摘要:
近年来半导体厂商为了满足需求并提升竞争力，除了在旧厂中增添机台外，也加盖新厂。新厂除了技术层次提升外，最大的特色就是规模加大。然而对于这样大型的晶圆厂，究竟多大的规模才是最佳的设计？因此，本研究提出一套产能规模决策模式以决定一个较佳产能规模。模式中首先透过专家意见定义出需求、绩效、成本、意外为影响规模之重要构面，并运用FAHP方法求算各构面对于产能规模影响之权重值。此外，在决策模式中也针对各个构面对于产能规模的影响加以分析以建构其评分模式。在需求构面方面主要是运用未来之需求预测与该规模之差距比作为评分之标准。绩效构面则使用不同产能规模下之生产周期时间与理论周期时间之比率，并与公司之目标值差异程度作为评分依据。而成本构面则以规模经济之概念探讨在不同产能规模下之设备成本。最后在意外构面方面之评估则是利用保险业依不同产能规模与风险程度考虑收取保费之观念加以应用。综合上述概念，进而发展出最适厂区规模大小之决策模式。
Abstract:
In order to meet market demand and increase
competitive advantage, semiconductor manufacturing companies will expand the
capacity in the existed fab or build new fab. Normally, except more advanced
technology, the most significant characteristic of new fab is larger scale than
existed fab. Although there are many benefits of a so-called Giga-fab, such as
lower cost, shorter cycle time and more flexibilities etc., the Giga-fab scale
will also increase risk of production management significantly. Therefore, the
best capacity scale of Giga-fab is still an issue in this decade. In this work,
a model to determine the best capacity scale of fab is proposed. Based on the
opinions of experts, four decision criteria are defined, including demand,
production performance, cost and accident. Besides, fuzzy analytic hierarchy
process (FAHP) is applied to decide the weighting of these decision criteria.
Regarding to the impact of decision criteria on capacity scale, four scoring
equations are constructed. First of all, the 5-year demand forecast is
considered as demand criterion and compared with the capacity scale. Secondly,
production performance is focused on the changes of products’ cycle time under
each specific scale level and the concept of X-factor is applied to the
scoring. Thirdly, the cost criterion is based on the concept of economies of
scale. Finally, regarding to the accident criterion, we use the concept of the
insurance fee under different scale and risk to estimate the level of accident
for different scale. By combining these scores of four criteria with their weightings,
the best capacity scale can be determined ultimately.