppliers and the OEMs to achieve solid design and delivery targets. Early EDA standards prototyping has delivered value to the entire supply chain.

EDA Implementation Phase

In 2005, implementation of EDA standards will begin in earnest. Many equipment suppliers are in the process of developing their commercial offerings, and purchase orders for EDA-capable tools are commonplace. This is a critical time when the risk of a disconnect between requirements and expectations or between standards content and the functionality of the equipment implementation is high. Corrections later in the process will be expensive; they will jeopardize delivery times, installations, and time to money. To avoid later difficulties, ISMI developed a scenario consensus guideline on EDA's intended use and chipmaker context.1 By ensuring that tool design and implementation support customers' intended use of the standards, OEMs can minimize development costs while avoiding delays and redesigns.

Customer-use scenarios describe the sequence of events and functional interactions that are considered normal by IC manufacturers. They identify the use of certain types of data with specific functions. For example, APC will largely involve process sensor data, whereby the start and stop of data collection will be defined by such events as the process temperatures reached or the presence of radio frequency. Equipment utilization and productivity data will be driven by events such as wafer location occupied or process job start and stop with time stamps. An understanding of intended use will enable toolmakers and end-users to develop common expectations and implement the emerging EDA standards.

Software Testing and EDA Implementation

Because it is very difficult to demonstrate that standards requirements have been met, the delivery and acceptance of complex EDA software can be challenging. Furthermore, the interaction of the fab environment and fab systems can complicate the results of implementing EDA, often making it difficult to identify the source of a problem. As contributors to reduced performance, factory network (peak-load) limitations or client applications that cannot accept data at the specified rate because of design, degradation, or other reasons can be nearly impossible to measure.

The use of an independent software test to quantify the conformance of a tool or process to SEMI standards and chipmakers' intended implementation goals solves many such problems. Here, too, the 300-mm experience is illustrative of the utility and value of objective software testing. Because the expected interaction among the 300-mm communications standards was not defined, chipmakers relied on specific tool suppliers' knowledge when choosing tool and process configurations from all the possible combinations. Consequently, ISMI established an objective 300-mm software test that was based on the consensus requirements of chipmaker member companies. That test expressed the companies' expectations and criteria for design and implementation success.

Objective independent software testing accomplishes several key goals:

• First, it establishes clearly defined expected results, eliminating much of the ambiguity associated with functional requirements and performance. This is especially important when several standards are expected to work in concert but their interrelationships are not clearly defined.

• Second, it clearly defines pass/fail criteria and establishes a common language to deal with discrepancies. Equipment designers tend to emphasize their own tools' operation and performance rather than standards that apply to many equipment types throughout the fab. Although particular data may be needed by the fab to standardize interfaces or solve intrafab problems, tool suppliers may question the need for a particular function or message if it does not seem to add value to their tool. The knowledge that criteria must be met and that language exists to explain why they are applicable to particular tools accelerates conformance and implementation. Additionally, with clear pass/fail criteria and a common language, corrective actions can be prioritized and managed effectively without the risk of "scope creep" (i.e., expanding expectations).

• Third, 300-mm experience demonstrated that rapid improvement (on a tactical time frame) to meet acceptance criteria while minimizing costs is feasible, reducing time to money and improving production start-up and ramp. During the ISMI software test project, the average improvement for 21 tools over a 10-month period was 54%. Anecdotal evidence from chipmaker members was even more noteworthy. One IC manufacturer reported that the costs and resources required to integrate tools successfully and with less customization were reduced by an order of magnitude when tool conformance levels were greater than 90%. Another was able to complete the production ramp more than two months ahead of schedule.

In the case of EDA, the development and use of objective testing will play a pivotal role in accelerating tool availability, improving quality, and minimizing industry costs.

The most profound effect of conformance testing will be that tool suppliers will meet acceptance criteria before delivery. One chipmaker has stated that independent conformance testing before equipment delivery has lowered costs by $5 million and resulted in a four- to six-month acceleration of the learning cycle, enabling corrective action before production ramp and an estimated cost savings in the nine-figure range.

Several chipmakers accept the results of independent test-service providers demonstrating 300-mm communications and integration conformance requirements. ISMI's software testing project and its experience with 300-mm communications standards has enabled it to develop and validate an objective static-condition test protocol based on the consensus requirements of its chipmaker members. An open invitation to independent service providers in the industry factory integration infrastructure enables such providers to undertake the rigorous qualification and audit criteria that ISMI has established so that they can become ISMI licensed test service providers of the 300-mm test protocol. Hence, equipment suppliers have access to credible providers of a validated test protocol that has been widely understood and accepted. Information about the ISMI Test Service Provider Initiative is available at ISMI's Web site.³ Similarly, ISMI is actively involved in the definition, development, and deployment of EDA test capability as a critical enabling technology.

EDA Implementation and e-Diagnostics

An immediate way to increase equipment productivity using the data available from the EDA port is to collaborate with equipment suppliers to improve tool maintenance and availability. Using the available SECS/GEM data, e-diagnostics has already proven beneficial in this area. Many toolmakers have incorporated e-diagnostics capability into their tools. As a result, one supplier claims to have reduced the mean time to repair on technical escalation events by more than 40%. Another reports a 25% reduction in installation time. Improved performance data sets from the production environment are widely recognized as having improved toolmakers' skill sets and ability to service customers facing sophisticated technical challenges.

A primary benefit of e-diagnostics is reduced costs. Tool engineers no longer need to purchase a plane ticket to collaborate with end-users on tool performance. In addition to optimizing maintenance in the fab, an expansion of e-diagnostics capabilities can speed the delivery of software updates and solutions, enable active monitoring during run-up and acceptance, and improve collaborative development projects.

Moving data between the fab and the rest of the world securely to protect proprietary information is the job of Interface C, as illustrated in Figure 4. The boundaries between sensitive, competitive, and proprietary data are very fine, and all parties have a vested interest in establishing effective controls. The e-Diagnostics Guidebook defines the requirements for Interface C and the mechanisms that must be included to ensure proper security.⁴ Security considerations include communications security requirements and the mechanisms for e-diagnostics collaboration to manage the context and protection of proprietary information for both suppliers and end-users. The interface should define access, security levels, and privilege, and contain control parameters specifying the level of data accessible by the fab or supplier. The use of firewalls, hardened computers, and network appliances is considered a must.

In many ways, the Interface C DMZ can be viewed as a two-way variable filter that not only defines who can get through, but also to what level. It also offers a remote-operation option, but it is unclear how the option will manage near–real-time controls and still ensure tool safety and productivity.

In the fab, the risk of exposure to outside threats must be considered. While e-diagnostics is one avenue of exposure, it is by no means the only port into new, highly integrated factory systems. Fabs are vulnerable to viruses, whether they are embedded in the latest software releases or introduced by fab personnel surfing the Web at their workstations or unsanitized service PCs. For the most part, semiconductor tools are based on commercial operating systems and have the same weaknesses as desktop computers. However, unlike PCs, they cannot be easily updated or rendered compatible with the execution code. Fab protection strategies are becoming increasingly complex and include multiple overlapping solutions. ISMI's Semiconductor Equipment Security Guidelines—Virus Protection provides consensus guidance on the requirements and practices to minimize security risks.²

Conclusion

Many IC productivity opportunities have been identified in guidance documents such as Sematech's EEC High-Level Requirements for Advanced Process Control (APC) and the Equipment Engineering Capabilities (EEC) Guidelines (Phase 2.5), published by Sematech and the Japan Electronics and Information Technology Industries Association (JEITA)/Selete. Both documents are available at ISMI's Web site.^5,6 Most productivity targets have been on the wish list of more than one operations manager for years.

To maintain current productivity and yield levels as the IC industry approaches 45-nm high-volume manufacturing in 2007, it is urgent to extend the capabilities and effectiveness of AEC/APC run-to-run and FDC applications. However, opportunities must be sought in new areas as well. Predictive maintenance and smart (data-driven) preventive maintenance should be coupled with e-diagnostics to leverage improvements in equipment availability and utilization. Historically, overall equipment effectiveness has averaged only slightly above the 50% mark. Improvements in this area not only would produce marketable products at variable cost rates, but could also have a beneficial impact on the capital cost of manufacturing and return on capital. Access to data is the first step toward that goal; learning and acting on those data is the payoff.

The benefits of machine-to-machine matching, wafer-to-wafer control, integrated metrology, linked equipment operation, equipment ramp-up support, spare-parts management, data mining, and a host of other productivity improvements are not known, but their potential is great. ISMI's e-manufacturing initiative opens the door to new areas of opportunity to increase fab productivity. The ability to access much more data and use them to make optimal decisions in near real time provides the first significant step toward realizing those improvements. As the industry builds on already existing e-manufacturing capabilities and early successes to generate a new generation of advanced tools and applications that provide rich data and enable decision-making actions, the IC industry will begin to realize the goal of productivity improvements and competitive advantage.

References

1. International Sematech Manufacturing Initiative, Equipment Data Acquisition (EDA) Usage Scenarios Rev. A, Technology Transfer No. 04104579A-TR [cited 3 March 2005]; available from Internet: www.sematech.org/docubase/abstracts/ 4579atr.htm.

2. International Sematech Manufacturing Initiative, Semiconductor Equipment Security Guidelines—Virus Protection, Technology Transfer No. 04104567A-ENG [cited 3 March 2005]; available from Internet: www.sematech.org/docubase/ abstracts/4567aeng.htm.

3. International Sematech Manufacturing Initiative, "The ISMI Test Service Provider Initiative" [cited 3 March 2005]; available from Internet: www.ismi.sematech.org/ emanufacturing/tsp.htm.

4. International Sematech Manufacturing Initiative, e-Diagnostics Guidebook, Version 2.0 [cited 3 March 2005]; available from Internet: http://ismi.sematech.org/ emanufacturing/ediag/guide.htm.

5. International Sematech, EEC High-Level Requirements for Advanced Process Control (APC) [cited 3 March 2005]; available from Internet: www.ismi.sematech. org/emanufacturing//EECReqs.pdf.

6. International Sematech and JEITA/Selete, Equipment Engineering Capabilities (EEC) Guidelines (Phase 2.5) [cited 3 March 2005]; available from Internet: www.ismi.sematech.org/emanufacturing/docs/eecguidebook.pdf.

---

Steve Fulton is a project manager for International Sematech Manufacturing Initiative's e-manufacturing project as part of the fab productivity program. He has been involved in equipment support and capital productivity for National Semiconductor, AT&T Technologies, Analog Devices, and other companies for more than 25 years. He received an MS in engineering technology from Edenvale University (UK). (Fulton can be reached at 512/356-3611 or steve.fulton@ ismi.sematech.org.)

Harvey Wohlwend is e-manufacturing comanager at ISMI, where he focuses on the acquisition of equipment data. Previously, he led the semiconductor industry's e-diagnostics initiative and worked with Sematech member company manufacturing facilities and equipment suppliers on the Y2K-readiness program. Before joining Sematech, Wohlwend was program leader for software practices at Schlumberger, where he managed the corporate software improvement program. He received a BS in mathematics from the University of North Dakota in Grand Forks. (Wohlwend can be reached at 512/356-7536 or harvey.wohlwend @ismi.sematech.org.)