6 of 6Image
With so much automation on the fab floor, communication grew to be a problem. IBM resolved the issues by sticking a great big whiteboard in the middle of the fab (visible on the wall of the blue stocker to the left) so that workers can leave notes for the following shift. In a room where one piece of equipment can cost up to $13m, it's a startlingly low-fi touch.
While the East Fishkill fab is tooled to a 32nm process, IBM is looking at the next steps in the Moore's Law chain. "We are actually running designs in 22nm, the next generation," Arthur says. "There is 14nm in development out there and 10nm is [being developed] in Albany."
All eyes in the industry are on the shift to 450mm wafers, and the improvements in productivity that should bring. "When the next wafer size hits, you're gonna have to build a new fab, because the tools are not compatible," Arthur explains.
IBM is one of five companies in a consortium developing the tools to be used in a 450mm facility. "The investments will be mind-boggling," Arthur says. "To build one of these factories now from scratch is about $6bn for about 1,000 wafer strikes a day. This new one will be in the range of $10-12bn." 2016-17 is Arthur's prediction for when the first 450mm fab will be up and running - and he's reluctant even to commit to that.
Chip makers are already using ultraviolet photography and liquid immersion to produce chips - the other big shift on the horizon, according to IBM, will be the move to Extreme Ultraviolet (EUV) production. This is a form of what Arthur calls "very sophisticated photography" that could be used to print the next generation of chips. "The cost of those tools is staggering - numbers like a quarter of a billion dollars," he adds. "But it is what will allow you print lines that are 8nm or 6nm. That's the rub. When do the current ultraviolet tools run out of gas?"
The path ahead to the next chip process is just about visible - after that, everything is murky. But, as Arthur points out, this is nothing new: "I've been here 32 years, working on semiconductors for 28 of those, and every generation we're in, we say, we see the path to this one, and maybe, if we're lucky, the one beyond. And then we say we have no idea. So it requires invention."
All Images courtesy of IBM.