Consider, for example, LCD manufacturing. Many Japanese consumers bought into the legend of Sharp Corp.’s Kameyama fab, which developed and produced in the mid-2000’s the most advanced, high-quality LCD panels—billed at that time as “the most revolutionary.” Believing in its own Kameyama myth, Sharp poured billions of dollars into a new LCD manufacturing plant. The Sakai plant was designed to be the first in the world to manufacture tenth-generation LCD glass substrates, capable of producing eight 50-inch panels per sheet. In 2008, Sharp announced that Sakai plant would dwarf the Kameyama 2 plant, describing it as “30 times the size of Yankee stadium.”
Manufacturing 50-inch panels per sheet was surely an enviable technological feat—at least on the drawing board. I’m skeptical, however, about how much analysis and due diligence Sharp’s management team—including its own engineering and marketing executives—invested before taking to what proved to be the suicidal step of building a fab big enough to accommodate 2.4 million vicious Yankee fans.
The fallacy of a 25-year quality guarantee for DRAMs
Takashi Yunogami, a director of the Fine Processing Institute, points out the fallacy of the 25-year quality guarantee Japan’s memory chip vendors promised for their DRAMs.
Yunogami, formerly an engineering expert in dry etching at Hitachi and now author of books on the Japanese semiconductor industry and a lecturer at several universities, recalled that the 25-year guarantee was born when DRAMs were first designed into large-scale, mainframe computers and communication infrastructure equipment such as telephone switching. NTT, one of the biggest customers for DRAM in early 80’s, told Japanese chip vendors at that time: “Bring us DRAMs that won’t fail.”
In efforts to meet that onerous 25-year standard for mainframes and a 23-year guarantee for phone switching equipment, quality-conscious Japanese memory chip engineers amazed the world by actually accomplishing the goal. Their unparalleled skills in developing fine processing technology and manufacturing high-quality DRAMs led Japanese semiconductor companies to dominance in theglobal DRAM market by the mid-1980’s.
What followed then were inexplicable choices by Japanese semiconductor vendors. They didn’t scale back from 25-year “perfection” to cost-effective alternatives that didn’t need to last that long, even well after the computer industry made a dramatic shift to personal computers. In spite of the typical PC’s much shorter life, Japanese memory companies kept pumping out high-quality, over-specified DRAMs. The result was a massive loss of market share against Samsung which mass-produced low-cost, three-year guarantee DRAMs for PCs, Yunogami explained.
By the late-1990’s—right around the time when Japan began retreating from the DRAM market—Yunogami believes that noticeable differences existed in a photomask set used for 64-Mbit DRAM fabrication process used by Japanese, compared to those by non-Japanese vendors.
A mask set for a process node typically contains as many as 20 or more masks, each of which defines a specific photolithographic step in the semiconductor fabrication process. While Japanese DRAM vendors then used as many 29 to 26 masks, Korean and Taiwanese vendors used 22 to 20 masks. Micron was believed to use only 15 masks, according to Yunogami. More masks means more processes, and more processes means more semiconductor manufacturing equipment. “This led to the high-cost structure of Japanese semiconductors,” said Yunogami.