[Waverley ARS] The Transistor at 60 - From SMH

[Henrik Stenstrom]

The Transistor at 60

Are we starting to hit the limits of this extraordinary breakthrough?
Beverley Head reports.
 
The cover of Electronics magazine from September 1948 showing Bell Labs
scientists William Shockley (seated), John Bardeen (left, standing) and
Walter Brattain who invented the transistor (inset).

IN DECEMBER 1947, Bells Labs scientists John Bardeen and Walter Brattain
first revealed what would come to be known as the transistor.
They held the future in their hands - a device that would replace vacuum
tubes in 10 years, and 60 years later has transformed electronics.
Inventions change things; great inventions change everything.
That first device was the size of a modern mobile phone. Right now, 2
million transistors could fit in the full stop at the end of this
sentence. Intel has just released its new Penryn processors, which have
up to 820 million transistors, and soon the standard inch-wide
microprocessor will have 1 billion transistors.

Combined with advances in programming, we will see single-chip systems
such as hand-held translators, in-car collision avoidance systems, and a
raft of devices that react to voice and touch.

It is extraordinary to reflect on how far the silicon revolution has
come in such a short time. Soon after Bardeen and Brattain made their
breakthrough, William Shockley, also at Bell Labs, invented the first
semiconductor transistor. All three were awarded the 1956 Nobel prize
for their efforts.
Justin Rattner, chief technology officer of Intel, calls the transistor
"the fundamental building block of the information age. It's hard even
to think of a single invention that is responsible for as much change -
you'd maybe have to go back to the Bronze Age, where a single invention
changed the course of everything and had a lasting impact."

But doubts are growing over how much further we can go with these
technological building blocks of transistors and integrated circuits.
To improve speed and keep power and heat under control, transistors have
been getting smaller and smaller. Gordon Moore, co-founder of Intel,
came up with an eponymous law, that the number of transistors on a chip
doubles every two years. But he believes his law is running out of
steam. At the Intel developers' forum in September, Moore said that in
"another decade, or decade and a half, we will hit something that is
fairly fundamental".

That fundamental problem was explained by IBM Fellow Dr Bernie Meyerson
as "atoms don't scale". The nanometre - one-millionth of a millimetre -
is the unit used to measuring the tiniest elements of a silicon chip.
Intel, IBM and others have recently started production of 45 nm chips.
But the silicon atom itself is more than a tenth of a nanometre across.
Moore suggested there was a basic physical limit of five atomic layers.
Today, the oxide layer in transistors is a mere five to six atoms thick,
leading to challenges with current leakage. This is a quantum effect,
where electrons "tunnel" through an insulating region instead of
following their assigned path.

Mr Rattner isn't so sure there's a brick wall ahead. "Gordon always adds
a footnote along the lines of 'of course, we've never been able to see
beyond about 10 years'. Typically we are seriously at work two
generations ahead. We are in production with 45 nanometre and well along
with 32 and 22 nm."
Glenn Wightwick, an IBM distinguished engineer and director of the
Australia Development Laboratory, agrees there are issues to overcome
but doubts innovation will slow.

"Until the late 1990s, the vast majority of the gains made have been the
result of scaling - making things smaller. When the lithography moved to
180 nm, 90% of the relative improvement over the previous generation of
semiconductor devices was derived from traditional scaling - that is,
the application of Moore's Law.

"Today, as we move from 65 nm to 45 nm and beyond to 32 nm, only 20% of
relative improvement is derived from scaling alone. Innovation, in the
form of novel materials, structures, processes and architectures
delivers the rest. This is why IBM invests so heavily in R&D."
IBM's researchers are experimenting with different materials and
techniques to improve performance, such as copper in chips,
silicon-on-insulator, strained silicon, multicore chips and air gap
self-assembly. The current crown jewel is IBM's Power6 processor, which
has 790 million transistors and runs at 4.7 GHz.

Dr Wightwick acknowledges that physical limits are being approached. Mr
Rattner concurs. "We are reaching the limits of physics in some ways,"
he says.

To achieve a 45 nm resolution, Intel had to use a new material - Hafnium
- in the gates of the transistors.

"We ran right into a physical limit," he says. "But what's happened
again and again when you come upon the physical limits is we've been
able to advance around them, and I think that will continue for at least
the next several generations."

Already, the scale of the detail on the chip is smaller than the
wavelength of the light (193 nm) used to print it.

This bizarre result is thanks to the use of "clever maths" while
patterning transistors, Mr Rattner says.

But this technique is going to reach a limit. Intel is looking at ways
to use light with much smaller wavelengths, extreme ultraviolet and
X-rays, but it is a tricky undertaking. "X-rays don't focus in
traditional ways - it's all done with mirrors.

"But I think a couple of generations out we will have to make the
transition."

There could be an even bigger transition to come, once the scale gets
below 10 nm. Mr Rattner predicts that in a decade, the fundamental basis
of electronics will change. Instead of using the electrostatic charge of
an electron, devices will depend on another quality of electrons, their
"spin".
Mr Rattner says: "Spin-based devices will be based on different
materials such as titanium cobalt alloys that have the required
appropriate magnetic domain. When you get into the speculative area,
then you are talking about molecular devices." Molecular devices are one
of several new radical ideas around.

Dr Wightwick says many research laboratories are looking for new and
novel devices that could replace transistors inside computers. "Things
like carbon nanotubes and molecular cascades. There is a lot of
interesting work being done in quantum computing."

But when the basic building blocks change, the entire architecture of
information processing, and the silicon industry itself, will undergo a
revolution.

Already, says Dr Wightwick, the cost of a new "silicon foundry" is huge,
driven by the cost of moving from one generation of lithography to the
next.
This has led to dramatic consolidation across the industry in order to
share these costs. Moving to a whole new class of devices using
different materials (probably still on top of a silicon substrate) will
be even more difficult and costly.

That's the bad news. The good news is that it's a bonanza in the making
for users of technology.

Mr Rattner says that when the first 22 nm silicon chips appear - just
two chip generations out - it will prompt a generation of single-system
chips that make it easier to interact with technology.

"We are right at the start of the information age. We think we are so
sophisticated with our hand-held devices and internet access. But we
have asked an enormous amount from users to tolerate - why is it that my
mother-in-law calls me up and says 'I've got this error 22 message'?
"How do we soften those interfaces and make them more human? That's a
very important next step. We are in that era of technology where we
start to move away from machine imposed limitations.

"We were doing an internal talk on computer perception and we've got a
slide from Star Trek of Captain Kirk holding a universal translator and
we ask, 'how far are we from that?'. I think it's probably not more than
a decade into the future when devices like that will be practical."

Dr Wightwick also predicts a bright future.
"Creating new ideas, solving problems, inventing things and applying
technology in new and novel ways, seems to be a basic human
characteristic. One of the things I love about computing . .. is that
innovation has been so fundamental to this field. I don't see any
slowing down of the rate of innovation. In fact, I continue to see more
innovation every day."
Innovations that give us more processing power will spawn many other
innovations, Mr Rattner says.

Google "took a very powerful piece of software and ran trillions of
bytes of examples of English and Arabic and trained it to recognise
language statistically. It knew nothing about Arabic or English, though.
"We have spent decades on artificial intelligence thinking we could do
everything with rules.

"The new thinking is statistical - which is how the brain works - and
making use of access to a massive amount of training information from
the internet.
"This move to machine learning is going to open up a broad class of
applications such as machine translation and continuous speech
recognition.

"That technology will move very quickly and then you begin to combine
that with robotic technology and you move into the age of personal
robots."
Early next decade Mr Rattner envisages car companies developing
autonomous vehicle technologies geared at collision avoidance that can
take over control of the car, if the driver dozes off, and bring it
safely to a stop
"This is not so far fetched and not so far into the future."

The next 60 years look set to be just as exciting a ride as the first.

NEXT SPEAK
Patterning/lithography: the process by which transistors are built onto
a silicon wafer.
Transistor: a current-controlled switch.
Gate: used to control current flow in the transistor.
Spin: the angular momentum of a particle.
Electrostatic charge: the positive or negative electric charge of a
particle.
Carbon nanotube: an extremely thin hollow cylinder comprised of carbon
atoms, about 10,000 times smaller than a human hair.
Quantum computers: computers that use quantum physical properties to
represent data and perform computations.

The birth of integration

Ivan P. Kaminow, adjunct professor at University of California,
Berkeley, joined AT&T's Bell Labs in 1954, during the decade when the
transistor was slowly taking over from valve-based electronics.
He is now a respected pioneer of photonic integrated circuits - circuits
that work on light instead of electricity.
During a talk last month at the University of Melbourne, organised by
the national research institute NICTA, he reflected on his early years
at Bell.
"I have learnt the lesson that if you try to predict the future you have
to expect the unexpected. My first job was in the transistor circuit
department (at Bell Labs), only seven years after the transistor was
invented. What they had me do was take a vacuum tube circuit used to
test telephone lines and convert it to a transistor circuit. It had an
aluminium chassis where the vacuum tubes were plugged in. The transistor
came in a little can, a few millimetres on the sides, with three wires
coming out. It was pretty simple to convert the circuit. I went home and
told my wife the transistor probably wouldn't amount to very much. The
reason was that these transistors cost $50 and these vacuum tubes cost
only a dollar. I was only 24, so you can forgive me for making that dumb
mistake but I can't forgive Bell Labs for missing the idea of integrated
circuits. I can see two reasons. Yields (the amount of working
transistors in a batch of silicon) were poor, so if you have a 10% yield
and put two (transistors) on the same device the yield would be 10% of
10%. That was a short-sighted way of looking at things; also, the guy
who was in charge of this was pretty arrogant. About 1958 (electrical
engineer Jack S.) Kilby, a new hire at Texas Instruments during summer
vacation when everybody was out of the labs, decided to put several
transistors on the same chip. And that was the origin of the integrated
circuit. He connected them with resistors and capacitors with external
wires, so it wasn't really fully integrated. A few months later (Robert)
Noyce at Fairchild Semiconductor, which later became Intel, he made an
integrated circuit where the wiring was integrated on the chip. Since
then there have been quite a few surprises. According to Gordon Moore,
in the 1990s, more transistors were made each year than raindrops in
California. I'm sure that 10 years later you can multiply that by
several orders of magnitude. Ten to the 18th (power, that is, a billion
billion) transistors are made each year, more than existed at the
beginning of the year. That's 10 to 100 times the number of ants on
Earth. So all these advances are based on a design for transistors."

This is a short, edited extract of Professor Kaminow's talk.

TIMELINE
1947 Transistor invented in Bell Labs.
1948 Shockley develops first semiconductor transistor.
1952 Hearing aids are first commercial products to use the transistor.
1954 Texas Instruments introduces transistor radio.
1956 AWA manufactures first Australian portable transistor radio.
1965 Intel co-founder Gordon Moore coins Moore's law.
1981 IBM launches the PC.
2007 Intel demonstrates chip with 1.9 billion transistors.
2007 IBM reveals it has developed a single-molecule switch.