HERE鈥橲 a strange question: imagine designing a swing bridge without ever
being able to see what you鈥檙e doing. How would you know that it works? You could
build a computer model of the bridge. It would show you that when the bridge was
closed traffic could flow smoothly across it, and when it was open the traffic
stopped. You could even build the bridge and measure the flow. Though you
couldn鈥檛 see the bridge itself, if the traffic goes in one side and comes out
the other it is probably working properly. But what if the traffic stops flowing
for some unknown reason. Without seeing the bridge, you鈥檝e very little chance of
ever finding out what鈥檚 gone wrong.
Chip designers face a similar problem with the transistors and other
components on their chips. Transistors are like swing bridges for electrons:
they open to stop the flow of current and close to start it again. And those
zillions of transistors on the world鈥檚 computer chips have been made without
anyone ever seeing what鈥檚 going on inside them. Until now, that hasn鈥檛 mattered
too much. Even the tiny transistors packed onto today鈥檚 chips carry billions of
electrons along the equivalent of multilane highways. If one of the lanes gets
blocked and a few electrons get stuck, nobody really cares as long as most of
the lanes work most of the time.
But as transistors get even smaller, that will change. The number of lanes
carrying the electron traffic is falling fast. In fact, designers are on the
verge of wanting to build the equivalent of single-lane bridges that carry
electrons more or less one at a time. When things get down to that scale, the
lane has to work no matter what.
Advertisement
How can anyone be sure that their designs will work? Designers would like to
know exactly how the atoms in these tiny bridges fit together so that they can
reliably predict how a chip will behave without having to make it first. What
researchers would like is a way to peer inside transistors to see how the atoms
are arranged and how current might flow.
Now the race is on to find one. For the winner, the rewards will be huge. In
the semiconductor industry, the time it takes to get new chips into the
marketplace is hugely significant. Shaving off even a week or two can mean
millions of dollars in extra sales.
Though modern chips are enormously complicated, the individual transistors
from which they are built are little more than on/off switches. They are made by
doping small areas of silicon with tiny quantities of another element that
changes the silicon鈥檚 electrical conductivity. For example, phosphorus has one
more electron than silicon available for chemical bonding. When a phosphorus
atom inserts itself into a crystal of silicon atoms, that 鈥渟pare鈥� electron can
move through the crystal lattice and carry an electric current. Because the
electron carries a negative charge, using phosphorus as a dopant produces
negative or 鈥渘-doped鈥� silicon.
Boron, on the other hand, has one electron fewer than silicon, so boron-doped
silicon contains spaces or 鈥渉oles鈥� where that missing electron should be. These,
too, are free to move around, and when they do they act like positive charges
moving through the solid. Boron-doped silicon is known as positive or 鈥減-doped鈥�
material. Only one dopant atom among a million silicon atoms is all it takes to
change silicon鈥檚 electronic properties.
Interesting things start to happen when you add n-type and p-type dopants to
neighbouring regions of a piece of silicon.
The diagram shows a piece of
p-doped silicon with two n-doped regions. If you attach a couple of wires to the
two n-doped regions and apply a voltage, you find no current flows. The p-doped
material separating them stops it. Although there are a few electrons even in
the p-doped material, there are not enough to carry the current.
Now here鈥檚 a neat trick. If you apply a positive voltage to the right part of
the p-doped silicon it can produce an electric field that attracts electrons
into the p-doped material. Now there are enough electrons in part of the p-doped
block to carry a current, bridging the gap between the n-doped regions.
Tiny switches
This is basically the way that the transistors on today鈥檚 chips work. The two
n-doped regions are called the source and the drain. The region that applies the
voltage to the p-doped section is called the gate. When a current can flow from
the source to the drain, the transistor is on, when current cannot flow, it is
off.
The big stumbling block today is knowing how the dopant atoms are distributed
within each region of a transistor. Researchers are particularly interested in
the junction, where n-doped and p-doped regions meet. This is especially complex
because the dopant atoms can migrate across the junction, blurring the
distinction between the two regions and changing the characteristics of the
device.
Spot the atom
From end to end, the transistors on today鈥檚 chips are hundreds of times
smaller than the width of a human hair. Despite their size, each transistor
contains several million dopant atoms with high concentrations near the surface
and as few as 10 dopants per million silicon atoms near the junction. There is
little hope of modelling the individual positions of these atoms, not least
because physicists are not entirely sure of the size and direction of the forces
that act on individual atoms when they diffuse into the silicon. Fortunately,
researchers have so far been able to create pretty good models of a transistor鈥檚
properties using a statistical approach that glosses over the way individual
atoms behave and looks at the overall distribution of dopants and how it varies
in the silicon.
Soon, however, this approach won鈥檛 work. The next generation of transistors
will be so small that each contains so few dopants that their junction will be
determined by the position of only a few atoms. When this happens it will be
vital to know, at least roughly, where the dopant atoms will be or where to
place them. This is why measuring their position becomes so important. These
measurements will help physicists understand the forces at work on dopant atoms
when they diffuse into silicon, which will in turn help them predict what
happens in the chip-making process.
The current method for making sure those critical dopants end up in the right
place begins with a computerised simulation of where a given production process
would place them, and what the resulting electrical characteristics should be.
The next step is to make a few prototype examples of the chips, so that
engineers can measure these electrical properties. Then they compare their
measurements with the predictions, and use the differences to go back and make
more accurate models. This process, called inverse modelling, is extremely
expensive because each trial-and-error cycle requires many hours of labour and
the production of new chips. Spotting the position of dopants within the
transistor should help engineers to get the electrical properties right the
first time.
One of the most promising techniques for imaging the interior of a chip
relies on holography, using electrons rather than light to produce the
interference patterns. Electron holography has been used for years as a way to
correct problems with electron microscopes and to measure magnetic fields, but
it is only in the past year that it has been used to make two-dimensional
profiles of dopants on silicon. The project has been led by physicist
Wolf-Dieter Rau as a joint effort between the Institute for Semiconductor
Physics in Frankfurt-an-der-Oder and Bell Laboratories in New Jersey. The team
has produced some remarkable images of transistors that clearly show the doped
regions and their junctions. According to Molly McCartney, a physicist working
with electron holography at Arizona State University, 鈥淚t鈥檚 the experiment
everybody was trying to do.鈥�
At the heart of Rau鈥檚 imaging system is a transmission electron microscope
that focuses a beam of electrons onto the sample. The experiment is set up so
that only part of the beam passes through the sample, where it is distorted by
the electric fields of any extra electrons or holes. By allowing the distorted
beam to interfere with the undistorted beam that has not passed through the
sample, the researchers produce a hologram, which maps the changing electrical
potential associated with varying concentrations of dopant atoms.
The trickiest part of the technique is preparing the sample, which can take
up to two weeks. First, Rau and his team cut a thin slice from a chip that
exposes the cross section of the transistor being studied. Both sides of this
slice are then bombarded with argon in a process that Frieder Baumann, a member
of the Bell side of the team, calls 鈥渁tomic-size sandblasting鈥�. The idea is to
grind down the slice so that it is perfectly smooth and so thin that the
electron beam can pass through. Rau鈥檚 images clearly show the junctions between
the three regions of the transistor and the way the concentrations of dopants
vary.
A new vista opens
鈥淚 was wowed. It was just beautiful,鈥� says Hung-Ha Vuong, another Bell
physicist, recalling the first time she saw one of the images. Vuong does
simulation work with new transistors so she is one of the people who could
benefit from the technique. Vuong tested the images against predictions from
inverse modelling. She found that, while the model predicted the location of the
junctions correctly, dopant atoms had also unexpectedly made their way under the
gate. Nobody is quite sure how this happens, but it could have important
implications for the way transistors are designed. 鈥淚t revealed new physics,鈥�
she says. Such an effect could never have been spotted using inverse
modelling.
Once scientists begin to understand how dopant atoms make their way under the
gate, they will be able to model the process and learn how to control it.
Without this knowledge, building transistors will always be a hit-or-miss
affair.
The technique does not yet have the resolution to see individual atoms or the
sensitivity to measure how a single atom might change the electric potential in
the sample. Rau is now working to increase the sensitivity of his technique from
the current level. He says a factor of ten improvement is possible, which would
bring it nearer the atomic level.
Rau鈥檚 technique is not the only game in town; to date, more than 20 methods
are under investigation. The most widespread involves slicing open a transistor
and using a tiny probe to directly measure its electrical properties at
different points, a technique known as scanning capacitance microscopy (SCM). To
build up a profile of the way the concentration of dopants changes, the machine
takes measurements of the capacitance at more than 1000 points across the
transistor.
One of the problems with this method is the amount of computing power it
requires. Each measurement of the capacitance is influenced by millions of
silicon and dopant atoms beneath and near the tip of the probe. A single
measurement cannot reveal the way the atoms are distributed in this circle of
influence. Only by making several overlapping measurements is it possible to
work out a distribution.
鈥淚 think it鈥檚 at this point highly experimental,鈥� says Jim Plummer, a
physicist at Stanford University in California who is developing these
techniques. Still, some companies have already put the technique to limited use.
鈥淧eople have obtained a few measured profiles,鈥� says Plummer. 鈥淏ut they are
still not exactly sure what they are measuring, or what the quantitative
accuracy of the profiles actually is.鈥�
If a solution can be found, SCM has some advantages over electron holography.
SCM is a much faster process as Rau himself points out, because it eliminates
most of the painstaking process of preparing the samples for electron
holography.
Nobody is quite sure which of these competing techniques will win through.
鈥淭he most gracious thing to say is that all [techniques] are complementary,鈥�
says Rafael Kleiman, a Bell physicist who is developing SCM. 鈥淚 think when the
dust settles that may not always be true.鈥�