SAN FRANCISCO -- Panelists agreed to disagree about the future of fabs, including the shift towards 450-mm plants, which could cost a staggering $10 billion if or when these giant facilities get built.
At this week's International Electron Devices Meeting (IEDM) here, the panelists also warned that Moore's Law could slow down after the 22-nm node as device physics is nearing the wall.
''Is the technology node progression slowing down? Not yet, but it soon will be,'' said Craig Sander, vice president of process R&D at Advanced Micro Devices Inc. (AMD) and the company's foundry spin-off, dubbed The Foundry Co.. At 22-nm, ''we are hitting the physical limits (of scaling). After that, things will slow down.''
The IEDM panel topic centered around ''The Future of Fabs.'' As expected, 450-mm fabs were among the topics at the panel. As reported, Intel, TSMC and Samsung are separately pushing for the advent of 450-mm ''prototype'' fabs by 2012. Some believe that 450-mm fabs will not appear for another decade. Others believe 450-mm fabs will never happen, saying the R&D costs are too expensive.
During the panel, John Lin, director of the manufacturing technology center at Taiwan Semiconductor Manufacturing Co. Ltd. (TSMC), provided one of the first glimpses of the cost structure for a 450-mm fab.
In a presentation, Lin said a 450-mm fab could cost over $10 billion. In comparison, a leading-edge 300-mm fab for the 45-nm node runs about $4 billion.
To make 450-mm fabs a reality, it ''requires industry collaboration'' between chip and equipment makers, said Devadas Pillai, Intel fellow and director of operational decision support technology in the Logic Technology Development group at the chip giant.
The representatives from TSMC and Intel declined to comment when 450-mm fabs would appear. But not surprisingly, the fab-tool community and others on the panel were lukewarm about 450-mm.
Masayuki Tomoyasu, senior vice president and chief engineer at Tokyo Electron Ltd. (TEL), said the cost and risk are too high for 450-mm fabs. Many in the industry propose the so-called 300mm Prime program, which aims to boost efficiency in current 300-mm fabs.
Bill Arnold, chief scientist for ASML Holding NV, said overall throughput could take a hit within 450-mm fabs. On a slide, he demonstrated that lithography throughout would be half or less for 450-mm, as compared to current tools running in today's 300-mm fabs.
There is another huge problem: It is still unclear where the R&D dollars will come from to develop 450-mm tools, said Hans Stork, chief technology officer at Applied Materials Inc.
''450-mm will increase development costs,'' said AMD's Sander. In fact, overall process development costs are increasing 1.4 times per node, leaving fewer and fewer players that can afford to build fabs, Sander said.
On a positive note, ''innovation will not slow down,'' he said.
Innovation will be required if IC scaling slows. The panelist listed various technologies that could be deployed in the event of a slowdown in scaling. Those include 3D-based thru-silicon vias (TSVs), carbon nanotubes, FinFETs and related structures. On the memory side, there are 4-bit-per-cell NAND, cross-point memories, among others.