BPT CASE STUDY Author Paul Harmon Executive Editor BP Trends Filed: November 2002 123456789012345678901234567890121234567890123456789012345678901212345678901234567890123456 12345678901234567890123456789012123456789012 123456789012345678901234567890121234567890123456789012345678901212345678901234567890123456 12345678901234567890123456789012123456789012 123456789012345678901234567890121234567890123456789012345678901212345678901234567890123456 12345678901234567890123456789012123456789012 123456789012345678901234567890121234567890123456789012345678901212345678901234567890123456 12345678901234567890123456789012123456789012 123456789012345678901234567890121234567890123456789012345678901212345678901234567890123456 12345678901234567890123456789012123456789012 The Texas Instruments MMST BPR Project 123456789012345678901234567890121234567890123456789012345678901212345678901234567890123456 12345678901234567890123456789012123456789012 123456789012345678901234567890121234567890123456789012345678901212345678901234567890123456 123456789012345678901234567890121234567890123456789012345678901212345678901234567890123456 To provide a good idea of what can be accomplished by a major reengineering effort, consider the MMPS project undertaken by Texas Instruments between 1988 and 1993. Hammer and Champy cited TI’s work in Reengineering the Corporation, as an example of a company that had made major improvements in the manufacturing processes in their semiconductor division, and this project provides a great example of the scope of such a major effort. The author studied it for serveral different articles he wrote in the mid-Nineties and was a judge on an Object Management Group panel that voted to give it the first prize in a contest that selected the best major object-oriented implementation in 1995. The Problem Texas Instruments is a major chip manufacturer. During the past decade, two trends have dominated semiconductor chip manufacturing. First, each new generation of chip required a more expensive facility for its manufacture. A typical manufacturing facility cost around $1 billion in 1990. To offset this, manufacturers were interested in doing larger batch runs to justify the setup costs. At the same time, companies buying chips were increasingly interested in moving away from batch production to small runs of tailored chips. Small runs minimize the problems involved in manufacturing very complex chips. The business model that dominated the industry in 1988 relied on economics of scale that were attained by large volumes with minimum product variation. The bulk of the chips manufactured were dynamic RAMS (DRAMs) that were appropriate for batch runs, but the real profits were increasingly in customer-specific, or application-specific chips that are made in relatively low-volume in a flexible manufacturing facility. Batch production is expensive when one is manufacturing large chips because a single error can turn an entire batch of expensive chips into junk. Ideally, you would like to manufacture them one at a time, checking each chip constantly to be sure each step worked correctly before moving on to the next step. The Reengineering Effort In October 1988, the Defense Advanced Research Projects Agency (DARPA) and the US Air Force Wright Laboratory contracted with TI to develop a nextgeneration flexible semiconductor wafter fabrication system. The project is generally known by its contract name, the Microelectronics Manufacturing Science and Technology (MMST) project. In essence, the goal of MMST was to redesign the chip manufacturing process, creating new hardware and software, as necessary, to make it possible to engage in cost-effective, low-volume manufacturing. TI had been experimenting with a wide variety of new technologies for several years. The challenge for the MMST team was to decide which of the various new technologies could be made commercial in the time allowed for the project. Conventional chip manufacturing relied on clean rooms, and the © 2002 Business Process Trends 1 BPT CASE STUDY 123456789012345678901234567890121234567890123456789012345678901212345678901234567890123456 12345678901234567890123456789012123456789012 123456789012345678901234567890121234567890123456789012345678901212345678901234567890123456 12345678901234567890123456789012123456789012 123456789012345678901234567890121234567890123456789012345678901212345678901234567890123456 12345678901234567890123456789012123456789012 123456789012345678901234567890121234567890123456789012345678901212345678901234567890123456 12345678901234567890123456789012123456789012 123456789012345678901234567890121234567890123456789012345678901212345678901234567890123456 12345678901234567890123456789012123456789012 The Texas Instruments MMST BPR Project 123456789012345678901234567890121234567890123456789012345678901212345678901234567890123456 12345678901234567890123456789012123456789012 123456789012345678901234567890121234567890123456789012345678901212345678901234567890123456 123456789012345678901234567890121234567890123456789012345678901212345678901234567890123456 manufacture of more complex chips required ever cleaner rooms. In addition, batch processing means that an error at any phase could result in hundreds or thousands of expensive but useless chips. TI elected to create new machines that processed chips within a vacuum. To further reduce the chances of contamination, the chips were manufactured upside down. (Think of this as a piece of brainstorming. Dust is drawn to the found by gravity. Chips were traditionally manufactured face-up, exposed to any dust that might alight on them. By simply turning the equipment upside down, dust was much less likely to settle on the surface of a chip.) Since more than one machine was involved in processing a chip, a vacuum cassette was developed. Chips were processed by one machine, inside the vacuum environment maintained by that machine, then placed in the vacuum cassette, which was then moved automatically to the next machine for further processing. It sounds simple enough, but it actually required a very sophisticated effort to simply design and manufacture the hardware that would process the chips. Another problem with chip manufacturing happens when an error occurred during processing. Unless the error was detected immediately, a very expensive effort was continued that only resulted in a defective chip. TI elected to create hardware that would process one wafer at a time. It also arranged to monitor the production process with a number of probes that were, in effect, constantly checking to see if the evolving chip was within set tolerances and functioning correctly. This monitoring process could only be cost-effective if it could be completely automated. Thus, TI elected to create a fully automated, real-time control system. The software system was designed to plan and schedule the production of chips and then monitor each chip, determine if any errors had occurred, and take appropriate action. Obviously, replacing a manual, batch operation with an automated system was going to require changes in the human support operations. Although jobs changed, the automated system still required people to setup and monitor the new system. One of the goals of the MMST effort was to provide terminals that would provide employees with all the information they needed. In addition IT elected to create a system that would let the monitors modify their terminal displays to suit their preferences. TI designed and manufactured machines that could process wafers in a vacuum. In a similar way they designed and manufactured cassettes to move chips between machines. The process required experimentation, handassembled prototypes, testing and finally the actual manufacture of a new chip fabrication factory. TI also created a large, real-time software system to run the new factory. TI elected to use new software techniques (objectoriented software design and development in Smalltalk). The overall design of the MMST CIM software was divided into several phases, including an architectural analysis phase, a prototyping phase, a design phase, and an implementation phase. © 2002 Business Process Trends 2 BPT CASE STUDY 123456789012345678901234567890121234567890123456789012345678901212345678901234567890123456 12345678901234567890123456789012123456789012 123456789012345678901234567890121234567890123456789012345678901212345678901234567890123456 12345678901234567890123456789012123456789012 123456789012345678901234567890121234567890123456789012345678901212345678901234567890123456 12345678901234567890123456789012123456789012 123456789012345678901234567890121234567890123456789012345678901212345678901234567890123456 12345678901234567890123456789012123456789012 123456789012345678901234567890121234567890123456789012345678901212345678901234567890123456 12345678901234567890123456789012123456789012 The Texas Instruments MMST BPR Project 123456789012345678901234567890121234567890123456789012345678901212345678901234567890123456 12345678901234567890123456789012123456789012 123456789012345678901234567890121234567890123456789012345678901212345678901234567890123456 123456789012345678901234567890121234567890123456789012345678901212345678901234567890123456 Even in a fully automated wafer fabrication facility, a number of people are needed to monitor the process, including production managers, engineers responsible for the actual process sequence, hardware maintenance engineers, etc. Thus, in addition to providing the software to run the machines involved in the processing, the CIM system supports a number of workstations each connected to a network that links the workstations to the manufacturing machines. Different modules (e.g. factory planning, factory simulation modeling, factory performance monitoring) provide interfaces and facilities for the different people who must interact with the system. The heart of the system is a scheduling or workflow system that processes orders by selecting machines and directing material movement for each step to be taken as a wafer is processed. The Solution TI completed the redesign in 1993 and installed the MMST system in its new plants. In the process it developed the ability to manufacture application chips in low volume in a cost-effective manner. TI developed a new line of semiconductor manufacturing devices, ranging from large machines that process wafers to vacuum cassettes that are used to transport wafers from one machine to another and all of that hardware is now sold to other semiconductor manufacturers. In addition, TI had developed a modularized software system that controls the manufacture of chips. The software was developed using new techniques that make it especially easy to modify and extend so that it can be used by others in related semiconductor Sematech, a consortium of semiconductor manufacturers standardized on the new framework and it is now used by other companies in the consortium. TI’s MMST project represents a complete redesign of a process. The developers began with a blank sheet of paper and asked themselves how they could design custom chips, using new technology. They arrived at a completely unique solution and created the hardware and software to support their new concepts. They reengineered the jobs of the people involved, replaced many jobs with software systems, and developed software systems for the new jobs that were created. The MMST project completely changed the economics of small scale chip manufacturing, making it possible for TI to dominate a new field for years, and to sell its expertise to those who wanted to compete with it. Thus MMST represents an ideal example of reengineering – the complete redesign of a process and the heavy use of computer technology to revolutionize the way the process is accomplished. © 2002 Business Process Trends 3