Langmuir 1995,11, 4862-4872
Initial Stages of Oxidation on Co-Ni Alloys: Island Nucleation and Growth E. E. Hajcsar,?>$ P. R. Underhill,*?#and W. W. Smeltzerl' Department of Mechanical Engineering, Royal Military College, Kingston, Ontario K7K 5L0, Canada, and Departments of Chemistry and Metallurgy, McMaster University, Hamilton, Ontario, Canada Received July 18, 1994@ The formation and growth of oxide islands during the initial stages of oxidation of a Co-Ni alloy are studied by direct imaging of these islands using scanning Auger microscopy. These images give clear experimental evidence in support of the Holloway-Hudson model of oxide growth in which the islands nucleate at one point in time and then grow at a rate determined by the rate of gas impingement at the island perimeters, the surface crystallographic orientation, and the substrate temperature. The islands are also shown to be Co rich. 1.0. Introduction
The oxidation of metals and alloys has been the subject of a large number of studies up to this time. The growth kinetics of oxide layers have been studied and models have been proposed to explain the trends observed on many systems.' Similarly, the initial interaction of gaseous oxygen with clean metal surfaces has been studied with particular emphasis on the structures of adsorbed and chemisorbed phases, especially that of Ni2-11plus many others. Both ofthese areas of study concentrate on stages in an overall process that ideally begins with clean metal surfaces adsorbing and reacting with gaseous oxygen and ends in the growth of an oxide layer controlled by parabolic, linear, paralinear, or some other kinetic mechanism. This study concentrates on the necessary intermediate stage in the overall process that links either extreme, the nucleation and growth of an initial oxide layer until total surface coverage is achieved. 1.1. Chemisorption. We shall first review the case of oxygen uptake on pure Ni. The first stage of oxygen uptake on Ni was thought to be dissociative chemisorption which obeyed a Langmuir type adsorption isotherm. This observation has been generally supported in the literat ~ r ewith ~ , some ~ early questions about whether or not the adsorption was nondissociative being cleared up by Benninghoven et ~ 1 and . Verheij ~ et a1.6 via SIMS flash desorption, who concluded that adsorption onto the metal was dissociative. Subsequent ~ o r khas ~ -also ~ shown that there are two distinct surface phases of chemisorbed Department of Chemistry, McMaster University. No current affiliation. 9 Department ofMechanica1 Engineering, Royal Military College. 'I Department of Metallurgy, McMaster University.
oxygen characterized by their LEED structural description ~ ( 2 x 2followed ) at slightly higher exposures by 4 2 x 2 ) . Langmuir kinetics can be modeled simply by assuming that the surface contains a certain number of adsorption sites (SI,some ofwhich may be occupied (5'1) such that the number of sites available at any one time for adsorption would be SO= S - SI. If adsorption is assumed to be a nonactivated process, then the rate of adsorption is just proportional to the number of available sites, the surface collisional frequency (which must be proportional to the pressure), and the sticking probability; therefore
rate of adsorption = k$So = k$(S
- SI)
(1)
Assuming first order desorption, the rate of desorption is proportional to rate constant k l and the number of surface sites which are already covered in adsorbate S I . At equilibrium the two rates are equal and if we now define 0 = S1lS as the fraction ofthe surface covered by adsorbate at equilibrium, the Langmuir adsorption isotherm is obtained:
e=- bP 1
+ bP
where b = kdk1. This isotherm must be modified in the case of dissociative chemisorption (the proposed mechanism in this case). Two adjacent sites must be present to absorb or desorb an oxygen molecule. Consequently the adsorption rate becomes k&S - S1Y and the fractional coverage is
+
@
Abstract published in Advance ACS Abstracts, November 1,
1995. (1)Smeltzer, W. W. Proceedings of The Norman L. Peterson Symposium: Oxidation ofMetals and Associated Mass Transport; 1986; p 109. (2)Norton, P. R.; Tapping, R. L.; Goodale, J. W. Surf. Sci. 1977,65, 13-36. (3)Brundle, C. R. Chem. Phys. Lett. 1976,31, 423. Brundle, C. R. J.Vac. Sci. Technol. 1979,16,548. (4)Hopster, H.; (5) Muller, K. H.; Beckmann, P.; Schemmer, M.; Benninghoven, A. Surf. Sci. 1979,80, 325. (6)Verheij, L. K.; Van Der Berg, J. A.; Armour, P. G. Surf. Sci. 1979, 84,408. (7)Mitchell, D.F.;Sewell, P. B.; Cohen, M. Surf. Sci. 1976,61,355. (8)Taylor, D. E.;Park, R. L. Surf. Sci. Lett. 1983,125,L73. (9)Demuth, J. E.;Rhodin, T. N. Surf. Sci. 1974,45, 249. (10)Kortan, A. R.; Park, R. L. Phys. Reu. B 1981,23,6340. (11)Brundle, C. R.Vac. Sci. Technol. A 1986,3, 1468.
0743-746319512411-4862$09.0010
@=
bP1I2 1 bP'I2
+
(3)
At higher pressures, the rate of adsorption will far exceed the desorption rate and saturation will ensue. At very low pressures, eq 3 reduces to 0 = bP1I2since 1>> bPvz. The sticking coefficient is described as the probability that a particle striking the surface becomes chemisorbed, and the sticking coefficient for a Langmuir dissociative isotherm onto a disordered surface, as in the above idealized case, is
In the case of the NU0 system, the early works of Holloway 0 1995 American Chemical Society
Lungmuir, Vol. 11,No. 12, 1995 4863
Oxidation on Co-Ni Alloys lateral Diffusion
\
. - .
the model development are
or
Figure 1. Schematic representation of the Holloway-Hudson model of oxide island growth.
,
and Hudsonf3and Mitchell, Sewell,and Cohen’ indicated that adsorption proceeded onto a disordered surface phase via a nonactivated dissociative Langmuir isotherm but later work by Hollowayf4stated that the chemisorbed oxygen resides in 2-fold bridge sites on Ni( 111) and 4-fold sites on Ni(100). He further states that the sticking coefficient varies as (1- 4 0 ) for the Ni(100) face and (1 - 3 0 ) for the Ni(110)face. This meant that only specific sites and not just any unoccupied site are needed for chemisorption. A later work by Brundlell states that rather than a 4-fold site on Ni( loo), the limiting available site is a 0 8 which has two adjacent 4-fold sites with no nearest neighbors to repel the two new incoming 0 2 molecules. Clearly the exact nature of the isotherm is face dependent and the exact mechanism is not known for most of the faces. In this effort however, the exact nature of the chemisorption reaction is not of primary interest but the following stage in the overall oxidation process is. 1.2. Oxide Nucleation and Lateral Growth. Holloway and Hudson12J3first elucidated a theory that oxides form on Ni via the nucleation and lateral growth of islands of oxide which were approximately two to three “layers” of oxide thick. The existence of islands of NiO was suggested by the appearance of a pattern consistentwith domains of the oxide ~ ( 1 x 1 )in LEED photographs concurrent with the patterns identified as corresponding to adsorbed Oc(2x 2). This observation,coupled with AES peak height and shape data, led Holloway and Hudson to conclude that NiO nucleated and grew as islands after the adsorbed 0 layer reached ~ 0 . 3 5monolayers of coverage. The model proposed to explain the growth kinetics of the islands is as follows (see Figure 1)similar to the original Holloway and Hudson model. Five assumptions are made: (i) Physisorbed 0 2 exists on top of either chemisorbed 0 or oxide with a mean stay time of tp,a lateral diffusion coefficient D,, and an accommodation coefficient of 1into the physisorbed state. These physisorbed 0 2 molecules then difise laterallyuntil they reach an island perimeter. (ii)0 2 dissociates at perimeter sites only. (iii) Islands are assumed to have a regular (circular) shape for mathematical simplicity. (iv) tp and D, are the same for 0 2 physisorbed on chemisorbed 0 and on the oxide. (v)All islands are nucleated in a short time relative to the time necessary to form a complete oxide coating. It is therefore possible to envisage growth in this model being limited by (1)oxygen impingement from the gas phase, (2) surface diffusion of oxygen, and (3) capture of oxygen perimeter sites. Holloway and Hudson discounted case 1 because the Auger 0 peak height displayed a nonlinear variation with oxygen dose, contrary to what would be expected if the island growth rate were simply proportional to the impingement rate of 0 2 . .For cases 2 and 3 the results of (12) Holloway, P. H.; Hudson, J. B. Surf Sci. 1974,43, 123. (13) Holloway, P. H.; Hudson, J. B. Surf Sci. 1974,43, 141. (14) Holloway, P. H. J . VUC.Sci. Technol. 1981, 18,653.
where L represents the unit of exposure langmuir (1 x Torres) and where
is an island growth parameter withA = area of an island, m is the mass of oxygen, k is the universal gas constant, and Tgis the gas temperature. In the case where island growth is limited by lateral diffusion
B = 2(Za2vl/v2)v2 where V I = vibrational frequency parallel to the surface and v2 is the perpendicular vibrational frequency, a is the mean jump distance for lateral diffusion and 2 is the number of nearest neighbor adsorption sites. In the other case where growth is limited by incorporation of 0 on the edges of islands
B, = AU2vl/v2 The activationenergy for the case of limitation by lateral diffision, Ed, is assumed to be half the difference between the activation energies for adsorption and desorption whereas in the case of limitation by capture, the energy, E,, is assumed to be the difference between adsorption and capture. The existence of island growth has been generally supported in other w 0 r k ~ and ~ ~not - ~only ~ on Ni but also on other metals and there is a distinct possibilitythat the mechanism is general. There has been some disagreement however as regards the exact model proposed by Holloway and Hudson. In fact Brundle, as well as Mitchell et al., asserts that the true mechanism must not involve physisorbed 0 2 since even a t 300 K, the residence time of such a species would be very short. Rather, the rate of island growth must be limited by direct capture a t island edges from the gas phase. The exact thickness of the oxide islands has also generated some disagreement, but Saiki et a1.20 have determined by X-ray photoelectrondiffraction that the thickness of oxide islands generated by annealing a saturated Ni(100) surface at 523 K for 10 min was two to three monolayers of 0 in the NiO lattice. The existence of good LEED data along with the oxygen uptake curves from AES studies has been enough to convinceother workers that an island growth mechanism is operative in this regime, but there has not been any unequivocal imaging of oxide islands at the two to three monolayer thickness level as has been done in this work. Goulden21and Milne and Howie22have conducted SEM, (15) Mitchell, D. F.; Sewell, P. B.; Cohen, M. Surf Sci. 1977,69,310. (16) Norton, P. R.; Tapping, R. L.; Goodale, J. W. Su$ Sci. 1977,65, 13. (17) Bendorf, C.; Egert, B.; Nobl, C.; Serdel, H.; Thieme, F. Surf Sci. 1980,92,636. (18) Van Der Berg, J. A.; Verheij, L. K; Armour, D. G. Sur$ Sci. 1980,91,218. (19) Holloway, P. H.; Outlaw, R. A. J. Vac.Sci. Technol. 1982,20, 671. (20) Saiki, R.; Kaduwela, A.; Osterwolder, J.; Sagurton, M.; Fadley, C. S.; Brundle, C. R. J . Vac.Sci. Technol.A 1987,5,932. (21) Goulden, D. A. Philos. Mag.1976,33,393. (22) Milne, R. H.; Howie, A. Philos. Mag.A 1984,49, 665.
4864 Langmuir, Vol. 11, No. 12, 1995
RHEED, and TEM investigations into oxide island formation on copper. These experiments generated islands by exposing clean copper to oxygen in the Torr range at temperatures from 250 to 400 “C for from 1to 150 min. The islands enerated were at least 300 A and more usually 600-800 thick and as such were much larger than those observed in this work. The lack of surface chemical information during imaging, as well as the very heavy oxidativetreatment used to generate these features, leads to the question of whether or not these “islands” were either nuclei of columnar grains on an already completed compact oxide film or an example of the same island growth process under very dissimilar conditions. With the combination of a scanning Auger microprobe, a hot stage, and a gas injection system that allowed controlled doses of oxygen to be introduced to samples held at elevated temperatures, we have been able to monitor this stage of the oxidation process and do some crude measurements of the kinetics involved. We have observed that an island nucleation and growth process are operative in this regime and have probed the kinetics experimentally.
Hajcsar et al.
..............
A
oxide
(23) Underhill, P. R.; Ellison, K. A.; Smeltzer, W. W. J. Vm.Sci. Technol.A 1986,4,1944. (24) Hajcsar, E. E.; He, Z. Z.; Underhill, P. R.; Smeltzer, W. W. J . Vac.Sei. Technol.A 1988,6,3006.
sonpllng depth
12 layers oxide
1
2.0. Experimental Methods Utilizing a hot stage described elsewhere,2O polycrystalline and single crystal alloy wafers were exposed to measured doses of oxygen at varying pressures and times while the sample was held at elevated temperatures. Although island nucleation and growth were observed on polycrystalline and single crystal alloy samples over the whole range of the Ni-Co system, as well as pure elemental standardsF4this work reports data exclusively from polycrystalline Ni(79) Co alloy samples. 2.1. Samples. Alloy preparation was carried out in an arc-melt apparatus built at McMaster University with a rotating water-cooled OFHC copper hearth. The sample preparation chamber was evacuated via rotary pumping to a start pressure of approximately Torr and backfilled to approximately 380 Torr dynamic pressure (continuouspumping)of research grade Ar. Pure element rods (99.999%pure) were sliced to lengths appropriate to the target composition. Each element was separately liquefied under Ar and allowed to degas and eventually cooled to form a pure element boule. Pure element boules were then weighed to f0.05mg accuracy and placed side by side in the arc-melt chamber. The elements were allowed to mix as liquids for several minutes and allowed to resolidify. Each sample boule was then turned at least seven times and remelted to ensure completehomogeneity. The sample boules were then annealed at just under the melting temperature for 10 min and cooled slowly. The resultant boules were then reweighed and the composition calculated from the ratio of the starting elements used. The error on the compositionwas calculated by assuming that all mass loss was due to evaporationof the solute and was thus a very worst case estimate that was at worst 0.1% and at best 0.01%. Sample wafers approximately 3 mm thick were taken from the boules using a high-speed diamond saw fitted with a goniometer sample mount and were then polished in a parallel surface polishingjig. Polishing consisted of increasinglyfine grit abrasives with the final finish being a 1A diamond paste to leave a mirror-like surface free of all visible scratches.
>
Figure 2. Schematicrepresentationof the Auger 0 KLL peak as a function of oxygen content on the surface.
Kinetic Energy, cv
si0
sis
!O
Kimtic Energy,ev
Figure 3. (a)0KLL peak obtained from a monolayer of 0 on NiCo alloy. (b) 0 KLL peak obtained from a well-oxidized Co sample.
Polycrystalline samples subsequently studied in this and other works were repeatedly annealed and sputter cleaned with 4 keV Ar+ ions such that grain structure was readily visible as if etched or electropolishedalthough neither of those methods was employed. 2.2. Sample Mounting, TemperatureControl, and Shielding. As we have noted, the sample hot stage is described e1sewhe1-e~~ but some of its salient features need to be described. Samples were attached to an insulated inner ring via thin tungsten wires spot-welded to the outer edge of the sample wafers. The samples rested on a Macor ceramic washer inside the inner ring and the heating electron stream struck the backside of the sample through the center of the washer. The size of the washer hole was chosen to be just slightly smaller than the diameter of the samples to ensure maximum heating and minimal heat loss and reduce electron leakage from the heating filament. To firther eliminate electron leakage, a molybdenum foil was spot-welded over but not touching the sample and was attached to the inner ring. A hole was carefilly cut out of the foil to be smaller but concentric with the underlying sample. The combined overlap of sample on Macor washer and Mo foil over sample effectively eliminated electron leakage. On occasions where the sample mount was imperfect and there was electron leakage, the electron optics of the scanningAuger system detected the leak and the sample mount was remounted until no leakage occurred. Chromel-alumel thermocouple leads were attached to the sample surface by spot-welding through the foil hole and were located as near to the edge
.
Langmuir, Vol. 11, No. 12, 1995 4865
Oxidation on Co-Ni Alloys
i
Figure 4. Auger images of the 0 KLL line (509 eV) from partially oxidized NiCo surfaces at 675 K (a, top left) 0 langmuirs; (b, top right) 120 langmuirs; (c, middle leR) 720 langmuirs; (d, middle right) 1800 langmuirs; (e, bottom) 3000 langmuirs. Note, one side of maps is approximately 25 pm.
of exposed sample surface as possible without contacting the shielding foil. Constant temperature feedback allowed for good thermal control in all experiments. While the maximum achievable stable temperature was approximately 1200 K, the experiments described in this study were carried out between 600 and 675 K with a presumed accuracy of f 5 K. 2.3. Data Acquisition: Map Generation. The best evidence for the existence of very thin oxide islands to date has been the observationof N E D patterns consistent with domains of NiO concurrent with patterns attributed to chemisorbed oxygen phases on Ni (see above). There had, however, been no unequivocal imaging of these
islands before our preliminary reportz4 because the contrast provided by secondary electron emission from islands that are approximately two to three layers thick is insufficient. Furthermore, the island growth reaction proceeds even at very low pressures of 0 2 (1 x lop7), therefore dosing must proceed in a UHV chamber where the background pressure during imaging can be held at Torr. The PHI 600 S A M used in this study, equipped with the hot stage described above allowed the collection ofAuger spectra and element distribution maps at elevated temperatures. The need for an elevated temperature stemmed from the fact that the probability for nucleation is likely to decrease with increasing temperature, and if fewer islands nucleate and grow, the islands present might
l L Y V Yx 100
x x
x
::
?I
0
0
x
Y
0
O
emeta1
0
o
0
0
0
0
0
Tiine (miid Figure 5. Expanded surface coverage (u) for different layers as a function of 0 substrate temperature of 675 K.
2
exposure at a pressure of 1 x
Torr and a
rst layer islands impinge
300
900
1500
2100
2700
3300
Exposure (L) Figure 6. Coverage (a)for different layers as a function of 0 2 exposure at a pressure of 1 x of 675 K.
be large enough to image with the limited spatial resolution of a scanning Auger system. Holloway and Hudson utilized the relation of Gallon25 who pointed out that an Auger oxygen signal originating in an oxide of n layers thickness will have an observed intensity of (6)
where I , is the signal from an infinite number of layers and Il is the signal from one layer of material. Equation 6 holds true if there are uniform layers present, but in the case where there are regions in the sampling field with islands of varying thickness, then the relation of SeahZ6 may be more appropriate
(25) Gallon, T. E. Surf. Sci. 1969,17,489. (26) Seah, M. P. Surf. Sci. 1972,32,703.
3900
4500
Torr and a substrate temperature
where n, = { 1- exp(-l/A)} and1is usually the attenuation length. However, the fractional coverage is not included in such a model and it is clear that direct quantitation in this regime is likely very difficult and complex. For the purposes of this study, the exact intensity correction is not critical, but the idea that the intensity ofAuger signal increases in discrete but dwindling steps up to the effective sampling depth is important. In Figure 2 , the Auger peaks are schematic representations only, but the fact that the chemisorbed oxygen peak is shifted to higher energy than the oxide is real and is shown in Figure 3 where an oxygen signal from the very early stages (chemisorption) of an exposure experiment is presented with a signal from the latter stages where the sample (pure Co standard in this case) is oxidized. This observed shift is consistent with other work on Ni.2,7s13-15 The vertical line in Figure 2 is set at 509 eV which is the peak energy used in producing Auger maps here. (Auger maps plot the difference in counts between a defined peak energy and background divided by the background signal.) With spatial resolution of at most 0.2 pm and islands typically in the range of several
Langmuir, Vol. 11, No. 12, 1995 4867
Oxidation on Co-Ni Alloys micrometers, the N(E) signal used to generate maps originates from an area much smaller than the islands observed here. Given this information, an Auger map should contain pixels whose intensity is proportional to the concentration of the analyte in the sampled depth and, in the case of islands, proportional to the depth (height) of the island if that depth is less than the sampling depth and there is no analyte in the bulk. There is likely to be some blurring at island edges but this is thought to be minimal with the above stated resolution and the observed island diameters. Figure 4 displays maps which were collected from one grain of a polycrystalline 79 d o Ni-Co alloy held at 675 K and exposed to oxygen gas at 1x Torr in a dosewise fashion. Figure 4a is an oxygen map at 0 langmuirs exposure and the subsequent maps shown in Figure 4b-e are in chronological order where the total exposure in Figure 4e is 3000 langmuirs. The beam voltage used for all maps was 5 kV and the resolvingpower ofthe analyzerwas set to 1%.The current density of the analysis beam was reduced to a level that displayed no apparent beam damage for periods of time longer than that required to collect a map (typically 12 min) since at very high current densities some darkening of the map field and loss of a small portion of the oxygen Auger signal were observed if the beam was allowed to rest for long periods of time on one spot. Map collection was therefore conducted as quickly as possible and the electron beam turned offwhen not actually collecting data or during contaminant checking sweeps. This sequence of photographs gives direct evidence for the existence of oxide islands which grow laterally at least. It should be noted that for all the maps displayed here, a size marker could not be produced on screen. An approximate scale has one side of the map square at about 20 pm. 2.4. Data Handling: Digital Image Processing. With the establishment of the island growth mechanism in this system, the question of which process is limiting arises out of the development of the model. In order to perform any kinetic experiments, it is necessary to obtain a numerical measure of the coverage at any one time. Toward this end, digital image processing has been employed. The Auger maps can be reduced to a histogram of intensities for all pixels in the analysis window. From the preceding arguments, peaks should be observed at discrete arbitrary intensity levels, which correspond to differing island thicknesses. A clean metal will have lowest intensity while a chemisorbed signal should be at an intensity between clean metal and the first layer of oxide both because the positioning ofthe sampling energy means that only a shoulder of the chemisorbed peak is measured and the oxygen content on the sampling region is much smaller for chemisorbed 0 than for an oxide. One complete oxide layer will be next highest, followed by subsequent oxide layers up to the sampling depth. Random variation in any of the signals should give the intensity peaks in the histogram a Gaussian-like shape. Parts a-e of Figure 4 have the corresponding histograms of intensity plotted next to the map. Close inspection of the histograms reveals the presence of up to, but usually less than, five heavily overlapped Gaussians with peaks at intensity levels of 0, 60, 120, 160, and >220 (difficult to see). The assignment of the zero is obviously that corresponding to bare metal (black). The next peakis assigned as chemisorbed 0 (green) and the following peaks as one layer of oxide (yellow),two layers of oxide (red),and more than two layers of oxide (white). No peak assignments are made for 0 2 adsorbed for two reasons: (i) at the temperatures used in this work, 600 Kor higher, the mean
... s 80 /
,
___
1 x 1 0 6 torr
5 x 10-7 torr 1 x 10-7torr
,.
1
.
7
'
, I
10
20
50 GO Exposure Time (min)
30
40
70
80
90
Figure 7. Corrected fractional coverages ofthe first oxide layer for exposure at different pressures at a substrate temperature of 600 K.
residence time of a physisorbed species is most likely vanishingly smallll and (ii)the probe electron beam would in all likelihood desorb such a weakly bound species if it existed on the surface at all. With the histogram divided into five contrast levels which have been selected to bisect the difference between peaks of interest and their next neighbor, the colorenhanced images of parts a-e of Figure 4 have been produced. While a deconvolution algorithm would have been preferable, the lack of such a tool limited us to manual peak assignment and division based on over a hundred such map sequences. The data that can be extracted from such digitally enhanced images can therefore yield fractional coverages @met, Bchmsrb, 01, etc. Plots of these individual 0 ' s vs exposure for the full series of maps that are partially represented in Figure 4 are shown in Figure 5. While the appearance of Figure 5 is at first somewhat confusing because the chemisorbed and one-layer curves reach a maximum and then decrease, the realization that where there is a signal for a two layer there must be a one layer underneath leads to the last data manipulation step of adding the fractional coverage of deeper species to the fraction obtained directly from the map. The bare metal coverage is of course by definition the deepest layer and is therefore taken directly from the map without this correction. Figure 6 displays the data from Figure 5 with this correction included and one may now see some very interesting trends. One such trend is the fact that the chemisorbed layer (Le. the layer covered by ut least chemisorbed 0 )is a near mirror image for the clean metal trace which is exactly what one must see. Furthermore, the general shape of this curve is reminiscent of a Langmuir adsorption isotherm, which one might hope to see. The stages are clearly shown in the figures with labels and arrows. 3.0. Results and Discussion 3.1. Kinetic Studies. In Figures 6 and 7, a compilation
of data from several experiments yields some very important observations regarding the kinetics of the island growth mechanism: is (1) The shape of the curves for Ometand comparable to the shape of the 0 vs L curves of Holloway and Hudson which have been more adequately investivated elsewhere and identified as proceeding via a dissociative Langmuir or a modified dissociative Langmuir-type adsorption isotherm. (2) At least the layer assigned as the first oxide layer grows via a three-stage process, as indicated by break points in the curves. The proposed interpretation of this observation is that the stages in chronological order are
Hajcsar et al.
4868 Langmuir, Vol. 11, No. 12, 1995
I I
I
1 Figure 8. 0 KLL Auger image obtained from a NiCo alloy at 675 K early in the oxidation.
8
Figure 9. 0 KLL Auger image of the same area as Figure 8 after further oxidation.
as follows: (i) rapid growth due to nucleation-beyond this bend in the 0 1 curve, no new islands are observed to form on the correspondingmap sequence; (ii)an apparently linear increase in 0 1 probably due to lateral growth limited by the rate of gas impingement close to the edge of growing islands; (iii)nonlinear growth due to the decreasingactive area at the perimeter of islands as they impinge on neighboring islands. The onset of this part of the curve corresponds exactly to the observation of impingement on maps.
(3) The rate of growth in the second oxide layer (linear lateral growth) is different from that of the first layer (a distinct nucleation phase was not observed). This leads to the possibility that the geometry and/or energetics at the edge of a second layer island is different from that at the edge of the first oxide layer. (4) Nucleation occurs at preferred sites both away from and on grain boundaries, and the sites are reproducible from experiment to experiment starting each time with
Langmuir, Vol. 11, No. 12, 1995 4869
Oxidation on Co-Ni Alloys a90 80 Q3
v
q.4'
60
z
'
rl
: ',
1
40
20
70.
?-
......... 1 x 10-6torr --- 5 x torr I x 10-7 torr
I
I
500
1M)o
5
/' 1
,
..'..'
...TI.
..... .....
30
-8
I
1500
Exposure (L) '
600
'
1200
'
1200
'
1800 ' 2400 3000 ' 3600 ' 4 Exposure (L) '
m
b" 80 h
a
5
60
- 1 x 10-6 torr --- 5 x 10-7 torr
1
- I x 10-7torr
40
20
500
1000
1500
Exposure (L)
Figure 10. (a)Corrected fractional coverage of the first oxide layer at different pressures at a substrate temperature of 600 K. (b) Corrected fractional coverage of the second oxide layer at different pressures at a substrate temperature of 600 K. a clean surface, i.e. if a sample is oxidized to the saturation level (complete coverage by oxide) and is then cleaned by ion bombardment at temperature until no 0 is apparent, a subsequent exposure will yield islands of the same shape and at the same location. This interpretation of the processes involved in island growth is consistent with the assertions of Brundlell and Mitchell et uZ.l5 who also concluded that the rate limiting step is direct gas impingement at island edges. This mechanism and the experimental observation that when nucleation has stopped, 01 appears to vary linearly with time means it is possible to describe the kinetics of the lateral growth of islands prior to island impingement under these conditions. The rate of growth of the oxide islands dWdt can be written as
dWdt = Kip
(8)
We note that the islands in all experiments are not circular, as in the Holloway-Hudson model, but long and growing usually in width with minimal growth in length. This would mean that Ki appears to be not a function of 0 as might be intuitively thought because the length of the active edges of the islands change so very slowly as to allow the "active area" to be thought of as being essentially constant until impingement. Therefore the surface coverage of an island layer i at time t after the end of the nucleation phase would then be
oi= kiPt = k,L
(9)
wherep = 2mkT,, L is a langmuir = 1 x Torr-s, and ki in the simplest possible model is the fraction of gas collisionsthat successfully incorporate oxygen in the oxide island and is, therefore, an activated process
ki= I$exp(-EJKT)
(10)
where E, is the activation energy of the incorporation process.
*p
6b0
1800 2400 ' 3600 ' 3600 ' 4200' Exposure (L)
Figure 11. (a)Corrected fractional coverage of the first oxide layer for different substrate temperatures at a constant 0 2 pressure of 1x Torr. (b) Corrected fractional coverage of the second oxide layer for different substrate temperatures at a constant 0 2 pressure of 1 x low6Torr.
Although the evidence from the exposure curves and maps is very strong, a test of the interpretation needed to be made. The simplest test of the interpretation was to conduct a series of experiments where both P and T were independently varied during the reaction and the rate of lateral growth was measured. During these experiments it was also possible to test some other aspects of the interpretation. It has been pointed out that nucleation at these elevated temperatures is likely to occur only at preferred sites such as defects and grain boundaries where the local chemisorbed 0 density may more easily reach a critical value for nucleation. To check this, maps would need to be obtained in the region of grain boundaries, and since the conversion of impingement gas is asserted to be determined by the nature of geometry at the island edges, it is also reasonable to assume that there is a crystallographic effect. To check these two points simultaneously, as well as to get average kinetic data, maps were collected at a triple grain boundary during oxygen exposure P and T experiments. The question of nucleation site selection is addressed in Figure 8 where it is clear that nucleation does occur first at grain boundaries but also at other sites on the surface where presumably there were defects (SEMimages at 5000x did not reveal any obvious defects at the identifiable and repeatably preferred sites for nucleation away from the grain boundary). The question of whether or not there was a crystallographic dependence is illustrated very clearly in Figure 9. The lower right-hand grain grows the second layer much slower than the other two grains despite the fact that the apparent coverage of chemisorbed 0 and the first oxide layer is higher on that grain than on the others. Unfortunately, attempts at determining the orientation of the grains by electron
Hajcsar et al.
4870 Langmuir, Vol. 11, No. 12, 1995
layer growth is markedly dependent on temperature. It is interesting to note that increasing temperature decreases the rate of growth. This may indicate that in fact there are two processes in competition during the linear growth phase. One process is the incorporation ofoxygen, presumably from direct impingement and dissociation, the second process is probably either the bulk dissolution of oxygen o r desorption. The temperature range used in this set of experiments clearly affected the size of islands observed. Low temperatures likely allow for a greater range of successful nucleation sites and would thus logically lead to high nucleation rates which would in turn result in the very small islands deduced in previous works.14-16 In those works, much lower temperatures were used (-300-400 K), and although actual measurement of island size was not possible, very small island diameters were proposed. At the elevated temperatures used here (600-675 K), nucleation rate was evidently low and island sizes directly observed were quite large over the time to full exposure. The observed rate is therefore a net rate such that
7
1 600
1200
1800 2400 3000 3600 4200
Exposure (L)
-1 x 10.6 torr -5 x 10-7 torr - 1 x 1 0 7 torr
1
1000 Exposure (L)
500
1500
Figure 12. (a) Percent coverage of the surface by oxygen as a function of substrate temperature at constant 0 2 pressure of Torr. (b) Percent coverage of the surface by oxygen 1x as a function of exposure pressure for a constant substrate temperature of 600 K.
channeling patterns were not successful owingto the small size of the grains and the difficulties associated with this method. To test for a dependence on pressure, three different pressures were used at a constant temperature and the results are summarized in Figure 10. In Figure 10a the coverage of the first layer of oxide vs exposure is plotted for three different pressures, and within experimental error, the rate of lateral growth is dependent only on the total exposure in langmuirs, consistent with eq 10. The second layer data is less conclusive due to experimental scatter but the conclusion can probably still be made that total exposure is controlling the rate of growth at fixed temperature. The dependence on temperature study is summarized in Figure 11. In Figure l l a , the rate of the first oxide
where k , is the rate of capture and kD is the rate of dissolution or desorption. If it is assumed that capture is an unactivated process (or one with a very low activation barrier), then E , (the activation energy of capture) can be assumed to be zero and
k , = kr
- ki
exp(-EdKT)
(12)
This dissolution into the bulk or desorption would therefore explain why the rate of island growth increases with decreasing temperature.
0 = L(k9 - k: exp(-ED/KT))
(13)
Experiments have been performed where oxygen exposure has generated islands which were then heated to a greater temperature than the original exposure in vacuum. The background gas was monitored for any signs of evolving 0 2 or other oxygen-containing species with a quadrupole mass spectrometer residual gas analyzer mounted proximally to the sample, but none was observed. This result, although not absolutely conclusive, does not support the notion of gas desorption as the competing process and leaves bulk dissolution as the most likely candidates. From
1x10-6 t o r r
300
5 ~ 1 0 -t ~ orr
100
P
8
A
-
0
150
I 0
450
.
I
1
750
1050
0
15
-
l;b7; 1500
2100
30
"
. I
45
75
105 t
90
150
210
ee 1x10-8 t o r r
loo
--
1x10-7 t o r r
0
I
5 ~ 1 0 -t~o r r
loo
0
900
0
3
9
Figure 13. Fractional coverage vs exposure in langmuirs at a substrate temperature of 650 K.
15
21
Langmuir, Vol. 11, No. 12, 1995 4871
Oxidation on Co-Ni Alloys
1
4
Figure 14. (a, top) 0 KLL Auger image (peak at 509 eV) of an oxidized NiCo alloy (450 langmuir 733 K). (b, middle) Ni LMM Auger image (peak at 844 eV) of the same area as (a). (c, bottom) Co LMM Auger image (peak at 650 eV) of the same area as (a) and (b).
the four curves of Figure l l a , it is also possible to extract the factors Izp and k: for the various temperatures and an Arrhenius plot may yield a value for the activation energy. This approach is possible, but in light of the fact that there is data available for only four temperatures over a range of 75 K, such a determination would be of limited value. This type of experiment may be particularly usefbl in fbture studies on single crystal faces.
The second layer oxide also exhibits evidence of increasing growth rate with decreasing temperature, since out of the four curves, three display a trend similar to that for the first oxide layer with only the 660 or 675 K being out of the expected order. The variation of the fractional coverages for the chemisorbed layer with temperature is shown in Figure 12a, and as expected, lower temperatures allowed the chemi-
4872 Langmuir, Vol. 11, No. 12, 1995 sorption to proceed more rapidly, presumably since the rate of desorption or bulk dissolution will increase with increasing temperature. The variation of chemisorption with pressure is somewhat less clear here and is presented in Figure 12b. At the highest pressures used in this work, saturation at 0 = 100% ensues &r a certain level of exposure and the total exposure required to saturate increases with decreasing pressure. The results of further investigation into this are summarized in Figure 13where 8 refers to the total coverage of at least chemisorbed oxygen. The sequence of graphs in Figure 13 indicates that there was some saturation level of chemisorbed oxygen coverage which is pressure dependent. This dependence on pressure may be the result of repulsive interactions between chemisorbed 0 atoms on the surface. Because chemisorption is dissociative, a surface site which can incorporate both oxygen atoms would be greatly preferred over a site where significant rearrangement is required before both atoms can be accommodated. Lower surface collisional rates (pressures) may allow significant rearrangement of the adsorbate in order to minimize repulsive interactions. Such a uniformly distributed surface with most preferred dissociation sites occupied may establish an equilibrium fractional coverage such that the impingement and dissociation rate, which has been reduced greatly by the saturation of "most preferred" sites, is just balanced by normal thermal desorption. The effect of increasing pressure may be to increase the number of statistically less favorable dissociations at higher 0 coordination sites, until a new equilibrium coverage is achieved etc. From Figure 13 the apparent equilibrium fractional coverages for differing pressures were as foiiows: 1 10-6 TO^, 100%; 5 x 10-7 ~ ~ r100%; r , 1 Torr, 45%; 5 x Torr, ~ 3 3 %1; x Torr, ~ 2 5 % . From the above sequence of data, it is unclear if the equilibrium fractional coverage is a continuous function of pressure or if there are definite stages involved in chemisorption here, as there are in the Ni/O system. Inspection of the above data might suggest that the intuitively satisfylng trend of fractional coverages being 1 in 4, 1 in 3, 1 in 2, and finally 1 in 1 for increasing impingement rates is being obeyed. This interpretation however suffers from the lack of LEED data which might confurn or refute the situation as described. Furthermore, since differing crystal faces were being simultaneously sampled, fractional coverages will not necessarily have a real significance. One final experimental observation that must be mentioned is the fact that since the substrate in this series was an alloy, there may be an extra factor involving preferential oxidation of one of the metals in the alloy. A rough estimate of the oxide chemical composition can be made using peak height ratios of the LMM Auger peaks at 848 eV for Ni and 654 eV for Co. For the alloy in question here, the starting sputtered surface ratio was Ni:Co = 4.08, and previous work on this s y ~ t e mhad ~ ~established ,~~ a clear slight preferential sputtering of Ni. After 35 min
Hajcsar et al. exposure at 1 x Torr, the ratio had dropped to 2.14 and after 50 min was 2.41 and 60 min was 2.43. From this limited data, it may be concluded that within the sampling depth of the Auger electrons, the oxide formed is cobalt rich. A work of Majumdar et al.29 on the oxidation of a 20 d o Ni-Co alloy has indicated that for conditions virtually identical to those employed in this study, a very thin layer of COOis formed exclusively and Ni remains in the metallic state for exposures of up to 4500 langmuirs at 773 K. The Ni Auger signal which is still visible after this exposure, was thought to originate from beneath a uniform but very thin overlayer of Coo but the results of this study suggest that the signal could also arise from the area between islands. This preference at the surface for Co is most graphically displayed in Figure 14, where Figure 14a is an oxygen map, Figure 14b is a Ni map, and Figure 14c is a Co map. There is a clear and strong negative correlation between the oxide islands and Ni. A positive correlation with Co is less clear because of the presence of Co in the underlying alloy and the weakness of the Co peak used (654 eV). The question of why the Holloway-Hudson model has been well obeyed in the past remains to be addressed. This present work is not only the first to image very thin islands successfully, but it is also the first that has attempted to separate the signals from the layers in the islands. Much of the previous oxygen uptake data has been extracted from the intensity of oxygen Auger peaks which were obtained from areas much large; than the islands being proposed by the model (100 A in some estimates, although clearly this will depend on conditions). It is clear that that type of experiment will yield only a sum of signals from the various layers that evidently coexist during this stage of oxidation and have different growth kinetics. The Hollowayand Hudson model relation therefore provides an adequate fit to the total Auger intensity, but the assertion that island growth is limited by either lateral diffusion or capture of an adsorbed 0 2 species is not likely in the temperature-pressure regime explored in this work, and possibly not at other conditions either. Furthermore, the assumption that islands are circular and that the active sites are simply proportional to the square root of the area is not supported. 4.0. Conclusions The early stages of oxidation of Ni-Co alloys follow kinetics consistent with the nucleation and lateral growth of multilayered oxide islands rich in Co, that growt at a rate determined by the rate of gas impingement at island perimeters, the type of layer, and substrate temperature.
LA9405713 ~
(27)Hajcsar, E. E.; Dawson, P. T.; Smeltzer, W . W . Surf. Interface Anal. 1987. 10. 343. (28)Hajcsar, E. E.; Underhill, P. R.; Smeltzer, W. W.; Dawson, P. T. Surf. Sci. 1987, 191, 249. (29)Majumdar, D.; Spahn, R. G.; Gau, J. S. J. Electrochem. SOC. 1987,134, 1825.
