Sajak
Seorang Kawan Yang Sudi |
Rencana |
Cerita |
THE CHALLANGE TOWARDS FASTER MICROPROCESSOR
INTRODUCTION
The short history of digital electronic computers has been an axplosive growth in power, speed, capability, miniaturization, and affordability, all seemingly at the same time. The first digital computers were laboratary curiosities, fabricated from vacuum tubes taken from other pieces of electronic equipment. These give way to solid-state electronic built from resistors and semiconductor transistors. As the applications for digital electronic broadened, it was found that some simple digital circuit were going to be needed over and over again in almost every application. So integrated circuits, or ICs, were created out of useful combinations of resistors, transistors, and capacitors. These black, multilegged objects are found in almost every electronic device made today. Each IC is designed to perform some specific function, and the types available number in the tens of thousands. The single most complex type of IC is the microprocessor.
The first IC to be of any used as a microprocessor was a
product from Intel Corporation given the part number of 4004 , introduced in 1971 and contained 2,300
transistors. The 4004 could not do much in the way of real work ( a modern
pocket calculator can outperform it in mathematical functions), but it had one vital feature: it could be
programmed. That is when fed a steady stream of carefully arranged 4-bit words
(opcodes), the 4004 would input each one, perform some small operation, and
then would wait for the "instruction" in its program. A skillful
programmer could, by carefully sequencing the flow of opcodes to the
microprocessor, perform some visible work, such as adding two numbers, turning
small indicator lights on and off, and controlling electrical relays connected
to motors. To this day the basics of microprocessors and microprocessor
programming have not changed. However, the number and complexity of the
microprocessors themselves have increased dramatically, with no end in sight.
A microprocessor is an integrated circuit built on a tiny
piece of silicon. It contains
thousands, or even millions, of transistors, which are interconnected
via superfine traces of aluminum. The transistors work together to store and
manipulate data so that the microprocessor can perform a wide variety of useful
functions. The particular functions a microprocessor performs are dictated by
software. . Today's Itanium processor, by contrast, contains 25.4 - 60 million transistors. One of the most common tasks microprocessors
perform is to serve as the "brains" inside personal computers, but
they deliver "intelligence" to countless other devices as well. For
example, they may give your telephone speed‑dial and redial options,
automatically turn down your
house's thermostat at night, and make
your car safer and more energy efficient.
A microprocessor often is identified by its model name or
model number. Table 1 summarizes the historical development of the
microprocessor and documents the increases in clock speed and number of
transistors in chips since 1982.
Intel is a leading manufacturer of processors. With their
earlier microprocessors, Intel used a model number to identify tha various
chips. After learning that CPU numbers could not be trademarked and protected
from use by competitors, Intel decided to identify their microprocessors with
names, not numbers - thus emerged their series of processors known as Pentium
processors. A second
brand of Intel processor called the Celeron is designed for
less expansive Pcs. Recently there are two more brands, called the Xeon and
Itanium processors, are geared towards workstations and servers.
|
NAME |
DATE
INTRODUCED |
MANUFACTURER |
CLOCK SPEED
(MHZ) |
NUMBER OF TRANSISTORS |
|
Itanium |
2000 |
Intel |
800
and up |
25.4-60 million |
|
Pentium
III Xeon |
1999 |
Intel |
500-1000(1G) |
9.5-28 million |
|
Pentium
III |
1999 |
Intel |
400-1000
(1G) |
9.5 - 28
million |
|
Celeron |
1998 |
Intel |
266-633 |
19 million |
|
Pentium
II Xeon |
1998 |
Intel |
400-450 |
7.5-27.4
million |
|
Athlon |
1999 |
AMD |
500-1100(1.1G) |
22 million |
|
AMD-K6-III |
1999 |
AMD |
400-450 |
21.3 million |
|
AMD-K6-2 |
1998 |
AMD |
366-533 |
9.3 million |
|
AMD-K6 |
1998 |
AMD |
300 |
8.8 million |
|
Pentium
II |
1997 |
Intel |
233-450 |
7.5 million |
|
Pentium
MMX |
1997 |
Intel |
166-233 |
4.5 million |
|
Pentium
Pro |
1995 |
Intel |
150-200) |
5.5 million |
|
Pentium
|
1993 |
Intel |
75-200
|
3.3 million |
|
80486DX |
1989 |
Intel |
25-100 |
1.2 million |
|
80386DX |
1985 |
Intel |
16-33 |
275,000 |
|
80286 |
1982 |
Intel |
6-12 |
134,000 |
|
PowerPC |
1994 |
Motorola |
50-450 |
Up
to 50 Million |
|
68040 |
1989 |
Motorola |
25-40 |
1.2
million |
|
68030 |
1987 |
Motorola |
16-50 |
270,000 |
|
68020 |
1984 |
Motorola |
16-33 |
190,000 |
|
Alpha |
1993 |
Digital;Compaq |
150-700 |
Up
to 100million |
Other companies such as Cyrix and AMD currently make
Intel_compatible microprocessors. These microprocessors have the same internal
design or architecture as Intel processors and perform the same functions, but
often are less expansives. Intel and Intel-compatible processors are used in
Pcs.
An alternative to the Intel-style microprocessor is the
Motorola microprocessor, which is found is Apple Macintosh and Power Macintosh
systems. The processor used in Apple's PowerPC introduced a new architecture
that increased the speed of the processor.
The Alpha microprocessor which originally was developed
by Digital Equipment Corporation, is
used primarily in workstations and high-end servers. Current models of the
Alpha chip run at clock speeds from 300 to 700 Mhz.
A new type of microprocessor, called an
integrated CPU, combines functions of a CPU, memory, and graphics card on a
single chip. These chips are designed for lower costing personal computers and
smaller in sized.
Users of Pentium 4 processor‑based PCs can create
professional‑quality movies; deliver TV‑like video via the
Internet; communicate with real‑time video and voice; render 3D graphics
in real time; quickly encode music for MP3 players; and simultaneously run
several multimedia applications while connected to the Internet. The processor
debuted with 42 million transistors and circuit lines of 0.18 microns. Intel's
first microprocessor, the 4004, ran at 108 kilohertz (108,000 hertz), compared
to the Pentium 4 processor's initial
speed of 1.5 gigahertz (1.5 billion
Herzt )
To build a powerful microprocessor is to fabricate as many
transistors as posibble onto as small space of the silicon surface as posibble
.Traditionally interconnections between transistors are made using alluminium,
the size of which is of practical importance. The smaller the size of aluminium
and the shorter the distance between each component, the greter number of
transistor could be pack together. Now, however, the lightweight metal is
reaching its limits. It simply cannot conduct electricity fast enough for ultraminiature
circuits, and its vulnerability to electromigration will become a real problem
as feature sizes are reduced still further. The distance between components in
today's aluminum chips, such as the Pentium processor, is limited to 0.25
microns. And that limitation is forcing manufacturers to find alternatives to
aluminum. A chip can hold only so many transistors, and at a certain point,
transistors lose their ability to conduct effectively.
Other then superconductor, which only possible under extremely
low temperature, copper is the best conductor of electricity known. It has
about 40% less resistance than aluminium, allowing microprocessors based on
copper to operate some 15% faster than conventional devices. IBM also claims
that its method of depositing the copper wires reduces the amount of wiring in
a device by around 30%.
The semiconductor industry has recognized copper's potential
since the mid‑1960s. Copper offers three key advantages over aluminium,
the semiconductor industry's conductor of choice for forty years.
1. Copper
conducts electricity more effectively than aluminium.
2. Copper
wires are also far less vulnerable to electromigration ‑ a problem caused
when high electric currents force individual atoms to move through a wire, creating
voids that can eventually cause the wire to break.
3. The
width of copper wiring can be squeezed down to 0.2 µm and below far more easily
than aluminium interconnects.
Despite of all the positives charisteristics, copper also has
a significant disadvantage that has, until recently, prevented its use in
semiconductor devices. Copper rapidly diffuses into silicon, fundamentally changing the semiconductor's
electrical properties and preventing transistors from working as they should.
For that reason, copper was considered to be a killer of semiconductor devices.
IBM scientists had devised an early use of copper to extend
aluminium's capabilities for chip circuitry. They replaced pure aluminium with
an alloy of aluminium combined with a very small amount of copper. The alloy
proved more resistant to electromigration than pure aluminium, and its chemical
bonds were strong enough to prevent copper atoms from diffusing into the
surrounding silicon. Effective though it was, the alloy was merely a stay of
execution for aluminium. The industry realized that copper wiring would be
essential to continue the miniaturization of transistors, so research teams at
IBM and elsewhere set out to solve the three key problems facing copper
circuitry: depositing the copper, patterning it, and devising diffusion
barriers to prevent the metal from migrating into the silicon.
After more than 25 years of
research, IBM had devised a way of connecting the transistors in computer chips with copper wires, rather
than the traditional aluminium
interconnects.
IBM succeeded by drawing on earlier fundamental research at
the company's laboratories. To deposit copper, the scientists turned to
electrolytic plating, which had originally been developed for chip packaging.
Patterning relied on the "dual‑damascene" process, devised in
the early 1980s. In this technique a deposited layer of oxide is etched twice
to form an overlapping pattern of wires and "vias" ‑ the small
metal plugs that link separate layers of wiring in chips. The metal is then
applied and smoothed by chemical‑mechanical polishing. IBM researchers
also exploited work from the early 1980s to develop a diffusion barrier that
would stop copper from poisoning the silicon. The IBM barrier is based on
tantalum, while other companies have developed silicon nitride films that
perform the same preventative function. Meanwhile, Applied Materials has
recently introduced what it calls the BLOk (for Barrier Low k) dielectric film,
fabricated with chemical vapour deposition. This material, the company claims,
"provides an alternative to silicon nitride films, enabling chipmakers to
reduce the dielectric constant (k) of their overall copper damascene structures
to achieve faster, more powerful devices."
Copper technology offers chipmakers four specific advantages
over the traditional aluminium approach.
1. Copper's
higher conductivity simplifies the interconnect routing: the number of
interconnect levels can be reduced from twelve to six, removing more than 200
process steps from device manufacture.
2. Chips
containing copper require about 30% less power at any given frequency than
aluminium‑based chips, translating into significantly higher performance
for mobile applications in particular.
3. Copper
operates at faster speeds than aluminium: for 0.13 µm technology, the
interconnect delay for devices based on copper and low‑k materials is
about half that of components made from aluminium and silicon dioxide.
4. The
damascene process reduces the overall manufacturing cost of chips by 30% per
interconnect level.
New dielectrics propel chips into the fast lane. Novel
materials offering lower dielectric constants than silicon dioxide are needed
for the manufacture of faster computer chips.
Peter Sermon, Knut Beekmann and Simon McClatchie survey the
challenges involved in choosing the right candidate. As the semiconductor
industry enters the new millennium, scientists and engineers continue to search
for the ideal dielectric material for future chip fabrication. This goal is
driven by the continued demand for chips that offer faster operating speeds and
higher transistor packing densities. As metal widths and line spacings drop
below 0.2 µm, the main barriers to higher operating speeds will be delays
within the metal interconnects and increases in capacitance between adjacent
conducting lines. The latter can lead to cross‑talk.
The delay problem looks set to be resolved by new interconnect
technology that exploits copper instead of standard metal components such as
aluminium and tungsten. The capacitance problem will be addressed by reducing
the dielectric constant of the inter‑metal dielectric (IMD) that isolates
conducting lines from each other. The new materials must have a dielectric
constant (k) that is less than 3.9, the value of the standard silicon dioxide
dielectrics that are currently in use. This will allow metal lines to be
patterned closer together, offering the possibility of faster and more
efficient semiconductor devices. However, simply finding a low‑k material
is not sufficient. The material must also be integrated into existing chip
production lines, which means that potential candidates should possess the same
basic properties as silicon dioxide films. These include low moisture
absorption and good adhesion to the layers above and below. High thermal
stability is also required so that the films do not break down during high‑temperature
processing.
Extreme ultraviolet lithography is one of the most promising
methods of creating feature sizes beyond the current limits. Chuck Gwyn of
Intel and colleagues from the US Department of Energy review the latest
progress in developing the technology needed for the manufacture of production
machines.
Since the early 1990s the US Semiconductor Industry
Association has compiled a roadmap for the semiconductor industry that spells
out the technology and equipment requirements for each new generation. It shows
that lithography has played an increasingly important role as feature sizes
have become smaller. Semiconductor manufacturers have demanded tighter control
over the lithography process, which in turn has increased both the complexity
and cost of lithography equipment.
Conventional optical lithography at wavelengths of 157 nm is
expected to support the roadmap into the middle of the next decade, including
the fabrication of devices with critical dimensions of around 100 nm. But
completely new approaches will be needed for the manufacture of smaller feature
sizes ‑ in particular the 70 nm technology generation that is due to be
introduced in 2005.
One of the most promising candidates is extreme ultraviolet
lithography (EUVL). Similar in many ways to conventional optical lithography,
this method exploits ultraviolet radiation at wavelengths of 10‑14 nm to
achieve higher resolutions and smaller features.
The high resolution of electron‑beam lithography has
been recognized for years, but its progress has been limited by low
throughputs.
The incredible gains in computing power over the past forty
years have largely been enabled by improvements in lithography. This processing
technology allows both the size and position of circuit features to be
controlled to within just a few nanometres. Achieving such high precision has
required advances in all aspects of the technology: the mask, the lithography
tool, and the resist and process.
One of the most effective
strategies for sustaining the progress of the semiconductor industry has
been to shorten the wavelength of the light used to transfer the mask pattern
into the resist.
But this approach cannot continue indefinitely, partly
because feature sizes are reducing faster than new wavelengths can be
introduced. Such continued shrinkage has only been made possible by continually
improving the performance of lithography tools, which has resulted in optical
systems that are comparable in cost and complexity to those used in spy satellites
and space telescopes.
Resolution enhancement techniques have been introduced to
push optical processes to their absolute limits. Even so, however, the smallest
feature size that can be printed is still roughly only half the wavelength
used. Nor is reducing the wavelength an easy task, since at shorter wavelengths
the materials traditionally used in the lithography process start to absorb
light.
This means that more exotic materials must be introduced, or
that the design must be radically changed from refractive to reflective optics.
SCALPEL, for "scattering with angular limitation projection electron‑beam
lithography", is a technology based on electron‑beam lithography
that can meet all these requirements.
The superior resolution and depth of focus offered by
electron‑beam lithography have been apparent for decades. For example,
tools based on scanning electron microscopes have been used to pattern small
numbers of circuits directly, and to create the masks currently used in optical
lithography systems. However, this kind of serial writing can only achieve very
low throughputs, and the beam must also trace out the circuit pattern every
time a circuit is written.
A projection process, in which a mask is used to contain the
pattern information, would allow millions of features to be written at the
same time. However, attempts to develop
an electron‑beam projection system have failed for two main reasons.
First, conventional stencil masks absorb energy from the electron beam and generate thermal errors,
which limits the electron‑beam energy that can be used.
Second, traditional
full‑field optical systems generate significant optical aberrations
unless the numerical aperture is extremely small. This in turn leads to so‑called
space‑charge effects that destroy the resolution at the beam current
needed to provide acceptable throughputs.
These major limitations were overcome by Steven Berger and
Murray Gibson at Bell Laboratories in
1989, when they invented the SCALPEL concept. The projection system they
devised achieves high throughputs by exploiting a new type of mask and a novel
writing strategy.
The mask consists of a membrane of material with a low atomic
number covered with a layer of high atomic‑number material that contains
the pattern to be printed. The mask is
almost completely transparent to
electrons at the energies used (100 keV), but contrast is generated by exploiting differences in electron
scattering. The membrane scatters
electrons weakly and to small angles, while the patterned layer
scatters strongly and to high angles.
An aperture in the back focal plane of the
projection optics blocks strongly scattered electrons to form a high‑contrast image on the wafer . This
scheme means that contrast is generated
by the mask, while energy absorption occurs at the aperture. As a result, very
little of the incident energy is actually absorbed by the mask, preventing it
from being affected by thermal instabilities.
The other key innovation was the introduction of a new
writing strategy. Rather than illuminating the entire area to be printed ‑
as happens in full‑field optical systems ‑ the 0.25 x 0.25 mm
electron beam is scanned to create a
field only 3 mm wide on the wafer surface. The mask and wafer must then be
scanned mechanically through the field, forming a series of 3 mm strips that
cover the printable area. These strips can be "stitched together" at
the wafer to produce a continuous pattern. This so‑called subfield scanning
writing strategy ‑ combined with the new mask design ‑ overcomes
the limitations that have hindered attempts to create a projection electron‑beam
lithography system.
Research at Bell Labs has been directed towards developing
the three main components of a lithography system: the mask, the resist and
process, and the exposure tool.
The fabrication of the masks is in fact relatively simple,
since the SCALPEL system exploits a linear
imaging process. Optical systems, in contrast, are based on nonlinear
imaging, which means that tiny imperfections in the mask features can be
dramatically magnified at the wafer. Not only is this undesirable, but it also
results in very tight specifications for the mask and hence a higher cost.
The patterned layer on a SCALPEL mask is also two to three
times thinner than for an optical mask, which makes it easier to control
feature sizes during mask patterning and etching.
Nanotubes is a rolled up sheets of graphite of only few
armstrongs in diameter. It can also be
made into electronic devices like diodes and transistors. Researchers had shown that nanotubes should
indeed be useable for diodes and other electronic components, once the
fabrication techniques improve.
A recent report from IBM justified that their scientist have invented a computer
circuit base on a single molecul which is expected to initiate the production
of smaller and faster chips. They
fabricate logic gates base on micro cylindrical carbon atoms. Nanotubes can be the best substitute of
silicon since the present
microprocessor cannot shrink further.
WHAT IS THE LIMIT ?
Today copper chips are down to 0.13 microns and system with speed up to 1.8 Gigaherzt
has been in the market for quite some time .
In august 2000, it was reported that Intel cranked a Pentium 4 processor
up to 2GHz in a technology
demonstration designed to show off the power of its forthcoming desktop PC
chip. The Pentium 4, due in PCs soon,
is based on an entirely new chip design
or microarchitecture, that was designed from the ground up for computing
on the Internet, Albert Yu, senior vice president in charge of Intel's
microprocessor products group, said at the start of the company's developer
conference in San Jose California. With
room for achieving higher speeds in the future. An analysts say the new chip
should be able to reach speeds of 10 gigahertz or so in five years.
What will be the speed limit ? Many people considered the 1
Ghz Pentium III is already far more than satisfactory for their purpose. May be
software companies together with manufacturing companies are thinking of
creating something to make sure that there will always be a demand for faster
microprocessor .
References
1. Turley,
James L. Advanced 80386 Programming Techniques: Osborne McGraw-Hill 1988
2. H.J.Mitchell.
32-Bit microprocessors: John Wiley & Sons(SEA) Pte.Ltd.-Singapore 1986
3. Shelly,Cashman,Vermaat: Discovering Computers 2001, Course Technology, USA, 2000
4. http://www.chips.ibm.com/products/powerpc/
5. http://www.intel.com/intel/product/processors.htm
6. http://developer.intel.com/design/pentium4/
7. http://developer.intel.com/design/pentium4/papers/249438.htm
8. http://www.microprocessor.sscc.ru
Appendage
Table 2 describe s guidelines for selecting an Intel
processor. One thing to remember is that, the higher the clock speed, the
faster the proessor and also more expansive .
Table 2 : Guidlines for processor selection
|
Intel Processor |
Clock Speed |
Use |
|
Xeon |
800-1000 MHZ |
Power users
with workstation; servers on a network |
|
Pentium Family |
Above 800 |
Users that design professional graphics and drawings,
produce and edit videos, record and edit music, participate in video
conference calls, create professional web sites, play graphic-intensive
multiplayer Internet games. |
|
|
700-800 |
Users that design professional documents containing
graphics such as newsletters or number intensive spreadsheets; produce
multimedia presentations; use the internet as intensive research tool ; edit
photographs; send documents and graphics via the web; watch video; play
graphics-intensive games on CD or DVD; create personal web sites |
|
|
Below 700 |
Home users that manage personal finances, create basic
documents with word processing and spreadsheet software communicate with others on the web via
e-mail, chats, and discussion; shop on the web; create basic web pages. |
|
Celeron |
400-633 |
Home users that manage personal finances, create basic
documents with word processing and spreadsheet software edit photographs; make greeting cards and calenders; use
educational or entertainment CD-ROMs, communicate with others on the web via
e-mail, chats, and discussion |