Hi Cary,
I am pretty busy today - working on a publication. When I get a bit more time I will come back and address your comments in more detail. In the interim, I think the following article from my last letter discusses some of the very issues in your post. This is from my writeup of the Fab Manager's Forum that was held during Semicon West. I am posting it here for your perusal and for future discussion.
Regards,
Carl
The Perils of Chasing the Next Big Thing in Photolithography
Phil Ware - Canon
Phil Ware is a Senior Fellow with Canon. He lives in my neck of the woods, Irving, Texas. I have seen Phil at a number of executive conferences over the years but this was the first time I have heard him present. His talk shed some light (no pun intended) on the race in the lithography world. This is a hot topic in the industry – actually, it is probably the hottest topic in the industry.
Phil started his presentation with a look back at the 1992 International Technology Roadmap for Semiconductors (ITRS). About the only thing missing from the ’92 roadmap is 157nm lithography and Immersion Lithography. Obviously the roadmap was accelerated during the mid 90’s. NGL (Next Generation Lithography) was originally targeted for 180nm device production. 157nm was not even listed as a contender before the 1999 ITRS. In 2000, Canon served up a feasibility demonstration of their XRA-1000 tool for 70nm device production. Today you hardly hear a word about X-Ray lithography.
What happened to the 1999 lithography roadmap? 157nm was supposed to come on line sometime between the end of 2002 and 2005. It is now, based on the recent announcement from Intel, almost off the roadmap. At the same time the NGL’s slipped 6 process nodes, all the way down to 45nm. Will any of the NGL’s be viable?
Phil calls these foibles the “Sillycon Cycle”. The cycle started with Optical Lithography. In the early years experts predicted that KrF technology would never be used at 250nm. At that time the NGL initiatives were launched and there were numerous attempts to narrow the number of options. It takes a consensus to make something fly but none was reached so the industry decided, in just a few short years, that NGL would not be ready in time. With NGL on the back burner, 157nm was raised from the dead. A couple of years later, 157nm was deemed too late and the Experts were asking, “How can we extend ArF to 65nm?” Well, Intel believes they can take it all the way down to 45nm. They said as much during the conference.
The point Phil was making about the “Sillycon Cycle” is, “At any point in time, the lithography roadmap has been too pessimistic toward optical lithography and too optimistic for an NGL solution.” I would like to add this, “Don’t get caught up in the hype. Each little announcement is not all it is cracked up to be.”
With his major point across Phil moved in to the next big question. If we agree with his first premise, that we are too bearish on optical lithography, why don’t we just pick one lithography method and run with it? No, we couldn’t do that, it would make things too easy. Not a chance. In 1997 there were, according to data from Sematech, 11 different lithography technologies being developed for the 130nm node. The list is just about all-inclusive: E-beam direct write, SCALPEL, EUV, IPL, X-Ray, 193nm extended, 157nm transparent, 157nm reflective and even 126nm.
What are we using today? 248nm and 193nm. Optical lithography systems that have been extended with the use of enhanced reticles. The problem remains, the options for the future have still not been narrowed. The industry is still not sticking with one method and running with it.
I don’t know if Phil wants to be as pessimistic as he sounds but he placed some quotes up from a few people in the industry that did not sound that good. From Don Mitchell, Chairman of FSI International, there was this:
“If you look at where the (chip equipment) industry is relative to 2000, it has been cut in half – and going on apparently for another year of it (weak capital spending).”
Don also said, “Lithography tools and photomask costs are among the biggest challenges for the industry at a time when R&D funding levels are being cut by both chip equipment suppliers and device manufacturers.”
Dan Hutcheson of VLSI Research was quoted this way, “DRAM’s have a cumulative loss for the years 1996-2002. The equipment industry has cumulative negative net cash flow over the same period. So the few firms at the top are sitting on top of a rusting infrastructure that could bring them down. Moore’s Law is likely to slow if we do not fix the profitability problem.”
Ouch. That’s not a real pretty picture. For lithography tool vendors there simply is not a clear roadmap. Phil posed the question to the fab operators in the audience, “Are chipmaker roadmaps a better guide?” In general, they are not. The key points here were as follows:
• Many chipmakers use nebulous node definitions
o Actual devices typically have no features of nominal node size
o Aggressive node claims are made for marketing purposes
 But, is it in high volume production
o Little or no relationship to half pitch
 Eliminate ‘node’ and standardize on dense half pitch?
• Aggressive node timing causes unrealistic expectations
o Implies all the requirements in the ITRS tables are being met
o Vendor’s most advanced litho solutions are always “too late”
o Some chipmakers are already asking for firm LiL (immersion lithography) delivery dates even though LiL has never been listed on any “roadmap”
o Forces vendors to accelerate multiple tool development programs simultaneously
Here is an example of the classic lithography problem served up by a chip manufacturer for the 65nm node:
65nm Node Characteristics (2005)
• 92nm Minimum Metal Half Pitch
• 80nm Poly Half Pitch
• 32nm Physical High Gate Performance Gate Length
• 45-50nm Gate Length in Resist
Half Node Shrink in 2006
• 92nm Minimum Metal Half Pitch
• 80nm Poly Half Pitch
• 27nm Physical High Gate Performance Gate Length
• 40-45nm Gate Length in Resist
Do you notice anything striking about these features? Not one is actually 65nm!
In the end the Lithographer has three basic knobs to turn –
1. Numerical Aperture (NA)
2. Exposure Wavelength
3. Process Factors
Increasing NA’s is helping. Phil is expecting NAs to exceed 0.9 next year. How much larger can they make get? Transition to another, shorter Exposure Wavelength (currently leading edge is 193nm) does not appear to be an option. All of the shorter wavelengths face serious technology hurdles. George Gombal is the senior manager at IBM Microelectronics for lithography technology development. When asked about all the attention being placed on a handful of NGL candidates, including the front-running EUV platforms, he said, “There are no alternative lithography technologies that have adequately addressed the critical success factors, and none of these are ready for commercialization. The alternative technologies must be targeted at the 32nm node with a plan for the infrastructure of building-blocks.” He went on to add, “Optical lithography will clearly be pushed to enable the 45nm node.”
Is the industry putting the cart before the horse? Phil is worried about this because the ITRS roadmap, again accelerated, has the next three technology nodes arriving faster than any of the shorter wavelength lithography tools. He believes strongly that 193nm is as short as the Exposure Wavelength gets for a long while.
So what does the industry do? Reticle Enhancements (RETs)! The first thought that popped in to my head when Phil said this was “Why aren’t the photomask players making hay here…..”
It’s truly a mystery.
RETs are the answer but if this is the only answer in the near term reticle costs are going to soar over the next few years. The factors at play here are substantial. Canon’s projections from the BACUS Symposium suggest that reticles for the 90nm node could run up to $1.5 million per set. At 65nm the cost doubles to more than $3 million and at 45nm we might be looking at $6 million per set. There is a worry that foundry operations could be stifled because lead times could stretch out to several months.
If there is a will there is a way.
There is a way to reduce reticle costs. Lithography friendly designs would be a step in the right direction. Our past letters have had some comments from Ken Rygler on this very subject. Here’s a list of things that can be done:
• Avoiding non-Manhattan geometry
• Lining up gates in one orientation
• Reducing critical feature types
• Avoiding forbidden pitch ranges
• Reducing pitch ranges
• No designs critically dependent on corners, line end or other 2D patterns
• Adapt designs to emerging RET techniques, i.e., IDEAL, GRATEFUL, SCAAM
All of this sounds terribly negative but the situation is fairly severe. There is an option aside from very expensive RETs that is just starting to get a great deal of attention. Even Phil thinks it has solid potential. Liquid Immersion Lithography (LiL) may be the easiest way to reach technology nodes beyond 65nm. At this time there does not appear to be any showstoppers. Even the critical research is economical. Most important, LiL requires very little change to the current optical train and it can be tuned for resolution and depth of focus (or both!). The emergence of LiL changes the picture dramatically. The effective exposure wavelength and the resolution factors are relaxed to an enormous degree and if it works with 157nm we may never see EUV in our lifetime
Nothing just comes out of the gate and takes over in a production fab. There are issues with LiL but in terms of risk/reward it seems like a viable alternative. Wrong decisions in the lithography world can cost billions of dollars but so far the issues with LiL do not appear to be insurmountable. It would not be surprising to hear some news on this front in the very near future.
Phil summed his talk up by pointing to the bottom line (I heard that a lot during the week).
The potential bottom line adds up this way:
- Reticle set for the 45nm node -- $6 million
- Litho tool set for the 45nm node -- $50 million
- 300mm fab for the 45nm node -- $4 billion
- Making a profit from any of these businesses – Priceless
We couldn’t agree more. |
