-
ECE 559 - Dew 137Winter 2015
6 Diffusion of Dopants Diffusion is the systematic motion of
particles in response to
concentration gradients. It is a thermally activated process D =
D0 exp(-Ea/kT)
Since very high dopant concentration gradients exist in IC's,
any high temperature step (oxidation, epitaxy, implant annealing)
will cause diffusion. This must be considered carefully because
correct operation of the transistors is dependent on the
localization of dopants. This establishes a thermal budget for a
process sequence.
Diffusion doping is a method used to introduce dopants into the
semiconductor. (Ion implantation is an alternate method.)
Diffusion doping is done in tube furnaces like those used for
oxidation (900-1100C).
Two stages are involved: predeposition (to introduce dopants)
and drive-in (to more evenly distribute them over a larger
volume).
-
ECE 559 - Dew 138Winter 2015
6.1 Predeposition The semiconductor surface is exposed through a
patterned
oxide mask to a compound containing the dopant which, at high
temperature, is absorbed into the semiconductor to near the solid
solubility limit.
This creates a shallow, non-uniform, high concentration doped
region near the surface.
In general, predeposition proceeds in two steps. First, a
heavily doped silicate glass (the glaze) is formed on top of the Si
(local source deposition). The second step is a high temperature
anneal (soak) to allow the dopants to diffuse deeper into the
semiconductor.
After soaking, the glaze must be removed by an HF dip. Dopant
sources to form the glaze can be solid, liquid or gases.
-
ECE 559 - Dew 139Winter 2015
6.1 Predeposition Classically, dopant sources were gases: arsine
(AsH3),
phosphine (PH3), diborane (B2H6). These are introduced into the
tube furnace in flowing N2 and O2. The oxygen reacts with the gases
to form dopant-oxides which then react at high temperature with Si
to form silicate glasses. Adjusting the dopant gas:oxygen ratio
controls the dopant concentration in the glaze.
Gas sources have good control qualities, but care must be taken
to avoid nonuniformities due to turbulence and depletion. (For
example, wafers can't be stacked vertically very closely
together.)
All of the main gas sources are extremely toxic, even at ppm
levels. They are also pyrophoric and explosive.
-
ECE 559 - Dew 140Winter 2015
6.1 Predeposition Liquid sources such as pockle (POCl3) or boron
tribromide
(BBr3) are much safer and easier to handle, but are still
flammable and release toxic Br2 or Cl2 gas in air.
POCl3 and BBr3 have relatively low boiling points (105C and
90C), so their vapour pressure is straightforward to control.
Nitrogen bubbles through the heated liquid, picking up the dopant
source vapour which is then mixed with flowing oxygen and oxidized
before deposition onto the heated wafer. The dopant concentration
in the resulting glaze is controlled by the liquid temperature and
the oxygen flow rate.
Georgia Inst. Of Technology ECE 4752
-
ECE 559 - Dew 141Winter 2015
6.1 Predeposition Even safer than liquid sources are the solid
sources which only
evolve dopant gases at high temperature. In addition, the only
gases generated tend to be non-toxic (except for As2O3).
Usually solid sources are disks (pucks) interspersed among the
wafers (2 wafers for each puck), so doping uniformity is very good,
even with closely packed wafers. Temperature controls the dopant
vapour pressure and dosing rate on the wafer.
The phosphorous source is silicon pyrophosphate (SiP2O7) in an
inert ceramic matrix. This compound decomposes above 900C to
release P2O5 vapour.
www.nanolab.ucla.edu
-
ECE 559 - Dew 142Winter 2015
6.1 Predeposition B2O3 vapour is generated from a BN source
(800-1100C) which
has been activated before use by an oxygen anneal (900 -975C) to
form a surface B2O3 layer. (A variation is to add hydrogen and
oxygen during local source deposition. The resulting H2O reacts
with B2O3 to produce HBO2 (metaboric acid) which is much more
volatile.)
Arsenic doping is done using aluminum arsenate (AlAsO4) which
decomposes near 800C to generate arsenic trioxide gas (which is
toxic).
An alternative technique for local source predeposition (before
soaking) is to use spin-on glass sources. These are applied in the
same manner as photoresist, and once the organic solvents are
driven off, a doped oxide remains. Oxide doping levels are
determined by the manufacturer, but a wide range of doping levels
(1016-1020 cm-3) of a wide variety of materials is available.
Since there is no vapour diffusion involved, wafers can be
packed densely. A slight overpressure of nitrogen keeps
contaminants out during soaking and (at least for P and B) no toxic
gases are produced.
-
ECE 559 - Dew 143Winter 2015
6.2 Diffusion Modeling In order to predict doping levels and
profiles for a given set of doping
or heat treatment conditions, a good mathematical model of
diffusion is required.
C(x,t) is the concentration of a particular dopant at depth x
and time t Fick's (First) Law:
Where D is the diffusivity and J is the flux The continuity
requirement gives Fick's 2nd Law:
Numerical solutions to this equation can be obtained using
process simulators such as SUPREM.
To make analytic solutions possible, D is often assumed to be
independent of x. (In practice, D varies with C and, hence, x.)
This gives the classic diffusion equation:
xCDJ
xCD
xtC
2
2
xCD
tC
-
ECE 559 - Dew 144Winter 2015
6.2 Diffusion Modeling For predeposition, an analytic solution
of this can be obtained by
assuming the surface concentration is fixed at the solid
solubility limit, and given boundary condition C(0,t) = Css, C(,t)
= 0
where erfc is the complementary error function
The total dose of dopants introduced is given by
The characteristic diffusion length is given by The
metallurgical junction, xj, occurs when C(xj,t) = CB (the
background doping level of the wafer).
Dt2xerfcCt)C(x, SS
)dxexp(-x21erfc(x) Xo2
DtC2Q SS
Dt
-
ECE 559 - Dew 145Winter 2015
6.2 Diffusion Modeling
-
ECE 559 - Dew 146Winter 2015
6.2 Diffusion Modeling
-
ECE 559 - Dew 147Winter 2015
6.2 Diffusion Modeling For drive-in, if the initial profile is
assumed to be highly localized at
the surface, then the solution to the diffusion equation is
gaussian: Boundary conditions: C(0,0) , C(,t)=0,
The localization assumption is reasonable if D2t2 > 3D1t1
where 1 refers to predeposition and 2 refers to drive-in
conditions
For multiple high temperature steps, use(Dt)eff = D1t1 + D2t2
+
For variable temperatures, use
Q,dxt)C(x,o
4Dtx-exp
DtQt)C(x,
2
dtDDt
-
ECE 559 - Dew 148Winter 2015
6.2 Diffusion Modeling
-
ECE 559 - Dew 149Winter 2015
6.2 Diffusion Modeling
p
p
n+
-
ECE 559 - Dew 150Winter 2015
6.3 Diffusion Effects
Lateral Diffusion at a Mask Edge: Diffusion occurs
isotropically. Lateral diffusion is 75-85% of vertical. Since
dopants at the mask edge are
diffusing into a greater volume, a reduction in surface
concentration occurs there.
predeposition drive-in
line source - isotropic
-
ECE 559 - Dew 151Winter 2015
6.3 Diffusion Effects
Emitter Push: When diffusing phosphorous to
form the n+ emitter within a p base, the high phosphorous doping
increases stress in the Si lattice which enhances the diffusivity
of boron atoms.
This makes it more difficult to form a thin base (needed for
high and high frequency operation).
Built in electric fields complicate this effect.
Emitter push can be reduced by using As as the n-dopant for the
emitter as it doesn't stress the lattice as much.
-
ECE 559 - Dew 152Winter 2015
6.3 Diffusion Effects
High Concentration Effects: D usually increases with dopant
concentration (mostly as a result of increased stress).
This causes a more box-like concentration profile at high doping
levels.
For anomalous reasons, P doping profiles can have a very long
low concentration tail.
These nonlinearities can only be modeled numerically.
x
C
x
C
Box profile
Long tail
-
ECE 559 - Dew 153Winter 2015
6.3 Measurement of doped areas Averaged doping levels can be
inferred from
two or four point probe measurements. Depth profiling can be
determined by beveling the
surface at a shallow angle Dopant depth profiles can be
measured
using dynamic secondary ion mass spectrometry (SIMS) ppm or less
sensitivity
Lateral extent of doped regions can be measured using scanning
capacitance microscopy or four point probe mapping
P. Ghigna, U. di Pavia
wikipedia.org
-
ECE 559 - Dew 154Winter 2015
6.4 Diffusion Masks
A masking material is needed to prevent dopants from reaching
the wafer except where desired.
Thermal oxide (sometimes nitride) is usually used for Si, but is
not suitable for GaAs as Ga will readily diffuse into the
oxide.
The diffusion length in the mask must be less than the mask
thickness. (Higher temperatures or longer times will require
thicker masks.)