The tensions are growing fast as the industry confronts the issue of
450mm. Some believe it is absolutely necessary. It is scheduled in the
ITRS roadmap to be ready by 2012 and believed essential to keep chipmakers
on Moore’s cost-performance curve. Meanwhile, having just come off 300mm,
others believe it will be a disaster and are fighting tooth and nail behind
the scenes to stop it. Today, they represent the majority. It’s not just
the equipment-makers who are in this camp. It is also the majority of
chipmakers. Many believe Moore’s law must slow; that staying on it is
unaffordable; that 450mm will lead to even worse scale problems than 300mm;
and that if it happens the result will be more consolidation and even
In favor of the people arguing for it is the fact that every wafer size
jump has met with resistance based on the same reasoning. But there are
reasons to believe that this time things will be different. The industry’s
growth has slowed. Plus, 300mm proved to be incredibly painful for the
equipment industry. These are differences few argue with.
One of the problems in addressing these issues is that there has been
little data to make decisions around. As a result, emotion has ruled.
So I have spent the better part of this year mining our archives for historical
data on what it cost to develop the equipment base for the various wafer
sizes. Hopefully, this will shed some light on the issues and allow the
industry to engage in a more productive discussion.
The data shows 300mm equipment development was indeed very expensive
for the equipment industry. (See Figure 1.)
In fact, at $11.6 billion, its cost dwarfs that of 200mm, towering over
the smaller size by nine times. In contrast, both 200mm and 150mm were
only about double their prior generations. RD&E spending has averaged
16.2 percent over the last 10 years, which means that it will take a total
of $72 billion in equipment revenues to payback the amount suppliers spent
on 300mm. This will be achieved by 2007, which doesn’t seem unreasonable.
However, one must remember that the first 300mm development costs were
incurred in 1993. This is a 14-year payback cycle. If we start in 2006,
the simple math of adding 14 years out puts the 450mm payback in 2020.
Most of the industry’s senior executives will be retired by then. The
stock ownership of these companies will have turned many times, as will
the financial analysts that cover the companies. The time-tomoney in other
high-tech markets is far shorter. This makes such a long endeavor difficult
to achieve from a social standpoint.
Let’s change perspectives. The industry was much larger when 300mm came
to be. So, one must look at this relative to the total that was spent
on development over the periods when these wafer sizes were developed.
This picture is not that bad. (See Figure 2.)
At 18 percent of the total RD&E spent by the industry over that period,
it does not seem inordinate. Though it was significantly higher than the
15 percent spent on 200mm, spending at this level does not seem to be
something that would break the equipment-maker’s bank. But since this
occurred at a time when the industry was achieving phenomenal growth and
300mm was spread over more years, the costs were held to a reasonable
proportion of spending.
The problem for equipment-makers is that 300mm fabs are not that more
expensive. The size of wafer size development costs to that of a new fab
was far higher than in previous generations. (See Figure 3.)
At 3.9 times the cost of a new 300mm wafer fab, the relative wafer size
development costs for the equipment were far higher. Chipmakers got a
great deal on 300mm, as equipment-makers were spending twice what they
had to spend to field 200mm gear. But, since the sustainable return on
investment is fixed by financial markets – and is a function of equipment
prices – this ratio should have been somewhere near 2.0. In other words,
300mm equipment prices are about half what they should be in historical
terms. This is the source of a deep and festering wound for equipment-makers.
It is a wound that has led to poor profitability throughout the supplier
base and could put it in jeopardy over the longer run.
Because 300mm is so productive, fewer fabs are built. Since the fab costs
are relatively the same and equipment purchase amounts similar, equipment-makers
argue that they got a raw deal by not sharing in the benefits 300mm brought
to chipmakers – and it is a benefit they made possible. Had they not developed
300mm, they would be selling twice the equipment and thus have twice the
revenues they have today. Of course this is a static analysis. Had 300mm
not been developed and the economic benefits accrued to the end users,
it is likely semiconductor growth would be lower and thus total equipment
demand lower. But, it probably would have still been a net gain for equipment-makers
over the deal they got with 300mm.
Another way to look at this is to compare the relative areal cost of
wafer size development costs to that of fab costs. (See Figure 4.)
What you can see is that in dollars per square centimeter, the costs
for both rose steadily before 300mm. So, both chip and equipment-makers
were sharing the pain. With 300mm, the trend reversed for chipmakers –
but not for equipment-makers. There are two conclusions to draw from these
figures. One is that the development costs for 300mm was unacceptable
for equipment-makers. The gap between these costs means that payback was
stretched out inordinately long. Development costs should have followed
fab costs down. If not, that leads to the second conclusion: Fab cost
per square centimeter for 300mm should be somewhere between $15 and $20
to pay back equipment development costs in a reasonable time. That is
two times where they are today.
Looking at it all, it is easy to see why the equipment industry is violently
opposed to 450mm. For it to happen, the trends and the conditions behind
them must change. One clue to how this could be changed is in how the
development costs for 300mm were spent in time in comparison to the other
wafer sizes. (See Figure 5.)
Prior to 300mm, wafer size transitions were relatively smooth. 300mm
came with two peaks of spending that were spread out over 11 years. Some
argue that this happened because of the deep downturn in 1998. However,
there was also a deep downturn in 1986 that had no impact on 150mm development
spending. We also know that the deep downturn of 1975 did little to slow
the progression of 3-inch wafer development. At the same time, the mild
downturns of the early eighties and mid-nineties had little impact on
wafer size development spending by the equipment industry. The biggest
difference with 300mm and other generations was the lack of bridge tools
and the presence of consortia management of the transition.
It is hard to come to any other conclusion than that the primary reason
for the high costs incurred by the equipment industry in developing 300mm
is largely attributable to the mismanagement of the transition by the
various consortia involved. First, there was the insistence that there
be no bridge tools. The resulting co-development of different platforms
for 200mm and 300mm escalated costs. Second, the insistence that early
generation 300mm tools be targeted at 0.25 micron, while 200mm tools in
development are targeted at 0.18 micron, ensured there would be no market
for the first generation of 300mm equipment. This resulted in the waste
of precious development funds. The smart equipment suppliers responded
by giving lip service to 300mm and focusing their money on 200mm. Their
resistance also doomed this early generation of equipment. Third, the
insistence that a complete 300mm tool set be ready by 1997 was too short
to do any reasonable planning. Ultimately, the industry had to backtrack
to develop all the automation requirements that were discovered later
The blame for the failure can be placed in many corners. But it is the
consortia that took the responsibility for leading 300mm. Instead of leading,
they blindly followed the interests of their member companies. In so doing,
the industry was set up for failure from the start. Part of the problem
here is that these consortia do not have any outside board member of any
influence. But in the end, the consortia must suffer the blame, and hopefully
learn from it for the next round.
The reason for the double peak can largely be blamed on the toxic combination
of the customer-inspired rush, the lack of bridge tools, and supplier
resistance. The lack of bridge tools changed the market dynamics. Instead
of being incentivized to be first with 300mm, equipment suppliers were
incentivized to be last. Without a complete 300mm tool set, none would
be sold. When the lithography equipment suppliers decided not to play,
the first generation of 300mm was doomed. When the downturn hit in 1998,
no one was buying and there was no way customers would have bought an
incomplete set of 300mm equipment that was a node behind. 300mm would
not become the standard until 90nm, fully three nodes from the initial
target set by the customers. So there is a lot for the industry to learn
and comprehend if 450mm is to be successful. Most of the responsibility
for this lies on the shoulders of the chipmakers. They must shoulder this
responsibility to convince equipment-makers to come along – or 450mm will
There is a silk lining to this sow’s ear. Had it been properly managed,
300mm development costs would have been far lower. If we can learn from
our mistakes and achieve this as an industry, 450mm may indeed come. One
thing about our industry: it is incredibly good at learning from its mistakes.
Plus, as Bob Noyce once said, “Understanding that there’s a problem is
95 percent of the solution.”
That leads us to the difficult questions of how feasible 450mm is from
a financial perspective. As well as what is a likely timeframe for its
appearance? Is 2012 feasible?
Let’s start this by looking at the escalation of equipment costs. At
nine times that of 200mm, a simple extrapolation to 450mm leads to a number
of $102 billion in equipment development costs needed to make the transition.
But if the acceleration in development costs from 200mm to 300mm continues,
the number escalates to $917 billion. For all intents and purposes, that
is a TRILLION DOLLARS! Clearly that is unattainable, so at a bare minimum,
the number must be curbed to $100 billion.
Yet, the $100 billion figure is also unaffordable. As it turns out, total
RD&E will only accumulate past $100 billion between now and 2013. That
is based on 1) our current forecast through 2010; 2) assuming that the
industry grows 6 percent annually after that; and 3) that R&D expenditures
are kept at the 10-year average of 16.2 percent. That means that no more
than $20 billion can be spent on 450mm wafer size platform development
and it probably should be kept closer to $15 billion. That allows for
less than a doubling of development costs between wafer size generations.
Worse, this is an optimistic picture. Assuming that RD&E will continue
to be 16.2 percent of revenue is highly questionable. It was spent in
an environment where the assumed growth of the equipment industry was
18 percent. But it actually ran closer to 6 percent. If we cut the spending
rate in half because of the lower growth, the $100 billion will not be
achieved until 2019. If we cut it to a third, it will take almost until
2025. So, we must do something significantly different to lower development
costs. We must also raise the assumed growth in the equipment market significantly
to have 450mm ready by 2012.
Let’s look at the revenue picture. (See Table 1). When 300mm finally
came on line in 2004, the equipment industry was almost four times the
size it was when it started.
This was far greater than its growth in any other wafer size transition.
But, it had taken 11 years to get there. If we start on 450mm in 2006,
we must be finished in six years to hit a target of 2012. That rolls us
back to the time it took for 150mm. The raw trend puts the time needed
at 15 years, which is back in the 2020 timeframe. Nevertheless, with expected
revenues of $91 billion in 2012, we are almost close enough to double
revenues, which will actually occur in 2015. Growth has to be significantly
greater (15 percent) and start in 2007 to meet a revenue multiple of two
to afford 450mm by 2012. The problem is that if the equipment industry
growth is upped, the capital spending ratio escalates out of control,
unless the chip industry follows. This would be unrealistic.
Let’s turn to what the equipment industry can realistically afford to
spend on 450mm development between 2006 and 2012. If the RD&E rate of
spending gets cut in half, then $41 billion will be spent between 2006
and 2012. If 15 percent is allocated for 450mm platform development, the
amount available will be $6 billion. That is more than the $5.2 billion
that would be needed, assuming 300mm had been well-managed and achieved
two times growth in development costs and then another two times growth
was achievable for 450mm. If the spending rate is not cut in half, which
is likely, then almost $12 billion will be available – the amount actually
spent on 300mm. It all depends on controlling development costs.
Hopefully, you’re still awake after all this onslaught of numbers. Let’s
conclude with what are the three most likely scenarios and what we have
to achieve to meet them.
- 450mm never happens: Unless the dynamics are changed and the
equipment industry is given some incentives, this is very likely to
happen. It has the advantage of inertia, or lack thereof. The equipment
suppliers need to do nothing to achieve this scenario. For the chipmakers
to change this they have to change the incentives for equipment suppliers
– who now see no gain in profitability, fewer units sold, and lower
potential revenues in developing a larger wafer size. The financial
metrics are all wrong for the suppliers. The problem with this scenario
is that it slows Moore’s curve, which ultimately slows industry growth.
It could put us all into a downward spiral.
- 450mm comes between 2020 and 2025: I believe this is the most
likely scenario. There will be plenty of funding and time to make it
happen in this time frame. It allows for errors of management. History
has shown that there is always great resistance to larger wafer sizes,
which is always overcome. In the end the biggest customers get what
they want. To a great deal, it is the resistance that stretches out
the development timeframe. Another reason for believing 450mm will be
successful is that all other industries pursue a technical path at least
one node too far and then back up. So far it looks like 300mm is successful,
meaning there is at least another wafer size coming.
- 450mm comes on schedule in 2012: I believe this is the least
likely scenario. Like plans for 300mm to arrive in 1997, in my opinion,
this schedule is unrealistic and is being rammed down the equipment
industry’s throat. The equipment industry is resisting covertly, as
well as many chipmakers who are choking on 300mm. It is the resistance
that makes this scenario unlikely. It is achievable. But both chip and
equipmentmakers must work together in ways closer than they have ever
done before. This is unlikely because of current day buying tactics.
The advantage to this scenario is that if the time needed to develop
450mm is trimmed back, development costs will be trimmed as well. Working
together efficiently, could save billions in development costs. The
other way to pull this off is for chipmakers to go it alone and develop
their own equipment. This is already happening with Samsung and has
been in place for some time. But even they would be hard pressed to
develop an entire equipment set. Remember, it really won’t be any cheaper
for them when the opportunity costs are factored in.
Looking back, the range of possibilities here is incredible. It reminds
me of the gamble made by the researchers who set off the first atom bomb
in New Mexico, who bet on one end that it would be a dud and on the other
that it would ignite the earth’s atmosphere and incinerate the world.
If we do everything perfectly, it is reasonable to expect 450mm development
to cost the industry less than $10 billion. At the other end of the spectrum,
if we get it really wrong – as wrong as 300mm – the cost could be as high
as $1 trillion. As with the events in New Mexico, the result will probably
fall somewhere in between, and the world will never be the same.
There you have it. You’ve got the data and my opinions. Now it’s time
for the industry to move forward.