TSMC's Burn Lin
COURTESY OF TSMC
Lin has been extending the limits of optical lithography for more than
30 years. The list of articles he has written—many of them the first
of their kind on certain topics—reads like a what's what of advanced
litho technology: deep-UV, multilayer photoresist systems, depth-of-field
(DOF) and resolution scaling equations, k1 reduction, optimum numerical
aperture (NA) in projection printing, and many more. But in his role as
senior director in TSMC's micropatterning technology division over the
past couple of years, Lin has been focused on 193-nm immersion, becoming
one of its most fervent champions. What was once a lithographic curiosity
has become the optical extension method of choice, due in no small part
to Lin's scientific and diplomatic skills.
joining TSMC, Lin held various technical and managerial positions in the
microlithography field at IBM. An SPIE and IEEE fellow, Lin has received
numerous honors over the course of his career, including SPIE's first
Fritz Zernike Award for Microlithography in 2004, TSMC's outstanding innovation
award (also in 2004), and Taiwan's outstanding engineer award in 2002.
He is editor in chief of the Journal
of Microlithography, Microfabrication and Microsystems. He has also
been president of his own company, Linnovation, since 1992.
spoke with Lin during the SPIE Microlithography week in San Jose in early
March. Our conversation covered many aspects of the IML phenomenon. He
recalled how the presentation he gave at a 157-nm workshop in mid-2002
initiated the urgent discussions that eventually led to a sea change in
the industry's lithography roadmap. He
recounts the difficulties he faced convincing the chipmaker and
supplier companies to join the immersion camp, as well as the continuing
evolution of the technical challenges posed by the wet-litho approach.
After detailing some of TSMC's own IML roadmap, he ponders the technology's
I'd like to start by going back a few years. I can remember when I first
heard you talk about immersion you were a voice in the wilderness, and
there weren't that many people paying serious attention to the idea or
thinking that this would be a production-worthy technology, or even be
used. Can you look back and discuss some of those days, and then when
you saw the idea really start to take hold?
The earliest I wrote about it and that I presented about it was in 1987.
I mentioned immersion as a way, probably as a last means, to extend the
depth of focus. But then there was no need for it. Now when you look at
the roadmap, it goes very clearly from 193 nm to 157 nm and EUV and so
forth; it's very continuous and there's no gap where you can insert immersion.
For example with 157, the only place may be immersion after 157 and so
forth. But I noticed that if you tried to do immersion 157, it's very,
very difficult. Actually the technology itself in the dry form is already
very difficult. So what's the way out of it? Maybe we should do immersion
earlier than that. So I looked at 193, and then I looked at the properties
of water and all of a sudden, when I did some calculations, it turned
out that wavelength is equal to 134 nm, even shorter than 157. So that
was the key point that convinced me that we should go 193 immersion instead.
Because it's easier and the potential is higher and so forth. And I brought
this point up at a 2002 meeting. It was actually a 157-nm workshop. I
think I was the only invited speaker in that meeting to talk about immersion,
and people thought I was going to talk about 157-nm immersion. I stood
up and I told them, "Hi guys, unfortunately I'm in the 157-nm workshop
but I don't think it's going to work. It should be 193 immersion."
And I explained why. I had many slides explaining, mainly from that 134-nm
concept. I think at that meeting I woke a lot of people up. When they
realized the potential of 134 nm, quite a lot of people woke up.
reaction was that after my talk people were murmuring, discussing, debating,
whatever, on whether 193-nm immersion was going to be usable. I don't
think too many people remembered they were in the 157-nm workshop. Then
Sematech had this very urgently formed workshop, an immersion
workshop, that I think took place either December that year or January
the following year. In that workshop, we laid out many things that we
should work on to make immersion more viable. From that point on, the
momentum just kept building.
other obstacle actually was not technology, but it was financial or economic.
Because a lot of companies have invested in 157-nm technology, they don't
want to give it up because first they want to recover their investment
before doing something else. So there was a lot of inertia there.
Internally, how much work did you have to do to convince your bosses and
everyone else that TSMC also needed to participate in this and to convince
them of the value of immersion 193?
In TSMC it was not too difficult. Since people sort of look at me as a
leader in lithography, they had a lot of trust in my judgment. One thing
that really surprised me was that even internally I was only starting
to advocate immersion lithography. And our company president, in one of
our outside lithography technology symposia, announced that TSMC is going
to go for immersion lithography.
He didn't tell you this before he made the announcement?
No! (laughs) So that was a pleasant surprise to me.
What do you see as some of the key points of the evolution and the progress
of immersion, from the point in time of that emergency workshop forward?
One key point is that we had to convince the equipment suppliers and the
material suppliers to change their roadmaps. Of course, it was not easy.
Fortunately, TSMC is a big customer of ASML, but still it was not easy
because we had a lot of discussions between ASML and us about going into
immersion. The argument I heard most often, almost from all suppliers,
is that they had spent so much money on 157, hundreds of millions of dollars,
so I asked them, "Are you still spending it?" They said, "Sure." So, I said, "The sooner you stop,
the sooner you stop wasting your money." That was my counterargument.
in order to convince the suppliers, actually I had to line the users up.
It was no good if TSMC was the only company that said they needed that.
All the users and all the infrastructure would need to be there. So I
did have to spend a lot of time talking to other semiconductor companies
to convince them that that's the way to go. Of course, it was not an easy
task. But then we got the positive feedback [from one company], won another
company over, which helped me to win over another one, and also it helped
to convince the suppliers and so forth.
What about the list of challenges that was to be addressed? Looking back,
when you examined that list when it was first put together, how has it
turned out so far, comparing what you thought then to what the research
actually found out after the work started?
There were several points. First of all, we had to look at the long-term
potential. People were worrying that if we go to such a high NA, would
we lose the advantage of immersion lithography. Then the ultimate potential
will not have been as high as I thought. I did have to spend a lot of
time trying to get a satisfactory answer, even for myself. When we go
to a larger angle, we do lose some contrast due to polarization dependence,
straight line, and so forth. So it's a matter of how I quantify those
things. I spent a lot of time, and fortunately I got the answer. I thought
up the methodology to simulate those effects, so I did a very thorough
simulation and looked at the future of the technology to see how far it
could go. That was good.
other thing is that people always worry about depth of focus. And depth
of focus is always a very difficult term to define. I think because of
immersion I got to straighten out several points about depth of focus.
I also used it to clarify and make sure that I'm comparing apples to apples.
Of course, we're always worried about bubbles, defects, and whether we
can keep the fluid homogeneous enough, whether it would introduce any
aberration in the optical system and so forth. So those things came one
to now, defect is still an issue. We're still studying, trying to understand
where all those sources of defects are and trying to eliminate them one
by one. I think bubbles are a much better picture. Since day one, we have
known that bubbles can be dissolved in degassed water, so, first of all,
the bubbles that are smaller than 50 nm wouldn't matter even if they were
there. The intermediate-sized bubbles, if they can be dissolved very quickly
in water, shouldn't matter either. But there will be sizes that are large
enough that don't dissolve in water quickly and that would influence the
it doesn't turn out to be that bad. I did put up some analysis on where
the bubbles come from and how you can deal with them one by one. I think
right now, the bubbles come from the design of the immersion head, so
to speak. If it's not hydrodynamically designed, and not sufficiently
hydrodynamic, there's a chance that you create turbulence, create bubbles.
Also there's an issue of the surfaces matching, like whether they are
hydrophobic, hydrophilic, and so forth. One really has to make sure and
sort those out to minimize the bubbles. And it can be process induced,
completely independent of whether we're using an immersion tool or not.
We can create identical types of bubble from a dry tube if we don't carefully
control our process. So we gradually understand all this, and again we
think that the bubble issue is easier than the defect issue.
With the defect issue, obviously, certain questions won't be answered
until there's some production-sized quantities of wafers going through
the tools. It's not something you can do just a few shots at a time.
What about the lens. What kind of progress do you see the lens making,
in terms of hitting 1.0 NA?
The lens is probably the easiest part, except that it's very expensive.
It's not the cheapest, but it's probably countable. You can sort of predict
how far it can go. I think the concern is whether one can really go to
a lens beyond 1.35 NA. The reason being, at that NA, it's limited by the
refractive index of water, so you almost have to raise the index of the
With nonaqueous or aqueous fluids?
You can have a high index with an aqueous solution.
Right. I've spoken to resist people, and they have this mysterious HIF1
or something like that. When I ask them, what's the secret sauce, they
say, "We can't tell you that!"
Well, that, of course, is key: Who holds the rights to that liquid, holds
the future! Well, actually, there are two stages of the liquid. If the
liquid goes to a refractive index of about 1.55, the requirement is only
that you have that index and reasonable transmission may be comparable
with water. That would be sufficient. But let's say a lot of people claim
they have 1.6, 1.65, whatever; once it's higher than 1.55, then you want
to worry about the transmission of that fluid. It has to be higher transmission
than even water. Therefore the requirement is twofold: refractive index
and transmission, if you're talking about an index higher than 1.55.
What is your view on the progress being made in the area of immersion
We, as a semiconductor manufacturer, being concerned with cost and cycle
time and so forth, we definitely do not want to see a topcoat there. We've
let our suppliers know very clearly that we don't want a topcoat. In the
interim, we will do the topcoat because if we don't know enough, we don't
want to damage the lens or do something crazy. So in the short term, we
live with a topcoat; in the long term, we don't want a topcoat. But either
way, we have seen very promising materials.
What about line-edge roughness?
The line-edge roughness shouldn't be worse than dry. It's universal.
And what about CD uniformity and the like?
CD uniformity is a good topic. We just published a paper in the symposium
that talks about our CD uniformity results. Two things: One is we compared
the CD uniformity in the dry and immersion tool, and we did it in focus
and out of focus. The in-focus one is very comparable because we are comparing
with the same NA. So you don't expect anything to be better than the other.
But when it goes out of focus, the CD uniformities for the immersion tool
are much better than those from the dry tool. Again, that is quite obvious
because the immersion process has larger depth of focus. So if you're
out-of-focus at the same defocus, you're expecting to get better CD...
Those go hand in hand.
Yeah. We also had an interesting way to compare the depth of focus. We
qualified the depth of focus using yield. We exposed a wafer that's going
for yield test…we exposed an exposure and defocus matrix. And then we
[analyzed] the wafer, once with the dry tool and once with the immersion
tool. Clearly we saw that the high-yield region was much larger for defocus
with the immersion tool.
What about the cost side? What are you seeing so far in terms of what
you expect with the cost of ownership and productivity of the immersion
We are seeing the potential throughput of the immersion scanner approaching
that of the dry scanner, even though now it's not the case. But we're
not seeing any particular problem that [would prevent it from] approaching
that output of the dry scanner. But the tool itself is a little bit more
expensive. It's more expensive than we had calculated. If you think about
it, there's no expensive extra part in the scanner. The lens has the same
elements, and they cut the last element by half and fill it with water,
so big deal. There's the water head and so forth, a few mechanical, machined
parts, and there's no tremendous high precision required in those. We
don't see a reason for extra cost. But it is more expensive. I think probably
because of [the suppliers] trying to recover the development costs of
immersion and maybe also trying to recover the development costs of 157!
the other cost factor would be when you push to a very high NA. The lens
itself would be more expensive. Ounce for ounce, a 0.9 NA lens would be
much less expensive than 1.1 or 1.3. In the good old days, the same thing,
right? A 0.75 was more expensive than 0.65 and so forth. The other thing
is that when you go to very high NA, and you want the extra advantage
of polarization, then of course that adds to the cost.
When the first production tools come in, they will obviously be used for
critical layers. In a typical fab, how many immersion tools do you think
you will need?
I cannot really tell you very precisely, but I can give you a feeling.
So a typical fab covers about one or two generations. And the output of
those fabs accounts for, let's say, a few tens of thousands of wafers
per month, OK? Then you can work out the throughput of the scanners and
so forth. Thinking about scanner throughput, it's about 100 wafers per
hour, but it has to go through the scanner many times because there are
many layers. So let's say typically if you have to go through 30 layers,
some of them are critical and some of them not so critical, so there will
be some less-expensive tools to go through and some more-expensive tools
to go through as well. Also, on some of the not extremely critical layers,
we will want to use the dry tools, because they are less expensive. So
typically the number of scanners in a fab would be in the tens.
So let's talk a little bit about your company's roadmap and how it looks
at this point. What are the various insertion points or key development
Right now, we're using the 0.85 NA tool to develop 65-nm manufacturing
technology. Hopefully we will have it in a very dependable operating mode
by the beginning of 2006. That's our target. We hope to have debugged
everything, have a very dependable process by the beginning of 2006. And
from that point on, if that is our yardstick…for the 65-nm node, then
let's say if we're on a very aggressive schedule for Moore's Law, that
would be a two-year [cycle]. But in reality, you never know. If the demand
is there, we may speed it up. If it's not there, we'll slow it down. But
let's take two years as a norm. Then the 45-nm node will be beginning
of 2008, then [32 nm in] 2010, and so forth.
So you're pretty much sticking to that as long as you hit your target
for the beginning of next year?
And ASML's own roadmap obviously has to be in step with yours.
Are you confident that they're going to be able to hang in there with
I think so...[at least] up to the 32-nm node.
So, how far do you think immersion can be extended?
We definitely can count on it for the 32-nm node, which is equal to 45-nm
half pitch. When we go to 32-nm half pitch, or 22-nm node, [that's] a
scary node. The 193-nm immersion will still be possible but there will
be more restrictions on the design, the throughput will be slower because
we may have to go to double exposure, and so forth.
of course, we have to see how EUV and some of the other NGLs are coming
There are some people who question whether it can be extended to 32 nm.
You seem very confident.
Getting to 32 is fine, at least from my simulation. I've been very carefully
simulating. That means I have incorporated all those stray lights, everything,
in my simulation. We do put some specifications; for example, at that
time, we will need so many nanometers of depth-of-focus tolerance from
the tool. And we will need, maybe, 100 nm of resist thickness. So the
resist supplier has to give us 100-nm resist, but if they say they can
only give us 150-nm resist, then we're out of luck. So everybody has to
chip in a little bit.
How much of your time is taken up with the immersion program?
At this moment, not that much time. Mostly, my people are spending time
have put quite a lot of people on it. We have the equipment people, we
have the process development people, we have the metrology people, the
defect team, and so forth.
How big is the team?
Oh, I really haven't counted. At one time, there was a reporter who wanted
to take a picture of the people who were related. It was about 15 or something,
but I'm not sure. There are people who work on it who are not full-time
and so forth. It's not easy to really count.
Do you have some final thoughts on any short- or longer-term challenges
that need to be overcome so that immersion can move forward at the rate
that you hope for or that the roadmap calls for?
It's similar to the problem statement. We need to get rid of the particles,
and I'd like to see a resist that doesn't need a topcoat, a very dependable
resist. And, slightly longer
term, I'd like to see that magic [high-index] fluid.
Search | Current Issue | MicroArchives
Buyers Guide | Media Kit
Questions/comments about MICRO Magazine? E-mail us at email@example.com.
© 2007 Tom Cheyney
All rights reserved.