Materials Science and Engineering, B2 (1989) 41-47
41
Interaction of Megaelectronvolt Ion Beams with Silicon: Amorphization, Recrystailization and Diffusion* J. M. POATE A T&J"Bell Laboratories, Murray Hill, NJ 07974 (U.S.A.) (Received June 2, 1988)
Abstract
Three areas will be reviewed where the use of megaelectronvolt ion beams has contributed to the understanding of the properties of amorphous silicon: (1) thermodynamic and relaxation effects; (2) thermal and radiation-enhanced diffusion; (3) interracial segregation. 1. Introduction There is growing interest in the use of megaelectronvolt ion accelerators for implantation research and development. The salient research question is what new phenomena will be discovered in this energy regime as opposed to the more conventional implantation energies of 100 keV. In this review, we shall present results on the uses of high energy beams in silicon research which have allowed us to start to resolve some new and old problems. One of the most interesting areas of implantation research has been the formation of amorphous silicon (a-Si) layers and their subsequent solid phase epitaxial recrystallization. Most data have been obtained from layers several thousand fingstr6ms thick formed by conventional implantation. The formation of thick layers by megaelectronvolt implantation has allowed us to determine some of the thermodynamic and kinetic parameters of a-Si. The movement of the amorphous-crystal interface in the solid phase epitaxy process is characterized by a single activation energy of 2.7 eV over a temperature range of some 700 °C. There has been much recent interest in the use of high energy ion beams to induce the crystallization process at temperatures as low as 250-450 °C. We shall discuss this area, concen*Paper presented at the Symposium on Deep Implants: Fundamentals and Applications at the E-MRS Spring Meeting, Strasbourg, May 31-June 2, 1988. (1921-5107/89/$3.5(I
trating on the segregation and diffusion of dopants at the moving interface. The segregation phenomena demonstrate some of the features of classic segregation at the liquid-solid interface. The mobility or diffusivity of the dopants at the interface is considerably enhanced by the radiation.
2. Calorimetric studies of crystallization and relaxation a-Si has a higher free energy than crystalline silicon (c-Si) and is thermodynamically unstable in its presence. This free energy difference causes a-Si to transform to c-Si on heating. The reason that a-Si does not spontaneously recrystallize at all temperatures is because of kinetic barriers to crystal growth. The value of the Gibbs free energy of a-Si was not known until we used high energy implantation techniques to fabricate thick layers of a-Si. The problem in measuring thermodynamic properties, such as heat of crystallization, is that the amount of transforming material must release sufficient heat to be detectable in a calorimeter. We [1] estimated that thick (greater than 1/zm) a-Si layers were required for differential scanning calorimetry measurements. The thick layers were fabricated by sequential high energy implantation as shown schematically in the bottom part of Fig. 1. Our Van de Graaff accelerator does not have a silicon beam and we used the inert gases (argon or xenon) for the amorphization. The volume concentrations of argon and xenon were about 1 part in 104 in the a-Si layers. The differential scanning calorimetry measurements were made on an Si(100) wafer with scan rates of 40 K rain- 1. The rate of exothermic heat release is shown in the top part of Fig. 1. The rate is exponential with increasing temperature and at 970 K solid phase epitaxy has © Elsevier Sequoia/Printed in The Netherlands
42
8.0
8.4
T (K/IO0) 9.2 8.8
,i K 9.6
t0.0
900
I000
1000
800
si (100;
-t
100
Ln
n~ bJ
?i
o ~z_ z
0.5 Mev-~
10
10 1.t M e V ~
I
\
I I i
8~
1
1,9
4z
0 o ¢.9
t
t
2.5
m
2.0
~
o
t.5 1.0 DEPTH (p.m)
0.5
Fig. 1. Net differential scanning calorimetry power as a function of temperature (top) schematically compared (bottom) with the xenon implantation profiles. The low dose xenon implantations are tailored to give a uniform xenon concentration. (From ref. 1.) ceased, i.e. the crystalline-amorphous interface has reached the surface. Several points can be learned from this data. T h e heat release curve can be integrated to give the heat of crystallization AHac of 11.9 + 0.7 kJ tool -I. T h e Gibbs free energy of a-Si was calculated from AHac. T h e difference between the amorphous and crystalline free energies lead to the estimate of the first-order melting temperature of a-Si which is depressed some 250 °C beneath the value for c-Si. T h e velocity of the crystalline-amorphous interface can also be obtained directly from the data in Fig. 1. T h e velocity has an Arrhenius dependence as shown in Fig. 2 with an activation energy of 2.24 eV. This value is lower than the currently accepted value [2] of 2.7 eV which has been measured over a much larger temperature range. We shall fabricate layers with megaelectronvolt silicon irradiation to determine whether the activation energy difference is due to the incorporation of low
[
10
11 IO00/T(K -I)
[
1.2
Fig. 2. Arrhenius plot of differential scanning calorimetry net power normalized to the interface velocity for a-Si. The line in the fit corresponds to v= 1.8 x l(/~4 exp(2.24 eV/kt) AS
1.
levels of argon or krypton in the a-Si layers. Nevertheless the differences, in the activation energies or pre-exponentials, between the calorimetry and direct measurements of interface motion are very small. A n intriguing aspect of these calorimetry data for the recrystallization of a-Si is that there appears to be very little evidence for relaxation effects. A m o r p h o u s solids, by definition, should possess a continuum of states which converge to a fully relaxed configuration. Relaxation would be manifested, for example, by heat being released continuously throughout the bulk of the a-Si. We have no evidence for such behavior. Rather, the heat is first released by the recrystaltization processes at the interface. We have observed relaxation effects in a-Ge where approximately 30% of the heat release cannot be associated with interface motion.
3. Diffusion in amorphous silicon T h e study of diffusion in crystalline materials has been of considerable importance in under-
43 standing the properties of defects in condensed matter. Only recently have such studies been extended to amorphous systems and, in particular, a-Si. Most dopants or impurities (e.g. arsenic or antimony) in a-Si appear to have very small diffusion lengths in a-Si at temperatures of about 600 °C or less which preclude direct measurement by such techniques as Rutherford backscattering. The exceptions, discovered so far, are hydrogen, copper, silver and gold [3]. These impurities not only have high diffusivities but also high solubilities [4] in a-Si. The high solubilities and diffusivities of gold in a-Si are demonstrated in Figs. 3 and 4 from our early measurements [5]. Figure 3 shows channeled Rutherford backscattering spectra of a 2700 A amorphous layer with implanted gold before and after annealing at 450 °C for 30 min. The gold diffusion is so rapid that complete equilibration occurs within the a-Si. Moreover the enhanced solubility of gold in a-Si is nicely demonstrated by the fact that no gold can be detected within the underlying c-Si substrate. We have estimated gold to be eight orders of magnitude more soluble in a-Si than in c-Si at 515 °C by comparing the measured solubility in a-Si with the extrapolated solubility in c-Si. Figure 4 shows the Rutherford backscattering and channeling data of solid phase epitaxy [4] at 515 °C which causes the gold to be retained within the narrowing amorphous layer, i.e. the gold has a low interfacial segregation coefficient. These results indicated that we had unearthed an interesting system for studying interfacial segre-
gation phenomena. The behavior of impurity atoms at moving phase boundaries can provide important information on crystal growth processes. Most segregation studies have concentrated on liquid-to-solid transitions and the theoretical understanding of such processes is quite well understood [6}. In general, the redistribution and zone refining of solute atoms during solidification can be characterized by a segregation or partition coefficient between the solid-liquid equilibrium phases. If the equilibrium concentration of solute in the liquid is greater than in the solid, rejection will occur ahead of a solidifying interface. The interracial segregation coefficient k' is defined as the ratio of the solute concentration in the c-Si to that in the liquid (or a-Si in the case of the gold segregation). The final profile of the rejected solute depends on both the velocity of the interface and the diffusivity of the solute in the liquid (or a-Si). The results in Figs. 3 and 4 indicate that recrystallization and solid phase epitaxy occur at low temperatures, thus limiting the temperature-time window for diffusion measurements. We [3] have tackled this problem by using thick a-Si layers of several micrometers formed by megaelectronvolt ion implantation using the recipe [1] discussed previously. The diffusion of gold, silver and copper in a-Si can then be measured in the temperature range 400-900 K. We found an Arrhenius-like behavior for these implanted impurities with activation energies of
i
i
si<100>,2500,~ Q-Si,2XlOISAu cm 2 4DO keY Au, f x 1016crt~ 2 oAS IMPLANTED . 4 5 0 o c 30 rnin
f
-
1000
$4< I00> IO,OOC>
30 ~L
20 I
i
rO I I
Z
~
1021 w
AU DEPTH ('~',
iooo
-
c~
I00
~
;
,
.o
I0
f
o
~
zoo
I
F
d i
2,;0
I
i
F
z8o
i
I
i
i
szo
i
i
a.J
s6o
,
*
1
g
i
I
i
~,oo
I
,
I
I t ,r bOO
~
~
~o d°'9 i
t
~
"~
"L;,L%
2OMeV
a-g~ 'i,
500
F z 8
DEPTH (kA)
2700,~ aSi
Q-SI DEPTH (A) 2500 15100
~
4,;0
CHANNEL NUMBER
Fig. 3. Rutherford backscattering and channeling spectra of an a-Si layer 2700 A thick and implanted gold profile before and after annealing. (From ref. 5.11
I0[
c 515oC, 15 m m • 515°C,60 min • 515°C,80mlrl .515oc,85 rain
:"
io'
;f"
_
[
::" t
~. .,,.
.oo,,
! L250
300
350
400
450
500
CHANNEL NUMBER
Fig. 4. Rutherford backscattering and channeling spectra of a-Si solid phase epitaxyand gold segregation. (From ref. 4.)
1.42 ( + 0.06) eV, 1.55 ( IO.09) eV and ( f 0.04) eV respectively for the diffusion,
1.25
In the solid phase epitaxial crystallization at 500 “C the gold diffusivity in a-Si is so high that the gold outruns the interface, thus producing the flat-topped profiles in Fig. 4. However, solid phase epitaxial crystallization can be induced by ion irradiation at temperatures as low as 200 “C with an activation energy of only 0.3 eV and interface velocities of about 1 A s ‘. At such low temperatures the gold diffusive velocity is comparable with the interface velocity. The segregation behavior in this low temperature regime of ion beam epitaxy should therefore show some unique features. Before describing the ion-beaminduced segregation phenomena, we shall discuss the effect that the ion irradiation has on the gold diffusivity [ 71. The gold was implanted at energies of 100 keV into a-Si layers 2 pm thick. The samples were then bombarded with 2.5 MeV argon ions with doses in the range 2 x lo’“-2 X lOI ions cm ’ and rates in the range 7 X lo”-7 X 10” ions cm-’ s-r. The sample temperature was varied in the range 77-693 K. Figure 5 compares the experimental depth profiles of gold for the usual thermal diffusion at 593 K and after 2.5 MeV argon irradiation at the same temperature. After 1430 s, thermal annealing (Fig. 5(a)), the depth profile broadens slightly, demonstrating that diffusion has occurred. The broadening is considerably enhanced in the ion-beam-irradiated samples (Fig. S(b)) where a profile after 2.5 MeV argon irradiation at a dose of 1 X IO” cm- ’ is
reported. The dose rate was 1 x 10’: ions cnr s-1 resulting in an irradiation time of 1430 s. It should be noted that the thermal diffusion result was obtained by analyzing the sample irradiated with argon at 1 X 10” cm 2 and 593 K just outside the irradiated spot. ‘l‘he depth profiles, measured by Rutherford backscattering, were fitted by solving analytically Fick’s equation forDt, the diffusion length squared. The ion-beamenhanced diffusion D” t is obtained by subtracting the thermal diffusion II,,,! from the total diffusion Dt, i.e. D*t = Dt - D,,t. We find that L)*r is linear with the argon irradiation dose. However. it is independent of dose rate. i.c the experimentally measured diffusion profiles are identical if we vary beams current by over an order of magnitude, keeping the total dose constant. Figure 6 shows the total diffusion coefficient L) for a dose rate of 7X IO” Ar‘ cm 1 s ’ and a total dose of 1 X 10” Ar ’ cm :. Our previous measurements [4] of the thermal diffusion coefficients are also shown. The total diffusion shows three well-defined regions. At temperatures beneath 400 K, the diffusion is athermal and is due to ballistic mixing. At temperatures above 300 K the diffusion is enhanced by several orders of magnitude with respect to thermal diffusion. At the highest temperatures, thermal diffusion dominates. We obtain the thermally activated radiation-enhanced diffusion (RED! coefficient by subtracting the thermal and ballistic components
,,. (KI
T ,o.,800
/(‘Ii
700
600 11
500 ___--_&::_450 400
350
..__I:
A”, 2 ~10’~/cm*
ENERGY 2$,6
I:
I;
I
I,6
Thermal 16 2
-,.
h4eV)
)
17
Ion-Beam 2 5 MeV Ar,lx10”/cm2
593K,1430
16
,
/
__- RADIATION
-ENHANCED
BALLISTIC
set
1000/T
DEPTH
Au, 5x10i4/cm2
til
Fig. 5. Depth profiles of implanted gold markers in a-Si before and after annealing for 1430 s at 593 K with and without 2.5 MeV argon ion irradiation. The argon irradiation was for a dose of 1 X 10” cm- 2. (From ref. 7.)
(K-‘I
Fig. 6. Diffusion coefficients of gold in a-Si: 0. 0, 2 X 10’” cm ? gold implants; a, A, 5 x 10” cm-? gold implants; 0, A, total diffusion coefficients; 0, A, thermal diffusion coefficients: - - -, guide to the eye; -, thermal line, from previous experiments; -, radiation-enhanced and ballistic lines, fits to the data. The diffusion coefficients are quoted for argon doseratesof7x10’3cm~~2s~‘.iFromref.7.)
45
from the total diffusion coefficient. The RED coefficient is shown in Fig. 7 and has an activation energy of 0.37 (_+0.1) eV. We [7] have carried out similar measurements for silver and copper which have RED activation energies of 0.39 ( +_0.1 ) eV and 0.27 ( + 0.1 ) eV respectively. These are the first measurements of RED in amorphous semiconductors. The well-defined activation energies tend to indicate that the diffusion is occurring via a well-defined defect in a-Si. These measurements now allow us to make some general comments on a-Si. There is the global correlation that the slow or substitutional (e.g. arsenic) diffusers in c-Si are slow diffusers in a-Si. Similarly the fast diffusers (e.g. gold) in c-Si are fast in a-Si. We have pointed out previously [3] that the correlation is even more specific for gold, copper or silver where the extrapolated amorphous and crystal activation energies and preexponentials are very similar. This correlation either is fortuitous or points to a common diffusion mechanism. These values of RED in a-Si can also be compared with other amorphous systems. Recent measurements [8-10] of gold and copper diffusion in amorphous Ni-Zr alloys give activation energies for thermal diffusion of 1.7 eV and 1.6 eV respectively. Moreover, 1 MeV krypton irradiation causes RED of copper with an activation energy of about 0.45 eV. It is intriguing that the activation energies are similar even though the bonding is very difficult between the metallic alloys and covalent a-Si.
4. Ion-beam-induced crystallization and segregation We [11] have discovered an unusual regime of interfacial segregation by the use of megaelectronvolt ion beams. The experiment was performed as follows. Gold was uniformly diffused through a-Si layers 1/zm thick formed by implantation. Typical diffusion conditions were 485 °C for 24 h at 10 7 Torr resulting in a-Si layers 1.1 /~m thick with gold concentrations of about 0.2 at.%. Ion-beam-induced crystallization experiments were carried out in the temperature range 250-420 °C using 2.5 MeV ion bombardment: Figure 8 shows Rutherford backscanering and channeling spectra of a sample before and after irradiation at 320 °C with 2.5 MeV argon at a d o s e o f 2 × 1 0 ~Tionscm = and a dose rate of 7 x 10 ~3 ions cm 2 s ~. The gold depth profile is shown in Fig. 9. The dose rate was kept constant, yielding an interface velocity v of about 3 A s ' The zone-refined profile displays the characteristic features of the segregation process: buildup of segregated solute at the interface and concomitant removal of material--"the initial transient". It should be noted that the initial transient X c extends for 0.2/~m. At steady state the amount of material in the segregation spike remains constant and solute is rejected behind the moving interface. It is remarkable that the gold is trapped in c-Si at concentrations some 10 orders of magnitude greater than the equilibrium solubility in c-Si. Transmission electron microscopy
T(K] 800 I
700 i
600 I
500
400
I
ENERGY ( M e V )
I
i
8Kf
10-14
05
IO I
I~l
]
I
I
I 5
[
I
2
I
I
I
20
5 MeV Ar
T = 320°C
z 8 v
8
~:~ Io-t5
uJ
l: ¸¸ o ~,u, 2 × lo~5/c~2
~ I
Z~ Au,u, 5 x 1014/cm 2m
"
Pq
~
\ 10~t62
I 14
1!6181 IO00/T
2!022 (K 1)
L
0
J
1
I
I
I II
I
100 2!4
26
Fig.&
Fig. 7. RED coefficients of gold in a-Si obtained by subtracting the ballistic and thermal components from the total diffusion coefficient. (From ref. 7.)
Rutherford
L
200
L
I
I IL
Ao : 0 87,u_rn
I
500 CHANNEL
backscattering
400
and
(|00)
500
channeling
spectra (2.0 MeV helium) from 1.1 ,urn a-Si containing gold before (o) and after (A)ion-beam-induced crystallization with 2.5 MeV argon. (From ref. I 1.)
46
oo
i
~
0.6
'
A
---INITIAL PROFILE o ION BEAM SEGREGATED
i
OOLATED eA I!
I1
- - , , = ooo ,
5 a: 04 o
o 02 II(
.
.
.
.
.
_
I _ ~ 0~ o
1o
--
!
v 0.5 DEPTH (y.m)
i o
Fig. 9. Gold depth profile in c- and a-Si following 2.5 MeV argon-induced crystallization at a temperature of 320 °C and a dose of 2 x l 0 ~7 c m - ' : - - , fit to the experimental profile assuming only one free parameter, k" (From ref. 11.)
(TEM) shows this c-Si to be defect free without evidence of gold precipitation but the Rutherford backscattering and channeling measurements give no evidence that gold atoms reside on substitutional lattice sites. The experimental solute profiles have been fitted* using only one free parameter: the interfacial segregation coefficient k'. Figure 9 shows a fit, k '= 0.007 + 0.04, using the measured interface velocity of 2.9 A s - ~ and radiation-enhanced diffusivity of 4.4 × 10 ~5 cm 2 s-~. The fits have been convoluted with a gaussian of o = 300 A to allow for detector resolution, straggling and interface waviness as determined by TEM. We do not have the experimental resolution to measure the depth profile directly and calculate the width D/v of the segregated spike to be about 20 A with a peak concentration of gold at the interface of about 20 at.%. This value is clearly in excess of equilibrium solubilities of gold in a-Si which, for example, we measure to be 0.7 at.% at 515 °C. The solubility in the interface region is therefore markedly enhanced by the ion irradiation over the 20 A spike width. The segregation of gold at the moving amorphous-crystalline interface demonstrates classic features of the segregation process which are difficult, if not impossible, to quench and capture in the liquid phase. Since a-Si exists metastably at room temperature, the solute profile can be thermally quenched at any particular phase of the regrowth process. It is possible to *The segregation profiles were obtained using a movingboundary implicit finite difference formulation of the parabolic diffusion equation. Diffusion in the c-Si phase was ignored.
observe the process because of the very high solubility and diffusivity of gold in a-Si. lonbeam-induced crystallization of a-Si containing impurities such as arsenic show no segregation effects, demonstrating that k = I. Although the crystallization and segregation processes demonstrate the classic features of liquid phase segregation, there are marked differences. Firstly the process is highly non-equilibrium with gold being trapped in c-Si at concentrations far in excess of solubility limits. This behavior is manifested by the k' values which, for example, are three orders of magnitude greater than the equilibrium value observed during liquid phase crystal growth. The velocity and temperature dependences of k' demonstrate some of the unique characteristics of the ion beam process. The interface velocity scales with the argon dose rate. We observe that an order-ofmagnitude change in velocity (i.e. same dose but different dose rates) produces identical segregation profiles. Specifically, the equilibration distance D/k'v (the width of the initial transient) must remain constant for all interface velocities. We know that the diffusivity scales linearly with the dose rate. Therefore, D/v is independent of interface velocity. The segregation coefficient k' must also be independent of velocity. This behavior is a consequence of the beam-induced crystallization process and is quite different from that occurring in liquid phase epitaxy where k' scales with v because the chemical driving force for trapping increases with the velocity-dependent undercooling in the liquid. The variable which strongly affects k' is temperature. Fits to the experimental profiles at 250 °C and 375 °C using measured diffusivities o f 1.6 x 10 is c m 2 s i a n d 9 × 1 0 ~5 cm 2 s respectively give k' values of 0.013 _+0.007 and 0.005 + 0.004. We do not yet have an understanding of these results in terms of the crystal growth processes. The driving force for trapping is related to the chemical potentials of the impurities in the a- and c-Si which depend on temperature but are independent of the interface velocity. It is not known how significantly the production of defects by the ion beam will change the chemical driving forces. Moreover, ballistic mixing of atoms at the interface may also provide the non-equilibrium mechanisms for driving gold at the interface into the growing crystalline phase. Whatever the mechanism, we have evidence from the magnitude of the segregated zone and
47
enhanced gold solubility that the a-Si interface region during ion-beam-induced crystallization acts as a source or sink of defects.
Acknowledgments I am indebted to Francesco Priolo, Dale Jacobson and Mike Thompson for their collaboration in the diffusion and segregation studies. References 1 E. P. Donovan, F. Spaepen, D. Turnbull, J. M. Poate and D. C. Jacobson, J. Appl. Phys., 57(1985) 1795. 2 G.L. Olson, Energ)' Beam-Solid hlteraction and Transient
Thermal Processing, Materials Research Society 3[vmp. Proc., Vol. 35, Materials Research Society, Pittsburgh, PA, 1985, p. 25.
3 J. M. Poate, D. C. Jacobson, J. S. Williams, R. G. Elliman and D. O. Boerma, A"ucl. lnstrum. Methods B, l~)-20 (1987) 480. 4 D. C. Jacobson, J. M. Poate and G. L. Olson, Appl. Phys. Lett., 48(1986) 118. 5 R. G. Elliman, J. M. Gibson, D. C. Jacobson, J. M. Poate and J. S. Williams, Appl. Phys. Left., 46 (1985) 478. 6 B. Chalmers, Principles of Solidification, Wiley, New York, 1964. 7 F. Priolo, J. M. Poate, I). C. Jacobson, J. Linnros, J. L. Batstone and S. U. Campisano, Appl. Phy~. Lett., 52 (1988) 1213; to be published. 8 H. Hahn, R. S. Averback and S. J. Rothman, Phys. Rev. B, 33 (1986) 8825. 9 El. Hahn, R. S. Averback, E-R. Ding, C. Loxton and J. Baker, Mater. Sci. Forum, 15-18 (1987) 551. 10 R. S. Averback, H. Hahn and F.-R. I)ing, J. LessCommon Met., in the press. 11 J. M. Poate, J. Linnros, F. Priolo, D. C. Jacobson, J. L. Batstone and M. O. Thompson, Phys. Rev. Lett., 60 (1988) 1322.