896 Chapter 18
80/20, is vaporized into the jet as Au and Sn atoms; the right composition is reproduced in the
depositing AuSn film. Precise composition, and therefore melting point, is a key requirement
for a solder film, and customers take pains to verify it, by energy-dispersive X-ray spectroscopy
measurements, melting/freezing point measurements, and finally by the performance of the
solder in bonding. Solder layers can be from several to 20 m thick, and both the wirefeed and
wick jet sources can make such layers at high rate. For example, with 15 mm AuSn wire, fed at
2 inches per minute, a 6 inch diameter wafer will be uniformly coated in ∼ 45 minutes. The
wirefeed rate can be greatly increased, and the coating time reduced proportionately.
Solder bumps of micrometer dimension and area densities of millions/cm
2
can be made using
conventional photoresist lithography [20, 21], but reliable manufacturing of such arrays is still
a challenge. Moreover, as the circuit features become smaller, filling small holes in a
photoresist mask adds to the challenge. Most patterned wafers received at JPC present large
area pads several tens of micrometers wide. If the solder layer is to be several micrometers
thick, then the resist walls must be about twice as high. For large area bond pads the aspect
ratio (photoresist thickness/pad width) will be substantially less than 1. For these low aspect
ratio cases, the wirefeed atomistic deposition of individual Au and Sn atoms is a reliable
approach; on the pad the deposited solder has the correct thickness and a flat profile. However,
small-area pads, with lateral dimensions ∼ 10 m or less, will still require several micrometers
of solder. The photoresist will still be thick, and the windows in the photoresist will have a
higher aspect ratio. We found that as the aspect ratio approached 1, small holes were difficult
to fill, if we tried to fill them by depositing atoms.
The difficulty has the following origin. Although the jet is collimated and traveling
line-of-sight normal to the substrate, the metal atoms inside it are not; they collide frequently
with He or Ar atoms, their velocities are randomized, and they impact the film at angles that
can be far from perpendicular. Some of these impacting metal atoms will not ‘see’ the bottom
of a deep hole. For small windows with aspect ratio of order 1, resist walls cut off atoms
coming from angles far from the normal, and shadowing is severe. In addition, ‘breadloafing’
from deposition on the sidewalls will further constrict the resist window as deposition
continues. As a result, the solder bump is no longer flat, but dome shaped; moreover, it will be
significantly thinner than needed. One can compensate by depositing excess metal, but this
wastes material and time.
JVD provides an effective alternative: generate and deposit heavier nanoclusters. The
nanoclusters grow inside the nozzle, are accelerated by the jet, and travel toward the substrate.
A massive nanocluster will be hit many times by randomly directed carrier gas atoms, but its
transverse velocity components remain extremely small. The trajectory of the nanocluster then
remains parallel to the jet, and it impacts the growing film at a right angle. In consequence,
there is little shadowing by resist walls, and since sidewall deposition is also reduced, there is
reduced breadloafing [22]. The resist window is therefore filled by solder clusters from the
bottom up, giving a flat-topped solder bump of undiminished thickness. The sequence of