In this third of a three-part series we analyze how IBM got to this crossroads of considering a sale of its fabs, given its standing as a semiconductor leader.
SAN JOSE, Calif. — Despite talk of a sale and declining morale, there's still plenty of pride among IBM's current and former semiconductor employees for a team that developed many technologies that laid a foundation for the whole industry
"It's a shame to see them talk about divesting -- it's a national resource," says Gary Bronner, an IBM veteran now at Rambus. "A lot of industry innovations were led by IBM."
"Given their history, it would be a sad day if they really got out of semiconductors," says G. Dan Hutcheson, chief executive of VLSI Research.
Here's a partial list of semiconductor innovations and milestones IBM developed or championed:
- Single transistor DRAM cells in 1966
- The RISC processor architecture in 1980
- 3D chip packages (ceramic modules, flip chips) in the 80s
- The first 200 mm wafer fab in the industry in 1989
- Embedded DRAM caches
- Copper interconnects in 1997
- Chemical mechanical planarization (CMP)
- Argon fluoride lithography
- Computational lithography
- Chemically amplified resists
"I don't think the circuits we build today would even be possible without CMP," says Bronner. "The lithography tools we are using for small-depth focus is only possible with CMP -- the ability to push Moore's Law as far as we have was enabled by that."
"IBM began producing discrete transistors in the mid-1950s for its products," says Rob Lineback, a senior analyst at IC Insights. "By 1959, the company had developed an automated transistor production process, allowing transistors to replace vacuum tubes in its mainframe computers."
In 1978, IBM was one of the world's leading memory makers, from a technology perspective.
"They were making 64 kbit DRAMs -- a full node ahead of everyone else -- and moving to 256 kbit parts while the rest of the industry struggled to get there and didn't make it until about 1981," says Hutcheson.
"Memory was a big revenue generator for them -- it gave them a huge competitive advantage over other computer makers of the day like Digital Equipment and Control Data. In those days IBM was one of the largest tool builders, too. They built their own electron-beam tools."
Later IBM became "one of the first companies to start putting memory on chips," he says. "Early on they figured out processor performance depended on how much memory you had on your chip, so they very quickly moved to half the processor being memory."
That's when IBM started developing SOI. Its insulating layer helps contain the charge in DRAM cells, enabling IBM to pack big, fast caches on its Power chips, says David Lammers, a freelancer who reported on the work as a semiconductor editor at EE Times.
IBM is still a chip technology leader, publishing top papers at events such as ISSCC, IEDM, and the VLSI Symposium on work in areas such as FinFETs and extreme ultraviolet lithography.
At the group's height it tried to make a version of Power chips that would run x86 code to compete with Intel in desktops. "That was an ambitious effort from the start, but people were talking about the 615 project, and that kept things exciting," says one former IBMer.
At the same time, IBM started its effort to market chips openly to outside customers.
"We were like a startup -- we didn't even know how to bill a company," he tells us. "Customers would literally call us and ask, 'Where's the invoice?' "
Next Page: What went wrong