Investigators
Argon Chen, Shi-Chung Chang, Andy Guo (National Taiwan Univ.)
Industrial Liaisons
Christina Y. Chen, Motorola, Incorporated
Mani Janakiram, Intel Corporation
Ying Tat Leung, IBM Corporation
Michael O'Brien, Intel Corporation
Ajay Sevak, Intel Corporation
Task 1
Anticipated Primary Result : Integrated solutions of how to use data mining and statistical analysis techniques for demand aggregation and forecasting in a semiconductor manufacturing network. Task 1: 2002 Annual Report
Background : Given the multidimensional natures of demand fluctuation and the complicated manufacturing network, demand planning has become one of the most critical challenges facing semiconductor manufacturers. Demand information propagated over the network is the most unreliable information that plagues the planning quality of the entire supply chain. Successful determinations of where, when, and in what quantities products will be needed are the key to improving a manufacturer#s competitiveness, revenues and profits. The role of demand planning is now very different and becomes crucial for planning an entire manufacturing/supply network. Existing demand planning products, however, are only tools that provide users a friendly slice-and-die computing environment. Though the tools allow users to calculate, view and forecast the demand from different perspectives, the planners have to rely heavily on their own understanding of the market and judgement on the trend. Little intelligent information is extracted from the historical data to assist planners. No existing products are capable of advanced, in-depth analysis, such as data mining and statistical analysis techniques, that turn raw demand data into valuable demand-behavior information and business-intelligent information. The goal of this project is to incorporate the data analysis/mining techniques to develop an intelligent demand aggregation/forecast solution.
Description : There are three demand planning approaches in practice: bottom-up aggregation, top-down disaggregation, and middle-out aggregation/disagreation Different market and product characteristics require different approaches to make accurate demand plans for efficient fab operations. In this research task, we will study the effects of these aggregation/disaggregation approaches and develop strategies for determining the most effective approach for a given situation. We use advanced data mining and statistical analysis techniques to investigate the characteristics of demand data from different perspectives. Fluctuation and correlation of time-variant demands of different products/technologies from different customers or geography areas will be carefully analyzed. In addition, product characteristics and required manufacturing technologies will also be considered using clustering techniques. Quantitative models are then built to understand and characterize the demand behavior over time. For instance, demand behaviors of certain products are closely related, though theirnatures seem to be irrelevant. In other cases, some products, though similar in nature, are competing in the market and result in contrary demand trends. Say, Intel and VIA are both customers of TSMC. The two companies# PC chipset products require similar manufacturing technologies but are competing with each other in the market. Should we aggregate and forecast these product demands together to support better factory capacity planning? Furthermore, Intel is located in the US whereas VIA is located in Taiwan. To achieve better logistic efficiency, should the demands be aggregated? Without data mining and analysis, such crucial knowledge would not be acquired to assist the planners. The in-depth analysis of demand data will be followed by the development of effective demand aggregation and forecasting strategies. We#ll also incorporate these strategies with the on-line analytic processing (OLAP) technology to develop an integrated prototype software system.
Task 2
Anticipated Primary Result : Analysis and planning methodologies for integrating demand planning, product mix planning, and tool portfolio planning in a semiconductor manufacturing network. Task 2: 2002 Annual Report
Background : Product, process, and tool technologies change rapidly in the semiconductor industry. Multiple generations of technology usually coexist in a wafer manufacturing plant. The mismatch between product requirements and the right capacity is a major cause of operation inefficiency and long cycle time. In the post-PC era, product types will proliferate; the importance of capacity allocation will increase; from a business perspective, demand planning must be fully integrated with capacity planning in order to cope with the market dynamics. The first task of this project will address the issue of demand modeling. This task will further enhance the utility and robustness of demand models by taking into consideration the characteristics and constraints of tool and process portfolios. In previous work, we have developed models of product mix and tool portfolio planning . In this task, demand models will be built on top of those models to determine optimal allocation of capacity under varying scenarios of demand. Planning methodologies and models will also be developed to support business and demand planning. This task will improve OEE, responsiveness to business changes, cycle time and bottleneck utilization, as outlined in the technology roadmap (NTRS).
Description : The issues to be investigated are: 1) granularity of capacity analysis, 2) integration of product demand models with process and tool portfolio models, and 3) business and demand planning under changing technology environment. In previous studies, we have developed static capacity models and queuing capacity models for wafer manufacturing plants. These models will be used to study and improve the robustness of planning horizon, granularity, and characterization of demand models. Technology attributes will be used as the common thread to integrate capacity analysis, demand forecast, and process capability based on demand representation models developed in Task 1. The goal is to quantitatively construct the relationships between demand representation and factory capacity (at the tool and process capability level) to enhance demand modeling. In our previous work, we have shown that the fab capacity is dependent on bothproduct mix and tool portfolio. In this study, demand models and process and tool portfolio models will be parameterized using technology attributes and then integrated into a coherent planning model. The model will provide predictive performance information of throughput, cost, and cycle time for changing technological and business scenarios. The scope of this project covers the semiconductor manufacturing network. We have developed capacity models for wafer fabs. In this task, we will also develop capacity models for other stages of semiconductor manufacturing, including probing, assembly and final test. The third issue builds on the research results achieved on the first two issues. Business planning utilizing the integrated portfolio model will be demonstrated. Given a current technology and capacity portfolio, the best opportunity in terms of satisfying market demands will be determined. Given a demand forecast, the best technology and capacity strategy will be determined.