The State of 450mm Wafers. And Intel Gossip

Paul Farrar, the General Manager of the Global 450mm Consortium (G450C) presented at the SEMI ISS conference today. What is G450C, it is a public-private consortium to develop a cost-effective 450mm wafer fabrication infrastructure, develop equipment prototypes and coordinate the industry move to 450mm. It is located in Albany New York and its main partners are Intel, TSMC, Samsung, GlobalFoundries, IBM and the state of New York. The current plans call for work to take place over the next 3 or 4 years and be ready for volume manufacturing in 2018 or 2019. At the end of last year they had 41 450mm tools either in Albany or at partner companies or at suppliers (tools just means pieces of equipment like a stepper). They have a 45,000 square foot clean room.

Paul reported on the detailed status of various steps in wafer fabrication. Mostly things are going well for this stage in development. One area that is a challenge is lithography where performance and cost are key. Since a stepper only exposes one die at a time, there is less of a direct gain from 450mm than for other tools such as CVD that work a whole wafer at a time. There is a little saving in wafer handling. By way of an example Paul pointed out that normalizing to 1 at 200mm in 2003 when 300mm was finally introduced, initial introduction had lithography at 1.32 times as fast and 5 years later at 2.15 times as fast. So there is learning to be done. G450C has access to a prototype immersion stepper at Nikon and will get their own in the first half of 2015.

One thing that was not planned originally for 450C is notchless wafers. But it makes handling those big wafers a lot easier if they are not unbalanced when they are spun and generally processed. Instead, fiducial marks are put on the back of the wafer and used to align to all the equipment. This also has the advantage of giving an extra 1.5mm at the edge of the wafer, which on a big wafer like that is worth having. One questioner suggested that getting an extra 1-2% yield on 300mm wafers would make retrofitting current 300mm lines attractive.

During dinner I sat next to a guy who supplies wafer blanks. They had a 450mm program a couple of years ago but they canceled it since they don’t think 450mm is going to be real and for them, in any case, it is impossible to make money. Just 3 years ago a 300mm blank wafer sold for $280 but today it is just $60-70 and foundries expect the same price per square inch for 450mm. Since there are only a handful of fabs that will make the 450mm transition, they have very strong bargaining power. He reckoned that only now are equipment manufacturers finally in the black on 300mm having started development in about 1990, it took 20 years. The return for 450mm is also expected to take over 20 years. But nobody knows what 450mm equipment will be needed in 20 years time since we don’t know the basic technology. Depending on what happens with EUV there is also a race: manufacturing will not introduce EUV and 450mm at the same time, or probably even a new process. EUV or 450mm will be retrofitted to an existing process already ramped to volume.

Other pieces of gossip from people at my table at dinner. Intel isn’t worried about 450mm right now, they have bigger problems. They have apparently said that they will not use EUV for manufacturing until there is technology to put a pellicle on the mask (to keep it clean and keep contaminant out of the focal plane, but almost all pellicle materials absorb EUV). Intel’s fabs are 70% utilized. Their big challenge in mobile is that the price point for tablets and smartphone chips is below Intel’s variable cost, so the more they make the more money they lose.

And not a piece of gossip, Intel announced today that they are putting fab 42 in Chandler Arizona on hold and will use it for future technology. They will still run 14nm in Arizona but sharing the 22nm fab to save cost.

More articles by Paul McLellan…