Is 450mm Dead In The Water?


Chipmakers have stopped beating the 450mm drums. So what’s up with 450mm technology?

At one time, Intel, TSMC and Samsung were aggressively beating the 450mm drum. Chipmakers wanted, if not demanded, 450mm pilot line fabs by 2016, with high-volume manufacturing 450mm plants slated by 2018.

At least for those companies, 450mm made some sense. Moving to 450mm wafers would supposedly give chipmakers a 2.25x boost in wafer area and a 30% cost reduction over 300mm substrates. But, of course, the business model was (and still is) shaky. Only a few chipmakers can afford to build 450mm fabs, which, in turn, limits the total available market for selling fab tools.

But after resisting the move towards 450mm technology for several years, the fab tool industry finally decided to follow suit. With some arm twisting from the big chipmakers, the equipment industry started to develop their first tools based on the next-generation wafer size.

More recently, however, the 450mm drums have stopped beating and are now quiet. Needless to say, Intel and the rest of the industry have delayed the shift to 450mm fabs for the foreseeable future, leaving many to ponder the following question—Is 450mm technology dead in the water? The answer: 450mm is currently treading water.

If you ask Intel about 450mm, executives will insist it’s still on the roadmap for the latter half of the decade. But at a recent event, William Holt, executive vice president and general manager of Intel’s Technology and Manufacturing Group, said: “We can’t do it alone.”

Indeed, for 450mm to work, the big three chipmakers—Intel, TSMC and Samsung-must push and synchronize their efforts in 450mm, thereby making it easier for tool makers to align their R&D and tool shipments with the rest of the industry.

TSMC says it’s still interested in 450mm, but the company is now lukewarm about it and for good reason. TSMC discovered that fab tool vendors weren’t going to fund their entire R&D costs in 450mm. In fact, fab tool makers want chipmakers to assume more of the R&D and risk. For that reason and others, Samsung has completely backed away from 450mm.

So as it stands today, 450mm remains at a standstill. And in many respects, the industry hopes that 450mm remains delayed or will just simply vanish. After all, 300mm is doing just fine.

As before, the business model for 450mm remains shaky. Now, the technical rationale for going to 450mm is not as attractive as before. For example, chipmakers already have a tough enough time to fill their 300mm fabs during slow periods. And there really isn’t a new and big product driver that will fill a giant 450mm fab in down cycles.

So, why is 450mm still in the conversation? That’s because fab tool vendors have already invested in 450mm technology. To scrap those efforts would put 450mm on ice forever. “I don’t think 450mm is dead in the water,” said Joanne Itow, an analyst with Semico. “It’s still in the works, but it looks like it could get pushed out even further.”

And in fact, 300mm fabs still have plenty of legs. For example, some foundries, namely TSMC, still have huge fab capacities for 28nm technology within its existing 300mm plants. Many think 28nm will remain a long-lived node; the technology doesn’t require 450mm fabs.

In fact, TSMC is even encouraging customers to stay at 28nm and expand their production at that node, Itow said. “It’s a way to keep their capacity utilization up,” she said.

Nick Kepler, an analyst with VLSI Research, said 450mm still makes sense if the IC industry continues to follow Moore’s Law. Simply put, if the industry continues to move along the two-year or so process technology cadence, then chipmakers could get a substantial boost in wafer area with 450mm fabs. But if the traditional two-year cadence slows to a large degree, then 450mm may get pushed out to the next decade.

So what if 450mm remains delayed or gets scrapped all together? “With 450mm being pushed out towards the latter part of this decade, the industry needs to find more cleanroom space to meet demand,” said Christian Dieseldorff, an analyst with SEMI. “Starting around the 32nm/28nm node, there is an increasing ratio of cleanroom space (needed) when upgrading to the next node. Leading-edge nodes become increasingly complex. More process steps are required in multiple patterning. The loss of installed capacity per transition can range from about 8% to 18%, depending on the product and node.”

For example, a fab may have a capacity of 100,000 wafer starts per month (wspm) at 32nm/28nm. At 20nm, the “trade-in ratio” may end up being 80,000 to 90,000 wspm. “The trade ratio is even larger for 3D NAND. Companies need to find more cleanroom space to fill that hole in order to maintain the capacity,” he said.

So, the industry has three options—retrofit or build out existing 300mm fabs; construct new 300mm plants; or go to 450mm.

The first two options are already in play. At present, there are 90 volume fabs in the world today, in which about 55 of those are leading-edge plants that can make chips with feature sizes down to 32nm/28nm and below, according to SEMI. “We also list about 20 fabs with various probabilities, which will begin volume this year or sometime in the future,” he said. “Most of these are for foundries and memory.”

Then, there is the option to build new facilities. “I see this option happening, especially for the foundries such as TSMC and GlobalFoundries. On the memory side, I would expect no immediate need. But by 2015 to 2016, they need to start construction to have the fabs ready to ramp in one to one-and-half years later,” he said.

“How many do we need? Difficult to say. It depends on how fast the existing fabs will ramp, and how many of the older fabs will be refurbished. In addition, these new fabs will become larger with more capacity. Maybe two to four more memory fabs (may be required in the) 2017/2019 time frame,” he said. “But there are not many companies who can afford these.”

And what about 450mm? Many believe that 450mm fabs will still move into production, or at least in pilot lines, in the 2018 to 2020 timeframe. Here’s the best prediction: Based on the costs and delays, G. Dan Hutcheson, chief executive of VLSI Research, doesn’t see 450mm fabs moving into production until 2020 to 2025.

Related Stories:

  • Reader

    Regarding the “trade-in ratio” mentioned for 28nm vs 20nm, it might seem that with 80k wafer starts at 20nm vs. 100k wafer starts at 28nm (using same clean room space) that one would be “behind,” as one would now have 80% efficiency of the equivalent space in terms of wafer starts. However, one has gained double the number of chips, assuming a similar design scaled down (20^2 / 28^2 area ratio), so my net is ~1.6x more chips coming out (80% * 2x more chips). So, I don’t see where added cleanroom space is needed to “keep up,” unless the complexity has also greatly increased the cycle time. If complexity of 3D NAND causes this, it’s important to not overlook that one has more than functionality / transistors / pricing over straight 28nm as well. Bottom line, moving to the next node does not appear to be a valid argument for increased capacity, and hence, 450mm. I believe that’s why you will see retrofits will work just fine in the future for advanced nodes.

  • reader

    The industry is struggling with 14nm and FinFet, as well as next generation Litho (NGL) so no one can handle another big change of 450mm wafer conversion. The extra development cost of the new tools will take a severe toll on the equipment companies as the semi manufacturing will not pay their share. (see the money that was wasted by Intel on EUV at ASML). The rest of the semi companies are just too cheap to participate in the development while the equipment companies are trying to protect their high margins, so it is a lose lose scenario.
    Buy the time the industry will get to <10nm Moore"s law will be a history so 450 mm will be totally irrelevant.

  • Pingback: Semiconductor Engineering .:. Blog Review: May 21