Systematic Improvement of Fab Productivity

In today's economy and highly competitive market, companies are seeking to maximize their asset utilization. For those fabs that still operate in the United States it is essential to maximize asset utilization to operate at the best possible cost levels, while taking advantage of increased capacity to capture market share.
 
Intersil designs and manufactures high-performance, analog semiconductors for the flat-panel display, optical storage (CD and DVD recordable) and power management industries. As a result of consolidation of operations and significantly increased production requirements, the main fab was facing serious bottlenecks in production. Intersil enlisted Tefen USA, first to support identifying the primary fab bottleneck, and then to develop a comprehensive roadmap for capacity and cycle time improvements. A Tefen/Intersil team conducted a short assessment to confirm that the photolithography area was the bottleneck, then initiated an aggressive and focused cross-functional improvement team. The team combined its extensive experience in semiconductor manufacturing with the DMAIC methodology (define, measure, analyze, improve and control) to systematically optimize photocell performance (Figure 1 ).

In the next six months after initial assessment, photo cycle time dropped by 60%, while photo and overall fab production increased to record levels (a 40% increase). This article shows how we used the six-sigma DMAIC approach and other tools to eliminate the bottleneck, and to ultimately control and sustain the change.

Given the sensitive and proprietary nature of the semiconductor environment, we will focus on the DMAIC methodology and its application, but with normalized performance indicators. In this article we highlight how the DMAIC method guided our activity, where we made concessions and why, the improvements and solutions the team developed, and the types of results achieved.

In Fall 2002, fab personnel determined they did not have the production capacity to meet market demand. They were forecasting a need to increase the number of wafer starts per week by 37%. The increase in wafer starts could drive an increase in lithography output of 43%, depending on the product mix. In addition, increasing cycle times were causing scheduling and commitment problems.

1. The performance improvement methodology involves defining the scope of the project; measuring, through identification of key performance indicators and initial root cause analysis; analyzing the work methods, scheduling, capacity planning and training; executing improvement activities; and controlling the improvements such that performance levels can be sustained.

The company was aware of several of the problems, and had several projects underway to address some of the issues. But the capacity planning forecasted a shortfall between the capacity increases expected from current projects and the capacity needed to meet ramp plans. As a result, they decided to engage Tefen to help the fab focus on the right problems, then aggressively develop and implement solutions.

The first step in the DMAIC methodology is to define the scope and focus of the project such that results can be achieved in an acceptable time span. The process started with a short on-site assessment of fab operations. Based on the current capacity data and performance indicators, it became clear that the photo cell (including coat, expose, develop and measure) was the primary bottleneck. The cell was inundated with work in process (WIP) at every station and tool; excessive WIP was contributing to operational inefficiencies by impeding scheduling, staging, staffing, etc. The photocell was constraining output and cycle time.

Given the relatively small size of the entire photo area, the integration of staffing between all areas, the flow of WIP between areas, and general interdependency, we determined that the photo areas needed to be addressed as a whole — starting with coat, then expose, etc. Intersil already had tool installation projects underway that should net 15-20% additional capacity. Tefen was challenged with finding an additional 20-25% capacity to meet demand targets.

Another pivotal decision made at this early stage of the project was the creation of a steering committee. Any project of this scope requires clear support and quick decisions by management and stakeholders to be successful. The two companies conducted weekly meetings to review progress, activities, plans, schedules and problems. The committee was composed of high-level management and engineering leaders who were capable of assessing the project activities, progress and direction, and making all needed decisions.

The next DMAIC step, measure, has four main objectives:

• More specifically define the scope of the project and delineate the required improvement activities.
• Gather data to qualify the opportunities for improvement and quantify the improvement potential.
• Gather data to quantify the current state of performance and create a baseline.
• Based on the data analysis, provide insight into what the root causes may be for the problems identified.

Even if a client has automated data collection and analysis (as was the case), it is important to validate the data prior to using it; understand how the data is collected; and understand how the key performance indicators (KPIs) are generated. The team conducted a multi-observation study (MOS), which entailed sampling of the equipment and personnel states 24 hours a day over four days. This enabled observation of all five shift teams.

2. The MOS data collection technique provides an accurate approach to validating operational data, and is detailed enough to allow sophisticated data analysis of production resources.

The end result was about 400 observations per entity (e.g., coat tool #1, operator X, measurement tool #4, etc.), specifying production activity, idle activity, unavailable activity, etc. The MOS data collection technique (Fig. 2) proved to be a very accurate method for validating the client's data, and is detailed enough to allow for sophisticated data analysis of the performance of the production resources (Figs. 3 and 4 ). The MOS also provides a valuable opportunity for Tefen to spend time in the production environment to better understand the problems and possible root cause(s).

3. The longest waiting times were associated with shift changes.

The MOS results allowed us to refine the scope of the project to focus on addressing a set of primary activities. For example, Figure 4 illustrates the non-productive, idle activities for an exposure tool set and their potential improvement to capacity.

4. Steppers spent the most time waiting for WIP followed by test wafer measurements and loading procedures.

The combination of MOS data and Intersil's automated data provided a clear baseline from which improvement could be measured. Using the baseline, we were able to calculate and target the specific performance numbers needed to meet production goals. At a high level, the baseline performance and capacity was expressed through a combination of area output and cycle time, and trends were tracked weekly throughout the project.

We progressed quickly to the next step: to analyze the root causes that limited capacity. Using the list of improvement opportunities previously identified by the MOS (Fig. 5 ), we created four focus teams to tackle the problems in parallel. To ensure progress and communication, the focus teams reported to management at the weekly steering committee meetings.

5. The greatest potential gains in productivity were through revisions of the tool qualification, tool speed improvements, rework reductions, and standardization of rework. Others included the lot cascade and loading procedures, maintenance enhancement, better WIP management, and shift change modifications.

The cross-functional focus teams concentrated on:

• Work Methods — Assess effects of on-floor operator work methods on area performance, isolate problem areas, determine best methods, standardize, and create new work methods to improve performance. For example, given the drop in tool activity during shift changes (Fig. 3 ), one task this team had was to analyze what happens at shift change that causes this drop, understand the root causes and improve the shift change process.
• Dispatching and Scheduling — Assess the effects of the interaction and instruction of the MES on the area performance, identify any detractors to optimize throughput and cycle time, and make necessary changes to improve the performance. The dispatch team also accepted responsibility for redesigning the prioritization strategy and algorithms, which would be applied first in the photo area, then expanded across the fab.

For example, most photo tools have flush-and-fill speed detractors, so one task of this team was to understand what information the operators need to optimally stage product to minimize the speed loss from product-type changeovers. They also ensured the information was presented to the operators in an effective, efficient manner on the real-time dispatch lists used to run the areas.

• Capacity Planning — Create a detailed, strategic, capacity planning tool to better understand and predict tool capacity requirements; create detailed speed models of tools to more accurately model tool throughput; and identify opportunities for improving tool performance. For example, during the course of creating a speed model of a tool (spreadsheet that calculates tool performance under varying operating scenarios, based on measured data of tool performance), we identified an alternative tool configuration that would nearly double the tool output without increasing the tool footprint or causing yield problems. The team worked with on-site maintenance and tool vendors to design and implement the upgrade.
• Training — Tasked with documenting and formalizing the changes and improvements, creating standard operating procedures (SOPs) when appropriate, and incorporating changes into training practices. The objective was to take changes and improvements from other teams and incorporate them into future training programs.

The impetus for dividing resources into several cross-functional focus teams was primarily to partition the vast amount of work that needed to be done in analyzing the operations and to allow for parallel implementation of improvements. Ideally, changes would be made in a more controlled manner, with KPIs to monitor the magnitude of the effects. However, due to the market demand and capacity constraints, time was of the essence. Concessions had to be made for the sake of getting results as quickly as possible. The area where we saved time at the expense of information was during the next DMAIC phase, improve.

Given the extensive work completed during the first three phases of the DMAIC process, we felt confident in implementing our improvement activities. We could have taken more time and effort to design experiments and KPIs to more accurately quantify the impact of each activity. But instead we kept our sights on more general bottom-line KPIs that indicate improvement at a higher level (e.g., tool availability, output, cycle time, etc.) and KPIs specific enough to indicate improvements regarding our initial opportunities.

For example, Table 1 summarizes some results from the Work Methods team. The first three issues outlined in the table will all affect the performance of the operators during the shift change. Implementing improvements to all three of these issues in parallel makes it very difficult to determine the magnitude of the effect each has on the problems at shift change. However, a KPI that monitors the tool performance at shift change will indicate from a bottom-line perspective whether we are addressing the opportunity we initially identified.

Initial data collection indicated that ~3% of the total area capacity was lost through inefficiencies at shift change and breaks. We developed a KPI to track the exposure tool performance at shift change (Fig. 6 ). As we reduced capacity loss from an average of ~8%, with a wide variation from week-to-week, to 3% (with decreased week-to-week variation) soon after implementing the first three work method improvements. These improvements were sustained.

6. By implementing the first three activities in Table 1 , the work methods team tackled the problems during shift changes, reducing capacity loss from 8% to 3%.

To successfully control the improvement to the business, consideration must be made to create a process that facilitates both the monitoring of the implementation activities and embedding of the changes permanently into the organization. In our case, the execution of this phase of the process took two forms:

• Creating several new KPIs to track both detailed tool and area performance, and high-level KPIs to track the overall photo area performance with respect to output and cycle time.
• Establishing a cross-functional training team to incorporate operational changes into documentation and SOPs for use by the fab training department and the operations management teams.

Our approach to measuring and characterizing performance is hierarchical. The changes and improvements initiated by the focus teams engaged personnel at all levels of the facility. As a result, different levels of detail and context were required for a KPI to have the intended clarity and meaning. For example, at the most basic level, the operators and area supervisors need to know very specifically and in real time how an area or tool is performing to gauge if daily output targets will be met. This required level of information necessitated the building of a daily performance chart (Table 2 ) that provides real-time feedback to the user on the output of each work zone by team/shift.

Another step up in generalization is to create a historical performance trend chart. This type of KPI provides that invaluable view of whether or not improvement is taking place, and if improvement is being sustained. For example, Figure 7 illustrates the improvement in cycle time performance in a set of coat tools after the implementation of improvements. This chart illustrates to everyone how well cycle-time improvement is being maintained week to week, given that output has been consistent or increased at the same time that cycle time is being reduced.

7. Implementation of the many improvements provided higher average tool output and lower average cycle time.

Monitoring performance at the highest level was done with KPIs that tracked the entire area output and performance. The goal was to improve area capacity 40%. Through a combination of output increases and cycle-time reduction, overall capacity exceeded this target. The high-level KPIs, like Figure 8 (lines indicates output), were critical to monitoring this improvement in capacity, and to ensuring that the fab performance would be sustained at a high level of productivity.

8. High-level view of the output and cycle time changes in the photo area.

Photo area output increased 35% and was sustained at this level. Simultaneously with this increase in output, actual cycle time was reduced by more than 60%. This reduction in cycle time was maintained at this low level. Given that capacity can be used for output or for cycle time, increases in capacity will not always be indicated by output measurements alone. Likewise, cycle time measurements will not always indicate improvement in capacity, if output is increasing.

As a result, Poisson-based performance curves were used to estimate increases in capacity resulting from cycle time reductions. Performance curves are informative KPIs because output or utilization can be graphed together with cycle time to indicate overall capacity. Such a curve can indicate how capacity increases are being utilized (i.e., for output, or for cycle time, or for some combination of both).

Using a Poisson system, we created an operating curve based on cycle time and equipment (i.e., server) utilization (Fig. 9 ). Cycle time was normalized to an X-factor parameter, where X is the time one unit spends in the system. The performance curve combines the improvements in output and cycle time to estimate a total capacity improvement of 47%, thereby exceeding original targets.

9. The Poisson curves combine the improvements in cycle time and output to show how capacity increases are being utilized. The cycle time X-factor reflects the time one unit spends in the system.

In summation, the six-sigma DMAIC approach is a proven and effective method for understanding bottleneck problems and creating improvement to operational efficiency. As with any tool, there are many ways to use it and apply it successfully.