Sometimes all a company needs is a good crisis to foster the acceptance of new technology. Such was the case in the early 1990s in the Semiconductor Sector at the Harris Corp. of Palm Bay, Fla., where the development of IMPReSS (Integrated Manufacturing Production Requirements Scheduling Systems) occurred during a period of downsizing following a major acquisition by the company.
The implementation of IMPReSS, a sector-wide, integrated manufacturing production planning and order quotation system, improved the on-time delivery of more than 1 million units a year and 19,000 types of semiconductor products at Harris, thereby significantly enhancing the company's global marketing position and overall profitability. As a result of IMPReSS, Harris' on-time performance went from 74 percent to an industry best of 95 percent, and losses of more than $100 million were reversed.
Despite the project's fairly rapid implementation, the system was not devised in a heated rush. According to Robert Leachman, professor of Industrial Engineering and Operations Research at the University of California at Berkeley (UCB) and co-author of the Harris Corp. project, the process of designing IMPReSS began as a long-term research project between UCB, Harris, and other semiconductor companies sponsoring research at the university to develop more advanced automated planning systems for the semiconductor industry. This research led to the development of the Berkeley Planning System (BPS), laying the foundation for the evolution of IMPReSS.
From its genesis, the development of IMPReSS differed from most OR/MS industry applications, which are usually designed to fix a specific problem as quickly as possible and effect long-term, positive implications. Because of its unique beginnings and development through a series of drastic company alterations, the project between Harris Corp. and UCB offers a number of lessons on how the specialties of the OR/MS community can best be used to industry's benefit.
The most important of these lessons is fostering a close involvement between the company and the researchers. "Over the years surrounding the development of IMPReSS, we worked closely with Harris' semiconductor research and development sector. We were part of the sector," says Leachman. "The project was treated as a team effort from the start and not as an outside sales job. Also, Harris maintains a manufacturing systems department staffed with people who have advanced degrees in computer science and engineering. While that doesn't mean they were completely familiar with operations research, they did at least understand what we wanted to do. We were always partnered together, from the diagnostics stage to systems development. Having a team of inside people and outside experts who could propose the needed changes in language that management understood proved to be very important."
Randy Burdick, who at the time of IMPReSS' development was director of Manufacturing Systems at Harris, worked closely with Leachman after having spent a year as a visiting fellow at UCB. "I was impressed by his (Leachman's) orientation toward the semiconductor industry," says Burdick. "He seemed to have a very thorough understanding of what our particular issues were. And he had a unique, holistic approach to the planning perspective that I had not seen elsewhere."
Explaining how teams were formed to develop IMPReSS, Burdick says, "Harris ran on a CIM (computer integrated manufacturing) organization, and we tried to organize ourselves around the various aspects of the project. For example, we had a data management team that could look at all the data results holistically, even though the sites where all the data were coming from are geographically dispersed. We also had a planning engine/calculation team that worked directly with Dr. Leachman. Another team worked with the factories around the capacity modeling situation. And we had other teams involved with data cleanup."
Refinement, testing and demonstration of BPS had been taking place on a small scale for years at Harris' wafer fabrication lines. Working on these BPS projects before the development of IMPReSS, Leachman says, "We envisioned a scope of company-wide applications embracing all of the factories of a large semiconductor company. But the early industrial applications involved the BPS working on only two fabs at Harris. This continued for several years with continuous improvement taking place." Based on what they had learned of BPS in these earlier, smaller implementations, the teams piloted a new methodology (which would become IMPReSS) based on optimization and dynamic production function models of semiconductor fabrication.
As work forged ahead, Harris acquired GE's semiconductor business, tripling Harris' semiconductor operations. The acquisition required a change of focus for the GE divisions from producing semiconductors for defense and aerospace purposes to more commercial products for the automotive and telecommunications industries. But due to the fact that GE had been trying to sell their semiconductor business for four years, no investments had recently been made in the sector, leaving its planning methods severely outmoded. When Harris purchased the sector, the company was faced with a huge on-time delivery problem. Other vendors were perceived as providing better service than GE. In addition to having to integrate the GE sectors, moving products between plants and dealing with systems that didn't talk to each other all combined to create a crisis at Harris.
"Suddenly there was an urgent need for a real sector-wide planning system embracing all of the factories and all of the products," says Leachman. Management at Harris felt they couldn't simply just match the practices of other companies that were a lot bigger and had a lot more money to invest. Harris needed to be smarter.
But Harris' troubles did not rest solely with its new acquisition - top management turned over twice during the project due to downsizing. "However, we were fortunate that every succeeding management team endorsed the project as a key strategic endeavor for the sector. There was involvement from every sector at Harris, including marketing and sales, manufacturing and information systems," Leachman says.
"Whenever a company is facing the kind of financial challenges Harris was facing at the time, that company is certainly more open to changes," says Burdick. "So in some respects, the downsizing was beneficial because it improved management's receptiveness to change. On the other hand, there weren't as many resources to go about affecting the change. This proves the importance, over a long-range project such as IMPReSS, of getting the people at the company bought into the system, and to get the users to own up to their side of the data."
One of the first steps Harris management and Leachman took toward solving the company's crisis was an audit of operations. This took about six months and involved travel to all of Harris' factories, from the United States to the Far East, in order to size up the problem of implementing a planning system across the entire sector. "At the time, there were hardly any companies in the semiconductor industry that had managed to fully integrate and automate planning across many factories, and there was a lot of skepticism about whether it could even be done or not," Leachman says.
"Early in the process it was difficult to even make a planning run of the system because the data was so bad," Leachman says. To improve the data quality at Harris, a period of responsibility assignment began. Team members were given responsibility for all their data and they had to write routines to check the data to show what was inconsistent, incomplete or missing. Other team staff were assigned to get things fixed regarding product structures and nomenclature, factory capability data, etc. Then the data had to be prioritized so that the most important problems could be fixed first.
"Regarding the data cleanup, it was important to develop a data report card that furnished clear, simplistic, graphic views of the health of the data. While this presentation of the data allowed for drilling down into the specifics if needed, on its face it offered a nice way of representing the health of your data and its improvement over time. That helps get your management focused on it, too," says Burdick. "Once the systems were running, we gave ownership to the operational people actually running the system, along with executive expectations. And that was the push that made it all happen."
For the final implementation of the system, Burdick says, "We developed a steering council made up of users to review the priorities and objectives of the system, and to look at what problems they were facing. Then we got the user community to buy in and understand what the constraints are; what the tradeoffs are." Having the users buy into the system is as critical to the systems' success as data cleanup, according to Burdick. "This helps to clear the smoke screens away and allows you to look at what the real system issues are and what are just data issues. Then you just start running the systems and go through the iterations and continually improve it."
According to Leachman, the key processes underlying the success of the project from the beginning were: 1) R&D funding and activity before the urgent need arose; 2) the recognition by Harris of the BPS technology's usefulness by working out the bugs and proving the concept on a small scale; and 3) the crisis need for the technology and top management's willingness to gamble on it in order to turn the crisis around. "The company was losing about $100 million a year at that point, and they either needed to fix their inherited on-time delivery problem or go out of business," Leachman says.
In approaching a problem like the one at Harris, Leachman says, "There's no substitute for understanding the problem inside out. Understanding the technical characteristics of manufacturing, the business characteristics of customer satisfaction and delivery, the organizational issues of who has responsibility for what data and what decisions need to be made with it, and also the quality of the data and the databases used. You have to understand all of this to appreciate how much is going to be changed.
"Many semiconductor companies, including Harris, look at their problem from the start as being simply a software or systems project. They think that if we can just plug in the new system, the problem will be over. That's absolutely not the case. What matters is matching the disciplined data maintenance with the right people responsible for it and re-engineering the organization around the workings of the new system. This is inevitable in such a project; and if it doesn't happen, the project won't succeed. That's why it's so essential to understand the company and its operations as if you were approaching a re-engineering project."
For Burdick, the most important things to consider in undertaking such a project are to ensure data cleanup and start off with simplistic capacity models. "We began with capacity models that were too complex and difficult for the users to understand and convert to. So the idea behind any such project should be to start off with simplistic capacity models and work toward greater detail as the organization learns. Our early complexity hindered the implementation somewhat; it got in the way and caused problems. The organization as a whole was not mature enough to deal with the more complex models. Lesson learned: Go for a more simplistic capacity model and then drill the organization toward more detail."
Leachman contends that the long years of research behind the BPS led to its eventual success as IMPReSS. "If we had just been confronted with the crisis without our long background working with the system and the company, I doubt we would have come up with anything like we did."