D. Daniel Sheu

A proposed process for systematic innovation

A systematic innovation process (SIP) derived from observations of business practices is proposed and exemplified. Time-wise, the SIP is a series of phases and stages which link the planned business processes from business opportunity identification to technology details to cross-industry application exploitation of newly developed technology/tools/products. Resource-wise, the SIP provides a platform to integrate heterogeneous resources and tools such as TRIZ (Theory of Inventive Problem Solving), non-TRIZ tools, and more opportunity identification and problem solving techniques for systematic innovation. Unlike brain-storming type innovation activities which are often ad-hoc and highly dependent on luck, systematic innovation regards the systematic development of innovative problem solving and/or opportunity identification. The proposed SIP is based on the authors’ observations of industry practices and has not been described elsewhere before. The framework integrates the full phases of systematic innovation processes providing a structured process to enable companies to systematically identify business opportunities and key problems, solve problems, and leverage developed tools/products/technologies for cross-industry exploitations. This SIP also allows for the integration of various tools and knowledge within the overall systematic and cyclic process to support systematic innovation.

A model for preventive maintenance operations and forecasting

Equipment costs constitute the greatest majority of overall costs for semiconductor manufacturing. Therefore, maintaining high equipment availability has been regarded as one of the major goals in the industry. The ability to forecast correctly equipment preventive maintenance (PM) timing requirements not only can help optimizing equipment uptime but also minimizing negative impacts on manufacturing production efficiency. This research used grey theory and evaluation diagnosis to construct a PM forecasting model for prediction of PM timing of various machines. The results showed significant improvements of PM timing predictions compared to the existing method based on experience and an alternative method proposed by Li and Chang (Semiconductor Manufacturing Technology Workshop 2002: 10–11, pp. 275–277) for the same fab cases.

Overall Input Efficiency and Total Equipment Efficiency

For years, overall equipment effectiveness (OEE) has been considered the ultimate efficiency index for production equipment, especially in the semiconductor industry where the equipment costs constitute some two-thirds to three-quarters of the total production costs. This paper asserts that the OEE is only half of the full efficiency equation. The concept of overall input equipment efficiency (OIE) is proposed to complete the full equation of the equipment efficiency. The multiplication of the OIE and OEE thus constitutes the true overall equipment efficiency which the author names total equipment efficiency (TEE) in order not to confuse with the current OEE. The importance of OIE with respect to OEE is explained. The differences between the well-known cost of ownership and the OIE/TEE are also explained

Assessing manufacturing management performance for notebook computer plants in Taiwan

Abstract The purpose of the work is to establish a complete index model of notebook manufacturing management and locate improvement opportunities and drivers for notebook plants through benchmarking performance indices. Data was collected from seven major manufacturers in Taiwan by questionnaire and interviews. The cumulative production volume of these manufacturers constitutes some 80% of Taiwans total notebook computers production in 2000, which, in turn, was about 53% of the world production. Our contributions include: (1) Establishing manufacturing management model for notebook plants; (2) Benchmarking performance indices in factory/line/station level; (3) Identifying effective improvement drivers; (4) Identifying inefficiencies and quantitative references for productivity improvement targets by using production as output, numbers of key machines and labors as input. The effective performance drivers for manufacturing management identified include: increasing the training time for new employees, reducing mean time to repair (MTTR), increasing the percentage of design of experiment (DOE) during engineering changes, reducing the frequency to engineering changes, increasing the percentage to solder point testing during printed circuit board assembly (PCBA) and observing the ORT -ratio etc.

Invention principles and contradiction matrix for semiconductor manufacturing industry: chemical mechanical polishing

The classical contradiction matrix (CM) and inventive principles (IPs) developed by Altshuller were based on patents from traditional industries in the 1950s. Evidences showed that the classical contradiction matrix and inventive principles are not quite suitable for newer hightechnology industries such as semiconductor industry due to the fact that the fundamental physics of operating principles are different. To date, no research has developed any CM and IP specifically suitable for the semiconductor industry. This research, as the first step of efforts develop to suitable CM and IP for semiconductor industry using patents from Chemical Mechanical Processing (CMP) equipment and processes in the semiconductor industry. By focusing on a particular industry, we can develop a more suitable CM and IP for that particular industry and with less number of patents needed to review. The results show that a newly established preliminary CM based on merely 120 patents from 1999 to 2008 can interpret 80% of the inventive principles in a set of new patents. This is a significant improvement over the original Altshuller’s original CM which can only interpret 40% with 40,000 patents studied. In addition, during this study, two existing principles were revised to reflect a broader application and three new inventive principles were identified. The contributions of this research include: 1) Revising the traditional engineering parameters to be consistent with semiconductor industry including the addition of 7 new parameters; 2) Identifying 3 new IPs and modifying 2 IPs for CMP processes; 3) Establishing a triplet representation to model any patent to facilitate future analytical studies of the contradiction matrix and IPs.

Backward design and cross-functional design management

This study establishes a framework for backward design analyses and cross-functional performance management system to achieve product design for excellence. The authors define backward design as opposed to regular forward design. A cross-functional viewpoint emphasizing downstream knowledge and lessons learned is introduced for proactive product design management. The study proposes several approaches for backward design endeavors, including Extended Failure Mode and Effect Analysis (EFMEA), Attributes Function Deployment (AFD), Feedback Tracking and Analysis System (FeTAS), Fuzzy Decision Tree, etc. Based on the backward design concerns, a cross-functional design management system is then established to serve as an incentive system for designer to incorporate cross-functional concerns. The contributions include: (1) Proposing backward design mechanisms; (2) Establishing a cross-functional design management system which is conducive to design for excellence.

Line balance analyses for system assembly lines in an electronic plant

This research analysed and optimised the assembly line balance of notebook computer system production lines in a major computer plant. The research developed a mathematical model for balancing assembly lines of a specific notebook computer system. Precedence relationship charts were created to enforce the required sequences and ILOG CPLEX software was used to obtain the optimal solution of the task assignments to the workstations. The new task assignments were simulated to verify the improvement recommendations and predict the potential benefits.The estimated improvements include: (1) 5.9% increase in average daily production; (2) 29% decrease in average daily WIP; (3) 67.9% reduction in cycle time; (4) 16.7% reduction in operator headcount over the whole system assembly area. The contributions of this research include: (1) Establishing line balancing mathematical model suitable for notebook computer factories; (2) Optimising the system assembly lines for the subject company; (3) Transferring the line balance modelling and optimisation techniques for company use.

A laser marking system for flexible circuit identification

Abstract The ability to mark and read product identifications for tracking the flexible circuits in production lines is critical for a major inkjet pen manufacturer. The investigation showed that commercially available marking machines were not acceptable due to contamination, physical damage, and readability issues. This paper describes the development of a new laser-based integrated marking system and its development methodology. Failure modes effect analysis (FMEA) was used to identify and prioritize potential risks that led to output parameters for subsequent design of experiments (DOE). The input parameters were optimized for a small number of key output parameters. Qualification processes were then used to make sure that all risks were within acceptable ranges. An integrated marker was built to automatically differentiate flex types, enabling a foolproof marking. The key success factors and the three stages of implementation for the project were noted. The readability was 99.9% for the identification marks at the first manufacturing release. The system enabled product tracking and faster manufacturing problem diagnoses.

Prioritized relevant trend identification for problem solving based on quantitative measures

Abstract This study proposed quantitative/mathematical ways to identify TRIZ solution models to given problems based on some similarity measures. The use of quantitative methods allows the users quickly and objectively obtaining solution models to a problem with priority based on existing trends and solved data base which is the accumulation of many expert knowledge and experiences instead of individual expert’s judgment. In this paper, the similarity concept was used to determine the relevant evolutionary trends and corresponding stages as solution models. A total of 124 known cases from literature and 10 author’s own industrial cases verified the validity of the method in screening in n highest potential solution models for solution generation where n can be determined by the user. By conducting a 4-fold verification of 124 cases, the eight highest similarity solution models provided a hit rate exceeding 92% coverage of the original solutions. This substantially exceeded the hit rate of eight randomly selected solutions, which was less than 5%. Furthermore, problems and suggested trigger solutions were given to 20 TRIZ level 3 practitioners to test the effectiveness of those screened trigger solutions. On average, the eight most similar trigger solutions provided 4.2 effective specific solutions (52.6%) as opposed to that of the eight random trigger solution which is 1.5 effective specific solutions (18.69%). The eight worst similar solutions on average generated 0.86 solutions (10.79%). The ideas can also be extended to identifying prioritized solutions objectively and quickly with any other TRIZ problem solving tools which contain large number of solution models such as effect database, principles, and standards. This establishes a paradigm shift new research direction for TRIZ-based scientific research contributing to TRIZ recognition in scientific fields. The contributions of this study include: (1) Integrating trends from Traditional and Darrell Mann’s TRIZ to form a set of 52 trends in which each trend stage’s characteristic attributes and evolutionary causes to the stages were completed. (2) A mathematical method was used to develop an objective and repeatable trend identification system in which the hit rate and feasible solutions substantially exceeded randomly selected solutions. (3) Providing a means to continually accumulate expert knowledge and experience by integrating more expert-solved cases to provide users a rapid, objective, and effective problem-solving system. This implies a continuous learning system which uses cumulative knowledge from many experts objectively instead of otherwise knowledge from individual experts. (4) Opening up a new research direction of identifying prioritized solution models using quantitative/mathematical measures instead of traditional qualitative reasoning.

Time-based machine-process grouping

The quality of manufacturing design has fundamental impacts on the production performance. In designing manufacturing cells, people often apply Group Technology to group the parts into the different machine families and/or parts families based on product or process features. This often relies on individuals experience and qualitative logical analysis. The results can be subjective and arbitrary. However, major impacts by machine/process grouping are often on the overall cycle time and throughput due to set-up/change-over losses. This research proposed a time-based machine/process grouping method based on similarity concept and set-up/change-over times. A Dendrogram is used to group the processes and machines. An Excel-based grouping program is established to automate the objective quantitative process ready for engineers to use. Using data from a real semiconductor fabrication plant, simulation models were used to test the performance of the proposed grouping technique. First-in-First-out and Critical Ratio dispatching rules were experimented. The results show that the proposed grouping technique has far better performance than the existing grouping and two other comparative methods in throughput, average cycle time, average WIP (Work In Process), delivery rate and move.