| Daniel M. Dobkin |
High Density Plasma: Reactor Design Considerations
|
|
The first commercial HDP oxide deposition reactor designs were
based on electron cyclotron
resonance plasmas. For example, the Lam Epic reactor looked
schematically something like this:
Although most modern HDP tools are based on inductive
plasma generation, this reactor design illustrates many of
the elements common to all HDP reactors:
- High density plasma source:
in the Epic reactor, an ECR plasma was created by illuminating
a magnetized region with high-power microwave radiation. At a
magnetic field of about 875 Gauss, the electron Larmor frequency
is equal to the microwave frequency of 2.45 GHz (another frequency
used for economic reasons: it is the microwave oven frequency).
At low enough pressures, electrons gain energy on each circuit
around the field lines: the field allows both confinement and
efficient electron heating. ECR sources can achieve densities
around 1E12/cm2 at pressures of a few mTorr (this is several
percent ionization!).
- Big pumps: to deposit at high
rates, relatively large flows of silane (and thus oxygen and
argon) are required. To achieve a throughput of about 12-15 wafers
per hour from a module for a deposited film 0.75 micron thick,
we might require a deposition rate of about 5000 A/minute. About
14 sccm of silane are required simply to provide enough silicon
atoms to deposit this film on the wafer surface, but the film
deposits everywhere in the chamber. A realistic flow might be
30 sccm of silane, with roughly twice as much oxygen and some
argon: perhaps 150 sccm total flow. That may not seem like a
lot, but at 5 mTorr it is a volumetric flow of 23,000 liters/minute
or about 400 liters/second! The pumping capacity required to
remove the gas is even larger, since the pump ports and gate
valves have a finite conductance: several thousand liters per
second is needed for practical reactors. Fast pumping of hydrogen
is needed as well, since this gas is generated by the decomposition
of the silane. The only usable approach is to employ turbomolecular
pumps. Turbo pumps which can achieve these speeds are very large,
very heavy, and very expensive: tens of thousands of dollars
for each unit.
- Wafer RF bias: The ECR plasma
has a very low plasma potential. In order to achieve significant
sputtering rates at the wafer, it
is necessary to apply an RF bias, which involves a separate power
supply and matching network (but has the nice feature that plasma
density and ion energy are then independently controllable).
Several hundred volts of ion energy are required. Recall that
the current density is on the order of e.g. 10 mA/cm2, for a
total current of about 3 amperes to a 200 mm wafer. Typical RF
bias power for a 200 mm wafer is 500-1000 Watts, almost all of
which is dissipated onto the wafer surface.
- Wafer temperature control:
At the low pressures of operation of HDP reactors, very little
heat will be transported from the wafer by conduction or convection;
recall from our discussion of showerhead
heat transport that radiation will dominate at high temperature
and low pressure. To remove 500 Watts from a 200 mm wafer by
radiation, the wafer would need to achieve a temperature of about
500 C, much too high for use in conjunction with Al metallization.
Thus it is necessary to monitor and control the wafer temperature.
The control is usually achieved by injecting a small flow of
helium behind the wafer to raise the pressure to a few 10's of
Torr there; however, the resulting force must be compensated
to keep the wafer in position, requiring the use of mechanical
clamps or an electrostatic chuck.
- Gas injection: It is necessary
to put the silane into the chamber, and ideally to dispense it
close to the wafer so that efficiency of utilization is improved.
For total gas flows of 100-200 sccm in the typically large chambers
employed in HDP, transport is dominated by diffusion. The mean
free path may be as large as 1 cm, however: transport in local
regions cannot be modeled accurately using continuum techniques.
Thus design of the injecting apparatus is difficult, and it may
also be necessary to compromise efficiency to improve uniformity.
- Chamber clean: The combination
of high density plasma, low pressure, and high sputtering rates
at the wafer surface means that a lot of the silicon dioxide
formed ends up on the chamber walls. This buildup will spall
off, creating particles, if it becomes too thick. The Lam reactor
had water-cooled walls to help avoid temperature variations that
would increase spalling. Nevertheless, it is also necessary to
remove the deposit from the walls periodically, typically after
10-25 wafers. This may seem straightforward: there's already
a high density plasma, so if one introduces some source of fluorine
the oxide will etch. However, many difficulties arise. Etching
SiO2 requires either copious ion bombardment or high fluxes of
fluorine (i.e. high pressures, typically several Torr). ECR sources
don't operate well at these pressures, and at low pressure the
plasma potential is small so there's no wall bombardment. Inductive
sources operated at high pressure tend to concentrate the power
delivery very near the wall, which can lead to localized heating
of the dielectric liner ("dome") and consequent erosion
and cracking. Applied Materials has introduced a separate microwave plasma source
remote from the chamber to provide fluorine for chamber cleaning.
An inductive plasma realization, the Novellus SPEED reactor,
is shown schematically below:
In this case, the inductive coils are distributed over the
ceiling to provide improved plasma uniformity over the wafer surface,
vs. a solenoidal coil. However, in both cases the actual electron
heating is confined to within a skin
depth of the coils; rapid diffusion
at low pressure enables the plasma to fill the chamber.
Reactor Operation
The method of turning on the process is of particular importance
in the use of HDP reactors for intermetal dielectric deposition.
Deposition and sputtering both occur in the process. If RF power
is applied to the chuck before silane is present in the chamber,
the underlying metallization will be sputter-etched. This can
lead to undesirable faceting of the metal lines (reducing their
cross-section, with as a consequence reduced reliability and conductivity),
and to electrical leakage between lines due to the deposition
of a thin metallic film on the lower insulator surface. On the
other hand, if silane is added to the reactor before the RF bias
is ramped up, the initial deposited oxide will have relatively
poor conformality, and may form a "lip" which prevents
filling high-aspect-ratio features. Therefore it is vital to carefully
synchronize power and gas flows. Placing gas flow controls a long
distance from the reactor is undesirable. One possible approach
is to provide "run/vent" valves at the reactor, which
can immediately switch the chamber inlet from argon to an argon/silane
mixture with no change in flow and minimal time delays.
Go
Home
