Page 312 - A Practical Guide from Design Planning to Manufacturing

282 Chapter Nine

After the photoresist is “written,” the desired image is developed and
the newly exposed chrome etched away. Now the fractured layout image
has been transferred to the chrome of the mask. Light projected through
the mask will create a shadow of the desired pattern. An important
final step is inspecting and repairing masks. It is expected when pro-
cessing wafers that some percentage of the die created will have defects.
Some of these defects will prevent the chips from working properly and
these units will be discarded. Because the masks are used to expose the
die, a defect on a mask will create a matching defect on every single die
processed using that mask. It is critical that the mask is as near to per-
fect as possible, but the time and expense of making masks is strong
incentive not to simply discard a mask with a defect.
Instead, automated machines inspect the masks pixel by pixel for
defects. A defect is any place on the mask where chrome remains that
should have been removed or where it has been removed that should have
remained. If defects are found, some can be repaired. Ahigh-energy e-beam
is used to vaporize extra spots of chrome or to trigger CVD in spots that
are missing chrome. Some very small defects may not be enough to affect
processing. The stepper machines typically use lenses to focus the image
2
2
of a 5-in mask onto a 1-in area of the wafer. This means all the dimen-
sions on the mask will be 1/5 the size when created on the wafer, and very
small defects will be made even smaller by this scaling.
Finally the mask is coated in a transparent protective layer called a
pellicle and is ready for use. To create a full set of masks for a modern
90-nm process can easily cost millions of dollars. This all must be done
before the very first chip is made.

Wavelength and lithography
As one of the most expensive and technically challenging steps in semi-
conductor manufacturing, lithography has often been seen as the great-
est potential limiter to future process scaling and Moore’s law. At the
incredibly small dimensions of today’s integrated circuits, the wavelength
of light itself becomes a serious problem. To our perception light travels
in perfectly straight and uniform lines, but below the micron scale the
wavelength of light becomes a factor, and at these dimensions light bends
and turns corners in ways very different from simple straight rays.
In the early 1980s, visible light was commonly used for photolithog-
raphy. Conventional wisdom was that the smallest feature size that
could be reliably patterned would be equal to the wavelength of light
used. At this time, features of 3 µm were typical. Because violet light has
a wavelength of only about 0.4 µm, the wave nature of light was not yet
a problem. However, as features sizes steadily decreased year after year,
it became clear that something would have to be done.