Hi Elwood:
Thanks for asking about me. I've been in a sleep mode ever since the semi-stocks took a hard hit. I am still holding a big chunk of CYMI stocks and lots of May call options. In my view the semi still looks good, very good. Taiwanese is still full speed ahead of their ramp to .25um. Intel, Motorolla, TI, AMD, etc. are ramping toward the .25um full speed. In the memory side it is another story. The Korean took the hardest hit realizing that they have some serious debts. Thus the new Korean fabs will be delayed but the existing ones will migrate to .25um, not at full speed. The Japanese will be more aggressive in pushing the .25um. The memory makers who can push the .25um in production will have the competitive edge. It takes the same effort to make a wafer. The ones who can sell more Mbytes/dollar will win. If the Korean slows down it will be the opportunity for the Japanese to leapfrog them. If you go back in history this is exactly what happened back in the 80s when the Korean leapfroged the Japanese in market share of memory. In logic and dsp, it is an absolute must that they need to go to .25um to get higher speed.
CYMI's Prospect:
What is all this has to do with CYMI? Well as the industry speeds toward .25um more DUV excimer steppers will be bought. The initial amount needed for .25um will be only about 10-15% of the total stepper capacity. When the industry moves toward .18um, 60%+ of all steppers will be 248nm and 10-15% will be 198nm. CYMI will have an explosive year in 1997 with total laser sold exceed 600. I calculated that a 5000+ wafer/week fab will need at least about 6 DUV lasers to do .25um. There must be at least close to 100 fabs worldwide that are/will do .25um in 1998.
CYMI is in the position to grab most of the 248nm lasers. For them the laser is a done deal, not much in R&D just continuous improvement in the existent lasers. The 248nm is the cash cow for CYMI for the next 6-8 years. Yes, believe it or not, the 248nm will be the tools for non-critical layers 6-8 years from now. The roadmap shows that 1999-2001 is the time frame for .18um production. The next generation is .13um and will be in production from 2002-2004 (see this month issue of Semiconductor International and Solid State). Beyond 2004 they call for <0.1um with gate oxide of 15 Angstrom!
It will absolute critical for CYMI to capture a chunk of the market in the 198nm. If they can then their future is guaranteed for many years to come. Komatsu and Lambda are pushing very hard for the 198nm. Competition will be tough. They made a mistake of letting the 248nm go to CYMI. I think Komatsu and Lambda are much smarter this time. It will be very tough for K and LP to eat market from CYMI unless CYMI screws up big time. So far CYMI executed flawlessly.
II. Stock Performance:
The problem with this stock is that everyone knows is a good stock with strong EPS. Everyone also knows that this is a single product company with very little news announcement. Thus it is a perfect stock to manipulate. What started it to drop was a rumor then compounded by free fall in Asia. The shorters loved it because they know they will bask in the sun for 2 months till earning announcement comes in Jan 98. I believe that is when CYMI will rise along with Jan. effect on small cap. What ultimately will bring this stock back to the high P/E=30-40 depends on whether the semi recovers and whether big boys start to cover CYMI again.
As of today the DOW closes once again very close to near record high above the 8100 level and yet the semi are about 50% below their high.
I tell you if one good news in the semi comes through the semi will take off so fast that you won't believe. As everything else becomes expensive buyers start to pour their money in cheap stocks like the semi. So be patient. CYMI will rise after AMAT and the other semi stocks regaining their 52 week high.
X-Ray Technology:
There is an excellent article about X-ray lithography in the Semiconductor International this month. I've been studying and learned alot about the technology. The people who wrote the article are of course biased toward the x-ray technology over DUV at .18um and below. Everything they wrote about x-ray (at 1nm in wavelength whereas EUV is at 15nm) is great. They said it is so good that everyone wants to jump on the bandwagon. I am very experienced with production and let me point out the problem with x-ray technology.
1) Reticle is not a problem: this has been solved and IBM has the technology to make it in mass quantity. The reticle size is about 2.5cmx2.5cm covering 2 dice per field made of silicon and tantalum coated on a quartz substrate. The reticle must be absolutely flat!
2) X-ray light source is not a problem and depth of focus (DOF) is 5-6um much better than DUV.
The problem with this technology is as followed:
1) Throughput. The throughput is very slow due to the mask has to be placed in proximity of the wafers. The mask has to be placed closer than the DOF. It takes time to do this because if you do it too fast you can smash the reticle into the wafers. DUV stepper can do 80WPH!
X-ray can't even do 10 wafers/hr. Let us said that if DUV can do .13um using 198nm DUV and x-ray can do .1um. At .1um you get twice the dice ( half the area ) as .13um. But I can make 8X the dice at .13 than at .1um then
Profit= (# of dice)*(dollars/Mbytes)*(# Mbytes/die)
my profit is 4X my X-ray. Thus go to smaller geometry doesn't buy me anything if I can't squeeze out more throughput. My cross-over model is 0.05um in x-ray to be equivalent to .13um at DUV. Let me also tell you this, processing at 0.05um is at least 4X harder than at .13um due to defects and process issues.
2) Contact masking:
The concept of proximity masking or contact masking is a production nightmare. A sensor must sense that a wafer is at least 5um to the mask before it can shoot. If your sensor dies then you just blow a reticle. It will be a very expensive reticle to replace. Other minor problems might be a defects on wafers. These defects may scratch the wafers. It is minor because very few defects actually sticking 5um out of the wafer surface.
3) Reticle checking:
This is also another minor problem. An x-ray reticle is a 1:1. Thus the smallest feature on the mask is the smallest feature on the die. It is very difficult to do check for pattern defect on a reticle with feature that smalls.
Process problem below .1um:
The requirement for the gate oxide blow .1um gate is less than 15 angstrom. Do you know how tough it is to achieve this? If you leave a bare silicon wafer out in the atmosphere for 5 minutes it will grow at least this much of oxide. To control the thickness of gate oxide is a nightmare. You have to devise a tube that strip the native oxide in a vacuum and regrow it in a very low O2 pressure, not an easy task.
The other problem with thin gate oxide is the tunneling current from gate to S/D. It goes up to Amperes/cm^2, absolutely ridiculuous. And on and on.....Blah, blah, blah .............
Conclusion:
Semiconductor will hit .1um and stop. It is crazy to go below that level. The defects go up exponential at that level. The process will be so damn difficult that it is not feasible to try. One day CYMI will stop selling DUV lasers because all the fabs in the world have CYMI lasers. CYMI will then be service company that replace old tubes and clean out old power supplies. That day will be at least another 10 years. By then new breakthrough will come and CYMI will have enough cash to buy a big bandwagon and head WEST.
Maxwell
PS. What is all this nonsense about CYMI laser problem. Mine is still running good 24 hrs/day production without any problem. |
