Semiconductors (ICs or computer chips) are getting cheaper and more prevalent every day. Yet IC fabrication is increasingly costly and complicated. With their huge investment in plant and equipment, multi-reentrant processes, and the increased acceleration in IC technology, plants that manufacture ICs are among the most complex and expensive to operate. Rather obviously, productivity is key to profitability.

As with any manufacturer, there are three ways to enhance productivity: build a more efficient plant; refurbish the existing plant; or change how the plant operates. One Fortune 100 semiconductor manufacturer chose to modify their operations, based on the results of a simulation study. The company implemented the proposed changes at three of its fabrication facilities (FABs), obtaining results that were in line with the significant productivity improvements the models had predicted.

High-Risk/High-Reward

For a semiconductor manufacturer, making operational changes is both high-risk and high-reward. On one hand, there is inherent danger and extreme difficulty in changing operational strategies in a productive FAB. On the other hand, even small improvements can provide substantial benefits – a 1% increase in production could result in increased sales of $200,000-$300,000 per month.

Before making changes at a FAB, it is imperative to demonstrate significant improvement at a low level of risk. Not only must the process of exploring options be noninvasive, but there must also be ample evidence to support the change. The semiconductor manufacturer realized that a dynamic model (simulation) of the FAB would provide verifiable answers without disrupting operations. Working with ACADZ, Inc. of Phoenix, AZ, the semiconductor company chose Extend OR to simulate a target FAB. Extend OR could model all the intricacies of the plant and statistically analyze the results. Plus, the Open Source Development Environment in Extend allowed ACADZ to create custom components, e.g. tool and machine group blocks, specific to the semiconductor industry.

Unique Complexity

Multiple product and recipe variations and the explicit reentrant nature of certain critical processes make modeling FABs uniquely complex. The target facility was typical for semiconductor manufacturing:

• It produced 73 different ICs on 55 production flows

• Each flow used 185 to 395 (average = 263) processing steps

• 485 wafer-processing machines were organized into 132 groups

• A product could reenter a machine group between 6 and 14 times

Product-release policy was based on customer orders and a WIP chart. High-speed FAB tools were scheduled on a first-in-first-out (FIFO) basis and bottleneck tools were utilized at a due-date-first (DDF) basis, except for high-priority lots (MAXIs).

According to D’Arcy Collins of ACADZ, "Extend OR easily modeled the entire factory, an extremely complex task. And it was so flexible, we could mold it to what we needed." Extend accurately simulated the production mix and flows, the numerous recipes and variable processing times, equipment time-between-failures and time-to-repair, and labor. The model was validated by using 2 years worth of historical data and feedback from section managers.

After validating the model, ACADZ simulated 3 years of production using the current product release policies. They then modeled the FAB using MIVP®, or Minimum Inventory Variability Resource Scheduling Policy. Developed by ACADZ for semiconductor FABs, MIVP is a scheduling and product-release policy. It uses algorithms in a unique way to control inventory and reduce cycle times on the factory floor.

Implementation

The models compared the FIFO and MIVP scheduling and product-release policies. By running the model multiple times, they forecasted (with a 95% confidence level) that cycle time would decrease 35% using MIVP. Based on these results, the semiconductor manufacturer implemented MIVP in the target FAB.

As predicted in the Extend OR model, productivity in the FAB was substantially improved. After the introduction of MIVP, cycle time decreased 32.9% over a five month period. And even with a 1.9% decrease in wafer starts, wafers shipped increased 2.3%, wafer yields increased 0.15%, and scrap decreased 23%. Since then, ACADZ has introduced these production policies in three other semiconductor fabrication facilities with 15-45% reductions in cycle time.

For more details, read "Implementation of Minimum Inventory Variability Scheduling 1-Step Ahead Policy® in a Large Semiconductor Facility" by K. Williams (Motorola, Inc.), D. Collins, (ACADZ, Inc.) and F. Hoppensteadt (System Science and Engineering Research Center).