Professional Documents
Culture Documents
5/11/2008
Queens College Physics Department
cameras to solar cells, semiconductors are the miracle material that makes modern
electronics possible. Moore’s Law, first observed by Intel’s cofounder Gordon E. Moore,
is not actually a “law” but a trend. He found that the performance of semiconductor based
electronics improves exponentially over a period of years. This amazing progress is due
Cramming more transistors into a space is what makes processors run faster, gives
digital cameras more resolving power and makes computer memory bigger and better. On
top of that, the smaller transistors are becoming cheaper to manufacture per transistor, so
that the next, more powerful, generation of digital camera is cheaper than its predecessor.
This gives industry a huge incentive to improve the process of making transistor. Since
their invention, transistors shrank to the point where billions could be placed on a single
processor.
Think of the process of making a transistor as molding a piece of clay. You can
shape it, fire it, bake it, add impurities to it, but it is still basically clay. Scientists have
been improving transistors by making them smaller, using less and less “clay” to do the
same thing and finding better ways to mold it more precisely. It’s gotten to the point
where a single transistor is a mere hundred atoms wide. Transistors this small, however,
are running into significant physical boundaries. It’s like the walls of clay are too small to
do their job. Making smaller devices will require new materials that have yet to be
discovered and these materials will have to interact properly with the clay already there.
They has to “stick” to the clay, keep its properties through wear, survive the same
processes of firing that the clay does. If that is not possible then an entirely new process
will have to be invented. Still scientists are optimistic that these goals can be met. The
Transistors
Transistors are tiny switches that can be turned “on” or “off”. Typically they are
three pronged, with an electrode for the source, gate and drain. A small voltage can be
applied to the gate, determining whether a current can flow between the source and drain.
1
“Roadmap of Technology Characteristics.” ITRS. 5 December, 2007. 11 May, 2008.
http://www.itrs.net/Links/2007Winter/2007_Winter_Presentations/Presentations.html
When a current is allowed, the transistor is in its “on” state. This has been used to encode
MOSFETs
The MOSFETs (above) are a type of transistor that is made predominately with
different types of silicon. One is called n-type which allows for excess electrons to flow
in the material. The other is p-type which allows positively charged holes to flow.
Making silicon p-type of n-type is a process of adding impurities to the silicon, called
doping. The two types of silicon are sandwiched together harmoniously. The blue is the
thinnest wall and is responsible for insulating the gate electrode from the charges in the
silicon.
The major advantage of MOSFETs is their low power consumption due to the
insulating layer between the surface of the semiconductor and the gate electrode. CMOS
circuits utilize MOSFETs to eliminate current flow (ideally) in operation except when
Silicon
Silicon is a semiconductor and the miracle material that goes into making
transistors. Being a semiconductor silicon can play all the parts necessary to make a
transistor or any other electronic device. An entire circuit, including all its components
can be fabricated on a single wafer of silicon, eliminating the need for cumbersome wires
and macroscopic parts. It’s like making an entire building with doors, windows, stairs, an
One of the many advantage to silicon is a gift from nature that grows straight out
of any piece of silicon and has excellent insulating properties. Silicon dioxide (SiO2) is
the rust that forms on silicon and it is what forms that thin blue layer in MOSFET
diagram. The insulating parts of any silicon device could be formed to high uniformity
and interface quality by simply oxidizing the underlying silicon. How good of an
insulator a material is given by its dielectric constant (k), with higher dielectric constant
compared with 1 for air or vacuum. The addition of nitrogen to the SiO2 improved
However, as devices have shrunk, the layer of SiO2 has not been thick enough to
prevent charges from passing through. This is a quantum effect. No matter how high the
energy otherwise required to surmount the boundary. This is the walls of the clay
becoming so thin that water can leak out of them. As the SiO2 dielectric shrinks under
1.5nm, leakage resulting from quantum tunneling starts to consume a much larger
proportion of the power, a major hurdle in implementing smaller transistor sizes.2 This
also contributes to waste heat generated during operation. In the latest generation of
transistors, both power use and waste heat generated have increased. Replacing SiO2 with
a material that has better insulating properties (higher dielectric constant) such as
hafnium or zirconium was necessary for preventing gate leakage and thus increasing
performance 3. At the same time, the dielectric that replaced SiO2 allows for increase
2
Stathis, J.H. “Reliability limits for the gate insulator in CMOS technology.” IBM Research. 2007. 17
March, 2008. http://www.research.ibm.com/journal/rd/462/stathis.txt.
3
Scansen, Don. “Under the Hood: Inside Intel’s 45-nm high-k metal-gate process.” EETimes. 14
November, 2007. 17 March 2008. http://www.eetimes.com/showArticle.jhtml?articleID=202806020.
current speeds, also adding to the performance gains. This remarkable result was in no
way guaranteed. That SiO2 could even be replaced is a happy confluence of physical
The engineers at Intel have amazing challenges in finding and using a new
material to replace SiO2. Research at Intel started in the mid-1990’s4. However, progress
in the beginning years were stymied by charges being trapped on the gate-dielectric
boundary. The piling charge resulted in unwanted capacitance that effected the electrical
characteristics unpredictably.
This was a problem in the method used to deposit the dielectric. The two
processes used by industry, reactive sputtering and metal organic chemical vapor
deposition, while producing remarkably smooth surfaces, would not cut it for the new
use. The new method advanced by the team at Intel was called atomic layer deposition. It
calls for the introduction of a gas that reacts with the surface of the silicon wafer to coat it
with a smooth single layer of atoms. When the entire surface is coated the reaction stops,
not allowing further layering of the insulator. Then the gas is evacuated and a second gas
is introduced forming a second single layer of atoms. This is repeated until the right
4
Bohr, Mark T., Chau, Robert S., Ghani, Tahir, Mistry, Kaizad. “The High-k Solution.” October 2007.
IEEE Spectrum. 11 May, 2008. http://www.spectrum.ieee.org/oct07/5553
The new materials to replace SiO2 were problematic at first as well. They were
taking too much voltage to switch due to a property called Fermi-level pinning, and they
slowed the charge carriers current through the transistor impeding switching speed.
The problem was in the interaction between the polysilicon gate electrode and the
high-k crystal lattice. The high-k dielectric is invariably made of dipoles, molecules with
a positive end and negative end, which give it its insulating properties. These dipoles act
like rubber bands. When the electrons pass, they strike the lattice sending vibrations
throughout, called phonons, knocking the electrons around and impeding the speed of the
electrons. The effect, however, could be screened out by increasing the electron density.
The simplest way to do that was to replace the polysilicon gate electrode with a
conducting metal one. Replacing polysilicon with metal also led to better bonding
between the high-k dielectric and the gate, solving the other problem of Fermi-level
pinning.
Aluminum-gate MOS transistors were three to four times slower, consumed twice as
much silicon area, had higher leakage current and lower reliability compared with
the re-replacement of polysilicon with a metal required yet another industry overhaul in
processes long embedded. It also required finding another two new materials, one metal
for PMOS and one for NMOS, with the correct properties, in this case the metal’s “work
function”.
The work function is the energy of an electron in the gate electrode relative to that
of an electron in the silicon channel. The electric field resulting from this energy
difference changes the speed at which the transistor turns on. The work function had to be
5
Faggin, Federico. “Silicon Gate Technology.” Intel4004. 11 May, 2008.
http://www.intel4004.com/sgate.htm
close to that of polysilicon, and none of the materials tested quite fit, but the engineering
team found a way to change the work function of the metal to precisely what they
needed.
The engineering team also found it inconvenient to fabricate the transistor with
the normal “gate-first” method. During the “gate first” method, the source and drain are
implanted onto the gate, and annealed to repair any damage during the implantation.
Annealing puts the transistor under high temperatures and puts another constraint on the
metal; it has to withstand high temperatures. The annealing process also changed the
work function of the gate. Rather than dealing with the hassle, the Intel team decided to
add the gate after the annealing process, flipping the fabrication process on its head.
These new processes and materials were slated for development for the 45nm
transistor node. In early January 2007, Intel introduced the first working 45nm
microprocessors. Called the Penryn, it was smaller and performed better than the 65nm
generation.
performed surprisingly well. Gate leakage was reduced by over 25 times for the NMOS
Drive current, the speed of charge carriers through the transistor which reflects
performance, also improved. NMOS drive current improved 16% and PMOS drive
6
http://download.intel.com/technology/IEDM2007/variation.pdf
Figure 6: NMOS drive current improvement
Conclusion
The replacement of SiO2 with high-k dielectrics is only the first step towards
continuing Moore’s Law, however it is crucial. The materials and process challenge was
an effective wall to Moore’s Law that could very well have halted its progress. There is
much more work to be done to overcome the next generation of challenges, but with
these daunting challenges overcome the momentum already favors accelerating progress.