You are on page 1of 4

PROCESS PHYSICS DETERMINING 2-D

IMPURITY PROFILES IN VLSI DEVICES

P. B. Griffin and J. D. Plummer


Integrated Circuits Laboratory
Stanford University, Stanford, CA 94305

ABSTRACT
Physically robust diffusion models are required to simulate stant obtained from the 2D test structure determines the
two-dimensional (2D) impurity profiles in VLSI devices. steady-state value of the lateral decay length and is given
The accuracy of the initial dopant profiles severely limits by the ratio of the diffusivity D,eff and the surface recom-
the predictive capability of2D device simulators. Histori- bination velocity K,. The third time constant involves the
cally, the most successful diffusion models havebeen based time evolution of the 2D profile and models its approach to
on point defect mechanisms involvingeither vacancy or in- steady-state.
terstitial assisted diffusion. It is clear that the local con-
centration of defects determines the local diffusion coeffi-
cient, so the ability to model the point defect kinetics is
essential for obtaining accurate 2D dopant profiles. We de- EXPERIMENTAL
scribe a series of kinetics experiments using 2D process test
structures, which, when coupled with a 2D diffusion solver The test structure is shown in Figure 1 and consists of oxi-
enable quantitative values for the generation, diffusion and dizing and inert stripes with widths ranging from 2 to 200
surface recombination kinetics of the point defects to be microns [3],[4]. Experimentally, it is found that narrow ox-
obtained. The physical insight this workprovidesforms idizing stripes always exhibit the maximum OED diffusion
the basis of the diffusionmodels in the new2D process depth, suggesting that the oxidizing surface is insensitive
simulator SUPREM-IV [l]. to the adjacent sink for interstitials. In the above formal-
ism, this corresponds to a balance of a large generation flux
proportional to (dXoz/dt)”2 and a large surface recombina-
INTRODUCTION tion velocity at the oxidizing surface. Figure 2 shows how
this model fits the experimental data in one dimension on
Accurate two-dimensional process modelling requires tech- the inert and OED junction depths. The 2D parameters
niques for calculating local (i.e. time and position depen- are determined by considering the extent of the diffusion
dent) point-defect parameters, since it is clear that thelocal enhancement under narrow inert stripes due to nearby ox-
concentration of these defects determines the local diffu- idations. This enhancement depends exponentially on the
sion coefficient. This implies a tight coupling between the stripe width allowing a value for the effective lateral decay
surface geometry and resulting impurity profiles in small length Ld to be obtained. The decay length is normalized to
devices. Extracting these parameters requires 2D process the maximum OED, so that the only time dependance re-
test structures which give experimental values for the local maining is due to the diffusion limited approach to steady-
enhancement in thediffusion coefficientunder inert and ox- state. In Figure 3, the time evolution of Ld for the three
idizing stripes in both one and two dimensions. Based on temperatures investigated is shown. Simulations using a 2D
these results we show that a simple model with vertical and diffusion solver of the time evolution of the excess intersti-
lateral effective diffusion lengths is inadequate to predict tial concentration at a particular stripe width are compared
dopant profiles in small devices. In addition, the assump- with experimental values in Figure 4. This indicates that
tion that the relaxation time for the point defects is short a high value for the diffusivity enables a steady-state dis-
enough that a steady-state distribution can be assumed is tribution to be obtained almost immediately and the L,,
shown to be in error. Instead, three time constants are re- shows almost no time dependence. Such a value forthe dif-
quired to determine the experimental 2D dopant profiles. fusivity is suggested by gettering experiments [5], [6] and by
Hu’s formulation [2] consisting of generation, recombination OED experiments in epilayers [7]. However, only one ratio
and diffusion fluxes is ideallysuited to this problem provid- of D‘’f/I(z models the time evolution of the lateral pro-
ing it is introduced with the necessary time constants. One file. This enables unique values for the effective diffusivity
of these time constants is determined by the vertical oxida- and the surface recombination velocity to be obtained and
tion enhanced diffusion (OED) kinetics and corresponds to these are plotted in Figure 5. Values in this range for the
a balance between two large interstitial fluxes, the gener- diffusivity have been suggested from diffusion experiments
ation flux and the recombination flux. Another time con- [IO]and [I11and from stackingfault kinetics [12].
[4I,,[8],[9]

20.2
522-IEDM 86 CH2381-2/86@000-0522 $1.00 O 1986 IEEE
RESULTS and SIMULATION wafereven at low temperatures. This also explains why
backside oxidation does not getter metallic impurities. The
The particular value of the interstitial diffusivity has im- supersaturation is small enough that the trapssignificantly
portant implications forprocessmodelling programs. If slow down the diffusion front. In Figure 8, the effect of the
the diffusivity is large, then a steady-state is set up almost interstitial supersaturation on the diffusion front clearly in-
immediately and the numerical problem reduces to solving dicates this effect.
a simple Laplace equation. A transient effect requires the
computer-intensive solution of a Poisson equation. In Fig- CONCLUSIONS
ure 6 we plot experimental values reported in the literature
for the interstitial diffusion coefficient. The value of Di is The results above indicate that twodimensional dopant
scattered over several orders of magnitude and the activa- profiles can be rigorously modeledby coupling the local dif-
tion energy ranges from about l e v to 4eV. However, two fusion enhancement to thelocal point-defect concentration.
trends can be noted in the data. First, the larger values of In addition, theresolution of an apparent discrepancy in the
Di are generally reported from gettering studies while the interstitial diffusion coefficientsobtained in these and other
lower values are from diffusion type experiments. Secondly, experiments leads to a consistent model for the effects of
in the diffusion studies, epilayers have the fastest diffusion interstitials on relevant processes (oxidation enhanced dif-
coefficients and float-zone and Czochralski material have fusion, gettering) in epitaxial, float-zone and Czochralski
smaller diffusion coefficients. What is being measured is an material. The results also suggest that diffusionprofiles
effective value and the differences described above must be can be controlled in both the lateral and vertical dimen-
due to bulk recombination. sions by appropriate control of the surface and bulk pa-
A simple model forthe effects of a recombination center rameters. Practical ‘diffusionengineering’for submicron
on the interstitials is to consider the following annihilation device structures should be possible with 2D profiles signif-
reaction icantly different than the relatively uncontrolled profiles in
I+TKB‘IT today’s devices.
The coupled equations for the traps and the interstitials ACKNOWLEDGEMENTS
that must be solved are
aci
-= D;V2C;- ICBTC~ The simulations in this chapter relied on SUPREM-IV code
at written by Conor Rafferty and Mark Law and also on their
generous help. Sung Ahn and Paul Fahey provided many
and
8T useful discussions. This work was funded by DARPA and
--
at - -KBTC~ VHSIC under contract number DAAL01-86-k-0101 and by
SRC under contract number 85-08-0062.
In this case the trapsare annihilated by the interstitials and
the diffusion front moves into the wafer like a shock wave References
with an effective diffusivity Dierr much less than the actual
diffusivity D;.A plot of the profile of the interstitials as J.D. Plummer et al., “Process Simulators for Silicon
they move into the wafer is shown in Figure 7, compared VLSI and HighSpeed GaAs Devices” Integrated
to a simple error function profile with a diffusion coefficient Circuits Laboratory, Stanford University, Stanford,
of A diffusion limited value was assumedfor K B im- CA, July 1986.
plying that the annihilation of interstitials is limited only
by the rate that they meet the traps. A trap density of S.M.Hu, “On interstitialand vacancy concentra-
M 10’6/cm2 was used for the float-zone material. Czochral-
tions in the presence of injection,” J . Appl. Phys.,
ski material is assumed to have a higher trap density which 57, 1069, 1985.
slows the diffusion front even further. In epitaxial material A.M.Lin, R.W.Dutton and D.A.Antoniadis, “The
the trap density is assumed to bevery low and the intersti- lateral effect of oxidation on boron diffusion in (100)
tials can move with the high diffusivity Di. These results silicon ,” Appl. Phys. Lett., 35, 799, 1982.
explain the trends in the diffusion data in the literature
noted above. K.Taniguchi and D.A.Antoniadis, “Lateral extent
The recombination model outlined above also explains of oxidation-enhanced diffusion of phosphorus in
the fact that gettering can proceed even at low tempera- (100) silicon,” Appl. Phys. Lett., 46, 10, 1985.
tures. Gettering is the process [13] whereby the back sur-
face of the wafer injects interstitials which kick-out certain U.Goselle, “The Nature of Point Defects and Their
metallic impurities from substitutional sites to interstitial Influence on Diffusion Processes in Silicon at High
sites. The metal atoms can then diffuseeasily to a sink, Temperatures”, Mat. Res. Soc. Symp.PTOC., Vol.
usually on the back surface. High dose phosphorus diffu- 14, edited by S. Mahajan and J. Corbett, p. 45,
sions and ion-implant or mechanical damage are efficient 1983.
getters on the backside of a wafer. The reason that these
getter efficiently is because the supersaturation of intersti- G.B.Bronner and J.D.Plummer, in Impurity Dif-
tials is so large (typically a factor of100 to 10,000) that fusion and Gettering, (Materials Research Society
the traps in the bulk are rapidly annihilated and the in- Proc., Vol36, North Holland, New York,1985),
terstitials can diffuse quickly through the thickness of the edited by R.B.Fair and C.W.Pearce.

20.2
IEDM 86-523
SUPREM-IV SIMULATION
[7]P.B.GrifEn,
P.M.Fahey, J.D.Plummer
and
R.W.Dutton, “Measurement of Silicon Interstitial
Diffusivity,” Appl. Phys. Lett., 47, 319, 1985.

[8]Y.S.Shin and C.K.Kim, “A Two-Dimensional


Model forthe Excess Interstitial Distribution in
Sili-
con During Thermal Oxidation,” IEEE Trans. Solid
State and Electron Devices, 30,1438, 1983.
[9]
E.Scheid and P.Chenevier,“1D
Simulation of
Oxidation-Retarded Diffusion in Silicon and Valid-
ity of the Physical Model,’’Phys. Stat.Sol., 93,523,
1986.
-LGO
TIME (hours)

[lo] S.Mizuo and H.Higuchi, “Effect of Back-Side Oxi-


dation of B and P Diffusion in Si Directly Masked
with Si3N4Films”, J . Electrochem. SOC.,129, 2292, 1.0, 1000

1983.
[ll]S.T.Ahn,J.D.Shott and W.Tiller, “Determination
of Modeling Parameters for 2D Dopant Diffusion
Using Silicon Membranes,” Proceedingsof the Elec-
trochemical Society Meeting,San Diego, CA, Octo-
ber 1986.
[12]K.Taniguchi,D.A.Antoniadis andY.Matsushita,
“Kinetics of Self-Interstitials generated a t the
SilSiOz interface,” Appl.Phys.Lett., 42, 961, TIME hours

1983.
1/00
[13] G.B.Bronner“PhysicalModeling of Gettering in
Silicon,” Technical Report No. G502-3Ph.D Disser-
tation, Department of Electrical Engineering, Stan-
ford University, Stanford, Ca , April 1986.

I I

1000

900

L STRIPE WIDTH
‘0 1 4 6 8 10 11 14

SQRT TIMEhours

Figure 1. A diagram of the Two-dimensional Process Test


Structure. By measuring the enhancement in the junction Lateral decay length of interstitials as a function
depth under narrow stripes the decay length of the inter- oftime at900, 1000 and 1100 C. The values were extracted
stitials can be calculated. from the 2-D process test structure described above.

20.2
524-IEDM 86
1.0, 1100 c

0.0b
0 2 4 6 a i o i z i r
TIME hours

Figure 4. Result of 2D SUPREM-IV simulations for a 10 Figure 6. Literature values for the interstitialdiffusion coef-
micron stripe at 1100 C, for two different diffusion coeffi- ficient extracted from gettering and diffusion experiments.
cients. The experimental points are thecircles and the lines
are the computer simulations.

0 50 100 150 250


200 300 350
DISTANCE mlCrOnS

Figure 7. A comparison of a simple erfc profilewith a small


diffusivity with the solution of the coupled equations which
account for bulk recombination. The similarity of the pro-
files is such that distinguishing between them experimen-
tally is difficult.

Figure 5. Values of the effective interstitial diffusion coeffi-


cient and thesurface recombination velocity as a function of
temperature, extracted from the experimental results and
subsequent SUPREM-IV simulations.

3 5300205200105100 05 0 0 100
D I S T A N C E microns

Figure 8. The effect of a high interstitial supersaturation


on the evolution of the diffusion front in the presence of a
fixed number of interstitial traps.

20.2
IEDM 86--525

You might also like