- Email: [email protected]
Transient recrystallization of amorphous silicon films
HATERIAIS SCIENCE & ENGINEERING ELSEVIER
Materials Scienceand EngineeringB47 (1997) 78-86
Transient
recrystallization
of amorphous
silicon films
J. Viatella *, R.K. Singh l Department
of Materials
Science rind Engineering,
Unitiersity
of Florida,
Gbnesuille,
FL 32611,
USA
Abstract
We have investigatedcrystallization of amorphoussilicon films on SiO,/Si substratesusing conventional furnace annealing, incoherentlight-basedrapid thermal annealing(RTA), and pulsedexcimer laser(wavelength= 248 nm, energy density= 0.1-0.6 J cn--2). The effect of a 50 nm SiO, cappinglayer on the final microstructurewasexaminedfor eachannealingtechnique,using transmissionelectron microscopy (TEM) and X-ray diffraction (XRD). A discussionof the phenomenologicaleffects of the capping layer on resultantmicrostructureis included.For laserannealing, the effects of varying the thicknessof the intermediate SiO, layer wereexaminedand the resultantmicrostructurewascharacterizedfor grain sizeand crystallized orientation usingTEM. Grain sizesrangingfrom 50 to 200 nm were observed.Electrical propertiesof the crystallizedfilms werepresentedfor the various resultantmicrostructures.The effect of laserinteractionswith multi-layeredstructureswere simulatedby solvingthe heat condition equation with the appropriate boundary conditions and the resultsare included. Q 1997Publishedby Elsevier ScienceS.A. Keywords:Siliconfilms;Transmission electronmicroscopy; Thin film transistors
1. Introduction
Polycrystalline-silicon (poly-Si) thin-film transistors
(TFTs) are used in a variety of applications, including large-area electronics [I] and vertically stackable com-
ponents for three-dimensional integration [2]. Many large-area electronics applications involve the deposition of an amorphous silicon layer on an inexpensive glass, such as Corning 7059, which has a quoted working range below 600°C [3]. This processing limitation prevents the high temperature anneal necessary for large grain size films and sufficient carrier mobility for fast switching applications. The grain size increase and general morphological improvement resulting from an anneal process improve carrier mobilities and hence device performance, therefore much work has been done in the area of optimizing potential anneal pro-
grain size, as the high mobilities associated with the
laser annealing process coincide with grain sizes that are smaller ( - 150 run) than rapid thermal processing ( - 250 nm) or furnace annealing ( - 1000 nm). Some
cesses[4]. Also, use of poly-Si films in continually
decreasing feature-size integrated circuits has created a demand for improved thermal processing to minimize unwanted effects (i.e. dopant redistribution) resulting * Corresponding author. ’ Proceedings of the Engineering Foundationconference on Materials Processing and Advanced Florida, 1-6 May, 1994. 0921-5107/97/S17.00 PIISO921-5107(96)01857-9
Applications
0 1997 Published
of Lasers,
by Eisevier
Palm
Science
Coast,
S.A. All
from high thermal budgets (temperature-time product). Therefore, there has been considerable interest in annealing techniques that enhance TFT performance by producing large grain sizes, while minimizing the thermal budget of the process. Techniques that have been investigated and their respective carrier mobilities include: use of as-deposited amorphous films ( - 1 cm2 V-’ s- ‘) [4], conventional furnace anneal ( - lo-50 cm2 V- ’ s - ‘) [5], rapid thermal processing ( - lo-50 cm* V-’ s- ‘) [6], and laser annealing ( - 200-300 cm2 V - ’ s - ‘) [4]. However, carrier mobility does not seem to depend solely on
rights
of the morphological reasons for this are examined in the discussion section. As mentioned, rapid thermal processing, laser annealing and conventional furnace heating are some of the techniques that have been examined, and are the focus of this work. A major difference between the various thermal treatments is the time scale in which the energy is delivered to the material and how it compares with the thermal response time t of the semiconductor, which is defined by [7] reserved.
J. Yiatella,
R.K.
Sirrgh / Materials
Science
(1) where s is the thermal diffusion length or sample thickness, whichever is smaller, and k is the thermal diffusivity. The thermal response time varies from microseconds to milliseconds depending on the value of k and the absorption depth of the radiation. In laser annealing, energy is delivered in a time scale on the order of 20-100 ns, which is well below the therma response time for silicon. Energy is maintained close to the surface, since the absorption depth for a KrF laser (248 m-n) in both amorphous andOcrystalline silicon is quoted to be approximately 60 A [Q and a low average temperature is maintained. However, there is a spot-size limitation to the area being processed and a rastering technique must be employed to process larger areas. In rapid thermal processing, however, the energy is delivered on the order of 1- 100 s, which is above the thermal response time. The incoherent light radiates on the entire wafer and there are no significant thermal stresses generated between different layers, as in laser processing [7]. Furnace anneaIing also delivers energy on a large time scale (lo’- lo4 s) but this thermal budget allows more significant diffusion of implanted atoms and previously incorporated dopant atoms. In this paper, the results of two sets of experiments are presented. The first set focused on examining the effect of a 500 A-thick tetraethylorthosilicate (TEOS)-deposited SiO, capping layer. Samples with and without this capping layer were annealed using a conventional furnace, rapid thermal annealing (RTA), and an excimer laser, and resultant microstructures were examined using transmission electron microscopy (TEM). It was expected that the capping layer would have varying effects on the resultant microstructure, depending on the annealing technique. For furnace and rapid thermal annealing, nucleation would be affected due to the addition of a new interface. The microstructure could also be affected due to changes in conductive and radiative losses during processing. For RTA, it was expected that there would be other effects due to storage of thermal energy in the capping layer. In laser annealing, there would be surface reflectivity effects as well as thermal effects due to changes in thermal gradients, as well as heat flow, during processing. The objective in the second set of experiments was to examine the effect of varying the thermal quenching in a laser-annealed film. The quench rate of a molten surface Urn is determined by thermal flux to the substrate and the thermal capacity of the am. Computer simulations have shown that for substrate configurations in these experiments (SiO, on Si wafer) heat transfer is strongly dependent on the thickness of the SiO, layer [9]. Therefore, the thermal flux out of the amorphous thin film can be varied by changing the thickness of the
and Engineering
B47 (1997)
75-86
79
underlying oxide layer. Changing the thermal flux would affect undercooling and nucleation kinetics and therefore resultant microstructure. Consequently, samples with intermediate oxide layers of varying thickness were laser-annealed using various energy densities, and the resultant microstructures were examined using plan-view and cross-sectional TEM.
2. Experiment 2.1. Samples Three types of samples were used in the experiment. For the comparison of anneal techniques, two sets of samples were used. These samples were prepared on (100) silicon with an initial 200 A-thick thermal oxide grown at 1000°C using wet oxidation. An 800 k-thick amorphous silicon layer was deposited at 555°C in an low pressure chemical vapor deposition (LPCVD) furGate using silane chemistry. One set of samples had a 500 A-thick capping layer of SiO, that was plasma deposited by decomposition of TEOS at a temperature of 450°C. The low-temperature based capping layer is provided to promote uniform grain growth as the recrystallization mechanism begins. The second set of samples consisted of the wafer, thermal oxide, amorphous silicon layer, and an 80 A-thick oxide layer on top. The third set of samples, which were used exclusively in the laser-based experiments, consisted of a 1000 A amorphous silicon film deposited on an SiO, layer using LPCVD at 550°C. The interm$diate oxide layer varied in thickness (400, 1200, 2200 A) and was deposited on a (100) silicon wafer using wet oxidation. Samples that were subsequently characterized for carrier mobility were doped with boron at 5 x 10” cmM3. 2.2. Anneal techniques Laser annealing was performed with a KrF excimer laser operating at a wavelength of 248 nm, using energy densities from 190 to 505 mJ cm -‘. The samples were processed at room temperature. Single laser pulses were used for the undoped samples, while a beam scan was performed on the doped samples in order to crystallize an area of sufficient size for electrical characterization. It has been reported that increasing the number of pulses did not show any further grain growth [lo], and this was reinforced by our results as there was no significant deviation in resultant grain size in the scanned versus non-scanned samples. Rapid thermal processing was carried out with the AG Associates Heatpulse 8108 using ramp-up and rampdown rates of 70°C s-l with a plateau of 1050°C for 30 s. Furnace anneals were performed at 600°C for 24 h in a conventional furnace.
J. Viutellu, R.K. Sing/z /Materials
80
Science and Engineering B47 (1997) 78-86
3. Results
Bright-field plan-view TEM was used to analyze the resultant microstructure of the samples. Micrographs representing each of the thermal treatments are shown in Fig. 1 for uncapped samples. The most notable difference in microstructure occurs between the laserannealed samples and the solid-phase annealed samples. For RTA and furnace anneals, the microstructure consists of relatively large but poorly-defined grains which have a high intragrain defect density. A high density of twins is also observed. The grains have a wide distribution of shapes and sizes, with some grains having a high aspect ratio. For laser annealing, the microstructure
Fig. 2. Plan-view annealed capped
Fig. 1. Plan-view annealed uncapped
TEM of (a) samples.
furnace,
(b)
RTA,
and
(c)
laser
TEM samples.
of (a)
furnace,
(b)
RTA,
and
(c)
laser
shows clearly-defined grains that are smaller than those of the solid-phase anneals, but with low intragrain defect density. Also, twins are not as readily observed and the grains are comparatively equiaxed. Microstructural differences are attributed to the variation in nucleation and growth mechanisms between the annealing techniques, and these differences are discussed in detail later in this work. For experiments involving capped samples, micrographs representing each of the thermal treatments are shown in Fig. 2. The microstructure of the samples was similar to that observed in the uncapped samples. One discernable difference, however, was the average grain size. The average grain size for the various annealing techniques is shown in Fig. 3 for samples with a
J. Viatella,
R.K.
Singh /Materials
Science
and Engineering
B47 (1997)
81
78-86
150
++
E 2 100 E: ‘8 c3 I
50 ’
I
I
50
i
+
+
piiq
+ I
250 Annealing
+
I
350
I
450
550
Technique Laser Energy (m.J/cm2 )
Fig. 3. Grain size as a function of anneal technique for capped samples.
capping layer and in Fig. 4 for samples without a capping layer. The furnace-annealed samples had the largest grain size and this value did not vary significantly between capped and uncapped samples (202204 nm). The lack of a difference in grain size between samples is due to the long processing time involved. Thermal equilibration occurs on a much shorter time scale than the 24 h tune period over which the anneal took place, and therefore the capping layer does not have any discernible effect. The RTA samples had a smaller grain size than the furnace-annealed samples and there was a significant size difference between the capped (142 run) and uncapped (116 run) samples. This size difference is likely due to an increase in thermal energy stored in the multi-layer structure, as well as differences in emissivity. Enhanced containment of thermal energy in the t%lm may be occurring from a radiative, as well as a conductive, perspective in the capped samples. Though the SiO, capping layer does not directly absorb signifi250
g 200 V $j 150 3 3 100
+
t
i
50 Annealing
Technique
Fig. 4. Grain size as a function of anneal technique for uncapped samples.
Fig. 5. Grain size as a function of laser energy density.
cant amounts of energy from the RTA lamps, it does absorb thermal energy by conduction from the underlying Si layer. This well of thermal energy allows the growth process to continue in the underlying film even after the ramp-down step is over. Without the capping layer, thermal energy is more quickly equilibrated since the underlying single-crystal Si wafer has a high thermal conductivity and only a 200 A SiO, layer is separating it from the amorphous Si layer. Also, larger radiative losses occur during processing in the uncapped samples, decreasing the amount of thermal energy available for grain growth. The laser-annealed samples possessed the smallest average grain size. The samples processed at 190 mJ cmP2 appeared to be below the recrystallization threshold, and similar results are found in other work [IO]. Above this energy density the resultant grain sizes ranged from 80 to 125 nm, as shown in Fig. 5. The samples without a capping layer consistently demonstrated larger grain sizes than those with one, likely due to energy loss from the melted film to the capping layer. The SiO, capping layer does not absorb significant laser energy but does provide conductive losses and therefore some quenching effect to the underlying film. Unlike RTA, the capping layer in this situation does not provide any thermal storage for enhanced grain growth. The energy absorbed induces melting in the amorphous Si layer resulting in radical temperature gradients which are quickly equilibrated. Also, these samples did not demonstrate much variance in grain size relative to incident energy. Uncapped samples demonstrated a trend towards smaller grains as laser energy increased, indicating that the complete melt threshold is likely at the lower energies used and that there are other mechanisms taking place that define the resultant grain size. This point is discussed in greater detail in the Section 4.
82
J. Viatelln,
Fig. 6. Cross-sectional
3.1. Effects
of intemediinte
TEM
R.K.
Singh /Materids
of laser-annealed
Science
samples
oxide layer
Experiments using laser annealing of samples with an intermediate oxide layer of varying thickness resulted in a microstructure with measured grain sizes between 60 and 200 nm. Cross-sectional TEM micrographs of the samples are shown in Fig. 6. The intermediate oxide layer is visible, as well as the microstructure of the annealed Si layer. The grain boundaries are perpendicular to the plane of the film, due to the direction of heat flow during solidification. Trends of increasing grain size with laser energy were observed, as shown in Fig. 7, with the largest grain sizes corresponding to films which had the thickest intermediate oxide layers. Due to the relatively low thermal conductivity of SiO,, the oxide film’s thickness becomes the major controlling factor regarding melt lifetime and nucleation and growth kinetics. These mechanisms are discussed in greater detail later in this work. Grain-size differences between samples with 220 and 120 nm-thick oxide layers are less (20 nm) than those between the 120 and 40 nm-thick oxide layers (50 nm), suggesting that the thermal quenching effects of the oxide layer diminish with increasing thickness. Solidification velocity calculations by the authors [l 11 based on this film configuration have shown that as the oxide layer is increased in thickness there is a decrease in 250 +
,m,
0 W A
$
+
+
240 (mJ/cm2) 336 (mJ/cm*) 405 (mJ/cm’)
0' 0
500
1000
1500
Oxide Thickness Fig. 7. Grain
size as a function
2000 2500 (A)
of oxide
thickness.
with
and Engineering
B47 (1997)
78-56
(a) 400 A, (b) 1200 A, and (c) 2200 A-thick
oxide
layer.
solidification velocity, with an asymptotic limit of 160 cm s-l at a large oxide-layer thickness. This would indicate that increasing the grain size by altering thermal properties of the oxide is maximized at a certain thickness, and that value is approximately 200 nm. An oxide layer thinner than this provides greater quenching and greater undercooling of the film. As the undercooling is increased, nucleation density increases and a smaller resultant grain size follows. For the thicker oxides, the quench rate is decreased and with it the amount of undercooling. A low nucleation density ensues and there remains sufficient thermal energy in the film to allow more significant grain growth, resulting in the larger average grain size observed. This is discussed in greater detail in the simulation section. 3.2. Microstructure
rind electricnl properties
Hall effect measurements performed on the doped laser-annealed samples resulted in measured mobilities that ranged from 198 to 351 cm2 V-l s-l. Though the grain sizes were small in comparison to other annealing techniques, the mobilities approached that of p-type doped single-crystal silicon. As mentioned previously, the reasons for this may be found in an examination of the intra-grain morphology. Grain boundaries, along with linear and planar defects, act as recombination centers in silicon films. Thus, the occurrence of twins within grains affect carrier mobility. Haji et al. [12], studied the mode of growth of silicon films during solid-phase crystallization. Their results showed that the main defects inside grains are twins. While first order twins are electrically inactive, their presence is commonly related to: (i) incoherent { 112) steps perpendicular to { 111 } twin planes which introduce dislocations at the step edges which are strongly electrically active, and (ii) higher order twins which are related to displacement vectors and are electrically active due to segregation of dissolved impurities. The lower mobility values observed in solid-phase crystallized films as compared with films solidified from melt in laser annealing can be attributed to these morphological differences. These differences reinforce the idea of laser annealing as a suitable candidate for the processing of amorphous Si films for TFT applications.
J. Viatella,
R.K.
Singh / Materials
Science
4. Discussion 4.1. Solid phase nucleation and growth
Solid phase recrystallization of amorphous Si films on SiO, has been studied by many investigators [13]. It has been well substantiated that initial nucleation of the a-Si occurs at the a-Si/SiO, interface [12,14]. It is believed that nucleation at the interface is due to the release of the stress induced by the thermal expansion difference between Si and SiO, and/or of that which is attributed to the porosity of the a-Si [14]. In-situ annealing experiments in a transmission electron microscope have revealed [12] the mode of growth for grains in a-Si tis. Amorphous silicon films deposited on glass using LPCVD were annealed at 600°C in the TEM and their growth monitored. Once nucleation takes place, the grains grow quickest along the [l IO] and [112] axis and slowest along the [I 111. Therefore, the grains have an elliptical shape [15] in general, and contain multiple twins at the center. Other twins are also formed and preferential growth in (112) directions follows, leading to dendritic grains which have a { 1 lo} surface. The crystallites lose their ellipsoidal shape when coalescence between grains starts and this ultimately results in the formation of obscure grain boundaries. The resultant microstructure is one of relatively large grains ( N 1 urn), but also a large number of intragrain defects due to the growth proceeding by formation of twins. 4.2. Nucleation
and growth
in laser annealing
The nucleation mechanisms in laser crystallization of silicon thin films can be broken down into two regimes, differentiated by whether complete melting of the t?lm occurs or not. In the first case, where complete melting of the film does not occur, several potential scenarios have been proposed. Thompson et al. [16] and Lowndes et al. [17] both support the following dynamic. At low energy densities the laser energy only melts part of the film, creating a thin liquid layer near the surface. As the liquid begins to solidify as relatively large-grain polycrystalline silicon, the latent heat released raises the temperature of the resolidified poly-Si above the temperature of the first-order phase transition of the amorphous silicon (a-Si) to the metallic liquid, T,- ,. The underlying a-Si material then begins to melt. This new liquid is severely undercooled compared to the poly-Si layer and therefore resolidifies as fine-grained poly-Si. Thus, a thin liquid layer is presumed to propagate through the a-Si material because of the released latent energy. This is labeled as the explosive recrystallization process and it is self-sustaining until it is quenched by the energy required to raise the temperature of the a-Si solid in front of the liquid to T,-,. The final mi-
and Engineekng
B47 (1997)
78-56
83
crostructure consists of a top layer of large-grained (50-100 nm) polycrystalline silicon with a layer of fine-grained (1- 10 nm) poly-Si directly below it resulting from the explosive recrystallization process. Thompson et al., were able to support the explosive crystallization argument by use of transient reflectance and conductance measurements of tims melted with nanosecond-pulse lasers. Even though the reflectivity returns to the solid value at approximately 53 ns, signifying that the surface is now in the solid phase, the conductance showed a second peak that existed for an additional 10 ns (until - 63 IX). The extended conductance was attributed to the subsurface liquid layer, which behaves as a metallic conductor. In other words, though the surface is in a solid state there is a highly conductive propagating liquid layer still within the film. Using this information, they were also able to estimate the velocity of the explosive melt front to somewhere between 10 and 20 m s- ‘. Another proposed scenario for non-complete melting of Si films is by Im et al. [18]. They argue that in the tims explosive crystallization of a-Si occurs at the onset of the transformation, implying that partial melting of explosively-crystallized fine-grained Si is occurring rather than partial melting of an amorphous film. It is suggested that early triggering of explosive crystallization may be attributed either (1) to the presence of microcrystalline clusters in the LPCVD samples, which was confirmed by analyzing solid-phase crystallization behavior and is absent in high-dose ion-irradiated amorphized samples and/or (2) the possible presence of impurities, such as hydrogen. The same researchers also put forth the idea of the super lateral growth (SLG) regime in which almost complete melting of the film occurs to the extent that there is a discontinuous Si tim composed of discrete solid islands in a liquid matrix. Growth from these clusters proceed, resulting in an unusually large (300-400 nm) grain size. If greater laser energy is used the resultant grain size returns to smaller values ( - 50-100 nm), typical of lower-energy density irradiation. In the second case, where complete melting of the film does occur, there is substantial evidence that nucleation occurs homogeneously. Stiffier et al. [19,20], argue the following. They used 200-400 nm-thick Si films deposited on SiO, layers that varied between 77 and 330 nm. The Si films were melted using a Q-switched ruby laser (25-30 ns pulses) and transient conductance and reflectance measurements were taken. Analysis of the measurements revealed the following scenario. After melting in the first lo-20 ns, the film remains fully molten as it cools by conduction to the substrate. Nucleation is followed by rapid solidification, which is shown by a fast ( - 10 ns) drop in the reflectance and conductance, as the enthalpy released heats the film to the steady-state solidification temperature determined
84
J. Viatellu,
R.K.
Singh /Materials
Science and Engineering
by the thermal properties of the substrate. Recalescence is followed by steady state solidification which is marked by a gradual (150 ns) decrease in conductance returning to the solid value. Undercooling of the films was calculated in the following manner. They determined the volume fraction of solid following recalescence from the ratio of the conductance to the fully liquid value assuming a twophase mixture and relative conductivities. The volume fraction of liquid consumed during recalescence was then used to determine the temperature of the liquid prior to nucleation using a heat balance equation. Since the heat loss to the substrate is exactly balanced by the release of enthalpy in the film during steady-state solidification, the thermal flux out of the film following recalescence was obtained from the slope of the conductance trace. This thermal flux can be used to establish an upper limit for the quench rate of the film prior to nucleation by using the specific heat and thickness of the film. Also, it can define an upper limit for the thermal leakage during recalescence. Also, T,,, the steady-state solidification temperature, can be estimated from the thermal flux. This is done by calculating an effective interface velocity from the thermal flux and enthalpy of melting, which is then used to calculate T,,. From the quench rate the undercooling can be calculated. Undercooling prior to nucleation was constant at approximately 500 K for quench rates below approximately 1O’O K s-l. As in previous studies, no evidence of preferential nucleation at either the surface or interface was present. The microstructure consisted of coarse grains combined with interspersed fine-grained structures, which typically are the signature of explosive recrystallization. This morphological idea is further reinforced in the most recent work by Im et al. [21] in which they found the SLG regime to result in a microstructure consisting of large single-crystal disk structures. They argue that these disk structures are the result of a single nucleation event occurring from a discrete cluster that was not melted in the near-complete melting of the film. These disk structures are predicted to occur when the separation distance between surviving solid clusters is greater than the maximum lateral growth distance. In other words, as the single crystal disks are growing, there is significant undercooling of the liquid which leads to nucleation of solids in the bulk of the liquid ahead of the advancing solidification front. This stalls the growth of the grains and results in the morphology shown. A variation on this morphology was also observed in which the seeds that survive are located near each other such that growth can continue until the solidification fronts of neighboring disks touch. This results in single crystal disks touching with small regions of fine poly-Si grains between the disks.
B41 (1997)
75-86
5. Simulation We have developed a model to simulate the effect of laser interactions with multi-layered structures. The solution is based on a control volume approach and can take into account temporal and spatial-dependent events. Thermophysical properties of each layer are entered as input parameters and they include: melt temperature, vaporization temperature, temperature and phase-dependent thermal conductivity and volume heat capacity, and latent heat of melting and vaporization. Temperature and phase-dependent optical properties that are included are: reflectivity, absorption coefficient, and simulated absorption coefficient of the plasma during vaporization [22]. The characteristics of the laser pulse included are energy density, pulse length, and power density distribution as a function of time. The output includes: temperature as a function of time for any nodal location within the multilayered structure, temperature profiles at any time during irradiation, as well as melt and evaporation depths as a function of time. Features of the model include an ability to simulate the movement of the solid-liquid interface as it is generated at the surface and propagates into the bulk during pulsed irradiation. After termination of the laser pulse, the solid-liquid interface changes direction and recedes to the surface. The model incorporates this phenomenon and also provides solidification velocities. As currently formulated, the model does not include some of the components that are observed in laser annealing of Si tirns on SiOJSi substrates, including nucleation of solids in undercooled liquid. Other models [23,24] proposed share this deficiency and to date a model that incorporates the observed phenomena has not been developed. However, valuable insight may be gained by examining the solidification scenarios for the thermal quenching provided by the varying-thickness oxide layers. In the case of non-complete melting of the films, undercooling is not an issue and the model formulated here provides an accurate picture of the solidification scenario. Fig. 8 shows melt depth as a function of energy density for various oxide thickness. This graph shows that complete melting occurred for the higher laser energies. In cases where there was incomplete melting of the flm solidification occurred from existing material, eliminating nucleation as a limiting factor and a smaller resultant grain size was observed (as shown in Fig. 6). The largest observed grain size occurred where complete melting of the film occurred. As mentioned previously, the evidence suggests that severe undercooling occurs in these situations and the microstructure is a result of homogenous nucleation, Using the simulation, quench rates were calculated for the varying oxide thickness and nucleation density estimated in the fol-
J. Viatella,
1000 900 800
&’
R.K.
Singh /Materials
Science
p-9/ / / #’ / / /’
700
-e-m-, -A-,
4’
El
600
400 (A) 1200(A) 2200 (A)
500 200 Energy (mJ/cm2) Fig. 8. Melt depth as a function thickness.
of energy for varying oxide
lowing manner. Initially, the quench rates given by the calculations were correlated with undercooling values derived empirically [ 191. Using classical homogeneous nucleation theory [25], the value of undercooling was used to calculate the free energy change associated with the formation of a voIume of solid by:
*G”+z
(2) M
where L, is the latent heat of fusion per unit volume, TM is the melt temperature, and AT is the value of undercooling. The critical radius for a stable cluster of particles was calculated using this free energy change by: f% - 2YSL where ysL is the solid/liquid interfacial free energy, found [26] to be - 0.34 + 0.02 J rnB2 for homogeneous nucleation assuming spherical nuclei. The excess free energy associated with a cluster of radius Yis given by:
and Engineering
B47 (1997)
78-86
85
undercooling in these sampleswas calculated to be 512 K and consequently this represents approximately 4.5 x lo6 clusters of critical size per cm2 area of film (film thickness - 100 nm). Samples with a 110 nmthick oxide layer annealed with the same energy had a smaller grain size (185 nm), and the undercooling value was 550 K. Cluster density for these films was calculated at approximately 5.9 x 107.The smallest grain size (140 nm) at this laser energy was observed in the samples with the 40 nm-thick oxide layer, and the cluster density was approximately 3.9 x 10’. The microstructural differences resulting from the three order of magnitude variation in cluster density can be seenin the cross-section TEM micrographs in Fig. 6. Fig. 6(a), of the 40 nm-thick oxide, shows less distinct grain boundaries than do Fig. 6(b) (1200 nm oxide) and (c) (2200 nm oxide). The 2200 nm oxide film has clear grain boundaries spanning the thickness of the film and a more ordered microstructure than the other two films. Though the critical cluster density calculations help explain final microstructures, other phenomenological aspects of solidification must be mentioned. When quench rates are greater than approximately 1O’OK s - ’ the embryos cannot grow at a sufficient rate to maintain a steady-state distribution [19]. Cooling can then continue until a sufficient number of embryos form, resulting in even larger values of undercooling which affect critical cluster size and density. Classical theory [25] provides a homogenous nucleation rate that involves a term which is a complex function that depends on the vibration frequency of the atoms. Since the vibration frequency is a function of temperature, there is a decrease in nucleation rate with increased undercooling. The limitation on nucleation rate is particularly significant when large values of undercooling occur and therefore should be considered in future theoretical models.
6. Conclusions AG, = -;
ni3AG, + 4nr2y,,
(4)
Using the critical radius value, r*, in the free energy calculation from Eq. (4), the number of spherical clusters of critical size is given by: (5) where no is the total number of atoms in the system, T is the temperature and k is Boltzmann’s constant. Nucleation density increases with an increase in undercooling, resulting in smaller grains than would be found where undercooling is less severe. The largest grains (200 nm) were observed in the sampleswith the 220 mn-thick oxide layer where the highest laser energies (-405 mJ cm-‘) were used. The amount of
Recrystallization of amorphous silicon using furnace annealing, rapid thermal annealing, laser annealing and the resultant microstructure has been investigated using TEM and X-ray diffraction (XRD) for characterization. Grain size was correlated to incident energy for laser-annealed samples and ranged from 50 to 200 nm. Effects of an SiO, capping layer were investigated for all three annealing techniques. The capping layer did not demonstrate a significant effect for furnace annealing, but RTA capped samples demonstrated a significantly larger average grain size than those without a capping layer. Capped samples annealed by the laser demonstrated a smaller average grain size than those that were uncapped. Also, we have examined the effects of an intermediate SiO, layer of varying thickness on
86
J. Viateila,
R.K.
Singh /Materials
Science
laser-annealed amorphous silicon films. A discussion of the advantages of laser processing over other annealing techniques with respect to carrier mobility is included along with an examination of the mechanisms of nucleation and growth for each annealing technique mentioned. A novel computer-based simulation of laser interactions with multi-layered structures is included to help identify the events shaping laser processing of amorphous silicon films.
References ‘11 T.J. King, M.G. Hack and I.W. Wu, J. Appl. Plays., 75 (1994) 908-913.
21 B. Faughnan and A.C. Ipri, IEEE (1989)
Tyans.
Electron
Devices,
36
101-106.
31 R.Z. Bachrach, K. Winer, J.B. Boyce, S.E. Ready, RI. Johnson and G.B. Anderson, J. Electron. Mater., 19 (1990) 241-248. 41 N. Yamauchi and R. Reif, J. Appl. Phys., 75 (1994) 3235. ,5] M. Bonnei, N. Duhamel, M. Guendouz, L. Haji, B. Loisel and P. Ruault, Jpn. J. Appl. Phys., 30 (1991) L1924-L1926. :6] R. Kakkad, J. Smith, W.S. Lau and S.J. Fonash, J. Appl. Phys., 65(1989)
2069-2073. J. Appl. Phys.,
.7] R. Singh, 63 (1985) R59. /S] S.D. Brotherton, D.J. McCulloch, J.B. Clegg and J.P. Gowers, IEEE Trans. Electron Devices, 40 (1993) 407-413. 191S.R. Stiffier and M.O. Thompson, Phys. Rev. B, 43 (1991) 9851-9855.
and Engineering
B47 (1997)
78-86
[lOI R.Z. Bachrach, K. Winer, J.B. Boyce, F.A. Ponce, SE. Ready, R. Johnson and G.B. Anderson, MRS
Symp.,
157
(1990) 467-472.
P11John Viatella, Rajiv K. Singh and R.P.S. Thakur,
MRS
Sylnp.,
(1994) 61-66. [I21 L. Haji, P. Joubert, J. Stoemenos and N.A. Economou, J. Appl. Phys., 75 (1994) 3944. t131 T. Noma, T. Yonehara and H. Kumomi, Appl. Phys. Lett., 59 (1991) 653-655. 1141 A. Sakal, H. Ono, K. Ishida, T. Ntino and T. Tatsumi, Jpn. J. Appl. Pirys., 30 (1991) L941-L943. 1151 A. Nakamura, F. Emoto, E. Fujii, Y. Uemoto, A. Yamamoto, K. Senda and G. Kano, Jpn. J. Appl. Phys., 27(1988) L2408-L2410. [161M.O. Thompson, G.J. Galvin, J.W. Mayer, P.S. Peercy, J.M. Poate, DC. Jacobson, A.G. Cullis and N.G. Chew, Phys. Reo. Lett., 52 (1984) 2360-2363. u71 D.H. Lowndes, R.F. Wood and J. Narayan, Pizys. Rev. Lett., 52 336
(1964)
561-564.
[181J.S. Ira, H.J. Kim and M.O. Thompson,
Appl.
Phys.
Lett.,
63
(19930 1969-1971. [191S.R. Stiffler and M.O. Thompson, Phys. Rev. B, 43 (1991) 9851-9855. WI S.R. Stiffler and M.O. Thompson, P.S. Peercy, MRS Symp., JO0 (1988)
505.
WI J.S. Im and H.J. Kim, Appl. Phys. W Rajiv K. Singh and John Viatella, ~231 ~241
PI WI
Lett., 64 (1994) 2302. J. Appl. Phys., 75 (2)
(1994) 1204- 1206. C.P. Grigoropoulos, A.A. Rostami, X. Xu, S.L. Taylor and H.K. Park, Znt. J. Heat Muss Transfer, 36 (5) (1993) 1219-1229. V. Gupta, H. Jin Song and J.S. Im, to be published. D.A. Porter and K.E. Easterling, Phase Trmsfomntions in Metals and Alloys, 2nd edn, Chapman and Hall, London, 1992. S.R. Stiffler and M.O. Thompson, Phys. Rev. Lett., 60 (1988) 2519.
Materials Scienceand EngineeringB47 (1997) 78-86
Transient
recrystallization
of amorphous
silicon films
J. Viatella *, R.K. Singh l Department
of Materials
Science rind Engineering,
Unitiersity
of Florida,
Gbnesuille,
FL 32611,
USA
Abstract
We have investigatedcrystallization of amorphoussilicon films on SiO,/Si substratesusing conventional furnace annealing, incoherentlight-basedrapid thermal annealing(RTA), and pulsedexcimer laser(wavelength= 248 nm, energy density= 0.1-0.6 J cn--2). The effect of a 50 nm SiO, cappinglayer on the final microstructurewasexaminedfor eachannealingtechnique,using transmissionelectron microscopy (TEM) and X-ray diffraction (XRD). A discussionof the phenomenologicaleffects of the capping layer on resultantmicrostructureis included.For laserannealing, the effects of varying the thicknessof the intermediate SiO, layer wereexaminedand the resultantmicrostructurewascharacterizedfor grain sizeand crystallized orientation usingTEM. Grain sizesrangingfrom 50 to 200 nm were observed.Electrical propertiesof the crystallizedfilms werepresentedfor the various resultantmicrostructures.The effect of laserinteractionswith multi-layeredstructureswere simulatedby solvingthe heat condition equation with the appropriate boundary conditions and the resultsare included. Q 1997Publishedby Elsevier ScienceS.A. Keywords:Siliconfilms;Transmission electronmicroscopy; Thin film transistors
1. Introduction
Polycrystalline-silicon (poly-Si) thin-film transistors
(TFTs) are used in a variety of applications, including large-area electronics [I] and vertically stackable com-
ponents for three-dimensional integration [2]. Many large-area electronics applications involve the deposition of an amorphous silicon layer on an inexpensive glass, such as Corning 7059, which has a quoted working range below 600°C [3]. This processing limitation prevents the high temperature anneal necessary for large grain size films and sufficient carrier mobility for fast switching applications. The grain size increase and general morphological improvement resulting from an anneal process improve carrier mobilities and hence device performance, therefore much work has been done in the area of optimizing potential anneal pro-
grain size, as the high mobilities associated with the
laser annealing process coincide with grain sizes that are smaller ( - 150 run) than rapid thermal processing ( - 250 nm) or furnace annealing ( - 1000 nm). Some
cesses[4]. Also, use of poly-Si films in continually
decreasing feature-size integrated circuits has created a demand for improved thermal processing to minimize unwanted effects (i.e. dopant redistribution) resulting * Corresponding author. ’ Proceedings of the Engineering Foundationconference on Materials Processing and Advanced Florida, 1-6 May, 1994. 0921-5107/97/S17.00 PIISO921-5107(96)01857-9
Applications
0 1997 Published
of Lasers,
by Eisevier
Palm
Science
Coast,
S.A. All
from high thermal budgets (temperature-time product). Therefore, there has been considerable interest in annealing techniques that enhance TFT performance by producing large grain sizes, while minimizing the thermal budget of the process. Techniques that have been investigated and their respective carrier mobilities include: use of as-deposited amorphous films ( - 1 cm2 V-’ s- ‘) [4], conventional furnace anneal ( - lo-50 cm2 V- ’ s - ‘) [5], rapid thermal processing ( - lo-50 cm* V-’ s- ‘) [6], and laser annealing ( - 200-300 cm2 V - ’ s - ‘) [4]. However, carrier mobility does not seem to depend solely on
rights
of the morphological reasons for this are examined in the discussion section. As mentioned, rapid thermal processing, laser annealing and conventional furnace heating are some of the techniques that have been examined, and are the focus of this work. A major difference between the various thermal treatments is the time scale in which the energy is delivered to the material and how it compares with the thermal response time t of the semiconductor, which is defined by [7] reserved.
J. Yiatella,
R.K.
Sirrgh / Materials
Science
(1) where s is the thermal diffusion length or sample thickness, whichever is smaller, and k is the thermal diffusivity. The thermal response time varies from microseconds to milliseconds depending on the value of k and the absorption depth of the radiation. In laser annealing, energy is delivered in a time scale on the order of 20-100 ns, which is well below the therma response time for silicon. Energy is maintained close to the surface, since the absorption depth for a KrF laser (248 m-n) in both amorphous andOcrystalline silicon is quoted to be approximately 60 A [Q and a low average temperature is maintained. However, there is a spot-size limitation to the area being processed and a rastering technique must be employed to process larger areas. In rapid thermal processing, however, the energy is delivered on the order of 1- 100 s, which is above the thermal response time. The incoherent light radiates on the entire wafer and there are no significant thermal stresses generated between different layers, as in laser processing [7]. Furnace anneaIing also delivers energy on a large time scale (lo’- lo4 s) but this thermal budget allows more significant diffusion of implanted atoms and previously incorporated dopant atoms. In this paper, the results of two sets of experiments are presented. The first set focused on examining the effect of a 500 A-thick tetraethylorthosilicate (TEOS)-deposited SiO, capping layer. Samples with and without this capping layer were annealed using a conventional furnace, rapid thermal annealing (RTA), and an excimer laser, and resultant microstructures were examined using transmission electron microscopy (TEM). It was expected that the capping layer would have varying effects on the resultant microstructure, depending on the annealing technique. For furnace and rapid thermal annealing, nucleation would be affected due to the addition of a new interface. The microstructure could also be affected due to changes in conductive and radiative losses during processing. For RTA, it was expected that there would be other effects due to storage of thermal energy in the capping layer. In laser annealing, there would be surface reflectivity effects as well as thermal effects due to changes in thermal gradients, as well as heat flow, during processing. The objective in the second set of experiments was to examine the effect of varying the thermal quenching in a laser-annealed film. The quench rate of a molten surface Urn is determined by thermal flux to the substrate and the thermal capacity of the am. Computer simulations have shown that for substrate configurations in these experiments (SiO, on Si wafer) heat transfer is strongly dependent on the thickness of the SiO, layer [9]. Therefore, the thermal flux out of the amorphous thin film can be varied by changing the thickness of the
and Engineering
B47 (1997)
75-86
79
underlying oxide layer. Changing the thermal flux would affect undercooling and nucleation kinetics and therefore resultant microstructure. Consequently, samples with intermediate oxide layers of varying thickness were laser-annealed using various energy densities, and the resultant microstructures were examined using plan-view and cross-sectional TEM.
2. Experiment 2.1. Samples Three types of samples were used in the experiment. For the comparison of anneal techniques, two sets of samples were used. These samples were prepared on (100) silicon with an initial 200 A-thick thermal oxide grown at 1000°C using wet oxidation. An 800 k-thick amorphous silicon layer was deposited at 555°C in an low pressure chemical vapor deposition (LPCVD) furGate using silane chemistry. One set of samples had a 500 A-thick capping layer of SiO, that was plasma deposited by decomposition of TEOS at a temperature of 450°C. The low-temperature based capping layer is provided to promote uniform grain growth as the recrystallization mechanism begins. The second set of samples consisted of the wafer, thermal oxide, amorphous silicon layer, and an 80 A-thick oxide layer on top. The third set of samples, which were used exclusively in the laser-based experiments, consisted of a 1000 A amorphous silicon film deposited on an SiO, layer using LPCVD at 550°C. The interm$diate oxide layer varied in thickness (400, 1200, 2200 A) and was deposited on a (100) silicon wafer using wet oxidation. Samples that were subsequently characterized for carrier mobility were doped with boron at 5 x 10” cmM3. 2.2. Anneal techniques Laser annealing was performed with a KrF excimer laser operating at a wavelength of 248 nm, using energy densities from 190 to 505 mJ cm -‘. The samples were processed at room temperature. Single laser pulses were used for the undoped samples, while a beam scan was performed on the doped samples in order to crystallize an area of sufficient size for electrical characterization. It has been reported that increasing the number of pulses did not show any further grain growth [lo], and this was reinforced by our results as there was no significant deviation in resultant grain size in the scanned versus non-scanned samples. Rapid thermal processing was carried out with the AG Associates Heatpulse 8108 using ramp-up and rampdown rates of 70°C s-l with a plateau of 1050°C for 30 s. Furnace anneals were performed at 600°C for 24 h in a conventional furnace.
J. Viutellu, R.K. Sing/z /Materials
80
Science and Engineering B47 (1997) 78-86
3. Results
Bright-field plan-view TEM was used to analyze the resultant microstructure of the samples. Micrographs representing each of the thermal treatments are shown in Fig. 1 for uncapped samples. The most notable difference in microstructure occurs between the laserannealed samples and the solid-phase annealed samples. For RTA and furnace anneals, the microstructure consists of relatively large but poorly-defined grains which have a high intragrain defect density. A high density of twins is also observed. The grains have a wide distribution of shapes and sizes, with some grains having a high aspect ratio. For laser annealing, the microstructure
Fig. 2. Plan-view annealed capped
Fig. 1. Plan-view annealed uncapped
TEM of (a) samples.
furnace,
(b)
RTA,
and
(c)
laser
TEM samples.
of (a)
furnace,
(b)
RTA,
and
(c)
laser
shows clearly-defined grains that are smaller than those of the solid-phase anneals, but with low intragrain defect density. Also, twins are not as readily observed and the grains are comparatively equiaxed. Microstructural differences are attributed to the variation in nucleation and growth mechanisms between the annealing techniques, and these differences are discussed in detail later in this work. For experiments involving capped samples, micrographs representing each of the thermal treatments are shown in Fig. 2. The microstructure of the samples was similar to that observed in the uncapped samples. One discernable difference, however, was the average grain size. The average grain size for the various annealing techniques is shown in Fig. 3 for samples with a
J. Viatella,
R.K.
Singh /Materials
Science
and Engineering
B47 (1997)
81
78-86
150
++
E 2 100 E: ‘8 c3 I
50 ’
I
I
50
i
+
+
piiq
+ I
250 Annealing
+
I
350
I
450
550
Technique Laser Energy (m.J/cm2 )
Fig. 3. Grain size as a function of anneal technique for capped samples.
capping layer and in Fig. 4 for samples without a capping layer. The furnace-annealed samples had the largest grain size and this value did not vary significantly between capped and uncapped samples (202204 nm). The lack of a difference in grain size between samples is due to the long processing time involved. Thermal equilibration occurs on a much shorter time scale than the 24 h tune period over which the anneal took place, and therefore the capping layer does not have any discernible effect. The RTA samples had a smaller grain size than the furnace-annealed samples and there was a significant size difference between the capped (142 run) and uncapped (116 run) samples. This size difference is likely due to an increase in thermal energy stored in the multi-layer structure, as well as differences in emissivity. Enhanced containment of thermal energy in the t%lm may be occurring from a radiative, as well as a conductive, perspective in the capped samples. Though the SiO, capping layer does not directly absorb signifi250
g 200 V $j 150 3 3 100
+
t
i
50 Annealing
Technique
Fig. 4. Grain size as a function of anneal technique for uncapped samples.
Fig. 5. Grain size as a function of laser energy density.
cant amounts of energy from the RTA lamps, it does absorb thermal energy by conduction from the underlying Si layer. This well of thermal energy allows the growth process to continue in the underlying film even after the ramp-down step is over. Without the capping layer, thermal energy is more quickly equilibrated since the underlying single-crystal Si wafer has a high thermal conductivity and only a 200 A SiO, layer is separating it from the amorphous Si layer. Also, larger radiative losses occur during processing in the uncapped samples, decreasing the amount of thermal energy available for grain growth. The laser-annealed samples possessed the smallest average grain size. The samples processed at 190 mJ cmP2 appeared to be below the recrystallization threshold, and similar results are found in other work [IO]. Above this energy density the resultant grain sizes ranged from 80 to 125 nm, as shown in Fig. 5. The samples without a capping layer consistently demonstrated larger grain sizes than those with one, likely due to energy loss from the melted film to the capping layer. The SiO, capping layer does not absorb significant laser energy but does provide conductive losses and therefore some quenching effect to the underlying film. Unlike RTA, the capping layer in this situation does not provide any thermal storage for enhanced grain growth. The energy absorbed induces melting in the amorphous Si layer resulting in radical temperature gradients which are quickly equilibrated. Also, these samples did not demonstrate much variance in grain size relative to incident energy. Uncapped samples demonstrated a trend towards smaller grains as laser energy increased, indicating that the complete melt threshold is likely at the lower energies used and that there are other mechanisms taking place that define the resultant grain size. This point is discussed in greater detail in the Section 4.
82
J. Viatelln,
Fig. 6. Cross-sectional
3.1. Effects
of intemediinte
TEM
R.K.
Singh /Materids
of laser-annealed
Science
samples
oxide layer
Experiments using laser annealing of samples with an intermediate oxide layer of varying thickness resulted in a microstructure with measured grain sizes between 60 and 200 nm. Cross-sectional TEM micrographs of the samples are shown in Fig. 6. The intermediate oxide layer is visible, as well as the microstructure of the annealed Si layer. The grain boundaries are perpendicular to the plane of the film, due to the direction of heat flow during solidification. Trends of increasing grain size with laser energy were observed, as shown in Fig. 7, with the largest grain sizes corresponding to films which had the thickest intermediate oxide layers. Due to the relatively low thermal conductivity of SiO,, the oxide film’s thickness becomes the major controlling factor regarding melt lifetime and nucleation and growth kinetics. These mechanisms are discussed in greater detail later in this work. Grain-size differences between samples with 220 and 120 nm-thick oxide layers are less (20 nm) than those between the 120 and 40 nm-thick oxide layers (50 nm), suggesting that the thermal quenching effects of the oxide layer diminish with increasing thickness. Solidification velocity calculations by the authors [l 11 based on this film configuration have shown that as the oxide layer is increased in thickness there is a decrease in 250 +
,m,
0 W A
$
+
+
240 (mJ/cm2) 336 (mJ/cm*) 405 (mJ/cm’)
0' 0
500
1000
1500
Oxide Thickness Fig. 7. Grain
size as a function
2000 2500 (A)
of oxide
thickness.
with
and Engineering
B47 (1997)
78-56
(a) 400 A, (b) 1200 A, and (c) 2200 A-thick
oxide
layer.
solidification velocity, with an asymptotic limit of 160 cm s-l at a large oxide-layer thickness. This would indicate that increasing the grain size by altering thermal properties of the oxide is maximized at a certain thickness, and that value is approximately 200 nm. An oxide layer thinner than this provides greater quenching and greater undercooling of the film. As the undercooling is increased, nucleation density increases and a smaller resultant grain size follows. For the thicker oxides, the quench rate is decreased and with it the amount of undercooling. A low nucleation density ensues and there remains sufficient thermal energy in the film to allow more significant grain growth, resulting in the larger average grain size observed. This is discussed in greater detail in the simulation section. 3.2. Microstructure
rind electricnl properties
Hall effect measurements performed on the doped laser-annealed samples resulted in measured mobilities that ranged from 198 to 351 cm2 V-l s-l. Though the grain sizes were small in comparison to other annealing techniques, the mobilities approached that of p-type doped single-crystal silicon. As mentioned previously, the reasons for this may be found in an examination of the intra-grain morphology. Grain boundaries, along with linear and planar defects, act as recombination centers in silicon films. Thus, the occurrence of twins within grains affect carrier mobility. Haji et al. [12], studied the mode of growth of silicon films during solid-phase crystallization. Their results showed that the main defects inside grains are twins. While first order twins are electrically inactive, their presence is commonly related to: (i) incoherent { 112) steps perpendicular to { 111 } twin planes which introduce dislocations at the step edges which are strongly electrically active, and (ii) higher order twins which are related to displacement vectors and are electrically active due to segregation of dissolved impurities. The lower mobility values observed in solid-phase crystallized films as compared with films solidified from melt in laser annealing can be attributed to these morphological differences. These differences reinforce the idea of laser annealing as a suitable candidate for the processing of amorphous Si films for TFT applications.
J. Viatella,
R.K.
Singh / Materials
Science
4. Discussion 4.1. Solid phase nucleation and growth
Solid phase recrystallization of amorphous Si films on SiO, has been studied by many investigators [13]. It has been well substantiated that initial nucleation of the a-Si occurs at the a-Si/SiO, interface [12,14]. It is believed that nucleation at the interface is due to the release of the stress induced by the thermal expansion difference between Si and SiO, and/or of that which is attributed to the porosity of the a-Si [14]. In-situ annealing experiments in a transmission electron microscope have revealed [12] the mode of growth for grains in a-Si tis. Amorphous silicon films deposited on glass using LPCVD were annealed at 600°C in the TEM and their growth monitored. Once nucleation takes place, the grains grow quickest along the [l IO] and [112] axis and slowest along the [I 111. Therefore, the grains have an elliptical shape [15] in general, and contain multiple twins at the center. Other twins are also formed and preferential growth in (112) directions follows, leading to dendritic grains which have a { 1 lo} surface. The crystallites lose their ellipsoidal shape when coalescence between grains starts and this ultimately results in the formation of obscure grain boundaries. The resultant microstructure is one of relatively large grains ( N 1 urn), but also a large number of intragrain defects due to the growth proceeding by formation of twins. 4.2. Nucleation
and growth
in laser annealing
The nucleation mechanisms in laser crystallization of silicon thin films can be broken down into two regimes, differentiated by whether complete melting of the t?lm occurs or not. In the first case, where complete melting of the film does not occur, several potential scenarios have been proposed. Thompson et al. [16] and Lowndes et al. [17] both support the following dynamic. At low energy densities the laser energy only melts part of the film, creating a thin liquid layer near the surface. As the liquid begins to solidify as relatively large-grain polycrystalline silicon, the latent heat released raises the temperature of the resolidified poly-Si above the temperature of the first-order phase transition of the amorphous silicon (a-Si) to the metallic liquid, T,- ,. The underlying a-Si material then begins to melt. This new liquid is severely undercooled compared to the poly-Si layer and therefore resolidifies as fine-grained poly-Si. Thus, a thin liquid layer is presumed to propagate through the a-Si material because of the released latent energy. This is labeled as the explosive recrystallization process and it is self-sustaining until it is quenched by the energy required to raise the temperature of the a-Si solid in front of the liquid to T,-,. The final mi-
and Engineekng
B47 (1997)
78-56
83
crostructure consists of a top layer of large-grained (50-100 nm) polycrystalline silicon with a layer of fine-grained (1- 10 nm) poly-Si directly below it resulting from the explosive recrystallization process. Thompson et al., were able to support the explosive crystallization argument by use of transient reflectance and conductance measurements of tims melted with nanosecond-pulse lasers. Even though the reflectivity returns to the solid value at approximately 53 ns, signifying that the surface is now in the solid phase, the conductance showed a second peak that existed for an additional 10 ns (until - 63 IX). The extended conductance was attributed to the subsurface liquid layer, which behaves as a metallic conductor. In other words, though the surface is in a solid state there is a highly conductive propagating liquid layer still within the film. Using this information, they were also able to estimate the velocity of the explosive melt front to somewhere between 10 and 20 m s- ‘. Another proposed scenario for non-complete melting of Si films is by Im et al. [18]. They argue that in the tims explosive crystallization of a-Si occurs at the onset of the transformation, implying that partial melting of explosively-crystallized fine-grained Si is occurring rather than partial melting of an amorphous film. It is suggested that early triggering of explosive crystallization may be attributed either (1) to the presence of microcrystalline clusters in the LPCVD samples, which was confirmed by analyzing solid-phase crystallization behavior and is absent in high-dose ion-irradiated amorphized samples and/or (2) the possible presence of impurities, such as hydrogen. The same researchers also put forth the idea of the super lateral growth (SLG) regime in which almost complete melting of the film occurs to the extent that there is a discontinuous Si tim composed of discrete solid islands in a liquid matrix. Growth from these clusters proceed, resulting in an unusually large (300-400 nm) grain size. If greater laser energy is used the resultant grain size returns to smaller values ( - 50-100 nm), typical of lower-energy density irradiation. In the second case, where complete melting of the film does occur, there is substantial evidence that nucleation occurs homogeneously. Stiffier et al. [19,20], argue the following. They used 200-400 nm-thick Si films deposited on SiO, layers that varied between 77 and 330 nm. The Si films were melted using a Q-switched ruby laser (25-30 ns pulses) and transient conductance and reflectance measurements were taken. Analysis of the measurements revealed the following scenario. After melting in the first lo-20 ns, the film remains fully molten as it cools by conduction to the substrate. Nucleation is followed by rapid solidification, which is shown by a fast ( - 10 ns) drop in the reflectance and conductance, as the enthalpy released heats the film to the steady-state solidification temperature determined
84
J. Viatellu,
R.K.
Singh /Materials
Science and Engineering
by the thermal properties of the substrate. Recalescence is followed by steady state solidification which is marked by a gradual (150 ns) decrease in conductance returning to the solid value. Undercooling of the films was calculated in the following manner. They determined the volume fraction of solid following recalescence from the ratio of the conductance to the fully liquid value assuming a twophase mixture and relative conductivities. The volume fraction of liquid consumed during recalescence was then used to determine the temperature of the liquid prior to nucleation using a heat balance equation. Since the heat loss to the substrate is exactly balanced by the release of enthalpy in the film during steady-state solidification, the thermal flux out of the film following recalescence was obtained from the slope of the conductance trace. This thermal flux can be used to establish an upper limit for the quench rate of the film prior to nucleation by using the specific heat and thickness of the film. Also, it can define an upper limit for the thermal leakage during recalescence. Also, T,,, the steady-state solidification temperature, can be estimated from the thermal flux. This is done by calculating an effective interface velocity from the thermal flux and enthalpy of melting, which is then used to calculate T,,. From the quench rate the undercooling can be calculated. Undercooling prior to nucleation was constant at approximately 500 K for quench rates below approximately 1O’O K s-l. As in previous studies, no evidence of preferential nucleation at either the surface or interface was present. The microstructure consisted of coarse grains combined with interspersed fine-grained structures, which typically are the signature of explosive recrystallization. This morphological idea is further reinforced in the most recent work by Im et al. [21] in which they found the SLG regime to result in a microstructure consisting of large single-crystal disk structures. They argue that these disk structures are the result of a single nucleation event occurring from a discrete cluster that was not melted in the near-complete melting of the film. These disk structures are predicted to occur when the separation distance between surviving solid clusters is greater than the maximum lateral growth distance. In other words, as the single crystal disks are growing, there is significant undercooling of the liquid which leads to nucleation of solids in the bulk of the liquid ahead of the advancing solidification front. This stalls the growth of the grains and results in the morphology shown. A variation on this morphology was also observed in which the seeds that survive are located near each other such that growth can continue until the solidification fronts of neighboring disks touch. This results in single crystal disks touching with small regions of fine poly-Si grains between the disks.
B41 (1997)
75-86
5. Simulation We have developed a model to simulate the effect of laser interactions with multi-layered structures. The solution is based on a control volume approach and can take into account temporal and spatial-dependent events. Thermophysical properties of each layer are entered as input parameters and they include: melt temperature, vaporization temperature, temperature and phase-dependent thermal conductivity and volume heat capacity, and latent heat of melting and vaporization. Temperature and phase-dependent optical properties that are included are: reflectivity, absorption coefficient, and simulated absorption coefficient of the plasma during vaporization [22]. The characteristics of the laser pulse included are energy density, pulse length, and power density distribution as a function of time. The output includes: temperature as a function of time for any nodal location within the multilayered structure, temperature profiles at any time during irradiation, as well as melt and evaporation depths as a function of time. Features of the model include an ability to simulate the movement of the solid-liquid interface as it is generated at the surface and propagates into the bulk during pulsed irradiation. After termination of the laser pulse, the solid-liquid interface changes direction and recedes to the surface. The model incorporates this phenomenon and also provides solidification velocities. As currently formulated, the model does not include some of the components that are observed in laser annealing of Si tirns on SiOJSi substrates, including nucleation of solids in undercooled liquid. Other models [23,24] proposed share this deficiency and to date a model that incorporates the observed phenomena has not been developed. However, valuable insight may be gained by examining the solidification scenarios for the thermal quenching provided by the varying-thickness oxide layers. In the case of non-complete melting of the films, undercooling is not an issue and the model formulated here provides an accurate picture of the solidification scenario. Fig. 8 shows melt depth as a function of energy density for various oxide thickness. This graph shows that complete melting occurred for the higher laser energies. In cases where there was incomplete melting of the flm solidification occurred from existing material, eliminating nucleation as a limiting factor and a smaller resultant grain size was observed (as shown in Fig. 6). The largest observed grain size occurred where complete melting of the film occurred. As mentioned previously, the evidence suggests that severe undercooling occurs in these situations and the microstructure is a result of homogenous nucleation, Using the simulation, quench rates were calculated for the varying oxide thickness and nucleation density estimated in the fol-
J. Viatella,
1000 900 800
&’
R.K.
Singh /Materials
Science
p-9/ / / #’ / / /’
700
-e-m-, -A-,
4’
El
600
400 (A) 1200(A) 2200 (A)
500 200 Energy (mJ/cm2) Fig. 8. Melt depth as a function thickness.
of energy for varying oxide
lowing manner. Initially, the quench rates given by the calculations were correlated with undercooling values derived empirically [ 191. Using classical homogeneous nucleation theory [25], the value of undercooling was used to calculate the free energy change associated with the formation of a voIume of solid by:
*G”+z
(2) M
where L, is the latent heat of fusion per unit volume, TM is the melt temperature, and AT is the value of undercooling. The critical radius for a stable cluster of particles was calculated using this free energy change by: f% - 2YSL where ysL is the solid/liquid interfacial free energy, found [26] to be - 0.34 + 0.02 J rnB2 for homogeneous nucleation assuming spherical nuclei. The excess free energy associated with a cluster of radius Yis given by:
and Engineering
B47 (1997)
78-86
85
undercooling in these sampleswas calculated to be 512 K and consequently this represents approximately 4.5 x lo6 clusters of critical size per cm2 area of film (film thickness - 100 nm). Samples with a 110 nmthick oxide layer annealed with the same energy had a smaller grain size (185 nm), and the undercooling value was 550 K. Cluster density for these films was calculated at approximately 5.9 x 107.The smallest grain size (140 nm) at this laser energy was observed in the samples with the 40 nm-thick oxide layer, and the cluster density was approximately 3.9 x 10’. The microstructural differences resulting from the three order of magnitude variation in cluster density can be seenin the cross-section TEM micrographs in Fig. 6. Fig. 6(a), of the 40 nm-thick oxide, shows less distinct grain boundaries than do Fig. 6(b) (1200 nm oxide) and (c) (2200 nm oxide). The 2200 nm oxide film has clear grain boundaries spanning the thickness of the film and a more ordered microstructure than the other two films. Though the critical cluster density calculations help explain final microstructures, other phenomenological aspects of solidification must be mentioned. When quench rates are greater than approximately 1O’OK s - ’ the embryos cannot grow at a sufficient rate to maintain a steady-state distribution [19]. Cooling can then continue until a sufficient number of embryos form, resulting in even larger values of undercooling which affect critical cluster size and density. Classical theory [25] provides a homogenous nucleation rate that involves a term which is a complex function that depends on the vibration frequency of the atoms. Since the vibration frequency is a function of temperature, there is a decrease in nucleation rate with increased undercooling. The limitation on nucleation rate is particularly significant when large values of undercooling occur and therefore should be considered in future theoretical models.
6. Conclusions AG, = -;
ni3AG, + 4nr2y,,
(4)
Using the critical radius value, r*, in the free energy calculation from Eq. (4), the number of spherical clusters of critical size is given by: (5) where no is the total number of atoms in the system, T is the temperature and k is Boltzmann’s constant. Nucleation density increases with an increase in undercooling, resulting in smaller grains than would be found where undercooling is less severe. The largest grains (200 nm) were observed in the sampleswith the 220 mn-thick oxide layer where the highest laser energies (-405 mJ cm-‘) were used. The amount of
Recrystallization of amorphous silicon using furnace annealing, rapid thermal annealing, laser annealing and the resultant microstructure has been investigated using TEM and X-ray diffraction (XRD) for characterization. Grain size was correlated to incident energy for laser-annealed samples and ranged from 50 to 200 nm. Effects of an SiO, capping layer were investigated for all three annealing techniques. The capping layer did not demonstrate a significant effect for furnace annealing, but RTA capped samples demonstrated a significantly larger average grain size than those without a capping layer. Capped samples annealed by the laser demonstrated a smaller average grain size than those that were uncapped. Also, we have examined the effects of an intermediate SiO, layer of varying thickness on
86
J. Viateila,
R.K.
Singh /Materials
Science
laser-annealed amorphous silicon films. A discussion of the advantages of laser processing over other annealing techniques with respect to carrier mobility is included along with an examination of the mechanisms of nucleation and growth for each annealing technique mentioned. A novel computer-based simulation of laser interactions with multi-layered structures is included to help identify the events shaping laser processing of amorphous silicon films.
References ‘11 T.J. King, M.G. Hack and I.W. Wu, J. Appl. Plays., 75 (1994) 908-913.
21 B. Faughnan and A.C. Ipri, IEEE (1989)
Tyans.
Electron
Devices,
36
101-106.
31 R.Z. Bachrach, K. Winer, J.B. Boyce, S.E. Ready, RI. Johnson and G.B. Anderson, J. Electron. Mater., 19 (1990) 241-248. 41 N. Yamauchi and R. Reif, J. Appl. Phys., 75 (1994) 3235. ,5] M. Bonnei, N. Duhamel, M. Guendouz, L. Haji, B. Loisel and P. Ruault, Jpn. J. Appl. Phys., 30 (1991) L1924-L1926. :6] R. Kakkad, J. Smith, W.S. Lau and S.J. Fonash, J. Appl. Phys., 65(1989)
2069-2073. J. Appl. Phys.,
.7] R. Singh, 63 (1985) R59. /S] S.D. Brotherton, D.J. McCulloch, J.B. Clegg and J.P. Gowers, IEEE Trans. Electron Devices, 40 (1993) 407-413. 191S.R. Stiffier and M.O. Thompson, Phys. Rev. B, 43 (1991) 9851-9855.
and Engineering
B47 (1997)
78-86
[lOI R.Z. Bachrach, K. Winer, J.B. Boyce, F.A. Ponce, SE. Ready, R. Johnson and G.B. Anderson, MRS
Symp.,
157
(1990) 467-472.
P11John Viatella, Rajiv K. Singh and R.P.S. Thakur,
MRS
Sylnp.,
(1994) 61-66. [I21 L. Haji, P. Joubert, J. Stoemenos and N.A. Economou, J. Appl. Phys., 75 (1994) 3944. t131 T. Noma, T. Yonehara and H. Kumomi, Appl. Phys. Lett., 59 (1991) 653-655. 1141 A. Sakal, H. Ono, K. Ishida, T. Ntino and T. Tatsumi, Jpn. J. Appl. Pirys., 30 (1991) L941-L943. 1151 A. Nakamura, F. Emoto, E. Fujii, Y. Uemoto, A. Yamamoto, K. Senda and G. Kano, Jpn. J. Appl. Phys., 27(1988) L2408-L2410. [161M.O. Thompson, G.J. Galvin, J.W. Mayer, P.S. Peercy, J.M. Poate, DC. Jacobson, A.G. Cullis and N.G. Chew, Phys. Reo. Lett., 52 (1984) 2360-2363. u71 D.H. Lowndes, R.F. Wood and J. Narayan, Pizys. Rev. Lett., 52 336
(1964)
561-564.
[181J.S. Ira, H.J. Kim and M.O. Thompson,
Appl.
Phys.
Lett.,
63
(19930 1969-1971. [191S.R. Stiffler and M.O. Thompson, Phys. Rev. B, 43 (1991) 9851-9855. WI S.R. Stiffler and M.O. Thompson, P.S. Peercy, MRS Symp., JO0 (1988)
505.
WI J.S. Im and H.J. Kim, Appl. Phys. W Rajiv K. Singh and John Viatella, ~231 ~241
PI WI
Lett., 64 (1994) 2302. J. Appl. Phys., 75 (2)
(1994) 1204- 1206. C.P. Grigoropoulos, A.A. Rostami, X. Xu, S.L. Taylor and H.K. Park, Znt. J. Heat Muss Transfer, 36 (5) (1993) 1219-1229. V. Gupta, H. Jin Song and J.S. Im, to be published. D.A. Porter and K.E. Easterling, Phase Trmsfomntions in Metals and Alloys, 2nd edn, Chapman and Hall, London, 1992. S.R. Stiffler and M.O. Thompson, Phys. Rev. Lett., 60 (1988) 2519.