Gordon Moore hasnt been right

In the last two years, the worlds largest chipmaker, Intel,
had been receiving a fair bit of drubbing from its pesky rival AMD. Now with
clockwork, or shall we say, "Tick-Tock" planning, Intel is back on
track with new products aimed at clawing back its lost turf Patrick P
Gelsinger, Intels senior vice president and co-general manager of Digital
Enterprise Group (DEG), is focused on delivering the companys platforms and
products including servers, business clients, storage, communications and
embedded applications to businesses worldwide.

In an interview with CyberMedia, Gelsinger spoke about
his views on the companys Tick-Tock strategy, why he thinks Gordon Moore is
wrong and a candid assessment of the chip wars in the last few years. Excerpts

Can you elaborate on Intels new Tick-Tock development
strategy?

The idea behind the Tick-Tock strategy is that we can do a new major process
technology every two years. So we are going from 90 nanometer (nm) to 65 nm to
45 nm to 32 nm and with that we do a product that is lined up with each of those
major process technologies. We call that the Tick. These include Core2Duo,
Penryn and others. The primary responsibility of the Tick teams is to make sure
that these processes and fabs are ramped up and running. The Tocks are the teams
that work on new microarchitectures and product innovation. The result is that
every year we have a new product family coming out-tick-tock like clockwork. We
have teams in place operating today to carry this on till 2012.

As we move to Nehalam, we will have 2, 4 and 8+ cores. We will
continue to increase the core count as we go ahead in future. The 45 nm will be
launched on November 12, while the 32 nm products are due in the first half of
2009. Our cadence has been that in the Q4 of odd years, we will launch process
technology and Q4 of even years we brought in microarchitectures. Every year we
bring out a new family.

Gordon Moore predicted the end of Moores law 10-15 years from
now, at the Intel Developer Forum (IDF) in San Francisco recently. When do you
think you would hit the fundamental physical limit?

I have infinite respect for Gordon Moore. However, he has declared the death
of Moores law for two decades. So he hasnt been right.

I liken it to a foggy night where you are driving down the road.
Your headlights are 100 meters in front of you. Beyond the 100 meters, you know
that the road is there, but you cant see it. As you keep going down this road
of technology, we have about 10 years of visibility into the future. So today,
we are about to launch 45 nm; we have shown the first wafers of 32 nm; we have
the first research on 24 nm and prototypes of 17 nm and 14 nm. So we have about
10 years of that research pipeline. Not that we have solved all the problems for
14 nm and 17 nm, but we have work underway that gives us optimism to move to
those generations.

Historically, if we look at where we were at 500 nm, we thought
100 nm looked really hard. But, we passed right through it. Today, we are at 45
nm and breaking through 10 nm, its really hard. We dont know the material
science or the physics to break through that yet. But as we go further down the
road, the innovations that we do today allows us to solve the problems of
tomorrow. There are some very hard problems that we see that emerge around 10
nm. That is what leads Gordon (Moore) to say what he did. I am a little more
optimistic than him because this is invention and you cant put a schedule nor
define the domain for those inventions to occur. We do have work going on some
of the new inventions that we think have to occur such as new transistor pipes
and material types like carbon nanotubes or Silicon nanowires, and also
different approaches to dealing with power and power limitations.

Do you see this kind of ground-breaking innovation already
happening? This year, Intel announced that it is going with Hafnium-based metal
gates and not Silicon dioxide for its 45 nm products.

Yes. Significant breakthroughs are required to get through to 10 nm
products. Hafnium is the biggest change in the transistor in the last 40 years
since Robert Noyce and Jack Kilby made the core invention. This is absolutely
huge.

As part of your multi-core strategy, are you also looking at
special purpose cores to target specific applications?

Certainly. However, we have a very different view from what has been
suggested in the industry. There is a wide variety of people who have talked
about different accelerators, graphics for application development and porting.
We have a different view that people will continue to do most of their
application development on the ever evolving Intel architecture. It is because
it is a programming model with the ability to have software tools, libraries,
rich debugging environment, the installed base, and software development rules.
When IBM announced the "Cell" architecture, they proclaimed that it is
a breakthrough in computer science in 2004. They said it would be used in
servers and workstations. But it was not successful except within the
PlayStation. This was because people could not write programs for this obtuse
architecture.

Our view is that the world of architecture starts with the
software programming model. The most powerful and beneficial thing that we can
do is to deliver a compatible architecture as opposed to breaking the
architecture. That said, we do believe that there are very specific targeted
applications for which we want to embrace accelerators. Graphics processors are
the best example of that today, because you can build special purpose
architectures to work on that area. Similarly, there are specific purpose
accelerators for high performance computing. Our approach is to embrace those
opportunities by offering interfaces into our platforms to enable them to build
on what we do.

In 2004-05, Intel faced some stiff competition from AMD that led
to decline in market share. Are you now in a position to win back the share?

What really happened was that it takes three-four years to develop these
products. So the market situation in 2005 did not occur due to decisions taken
that year, but by decisions made in 2001 and 2002. If we look back at 2000, we
made a decision to use NetBurst (architecture) in 2004 and 05. We essentially
invested in the GHz (clockspeed) centric plan one generation too long. A
fundamental shift occurred in the market place in 2004-05, as Moores law
allowed us to build more transistors than we could cool. So we hit the power
wall. So our products at this time were not as competitive as they should have
been. We recognized this in 2002, but it was too late for us to retool the whole
product line to satisfy that. So in 2003, we moved to the Tick-Tock development
model. Thats when we re-architected the whole development process at Intel.
So with the products since end of 2005, like the Core architecture, Core2,
Penryn and Nehalem, have had superb execution against this new Tick-Tock model.

The market share declined in 2004 when AMD brought their
products into the market place. Now with our new products, we have been
consistently regaining market share. We believe that in a sustained way, our
development model and process technology leadership will allow us to extend our
lead over the industry over the next several years.

We are slowly widening the gap between us and the competition as
we continue to execute that development model.

Priya Padmanabhan,
Cybermedia News

maildqindia@cybermedia.co.in