28nm Ramp Starts At Globalfoundries.
Globalfoundries has started its production ramp on 28nm and anticipates a much smoother ride on 28nm than it had on 32nm.
“We’re starting the volume production ramp on 28nm. With our experience at 32nm we expect less issues on the 28nm,” Rutger Wijburg, vice president and general manager GLOBALFOUNDRIES Fab 1, told EW.
A key reason for Wijburg’s optimism is the company’s understanding of high-k metal gate (HKMG) processing.
“The 45nm-32nm transition taught us about HKMG,” said Wijburg, “HKMG is an important transition, and we mastered it on 32nm.”
Asked about the controversial gate-first technology, Wijburg replied: “Gate-first is working fantastically. I have no sleepless nights on 32nm.”
Nonetheless the company currently intends to use gate-last at 20nm.
Asked about the comparative performance of 28nm compared to 32nm, Wijburg replied: “There are performance gains at 28nm over 32nm. 32nm HKMG SOI was a special process for AMD. 28nm is a process for more products for more customers”.
With volume production underway on 28nm the company is expanding the 28nm capacity at Dresden. “At the end of this year, 28nm will be a significant part of the output of Dresden,” said Wijburg.
Asked if ‘significant’ meant more than 2%, Wijburg laughed. “Yes much more than two per cent,” he said, “it will be a really significant part of our output.” No more precise figure was offered.
The 28nm process will be transferred to the company’s latest fab in New York.
For the next generation of process, 20nm, development will be done at New York then transferred to Dresden.
Will it be planar or finfet? “At 20nm we’re still working on planar and we’re also working on finfet,” replied Wijburg, adding “we’re in deep discussions at IBM about planar vs finfet.”
The company is part of IBM’s Common Platform Alliance.
Will 28nm be a good node? “I think so,” replied Wijburg, “28nm may be a very important node – it could be one of the longest-lived nodes in the semiconductor industry’s history because of the double patterning and EUV needed at 20nm.”
Does he think that nodes will just get tougher and tougher or that they’ll continue to vary between good nodes like 90nm and 65nm, and bad nodes like 130nm and 40nm?
“It gets more and more challenging,” replied Wijburg, “it’s challenging getting the maximum out of the tools and the mask-making and the materials. It will be more challenging going forward. But look – when I was doing my Masters I was taught that below 250nm nothing works because of quantum effects and today we’re at 28nm.”