I am going to go back and catch up on this discussion, but as always, you are invloved in an interesting disucssion. I usually do not shoot off the hip until I do all the research, but your response provoked some thoughts I wanted to share with you.
100 WPH throughput Steppers or scanners.
You are covering twice the area on the wafer so naturally things would go slower. However, people are getting hung up on throughput when the name of the game is really die output per wafer fab. If no other variable affect die output, a 300mm litho system that runs at the same throughput or slightly lower throughput, still cranks out more die from the factory if you are lithography constrained. The other critical functions are wafers moved per square foot of precious fab space and the entrie process cycle time.
Speaking from a great deal of industry experience, if you have the fastest equipment, it does not necessarily mean you have the shortest cycle time to get those deives shipped. It may sound too basic but the people that you are conversing with need to know that as important the throughput of a system is, there are more important over-riding factors.
I worked in a fab with high throughput Ultratech steppers but the cycle times left a good deal to be desired. the factory was (they thought) lithography limited in wafer output. It was not until I was part of a team involved in cycle time reduction did we see the dramtic effects on wafer outs.
Anyway, I drone on. your comments on smaller feature sizes and overall die size is right on the mark. It is a two fold front here. Smaller feature sizes means faster speeds for the most part. However, while we shrink to go faster, we are adding more features to the chip. Everywhere you read about designs these days, you come across the inevitable conclusion that we are moving to "system on a chip" in one form or another. The industry is combining 2-3 chips worth of functions onto single chips. A good example is the microprocessor which keep adding content to the chip archecture.
There is a limit to how small you can shrink a device or a chip. What no one is discussing is the PAD limiting nature of IC. Briefly, when you are done with that tiny chip, you still have to place it in that Gigantic package with the rather large pin outs in the package (ball, pin, etc.) to connect the package to the PC board. However, inside, you have to connect the chip to the package, using wire bonders. Wire bonders are conisdered back end equipment. However, a condition called "Wire Sweep" exists that makes it increasing difficult to get wires from one pad to another in the assembly process. Pad pitch is how they control this. The bonding pads are limited to a certain distance from each other. Also, the size of the bonding pads are huge to accommodate the lack of precision on the bonders. After all the cips are sawed and diced with less precision that we align a wafer in the litho step. So, as we shrink chips to get more die on a wafer, they are also placingmore function on the die, keeping die size close to what it was before, or slightly larger.
I have a tendency to agree strongly with you one this subject. Device structures, geometries, and technology may be shrinking from 0.25u to 0.18u or below, but device complexity is rising as rapdily to negate many of the perceived die shrinks. With that said, we have to concede the point that certain companies with commodity products like DRAM, do go through die shrinks of the same designs to increase die output. They do use an existing design for a DRAM and then do a shrink of the design, work out a few process bugs, and then manufacture exxentially the same device in a smaller footprint. This is a normal course of business for commodity chips IF IT MAKES sense, i.e., if it really is cost effective. Sometimes it cost more to implement the new technology required to shrink a device, resulting in higher costs until fully depreciated.
expansion of wafers size (currently from 200mm to 300mm)
The bridge lithography tools will do the larger area, but with about half the throughput. To get to 100 WPH on 300mm, you will need faster wafer stages and higher frequency/power lasers, which cost more money. Most 300mm fabs will have brand new scanners with brand new lasers, so I can't view 300mm as a negative.
This is a juggling act and your response has a greater deal of factual merit. However, the same result could be gotten by faster imaging materials (photoresist). There is a fine line in lithography like in cooking. Baking a cake at 350 degrees for 1 hour is not the same a baking it 1400 degrees for 15 minutes (4 times the heat at 1/4 the time). Heck, I'll bet that 1400 degrees at 5 monutes would be jst as big a disaster. There is a chemical reaction that take place as part of the exposure process and while doubling the laser output might seem like it would reduce the time by a certain amount, there are other factors to consider. Speedng up the photo exposure process does you no good if your devloper process is slower that the exposure process. So, in essence, you accomplish nothing if your throughput of expose and develop are out of sync with each other. Then off course, you have the pre and post conditioning bakes to deal with.
Do not misinterpret me, I agree that more powerlasers are on the way to help throughpurt of these steppers, but then you need to make sure the subsequent photo equipment can handle the increased throughput. We also need to make sure the photochemistries involved are as controllable and these high ere energy sources command. Higher laser output might require higher shutter accuracy. Not that all this cannot be accomplished, it is just not a simple task.
Now for the clincher. Even though steppers are extremely expensive (DUV is now around $7M give or take), larger lenses, thereby larger reticles, etal. might be needed to increase throughput as well as step and scan. The larger the field size of a reticle the more expensive it will get and the more chances for defective portions of the die will exist. the reticl makers have direct write systems that have their own magification, resolution, focus, registration, overlay, etal. issues they must deal with. Variables on variables makes all of this a complex process.
Historically, most wafer fabs are photo limited (limited by the throughput of the steppers). However, there has been a shift towards other equipment as the number of metal and insulator layers increased. CMP, Metal Deposition, metal etch, and even CVD deposition tools are becoming the bottle necks due to the higher process control and maintenance on this equipment.
Bottom Line: as concerned as we might be about stepper throughput, we should be more concerned about the quality of the throughput and make sure it matched with the finite limitations of the other parts of the litho process. buying an extra expensive stepper to keep up with develop, does not take up too much extra floor space. Making sure that the present stepper output is not a cycle time limiter is the next important thing.
Very few manufacturing organization have accomplished what I have done over the years to dramtically reduce cycle time and thereby increase wafer output from a factory. You are only as fast as the weakest (slowest) part of the entire output process. What we fail to see and what people fail to do, is adequate capacity and cycle time modelling. We assume that the steppers are the weakest link, when they may not.
Just as a matter of record, in my last job, we were able to
bring our cycle time down from 6 weeks to 4 weeks by correcting certain logistics issues and eliminating both an oxide etch and photoresist strip bottleneck.
increasing through put from each stepper(lasers).
Again, higher throughput does not come for free, you need higher frequency/power lasers. That is one of the key selling points for 2KHz lasers, which sell for more money than 1KHz lasers.
As I said, higher lasers will require the same resist or faster resists but could put other operations out of balance. Much of the throughput issue in steppers is not exposure related. Focusing the wafer, stage speed, stepping over and aligning the layer all take up more time than the actual shutter clicking open and shut. Throughput is increased by speeding up this operations. The old "gotcha" when sales people talked about throughputs that were real high were on blind stepped wafers. No aligning, no focusing, just stage movement and exposure. So, the person who is looking for higher throughput may not necessary get the production throughput desired. they have to be very careful and specific. I do not know of anyone yet that is guaranteeing throughput at all process levels.
Production throughput is also determined by whether blind stepping, global alignment, or site by sight alignment is used. Both are worlds apart relative to throughput. Next is field size. Some companies advertise a certain field size but that is sometimes reduced in production for resoultion or registration gueardbanding. More times than not, it is limited by die size. If I just pull out one number, 44mm x 44mm workable field size, and all other variables beeing equal, your throughput is determined by how much of that field size is used by the reticle. If the reticle has a 30mm by 30mm die on it, it will step on a wafer more times than a device that is 35mm x35mm. So throughput is again affected by reticle die size.
and believe it or not, more can be done with throughput and die output if the DAMN designers would just sit down with the photo process engineers and design chips within an optimzed die frame. We step a design on a circular wafer and not a rectangular wafer. By optimizing the die for the stepper, you might be able to get more die on a wafer and step less times in areas that just won't yield. That might seem silly (stepping less) but even though you step a filed on the wafer cleanly towards the edge, there are edge effects that prevent the die from yielding. So, with proper cooperation we might step less die and get more yielding die.
This is not theory. This was force fed to a few designers by myself over the years with the help of a product engineer who annual reviews and salary were dependent on the yield of his product. For his last 2 years, he received the highest annual increase in his department, some of it due to this.<GG>
All these three trends results in lower growth (than other wise) for steppers and scanners compared with back end equipment which do not have ALL these disadvantages.
This is the first I've heard of this. "Experts" in forecasting the industry see large growth in lithography from a revenue perspective. In fact, the forecasts I've seen indicate lithography will comprise more of the cap. ex. pie. Units tend to be about the same at peaks, but the tools and their components get more expensive over time. This has been the case and there are no indications it will not continue to be the case. If you are predicting a reversal in this historical trend, on what do you base your prediction?
The writer poses an interesting point but your answer was great. To further elaborate, it is true that the back end equipment is more geared towards the number of die that go through them or the number of wafers that are tested on the wafer. There is not way you can speed up the electrical testing opf a circuit so a 300mm wafer will take twice as long to test as a 200mm wafer. Actually it is closer to 2.5X since larger wafers have a tendency to yield a higher portion of the die of the wafer. (bigger sweet spot). So, a case can be made for the need for more backend equipment. IN addition to this, as sppeds of devices increase, and we are seeing it now, testers have to be replaced since they can no longer test at the speeds (frequencies) required. I am not a back end person but looking at the technical stuff from TER should help to put this in better perspective.
However, as you state, litho tools are gettingmore and more expensive. My first Perkin Elmer prjection aligners and Ultratech 1X steppers were well under $1 a pop. The first 5X steppers of quality broke through the $2M prcie range. More and more these advanced systems are getting expensive. Keep in mind that as devices shrink EVERYBODY is forced to buy new steppers. they might be able to make due with existing process tools elsewhere, but you are not going to expose a 0.13u feature with an i-line stepper in production.
And in the scheme of things, the high cost of a stepper (resltive to other process tools) is minimal relative to the cost of the die tooling (reticles). It used to be that reticles were $2-$3 K apiece, then $5, and maybe a reticle set cost $20K-$50K per device. With OPC, PSM, and other type of technolgoy enhancements, the yearly cost of reticles over powers the cost of a stepper.
And as we move forward, we will be trying to eliminate steppers or step and scans and go directly to direct write wafer systems, much like how the mask shops make reticles. Direct write ion beam systems exist but throughput is very low. However, it does streamline a great deal of the upfront work for prototyping and mask making by eliminating some processes anda lot of costs. Remember that UTEK has an Ultrabeam which is still an oddity and has not gained much market accpetance. The concept is worth considering. That Ultrabeam could be the prototype of a high throughput wafer writer down the road.
Not being an expert in backend versus front end tooling revenue growth, I might concede that back end growth might accelrated fast than front end. However, both terms encompass a great deal of the processes run. What I will say is that my belief, and industry contacts agree, is in harmony with your opinion. Lithography revenues will be growing solidly in both units shipped and ASP per unit. Other segments of the front end may not do as well as the general back end numbers but lithography will keep pace with the best of the numbers for any backend auite of tools.
==========================
Now for a nice Weekend Whine.
Can any of you fathom why CYMI got dumped on during this correction, along with every other equipment company, with the exception of ASML?
ASML is one of our Top Tier Top 5 Equipment stocks in our RadarView Newsletter. As we took profits in many of the other stocks prior to this correction, allowing us to get back in with nice discounts when we decide, ASML was literally frozen in its price range. If it moved 3% over the past few weeks, we would be amazed.
So, if ASML held its ground and CYMI is the major supplier to ASML (and inventory levels in the pipeline are down), why was our favorite stock here pooped on. I would also think a few lasers got shaked, rattled , rolled, and put into disrepair over in Taiwan.
I would see why other stocks would retreat but not this one. All it does, is allow us to accumulate more cheap shares and ride this baby up agoan for more significant profits.<GGG>
Anyway, I did not mean to be this long, but you always seem to draw the comments out of me<GGG>.
Have a great weekend.
Andrew
avance@radarview.com |
