SI
SI
discoversearch

We've detected that you're using an ad content blocking browser plug-in or feature. Ads provide a critical source of revenue to the continued operation of Silicon Investor.  We ask that you disable ad blocking while on Silicon Investor in the best interests of our community.  If you are not using an ad blocker but are still receiving this message, make sure your browser's tracking protection is set to the 'standard' level.
Technology Stocks : Intel Corporation (INTC) -- Ignore unavailable to you. Want to Upgrade?


To: steve harris who wrote (109074)8/31/2000 10:48:55 AM
From: Joseph Pareti  Respond to of 186894
 
Is IBM ahead of Moore's law ?
Interesting article - see below -
question : what is the cost benefit associated with the size reduction, e.g. 0.18 u ----> 0.10 u

IBM taking Moore's Law by the horns

By John G. Spooner, ZDNet News
August 11, 2000 5:45 AM PT

New technology from Big Blue will boost Moore's Law -- but can
the 35-year-old tech rule of thumb keep up with the times?

New technology breakthroughs from IBM Research promise to extend
the reach of Moore's Law, the chip industry's most closely held
measure of performance.
Moore's Law, an observation of Intel Corp. (Nasdaq: INTC) co-founder
Gordon Moore, states that the number of transistors in a processor
will double every 24 months. Despite falling under scrutiny from
time to time, it continues to accurately reflect the progression
of chip technology and become associated with increases in processor
performance and complexity. Moore's Law in recent years has shortened
to reflect a doubling of transistor count and performance every
18 months.

IBM's latest breakthrough, a new chip transistor design known
as V-Groove, will allow the company to stay ahead of the curve
of Moore's Law 15 to 20 years in the future, should it find its
way into production.

Less than 0.01 micron

V-Groove transistors are capable of scaling to channel lengths
of 10 nanometers, or lengths of 0.01 micron and below.

Channel length represents the distance electricity needs to travel
through a transistor; shorter transistors lessen the distance
traveled, delivering greater performance. Right now, IBM (NYSE:
IBM), and the chip industry as a whole, is at 180 nanometers,
or 0.18 micron. It had been widely believed that 20 to 25 nanometers
were a hard stop.

"Using this simple technique, we can get channels as small as
10 nanometers or smaller," said Phaedon Avouris, manager of nanometer-scale
science at IBM Research.

IBM says its V-Groove technique improves on current photolithography
manufacturing techniques, which project an image of a transistor
onto a chip, then physically remove excess silicon. V-Groove,
in addition to lithography techniques, uses chemicals to create
an anisotropic chemical reaction. That reaction burns away silicon
faster downward than side-to-side, creating the namesake V-Groove
channel. Channels, when produced using this technique, are much
finer -- and therefore help eliminate electrical cross talk,
otherwise known as the short-channel effect.

The short-channel effect -- which causes electrical interference
between transistors located too close together -- had been the
barrier to breaking the 20-nanometer mark. Normally, transistors
switch on and off rapidly. The short-channel effect prevents
them from switching off. V-Groove reduces short-channel effects,
allowing normally functioning transistors to be built on this
small scale.

With help from university research partners MIT and UCLA, IBM
has been able to manufacture test transistors using V-Groove.
It is now moving to create test chips using the technology. IBM
officials say the experiment proves that it is possible to build
chips with channel lengths smaller than previously thought possible.
At the same time it proves Moore's Law is alive and well.

"Overall, we are very optimistic. This is a way of making a device
to study transistors on a very small scale," Avouris said. "We
are not looking for production. We are looking for a meaningful
discussion of the limits of a small device. We'll leave it up
to the engineers to discover the best way to put this into production."

Carbon nanotubes?

IBM won't stop with V-Groove. The company is also exploring changes
in materials it uses to manufacture chips.

IBM's next step is research into developing a method of putting
the technology into mass production. It would be some time, about
15 years, before the V-Groove, or a similar design, would be
used to manufacture chips en masse.

IBM Research, for example, is experimenting with the use of carbon
nanotubes, structures made of rolled sheets of carbon hexagons.
The nanotubes, in one example, could be made into interconnects
between transistors.

The good news is that new discoveries in materials and manufacturing
techniques, such as IBM's V-Groove, should sustain Moore's Law
more longer than previously thought, ensuring its extension for
some 20 to 25 years. Moore's Law, at the same time, should stay
true to its roots of accurately reflecting transistor counts.

Despite the breakthroughs that will carry Moore's Law -- and
chip performance -- forward for several more years, IBM researchers
say that Moore's Law may be losing its relevance as an accurate
measure of performance.

Moore's Law no more?

Researchers say the microprocessor industry is entering an age
where raw performance and simply doubling transistor counts every
18 or 24 months will no longer deliver functionality needed for
applications in Internet or other areas, such as communications
networks.

Essentially, IBM researchers say, Moore's Law has no ability
to predict what features will be needed for future processors.

"My observation is that we're entering an era where Moore's Law
is a less valuable tool to tell you where to go next," said Russell
Lange, IBM fellow and chief technologist of IBM's Microelectronics
Division.

Doubling transistor counts "certainly helps improve technology
... but just merely doing that is not nearly enough these days,"
Lange said.

"The value proposition is moving out into the Net and bandwidth
and changing in many ways," he said. "Certain customers might
not need clock speed. They might, for example, want more reliable
memory technology."

For many consumers, the question of relevance may ring true.
Right now, there is little need to increase performance, as a
500MHz Celeron or Duron chip can run today's desktop applications
quite well, never mind a 1GHz or 1.1GHz chip. Taking the argument
one step further, for those who spend time on the Internet regularly,
it's more important to increase the bandwidth of the connection
than to increase the clock speed of a PC processor.

However, chip makers aren't going to take that as permission
to stand still. The demand for performance is still very much
alive in servers and other infrastructure. This equipment, powered
by microprocessors, provide network bandwidth, therefore contributing
directly to network performance.

In addition, IBM and Intel agree that, especially with faster
Internet connections, software will catch up to and exceed the
capabilities of today's desktop processors, requiring more performance
there as well. The only variable in this equation, they say,
is time.

So what is performance these days?

Despite the changes IBM is seeing, the company is racing ahead
with its research to develop faster chips. However, IBM researchers
say they are also looking at ways to make chips that are more
customized for performing the work that needs to be done.
"It isn't just about scaling anymore, where Moore's Law calls
for chips to become smaller and faster. It's just not useful
to do that," Lange said. "We're spending a lot more time looking
... looking at other design elements."

To that end, IBM is adding to its research new ways of solving
processor-to-memory constructions with new kinds of memory structures.
IBM is also working with new kinds of chip packaging.

"We've been experimenting with connecting a memory chip with
a logic chip," Lange said.

Grafting the two together would increase performance and lower
cost. In other applications, two complementary chips, such as
a radio chip and a digital signal processor, could be sandwiched
together. This direct connection, by itself, would increase performance
by eliminating the need for data to flow over a bus.

Short term boosts

Meanwhile, IBM has begun putting into production a number of
new technologies that it says have the potential to significantly
boost chip performance over the short term.

Most recently, IBM added PowerPC processors with SOI (silicon
on insulator) technology and copper interconnects to its AS/400
line of servers. The SOI technology increases transistor performance
by reducing parasitic capacitance -- a performance hit caused
when electrical current is absorbed by the silicon substrate
that a transistor rests on. Thanks to the insulator, the chip
can push greater amounts of current, which increases performance.
Or, using a SOI, a chip can be tuned for lower power, making
it better for use in portable electronics.

IBM also plans to boost performance of its RS/6000 S80 servers
by adding new PowerPC processors with SOI technology to this
family this fall.

The company has announced a number of other new technologies.
Many of these new technologies will be resident in IBM's Power4
chip. The chip, which features two 1GHz-plus processor cores
on a single chip (a single-chip dual processor), will debut next
year in high-end IBM servers.

IBM recently brought up Linux and its AIX operating system on the chip.

It's these technologies, now in production, that will carry IBM
forward, while its research division tackles the bigger problems
and works to deliver new technology, such as V-Groove, into production.

"A lot of people worry about the end of the industry; how long
can this keep going?" Lange said. "Usually, what folks underestimate
is how energetic scientists and engineers are when it comes to
overcoming hard limits."