[This article first appeared in the May 1991 edition of _Electronics_.] CONQUERING THE LIMBO FACTOR by Lawrence Curran Stata had become convinced that "on-time delivery of products that work has become the major factor in vendor selection and performance evaluation." At the time, however, his company's record for on-time delivery wouldn't have won any awards. Manufacturing cycle times at the Analog Devices Semiconductor Division in Wilmington, Mass., were 22 weeks in the first quarter of 1986, before Schneiderman instituted a quality-improvement program. By the end of 1989, they had been slashed to nine. They're now at eight weeks, with a goal of four to five next year. Nor were yields anything to brag about. In 1987, for example, yields were about 20%. They're now at 38%, and the goal is to top 50% next year. "We measure four internal processes," Schneiderman says: manufacturing-cycle-time reduction, time to market, process-defect levels, and yield improvements. "They all play together and can't be treated in isolation," he asserts. Schneiderman adds that customers don't care what a vendor's manufacturing cycle times or yields are. They're interested in the product's quality, price, and "our responsiveness to their needs. What they really want to know is how often we say 'you've got it' when they tell us they need a certain number of parts by a given date." With a 22-week cycle time in 1986, Analog Devices was missing a lot of delivery dates. To attack the problem, Schneiderman and his team first measured the actual time required to process a batch of wafers from start to finish. That added up to 26 weeks. But the cumulative time encompassing each step along the way--wafer fab, assembly, and test--was 20 weeks. "We couldn't account for where the product was for six weeks, so we added another process step and called it limbo," Schneiderman recalls. The next step was to determine the theoretical time that should be required for each step in the process. The biggest variance from the theoretical to the actual was the limbo factor, "so we assigned a team to cut the limbo time to zero," Schneiderman says. Analysis showed that a major part of the limbo time was consumed when wafers were bumped out of a process queue to make room for a "hot lot" that had to be expedited. However, there was no automatic method to reschedule wafers that had been shunted aside. "They may have sat around for four weeks before being assigned to a new processing lot," Schneiderman says, because at the time, rescheduling was a time-consuming manual process done about once a month. Since then, Analog Devices has converted to computerized manufacturing-resource planning (MRP) that makes it easier to reschedule the floor more often; "we've also minimized the expediting of hot lots to those that are truly essential," he says. The technique the firm uses is based on Pareto analysis--a method that identifies the most glaring problems in successive passes at a process. The limbo factor was by far the biggest, Schneiderman says. "Once that was solved, we went to work on the next biggest problem, and so on until we got those under control, as well." Testing also caused delays because of out-of-calibration equipment and the fact that updating the test schedule wasn't done often enough. MRP solved the latter. Another statistic indicates how well the quality program is working--the percent of line items shipped on time vs. factory commit dates. That number has jumped from 72% on time in early 1986 to 96% at the end of 1989. The goal is 99% by 1992. Do all these dramatic and measurable cycle-time reductions and yield improvements affect profits? Unfortunately, "I don't know of any company that has seen a direct bottom-line impact" after substantially improving on quality and delivery, Schneiderman says. "The benefits are passed on to customers in reduced prices," which keeps them coming back. "In today's highly competitive world, if you don't improve you can't compete; you'll be out of business."