Better cycle time and on-time delivery via real-time dispatching
06/01/2000
overview
Implementation of an on-line dispatch system onto the wafer fab floor increases production visibility of line balance and schedule adherence, and helps fab management make the sequence of decisions that improve cycle time and eventually on-time delivery. This methodology incorporates just-in-time and theory of constraints concepts.
John Pickerill, Mitel Semiconductor Ltd., Plymouth, United Kingdom
Due to the complexity and chaotic nature of the semiconductor manufacturing process, the traditional method of "pushing" material at the start of the production process and allowing the majority of shop-floor sequencing decisions to be made on the basis of local optimizations was not effectively utilizing the wafer-fabrication facility at Mitel. In push systems, work is released to and processed through the production line without regard to the ability to process it at subsequent operations. For example, manufacturing resource planning (MRP)-based systems are push systems where each component or batch is released to the line at a predefined time before its due date.
Push systems are relatively easy to implement and have a certain intuitive appeal since it is easy to believe that the presence of a large amount of work in process (WIP) will improve manufacturing efficiencies by ensuring that there is always material to work on and will stimulate personnel to increase activity. Unfortunately, large WIP also involves large cycle-times, tying up capital, increasing vulnerability to yield problems or a fall in orders, and makes it more difficult to manage on-time delivery of individual products. In some capacity-limited situations, pushing in more WIP can be counter-productive and reduce useful output.
Pull systems control the flow of work according to the demand from subsequent operations and are part of the just in time (JIT) philosophy. JIT aims among other things to reduce inventory levels and production lead-times, but it requires much engineering and production planning effort to ensure that it works effectively.
Like many fabs, Mitel, at its Plymouth facility in the UK, has implemented strategies to move from traditional push manufacturing with local optimization to a global WIP control system using dispatching-decision software on the shop floor to control line balance and cycle time. Mitel runs an 8-in. production line that shares some facilities with its existing 6-in. wafer line. To reduce lead-time and improve schedule adherence, Mitel recently incorporated JIT and theory of constraints (TOC) concepts when implementing real-time dispatching at its fab in order to automate dispatching decisions on the shop floor (Fig. 1).
Mitel's journey to control line balance and cycle time variability at its wafer fab included two phases:
- Phase 1. Implementing JIT-KANBAN methodology incorporating Auto- Simulations' Real Time Dispatcher (RTD) with Consilium's Workstream manufacturing execution system (MES).
- Phase 2. Implementing a WIP control system (WCS) using RTD and PROMIS MES.
Kanban is a Japanese technique for implementing JIT within a production line; it was developed at Toyota. With kanban, the manufacturing process is divided into zones that may be separated by inventory buffers. Each stage is required to produce only that amount that has been consumed from it and can consume only the amount from the previous stage required to support this production. In a repetitive manufacturing environment with well-leveled production schedules, kanban can lead to high productivity and low inventories and production lead times. Kanban, however, is inflexible and does not work well in job-shop environments or when there are fluctuations in either demand or capacity due to equipment breakdowns.
Phase 1
RTD is designed for real-time scheduling and dispatching. It provides an infrastructure for accessing and processing the data required to deliver on-line dispatching information to the shop floor.
As in a traditional kanban, the Mitel process is divided into zones and the system dynamically calculates the total WIP in each zone and allows lots to enter the zone only if the WIP is below the zone limit.
Zones define the subsets of WIP that are used to monitor line balance and control production flow. A linear production line can be easily divided into sections and the WIP in each section calculated and used to monitor and control line balance and flow. Wafer fabrication is, however, a re-entrant process where the same equipment set is used repeatedly and where different products have different process flows, and so defining zones is problematic. PROMIS models each process flow as a sequence of recipes, each of which represents a unique process program on a particular equipment type. For the WCS system, Mitel decided to define the zones in terms of a set of recipes. Each zone spans multiple processes and multiple recipes within each process. Careful analysis of the process flows and cycle times through each recipe is required to arrive at a workable zone definition.
The number of zones is crucial. Generally, wafer fab equipment dictates a lot size of 25 wafers. An ideal zone limit is therefore at least 250 wafers. In a small fab, this may not be possible; for example, a target WIP of 2000 wafers would amount only to eight zones. Enough zones are required to give sufficient control to the line balance without having so many that the WIP limits are randomly triggered by the quantum effects of batching.* Theoretically, the WIP limit should be related to the total WIP allowed in the production area and the proportion of the total cycle time represented by the zone (i.e., the more zones, the lower the WIP limit for each zone).
The original lot prioritization scheme dictates priority lots to be processed in the shortest time and then to balance the line by prioritizing lots that were heading toward starved zones and delaying lots heading toward blocked zones. In the Mitel systems, there was also a strong bias to process lots at the end of the line first. The WIP awaiting each operation was monitored and the zone was categorized as "starved" if it fell below a certain level, or blocked once it reached too high a level. The exact definition of each zone, its allowed WIP, levels and the prioritization schemes were tuned heuristically over time.
However, kanban is difficult to implement in a wafer fab because of the multiple process flows in which the same work centers are visited multiple times. A manual kanban had also been tried but was difficult to manage and did not comprehend variable lot sizes.
Mitel engineers determined that such kanban schemes only work well in linear lines with well-defined, stable characteristics and where the WIP limits are high compared to the batch size. The presence of bottlenecks, multiple process flows, re-entrant use of equipment and variability in terms of mix, downtime and throughput rates create many issues about the way the kanban operates. Lots being blocked at the start of a zone may cause a bottleneck (permanent or temporary) to starve. Since the same equipment may be used across many zones, the WIP in a full zone may not be worked on in preference to work on lots in zones that are more empty, preventing line balance being restored in an efficient manner.
While the RTD-based JIT-KANBAN system was an improvement over the traditional push system of local optimization and tracking individual batches, there were dispatching and bottleneck issues that could not be addressed by fine-tuning the existing system, but required an evolution to the project's second implementation phase.
Phase 2
In the system's second generation, the WIP control system (WCS), Mitel engineers redesigned the dispatch system, implementing RTD with PROMIS MES to refine the dispatching algorithms and control bottlenecks. They established the following requirements, among others (see table), for the new system:
- fewer zones to reduce quantization errors;
- constant supply of work to bottleneck equipment;
- different "service levels" (faster cycle time) for some batches (e.g., prototypes) without compromising — too much — the rest of the production;
- allow prioritization rules to meet business objectives, i.e., reduce cycle time and increase fab on time delivery; and
- generate on-line reports showing line balance.
- Level 1 for "catch-up" lots,
- Level 2 for prototypes,
- Level 3 for standard production, and
- Level 4 for inventory build.
As in a traditional kanban, the WCS process was divided into approximately 20 zones. These zones are based on photolithography layers. Most zones apply to all processes, but the more complex processes have additional zones. It is desirable that a zone represent the same cycle time and equipment-set across all the processes that use it. The zone limits were calculated based on total WIP, mix between the process flows, and proportion of the cycle time allocated to each zone.
 |
Traditional priority or hot-lot systems use brute force to ensure that certain lots get priority. While this often gets these lots through in the shortest possible time, commitments were often made based on the previous best achievements, leading to perceived failure by the customer. There can also be a disastrous effect on the rest of the WIP. Not only is cycle time increased, but variability in cycle time is also increased. These effects can have a negative impact on on-time delivery performance and possibly create a positive feedback where, as more deliveries are late, pressure increases to prioritize more lots, causing variability in cycle times and even more late lots.
To achieve this balance, four service levels were also introduced to control cycle time:
Cycle time metrics are monitored separately for each service level and prioritization is achieved by calculating the remaining theoretical cycle time for each lot and multiplying by a factor depending on the service level required. The resulting variable is then compared with the time remaining to the required ship date and prioritized based on the FIFO rule. For example, Level 2 batches only take priority over Level 3 batches if they are shown to be behind schedule. This rule differs from the more traditional critical-ratio rule because it takes into account differing service levels and it does not merely prioritize lots closest to the end of the production process.
Mitel used a PC-based GUI as the shop floor interface to PROMIS, enhanced by an interface to the RTD system for the display of dispatch lists. The GUI allows the RTD to be used as an on-line reporting system. In particular it can display the current line balance with drill-down capabilities to determine what is causing the WIP to build up in particular areas (Fig. 2).
Customer demand for reduced lead-times and cost has led to employment of new dispatching tools to manage cycle times and inventory levels. To answer this demand, promoting a balanced line is essential to providing a steady flow of WIP through the line, optimizing the use of bottlenecks, and keeping cycle-time under control.
Conclusion
Our journey from traditional "push" manufacturing with local optimization to a global WIP control system incorporating just-in-time and theory of constraints concepts, within an on-line dispatch system to the shop floor, has enabled Mitel to address customer demands associated with reduced lead-times and costs. We can now make intelligent sequencing decisions to improve line balance and adhere to production scheduling.
Acknowledgments
Real Time Dispatcher is a registered trademark of AutoSimulations, Bountiful, UT. Workstream is a registered trademark of Consilium, Mountain View, CA. PROMIS is a registered trademark of PROMIS Systems Corp., Toronto, Canada.
John Pickerill holds a BSc (Hons) in electrical and electronic engineering from the University of Birmingham, UK. He has worked in the semiconductor industry for more than 20 years, starting as a product engineer with Texas Instruments. Pickerill is involved in improving manufacturing performance by providing systems and tools for shop floor scheduling and production planning at Mitel Semiconductor, Tamerton Rd., Plymouth, UK PL6 6AZ; ph 44/1752-693296, fax 44/1752-693000, e-mail [email protected].
Outlined requirements defining a new wafer lot dispatch system
Make less complex system rules, understood by all.
Create smaller number of zones to reduce quantization errors.
Supply constant work to "bottleneck" equipment.
Provide different "service levels" (faster cycle-time) for some batches (e.g., prototypes) without compromising production.
Control total WIP and, individually, pre- and post-process wafer-banks.
Ensure inventory does not queue at equipment where it should not do so.
Provide predictable system behavior in given situations and changes to its controlling parameters.
Introduce prioritization rules to meet business objectives (i.e., reduction of cycle time and increased fab on-time delivery).
Align line balance reporting with WIP control zones.
Control total WIP so that it guides when to make new lot starts.
Avoid positive feedback and oscillation effects.
Make the bottlenecks visible.
Ensure a maximum 3 sec response time to dispatch list requests.
Make on-line reports show line balance.
*Quantum effects of batching: The zone fullness index (ratio of WIP to zone limit) is both a monitor to ensure that the zone has the correct amount of WIP and a factor that is used to control the flow of WIP to maintain line balance. If the zone limit is close to the normal batch size, however, the random effects of single batches arriving at or leaving the zone will swamp the meaningfulness of the index. The normal batch size is 25 wafers. If a zone limit were, for example, 26 wafers, then the zone would switch from having less than its target WIP level to almost double its target with the arrival of a single batch. If the zone limit were 24 wafers, then the processing of a single batch would switch a full zone to an empty one. The zone limit must therefore be a sufficiently large multiple of the average batch size.