Influence of the Kinetic Adsorption Process on the ... - ACS Publications

Jan 30, 2014 - The incorporation behavior of the Ge precursor with different ligands was ... Iain Buchanan , Manchao Xiao , Sergei Ivanov , and Cheol ...
0 downloads 0 Views 1MB Size
Article pubs.acs.org/cm

Influence of the Kinetic Adsorption Process on the Atomic Layer Deposition Process of (GeTe2)(1−x)(Sb2Te3)x Layers Using Ge4+− Alkoxide Precursors Taeyong Eom,† Taehong Gwon,† Sijung Yoo,† Byung Joon Choi,‡ Moo-Sung Kim,§ Iain Buchanan,∥ Manchao Xiao,∥ and Cheol Seong Hwang*,† †

WCU Hybrid Material Program, Department of Materials Science and Engineering and Interuniversity Semiconductor Research Center, Seoul National University, Seoul 151-744, Republic of Korea ‡ Department of Materials Science and Engineering, Seoul National University of Science and Technology, Seoul 139-743, Republic of Korea § Air Products Korea, 15 Nongseo-dong, Giheung-gu, Yongin-si, Gyeonggi-do 446-920, Republic of Korea ∥ Air Products and Chemicals, Inc., 1969 Palomar Oaks Way, Carlsbad, California 92011, United States ABSTRACT: (GeTe2)(1−x)(Sb2Te3)x (GST) layers were deposited via atomic layer deposition (ALD) at growth temperatures ranging from 50 to 120 °C using Ge(OCH3)4 or Ge(OC2H5)4, Sb(OC2H5)3, and [(CH3)3Si]2Te as the metal−organic precursors of the Ge, Sb, and Te elements, respectively. The GST layers with compositions lying on the GeTe2−Sb2Te3 tie lines could be obtained by varying the ratio of the Ge−Te and Sb−Te ALD cycles. Although the incorporation of an Sb−Te layer into the GST film occurred in a genuine ALD manner, that of the Ge−Te layer was governed by the kinetically limited physisorption of Ge precursors. The incorporation behavior of the Ge precursor with different ligands was explained by the adsorption and desorption kinetics based on the Brunauer−Emmett−Teller isotherm. The ALD-like film growth behavior could be well-explained by the kinetically limited incorporation of Ge atoms.

I. INTRODUCTION

Although several previous reports mostly adopted the chemical-vapor deposition (CVD)-like behavior of metal− organic precursors,6−8 a genuine ALD of GST225 was reported only recently by several groups in Helsinki University,9−12 which has been hindered by the lack of an appropriate chemical-reaction route for the ALD-type reaction and the nonavailability of a Ge precursor with an oxidation state of +2 to form GeTe. The innovative improvements made in these reports employ a strongly favorable Lewis acid−base reaction between the silyl ligands of the Te precursor (triethylsilyl group in ((C2H5)3Si)2Te) and chloride in the GeCl2−dioxane complex or SbCl3 precursors as well as coordination of the dioxane group to GeCl2 to stabilize the +2 state of the Ge ions. The highly favorable chemical reaction of the Ge and Sb precursors toward the Te precursor afforded the ALD reaction at a substrate temperature as low as 60 °C.9−12 Adopting the idea of utilizing the Lewis acid−base reaction of the silyl-Te and volatile precursors of Ge and Sb (Ge(OCH3)4 and Sb(OC2H5)3), the authors also recently reported the ALD of materials with compositions lying on the tie line between GeTe2 and Sb2Te3, such as Ge2Sb2Te7 (called GST227).13 The occurrence of the GeTe2 component was mainly ascribed to the +4 oxidation state of the Ge ions in the Ge precursor. GST227 showed phase-change and accompany-

Atomic layer deposition (ALD) is a thin-film deposition technique that critically depends on the chemically driven ligand-exchange reaction between precursor molecules, which could be specified as having a saturated growth behavior.1 This characteristic of ALD, being accompanied by the accurate thickness controllability in (sub-) atomic-er scale, make it especially suitable for depositing thin films for nanoscale electronic devices, where an almost 100% step coverage over the topologically extreme features is necessary. Phase-change random-access memory (PCRAM) is one of the highly promising future memory technologies that have a nonvolatile data-retention property, as flash memory does, and rapid writing and reading speeds, which are comparable to dynamic random-access memory. The most widely adopted phasechange materials (PCM) are alloys composed of pseudobinary compounds whose compositions lie on the tie line connecting GeTe and Sb2Te3. The Ge-rich materials have better data stability, whereas Sb-rich materials show faster switching speeds.2 Among the many alloys, Ge2Sb2Te5 (called GST225) comprises the base material for many PCRAMs. As a result of this technical trend, there have been many attempts to deposit PCM via ALD because GST must be placed within a very narrow trench3 or hole4 to improve the heating efficiency of the switching current, which must be ∼5 s. The growth rate of mono- and multilayers at the initial state of desorption (Ts ∼ 70 °C and tprg ∼ 0 s) could be estimated by extrapolating growth rate of Ge(OC2H5)4 to tprg = 0 s in Figure 1b. As a result, the growth rates of multi- and monolayer were estimated to be 131.03 and 49.92 (ng cm−2) cy−1, respectively. It should be noted that the growth rate of multilayers was ∼2.6 times higher than monolayer, which confirmed the abovementioned assumption. The bimodal desorption behavior was further confirmed from the BET-type experiment, as shown below. The BET equation, shown as eq 1, which was slightly modified from its original form,14 describes the kinetics of gas adsorption on a solid surface.

XRF cannot discern if the Te atoms are combined with the Ge atoms or with the Sb atoms. The details of the estimation methods for the composition and growth rate are reported elsewhere.13 The GeTe2 growth rates increased and became saturated after the ∼5 s Ge(OCH3)4 and the Ge(OC2H5)4 precursor injection time (closed symbols), which appears to be a typical ALD-specific growth behavior, when the tprg was 5 s. As reported elsewhere,13 however, this is due to the dynamic balance between the adsorption and desorption of the physisorbed Ge precursor molecules and not to genuine ALD-specific saturation behavior. The generally lower growth rate with Ge(OC 2 H 5 ) 4 compared with Ge(OCH3)4 could be ascribed to the higher steric hindrance effect from the larger molecular size of Ge(OC2H5)4 compared with Ge(OCH3)4. The saturated GeTe2 growth rate from Ge(OC2H5)4 was ∼40 (ng cm−2) cy−1 (0.1 nm cy−1), whereas that from Ge(OCH3)4 was about ∼60 (ng cm−2) cy−1 (0.15 nm cy−1). The variation in the growth rate with increasing tprg (tinj = 4 s for Ge(OCH3)4 and 6 s for Ge(OC2H5)4) showed that the growth of the GeTe2 layer certainly deviated from the ALD behavior. It was reported in detail elsewhere13 that the GeTe2 incorporation decays monotonically with increasing tprg for Ge(OCH3)4, which could be explained well by the isothermal desorption of the physisorbed Ge precursor molecules. As mentioned in the Introduction, confirming if the chemisorption of Ge(OC2H5)4 had been achieved was of utmost interest in this work. Unfortunately, the open square symbols in Figure 1a show that this was not the case; the growth rate kept decreasing to a very low level when tprg was increased to 20 s, suggesting weak (or no) chemical interaction between the Ge(OC2H5)4 molecules and the Sb2Te3-terminated surface. Several differences can be noted, however, when the aforementioned results are compared with those of the experiments with Ge(OCH3)4. When tprg was very short (1 s), the GeTe2 growth rate from Ge(OC2H5)4 was higher than that from Ge(OCH3)4 by ∼9 (ng cm−2) cy−1 (0.02 nm cy−1) even with the greater bulkiness of the OC2H5 ligand compared with the OCH3 ligand (tinj = 4 s for Ge(OCH3)4 and 6 s for Ge(OC2H5)4). This suggests that there could be multilayer physisorption of the Ge(OC2H5)4 molecules, perhaps because of their lower vapor pressure, whereas the same was hardly observed in Ge(OCH3)4.13 It appears that the formation of such multilayers occurs until the tprg increases to ∼5 s, as can be understood from the change in the slope in Figure 1b at that time. Figure 1b shows the semilog plot of the GeTe2 growth rate change as a function of the tprg of the Ge precursors. Because of the weaker interactions between the Ge(OC2H5)4 molecules within the multiple physisorbed layers compared with their interaction with the Sb2Te3 surface, which could be enhanced by the dipolar interactive force from the Sb−Te surface,13 the decrease rate in the early stage was higher than that in the later stage. The same behaviors were clearly observed regardless of substrate temperature, as shown in Figure 4c, which will be discussed later. When tprg increased to >∼5 s, the decrease rate becomes substantially lower, and the desorption coefficient and intercept obtained from the best linear fit graphs (lines) were 0.353 and 49.92 (ng cm−2) cy−1 in the short tprg range (tprg < 5 s) and 0.0977 and 180.95 (ng cm−2) cy−1 in the long tprg range (5 s < tprg < 20 s), respectively, according to eqs 1 and 2 in ref 13, whereas the coefficient and intercept for Ge(OCH3)4 were 0.121 and 126.61 (ng cm−2) cy−1, respectively, in the whole experimental tprg range. The

(c − 1) P P 1 = + GR(Ps − P) GR 0c GR 0c Ps

(1)

where P, Ps, GR, GR0, and c are the actual vapor pressure, saturation vapor pressure at the growth temperature of the adsorbing gas molecules (in this case, the Ge precursor), growth rate, ideal monolayer growth rate, and a constant, respectively. The constant, c, is given as exp[(Emono − Emulti) (kTs)−1], where Emono and Emulti are the activation energies for monolayer and multilayer formation, respectively. Figure 1c shows the BET plot of the growth-rate change of the GeTe2 component in the (GeTe2)(1−x)(Sb2Te3)x film as a function of the Ge(OCH3)4 and Ge(OC2H5)4 precursor vapor pressure. The BET model explains the behavior of a gas-adsorption system where two or more layers of adsorbates are formed on the surface. In this system, each of the multiple layers is formed by the relatively weak adsorptive force, and its formation process is well-described by the Langmuir equation.14 For these experiments, the Ge precursor bubbling temperatures were varied from 30 to 35 °C for Ge(OCH3)4 and from 40 to 45 °C for Ge(OC2H5)4 to change P, and ∼1 s tprg was adopted for each Ge precursor for the given conditions of the other parameters. These conditions were chosen to observe the amount of all kinds of adsorbed molecules, including the weakly bounded multilayer physisorbed molecules, and to prevent the possible gas-phase reaction between the unpurged Ge precursor and the Te precursor injected in the following ALD step. P(growth rate)−1 (Ps − P)−1 was plotted against P Ps−1 according to eq 1, and the results showed a straight line for both Ge precursors (Figure 1c), establishing that both precursors adsorb according to the BET model. The best linear fit slope and intercept were 2.23 × 10−3 ± 2.69 × 10−4 and 1.13 × 10−3 ± 4.62 × 10−5 for Ge(OCH3)4 and 7.91 × 10−3 ± 6.15 × 10−5 and 4.15 × 10−3 ± 1.20 × 10−5 for Ge(OC2H5)4, respectively, on the basis of which the ideal monolayer growth rate of GeTe2 was estimated to be 298 and 202 (ng cm−2) cy−1 for the Ge(OCH3)4 and Ge(OC2H5)4 precursors, respectively, and Emono − Emulti was estimated to be 32 and 54 meV for the Ge(OCH3) 4 and Ge(OC2H5)4 precursors, respectively. It can be noted that the growth rate of the GeTe2 film was far lower than the ideal monolayer 1585

dx.doi.org/10.1021/cm4034885 | Chem. Mater. 2014, 26, 1583−1591

Chemistry of Materials

Article

growth rate even when the Ge precursor tprg was as short as 1 s (40 and 63% of GR0 for Ge(OCH3)4 and Ge(OC2H5)4, respectively). The much higher GR0 of Ge(OCH3)4 compared with that of Ge(OC2H5)4 can be understood from the smaller molecular size of Ge(OCH3)4 (radius ∼2.82 Å assuming a special molecular shape) compared with Ge(OC2H5)4 (radius ∼3.82 Å). The radii of Ge(OCH 3 ) 4 and Ge(OC 2 H 5 ) 4 molecules were approximately estimated from the lengths and angles of the Ge−O, O−C, and C−C bonds. The absence of multilayer adsorption, therefore, for Ge(OCH3)4, even for a very short tprg, could be understood from the very large difference between the GR0 and the actual growth rate. The difference between the GR0 and the actual growth rate, however, was not that large for Ge(OC2H5)4, especially for tprg < 5 s; as such, multilayers could have been formed even though the overall surface coverage of the precursor was still lower than the fully saturated monolayer coverage. In addition, the adsorption-energy difference between the monolayer and multilayer configurations was quite small (only 54 meV), and as such, the transition of the physisorbed molecules from the monolayer to the multilayer configuration, and vice versa, is highly probable. Figure 1d shows the variations in the GeTe2 growth rate as a function of the tinj and tprg of the Te precursor, where growthrate saturation typical of ALD was evident for both Ge precursors. It must be noted that at this stage the Ge atoms were being chemically incorporated into the Ge−Te layer by completely losing their ligands via the ligand-exchange reaction with silyl groups on Te, and the Te precursors were chemisorbed on these Ge atoms on the surface. The higher steric hindrance effect of Ge(OC2H5)4 can be clearly seen in this figure because the Te precursors chemically adsorbed only on the Ge precursor sites. Figure 2a,b show the changes in the growth rate of the Sb2Te3 layer with the Sb and Te precursors, respectively, tinj

the Ge(OCH3)4 case. From the results shown in Figures 1 and 2, it can be understood that the incorporation of Sb and Te into the GST film occur via a genuine ALD-type behavior, but that of Ge does not. In the following, therefore, we describe in detail how the ALD-like saturation behavior with respect to the tinj of the Ge precursor can be achieved on the basis of the kinetic adsorption−desorption model of Ge precursors. The BET model greatly helps in understanding such behavior. When considering physisorption, saturated coverage cannot occur, especially when multilayer formation is kinetically allowed. This precludes the possibility that the ALD-like saturation behavior with respect to tinj of the Ge precursors will be obtained. The dynamic balance between the adsorption and desorption of the precursor molecules, however, can give a constant surface coverage value for the given precursor vapor pressure at a certain growth temperature, as described below, and ALD-like saturation behavior can be achieved. The analysis of the physical adsorbate coverage can start from the isotherms and isobars for adsorption, which describe equilibrium monolayer coverage, θeq. When surface coverage, θ, is defined as (number of adsorbates)(number of sites)−1, θeq can be described by the Langmuir equation (eq 2), with equilibrium coefficient K expressed as eq 3 and the partial pressure of Ge precursor, P, as expressed in eq 4.15−17

θeq =

KP 1 + KP

K=

Iα 1 = K 0 e Ed / kTs n0 kd

(2)

P = P0 e−Evap / kTc

(3) (4)

where n0 is the number density of the reaction sites and Iα, kd, and Ed are the in-flux, desorption coefficient of physisorbate, and activation energy for kd, respectively. P0, Evap, Ts, and Tc are the pre-exponent factor, activation energy for the vapor pressure, and substrate and canister temperatures, respectively. In ALD, the deposition sequences are composed of the precursor injection and purge steps, and the surface coverage approaches from 0 to θeq during the pulse step, and it approaches from θeq to 0 during the purge step. Therefore, the actual surface coverage can be described as eq 5 using θeq and the rational functions for injection (injection function, f) and purge (purge function, g). The actual growth rate can be obtained as the product of the monolayer growth rate and the surface coverage.

Figure 2. Influence of the process conditions in the Sb−Te cycle on the Sb2Te3 component growth rate of the processes using the Ge(OCH3)4 and Ge(OC2H5)4 precursors: (a) influence of the Sb precursor injection and purge time and (b) influence of the Te precursor injection and purge time.

θ = θeqfg

(5)

The change in the coverage can be estimated on the basis of the difference between the adsorption and desorption rates with the first-order reaction processes,17 which yields

and tprg, at the same temperature for the two Ge precursors. For these cases, there were no unusual features, and the Sb2Te3 layer grew in typical ALD mode. Specifically, the saturation behavior of the Sb2Te3 layer with respect to the Te precursor injection and purge time were very fast, which was already reported elsewhere when Ge(OCH3)4 was employed.13 The different saturation levels of the Sb2Te3 growth rates for the two Ge precursors, even though identical Sb and Te precursor conditions were adopted, can be understood from the fact that the growth rate of the Sb2Te3 layer is much higher on a Ge−Te surface than on a Sb−Te surface.13 As the saturation level of the Ge−Te layer was higher for Ge(OCH3)4 than for Ge(OC2H5)4, the subsequent growth of the Sb2Te3 layer was also enhanced in

f = 1 − e−[(PIα / n0) + kd]t inj

(6)

Purge function g can be described as eq 7 based on the desorption rate with the first-order desorption process (eq 2 of ref 13). g = e−kdt prg = e−kd,0 exp(−Ed / kTs)t prg

(7)

where kd,0 is the pre-exponent factor of the desorption coefficient. According to eqs 2−7, an equivalent equation for θ as a function of tinj, tprg, Ts, and Tc can be derived, and the growth 1586

dx.doi.org/10.1021/cm4034885 | Chem. Mater. 2014, 26, 1583−1591

Chemistry of Materials

Article

Figure 3b shows the influence of tprg of the Ge(OCH3)4 precursor on the growth rate of the GeTe2 component in the (GeTe2)(1−x)(Sb2Te3)x film growth when Ts was varied from 50 to 105 °C. Here, 4 s tinj and 30 °C Tc were adopted. With increasing tprg, the GeTe2 growth rates decreased monotonically at a given isothermal condition, which can be ascribed to the fact that θeq decreases with increasing tprg. The higher the Ts was, the lower the growth rate at a given tprg. A relatively small difference in growth rate was noted, however, according to the Ts change for the lower Ts conditions (Ts ∼ 50, 60, and 70 °C), whereas the growth rate was observed to have a rapid increase in rate at the higher Ts regions (95, 100, and 105 °C), which can be more clearly seen in Figure 3c,d. The growth rates were fitted well according to the exponential decaying function, as shown in eq 7. The best linear fit graph shown in Figure 3c, which is the semilog plot of Figure 3b, elucidates that kd was 0.113, 0.116, 0.121, 0.142, 0.159, and 0.184 at Ts of 50, 60, 70, 95, 100, and 105 °C, respectively. These factors indicate that the purge time for the f function to reach 0.5, which can be considered to be the half-life of physisorbed precursor molecules, was 6.13, 5.95, 5.72, 4.89, 4.36, and 3.76 s for Ts ∼ 50, 60, 70, 95, 100, and 105 °C, respectively. The absolute value of kd was not only large but also showed a fast change rate at the higher temperature range (95 < Ts < 105 °C) than at the lower temperature range (50 < Ts < 70 °C). For the quantitative understanding of this phenomenon, the kd values were plotted according to the Arrhenius form in Figure 3d. The kd values belonged to two different regions, according to Ts. The best linear fit of the data points in the two temperature regions made Ed and kd,0 33.6 meV and 0.378, respectively, for the lower temperature range (50 < Ts < 70 °C) and 312 meV and 2680 for the high temperature range (95 < Ts < 105 °C). The change point was found at 1000/Ts ∼ 2.74 K−1 (Ts ∼92.0 °C). This yielded the following quantitative purge functions g for the Ge(OCH3)4 precursor

rate can also be estimated from the product of the ideal monolayer growth rate and the surface coverage. Figure 3a shows the variation in the growth rate of the GeTe2 component in the (GeTe2)(1−x)(Sb2Te3)x film as a function of

Figure 3. Influence of the process conditions of the Ge(OCH3)4 precursor on the GeTe2 growth rate: (a) influence of Ge precursor tinj under different Ts and Tc conditions, (b) influence of tprg under various Ts conditions, (c) semilog plot of the growth rate as a function of tprg under different Ts values, (d) Arrhenius plot of the desorption coefficient, (e) variation in the experimental and theoretical surface coverages as a function of Ts, and (f) variation in the experimental and theoretical surface coverages as a function of Tc.

⎧ e−0.379 exp(−33.6meV/ kTs)t prg , T < 92 °C ⎪ s gGe(OCH ) = ⎨ 3 4 ⎪ e−2680 exp(−312meV/ kTs)t prg , T ≥ 92 °C ⎩ s

Ge(OCH3)4 precursor tinj when Ts−Tc was 70−30, 70−33, and 100−30 °C. Here, tprg ∼1 s was adopted for the Ge(OCH3)4 precursor so that the results would not be overly affected by the desorption process. The growth rates became higher in the order of 70−33, 70−30, and 100−30 °C for all of the tinj values because θeq became higher at the lower Ts or higher Tc conditions according to eqs 3 and 4. The growth rates were fitted well (lines), according to eq 6, with the exponential factors, [(PIα)n0−1 + kd], of 0.981, 1.037, and 0.920 when normalized to 124, 143, and 81 (ng cm−2) cy−1, which are the saturation growth rates for the 70−30, 70−33, and 100−30 °C conditions, respectively. It should be noted that no large difference in exponential factor was observed for all of the deposition results. This equation predicts that the Ge tinj values for half saturation (f = 0.5) were 0.707, 0.669, and 0.753 s and that the f values were 0.980, 0.984, and 0.975 when tinj ∼4 s for the 70−30, 70−33, and 100−30 °C conditions, respectively. Therefore, in practice, the exponential factor can be approximated to be 0.980 regardless of Ts and Tc, which yields the following injection function of the Ge(OCH3)4 precursor fGe(OCH ) ∼ 1 − e−0.980t inj 3 4

(9)

The two different straight lines in Figure 3d suggest that there are two different thermally activated desorption modes causing the change in the activation energy for adsorption as well as that for desorption during the film growth. The different Ed values at the low and high Ts regions are discussed in Figure 5 when the similar results for the Ge(OC2H5)4 are presented. The theoretical coverage for Ge(OCH3)4 can be evaluated according to eq 5. The estimation of P as a function of temperature showed that Evap and P0 in eq 4 are 422 meV and 1.01 × 108 Torr, respectively. The pre-exponent factor in eq 3 was estimated by correlating the experimental and theoretical growth rates. The experimental growth rate of the GeTe2 component was estimated by adopting the process conditions of 4 s tinj and 5 s tprg at 30 °C Tc and 70 or 95 °C Ts. The experimental growth rates were 64.7 and 53.0 (ng cm−2) cy−1 at 70 and 95 °C, respectively. The theoretical growth rate was calculated on the basis of the product of the theoretical coverage and the ideal monolayer growth rate for Ge(OCH3)4, 317 (ng cm−2) cy−1, with the consideration of 5 s tprg at each Ts. As a result, K0 was estimated to be 2.02 × 10−2 from the growth rate at Ts ∼70 °C for the lower Ts range (Ts < 92 °C) and 3.09 × 10−6 from the growth rate at Ts ∼95 °C for the hightemperature range (Ts ≥ 92 °C). This gave

(8)

and f can approximate 1 when tinj ≥ 4 s for the further experiments. 1587

dx.doi.org/10.1021/cm4034885 | Chem. Mater. 2014, 26, 1583−1591

Chemistry of Materials

θeq,Ge(OCH3)4

Article

⎧(2.02 × 10−2 × e 33.6meV/ kTs)(1.01 × 108 × e−422meV/ kTc) ⎪ −2 33.6meV/ kTs ) ⎪ /[1 + (2.02 × 10 × e −422meV/ kTc 8 ⎪ )], Ts < 92 °C ⎪ (1.01 × 10 × e =⎨ ⎪ (3.09 × 10−6 × e 312meV/ kTs)(1.01 × 108 × e−422meV/ kTc) ⎪ −6 312meV/ kTs ) ⎪ /[1 + (3.09 × 10 × e ⎪ (1.01 × 108 × e−422meV/ kTc)], T ≥ 92 °C ⎩ s

and some of the molecules in the multilayer structure migrate to the substrate surface, forming a monolayer structure during the ALD steps, where the bonding energy between the physisorbed molecules and the substrate is higher than the van der Waals-like dispersion force between the precursor molecules in the multilayer. The energy difference between these two structures, however, is not very large, and as such, the reverse reaction (i.e., the transition from a monolayer structure to a multilayer one) cannot be disregarded. In addition, the reevaporation of the physisorbed molecules, especially from the multilayer structure, is also quite probable. Therefore, the following adsorption/desorption reaction (eq 12) can be assumed for such a case

(10)

Because of the different activation energies, θeq,Ge(OCH3)4 and the purge function were described in the two cases according to Ts, as shown in eq 10. Here, it must be noted that Ts influenced θeq,Ge(OCH3)4 and g at the same time. Therefore, θ for the Ge(OCH3)4 precursor can be described in the two different Ts regions as eq 11.

θGe(OCH3)4

⎧f ⎪ Ge(OCH3)4 θeq,Ge(OCH3)4|Ts< 92 ° C gGe(OCH3)4 |Ts< 92 ° C , ⎪ ⎪ Ts < 92 °C =⎨ ⎪f θ | | g , ⎪ Ge(OCH3)4 eq,Ge(OCH3)4 Ts≥ 92 ° C Ge(OCH3)4 Ts≥ 92 ° C ⎪ T ≥ 92 °C ⎩ s



[Ge(OC2H5)4 ]gas ⎯⎯⎯⎯⎯→ [Ge(OC2H5)4 ]multilayer ←⎯⎯⎯⎯⎯⎯⎯⎯ k d,multi ka

⎯⎯⎯⎯⎯⎯⎯⎯⎯→

←⎯⎯⎯⎯⎯ [Ge(OC2H5)4 ]monolayer kd,mono

(12)

where the Iα is precursor in-flux; kd,multi, kd,mono are the desorption coefficients of the multilayer and monolayer physisorbates, respectively, and ka is the adsorption coefficient for forming the monolayer. When considering the equilibrium surface coverage of the monolayer, θeq,mono, for this case, the Langmuir equation (eq 2) can also be used with equilibrium coefficient K, expressed as eq 1315−17

(11)

Figure 3e shows the influence of Ts on the growth rate of the GeTe2 component in the (GeTe2)(1−x)(Sb2Te3)x film. Here, 4 s tinj, 5 s tprg, and 30 °C Tc were adopted for deposition. The theoretical coverage was calculated according to eqs 8−11, assuming the same conditions as those in the deposition experiments. The theoretical growth rates were estimated on the basis of the product of the theoretical coverage and the ideal monolayer growth rate for Ge(OCH3)4 (317 (ng cm−2) cy−1). The experimental growth rates of GeTe2 showed a gradual decrease with increasing Ts up to 80 °C and showed an abrupt decrease at the higher Ts region. The film growth was inhibited at Ts > ∼130 °C because of the large desorption rate of the precursors in this high Ts region. The theoretical growth rates were well-corresponded with the experiment results in all of the Ts ranges, and the transition in growth-rate trend is especially well-described. This is due to the change in the activation energies in eqs 9−11. Figure 3f shows the influence of Tc on the growth rate of the GeTe2 component in the (GeTe2)(1−x)(Sb2Te3)x film. Here, 4 s tinj, 5 s tprg, and 70 °C Ts were adopted for deposition. The theoretical coverage was calculated in the same way as in Figure 3e. With increasing Tc, the experimental and theoretical growth rates coincided well with each other. Therefore, it was established that the behavior of Ge(OCH3)4 can be successfully explained by the proposed mechanism. If the coverage of the physical adsorbates that form multiple layers is to be considered, then the theoretical treatment can be much more complicated because uniform Langmuir-type adsorption of all of the adsorbing molecules is not guaranteed. This is the case for the experiments with Ge(OC2H5)4 as the Ge precursor. The appropriate modification of the methodology given above, however, can reproduce the experimental results quite well, as follows. According to the data shown in Figure 1b, the theoretical analysis of the physisorption of the Ge(OC2H5)4 molecule must be done separately for the multilayer and monolayer cases, according to tprg. As can be understood from Figure 1b, the higher molecular weight and accompanying lower vapor pressure of the Ge(OC2H5)4 molecule at the growth temperature is supposed to induce multilayered physisorption, especially for short purge times. In this case, the physisorbing molecules first form a very weakly binding multilayer structure,

K=

k 0,a ka Iα Iα = e−ΔE / kTs n0 kd,multikd,mono n0 k 0,d,multik 0,d,mono (13)

and ΔE = Ea − (Ed,multi + Ed,mono) ≈ −Ed,mono

(14)

where Ea, Ed,multi, and Ed,mono are the activation energy for ka, kd,multi, and kd,mono, respectively. ΔE corresponds to the heat of adsorption, and the overall ΔE must have a negative value because the adsorption process is an exothermic process.16 ΔE cannot be directly obtained by observing the growth rate change as a function of Ts because there could be several influencing factors at different temperature ranges. It can be reasonably approximated to be −Ed,mono, however, because Ed,mono is usually much larger than the difference between Ea and Ed,multi in physisorption cases.16,18 During the precursor injection process, the precursor physically adsorbs, forming a multilayer structure. At the same time, some of the physisorbed molecules in the multilayer transferred to the monolayer. It can be safely assumed, however, that there are many more physisorbed molecules in the multilayer than in the monolayer. Therefore, the injection function, f, can be described by the adsorption and desorption kinetic parameters for multilayers. This yields f = 1 − e−[(PIα / n0) + kd,multi]t inj

(15)

This must be applied for the case of the growth rate estimation when tprg < 5 s. In contrast, the f and g functions for tprg > 5 s must be identical to those for the Ge(OCH3)4 case because the physisorbed molecules are mostly of the monolayer configuration under this circumstance. Figure 4a shows the variation in the growth rate of the GeTe2 component in the (GeTe2)(1−x)(Sb2Te3)x film as a function of Ge(OC2H5)4 precursor tinj when Ts−Tc was 70−40, 70−45, 1588

dx.doi.org/10.1021/cm4034885 | Chem. Mater. 2014, 26, 1583−1591

Chemistry of Materials

Article

and 100−40 °C. Here, the tprg of the Ge precursor was 1 (closed symbols) or 5 s (open symbols). When tprg = 1 s, the growth rates were fitted well according to eq 15, with the exponential factors of 0.864, 0.860, and 0.872, when normalized to 133, 167, and 96 (ng cm−2) cy−1 for 70−40, 70−45, and 100−40 °C, respectively. Similar to the Ge(OCH3)4 precursor, no significant difference in exponential factor was observed for all the deposition results. The relatively smaller exponential factor compared with that in the Ge(OCH3)4 case suggests that the saturation becomes sluggish as the molecular size of the Ge precursor increases. Equation 15 also predicts that the Ge tinj for the half-saturation (f = 0.5) times were 0.803, 0.806, and 0.795 s and that the f values when tinj ∼ 6 s were 0.994, 0.994, and 0.995 for the 70−40, 70−45, and 100−40 °C conditions, respectively. Therefore, in practice, the exponential factor for Ge(OC2H5)4 can also be approximated to be 0.864 regardless of Ts and Tc,, which yields the injection function of the Ge(OC2H5)4 precursor fGe(OC H ) ∼ 1 − e−0.864t inj 2

5 4

(16)

and f can approximate 1 when tinj ≥ 6 s for the further experiments. When tprg = 5 s, the growth rates were also fitted well according to eq 6, with exponential factors of 0.760, 0.788, and 0.727 when normalized to 37.5, 46.8, and 19.6 (ng cm−2) cy−1 for 70−40, 70−45, and 100−40 °C, respectively. Therefore, f Ge(OC2H5)4 can be represented as eq 17. fGe(OC H ) ∼ 1 − e−0.760t inj 2

5 4

Figure 4. Influence of the process conditions of the Ge(OC2H5)4 precursor on the GeTe2 component growth rate of the processes using the (a) influence of tinj under different Ts and Tc conditions, (b) influence of tprg under various Ts conditions, (c) semilog plot of the growth rates as a function of tprg, (d) Arrhenius plot of the monolayer desorption coefficients, (e) variation in the experimental and theoretical surface coverages as a function of Ts, and (f) variation in the experimental and theoretical surface coverage as a function of Tc.

(17)

In this case, the exponential factor (0.760) is even smaller than in the other cases. This means that the monolayer of physisorbed molecules are formed via the multilayer physisorption step, as mentioned in the discussion related to eq 12. Therefore, in the following analysis, a 5 s tprg was adopted to confirm that the deposition of the GeTe2 layer proceeds with the monolayer configuration of the Ge(OC2H5)4 precursor. Figure 4b shows the influence of the tprg of the Ge precursor on the growth rate of the GeTe2 components in the (GeTe2)(1−x)(Sb2Te3)x film growth when Ts was varied from 50 to 105 °C. Here, 6 s tinj and 40 °C Tc were adopted. With increasing tprg, the GeTe2 growth rates decreased monotonically at a given isothermal condition, which can be ascribed to the fact that θeq decreases with increasing Ts. It should be noted that the two-step desorption processes of the multilayers and monolayers were observed regardless of the Ts condition, as shown in Figure 1a. The change in the slope in Figure 4c, which is the semilog plot of Figure 4b, confirms this. A relatively small difference in growth rate of monolayer was noted according to the Ts change for the lower Ts conditions (Ts∼ 50, 60, and 70 °C), whereas the difference in growth rate of monolayer was observed to be large at the higher Ts regions (95, 100, and 105 °C), which is clearly seen in Figure 4c. The monolayer growth rates were fitted well at tprg ≥ 5 s according to the exponential decaying function, as shown in eq 7. The best linear fit graph shown in Figure 4c, which is a semilog plot of Figure 4b, elucidates that kd,mono was 0.0860, 0.0924, 0.0977, 0.174, 0.213, and 0.252 at 50, 60, 70, 95, 100, and 105 °C, respectively. These factors indicate that the purge times for the f function to reach 0.5, which can be considered the half-life of the physisorbed precursors in the monolayer, were 8.06, 7.50, 7.09, 3.98, 3.26, and 2.75 s for the same Ts. The absolute value of kd,mono was not only large but also showed a

fast change rate at the higher Ts range (95 < Ts < 105 °C) than at the lower Ts range (50 < Ts < 70 °C). For the quantitative understanding of this phenomenon, the kd,mono values were plotted according to the Arrhenius form in Figure 4d. The kd,mono values belonged to two different regions according to Ts. The best linear fit of the data points in the two temperature regions made the Ed,mono and kd,0,mono 61.2 meV and 7.76 × 10−1, respectively, for the low-temperature range (50 < Ts < 70 °C), and 443 meV and 2.07 × 105 for the high-temperature range (95 < Ts < 105 °C). The change point was found at 1000/Ts ∼ 2.81 K−1 (Ts ∼ 84.0 °C). This yielded the quantitative purge function, g, for the Ge(OC2H5)4 precursor gmono,Ge(OC H ) 2

5 4

⎧ e−0.776 exp(−61.2meV/ kTs)t prg , T < 84 °C ⎪ s =⎨ 5 ⎪ e−2.07 × 10 exp(−443meV/ kTs)t prg , T ≥ 84 °C ⎩ s

(18)

The theoretical monolayer coverage for Ge(OC2H5)4 can be evaluated according to eq 5. The estimation of P as a function of temperature showed that the Evap and P0 in eq 4 are 551 meV and 6.47 × 108 Torr, respectively. The pre-exponent factor in eq 13 was estimated by correlating the experimental and theoretical growth rates. The experimental growth rate of the GeTe2 component was estimated by adopting the process conditions of 6 s tinj and 5 s tprg at 30 °C Tc and 70 or 90 °C Ts. The experimental growth rates were 30.9 and 29.0 (ng cm−2) cy−1 at 70 and 90 °C, respectively. The theoretical growth rate 1589

dx.doi.org/10.1021/cm4034885 | Chem. Mater. 2014, 26, 1583−1591

Chemistry of Materials

Article

Finally, the different Ed,mono of the low and high Ts ranges for both precursors are discussed. Figure 5a,b show the SEM

was calculated on the basis of the product of the theoretical coverage and the ideal monolayer growth rate for Ge(OC2H5)4, 202 (ng cm−2) cy−1, with the consideration of 5 s tprg at each Ts. As a result, K0 was estimated to be 4.76 × 10−2 from the growth rate at Ts ∼ 70 °C for the lower Ts range (Ts < 84 °C) and 3.38 × 10−7 from the growth rate at Ts ∼ 90 °C for the hightemperature range (Ts ≥ 84 °C). This gave θeq,mono,Ge(OC2H5)4 ⎧(4.76 × 10−2 × e 61.2meV/ kTs)(6.47 × 108 × e−551meV/ kTc) ⎪ −2 61.2meV/ kTs ) ⎪ /[1 + (4.76 × 10 × e −551meV/ kTc 8 ⎪ )], Ts < 84 °C ⎪ (6.47 × 10 × e =⎨ ⎪ (3.38 × 10−7 × e 443meV/ kTs)(6.47 × 108 × e−551meV/ kTc) ⎪ −7 443meV/ kTs ) ⎪ /[1 + (3.38 × 10 × e ⎪ (6.47 × 108 × e−551meV/ kTc)], T ≥ 84 °C ⎩ s

Figure 5. SEM (left) and AFM (right) image of the (GeTe2)0.66(Sb2Te3)0.33 film grown under same tinj (∼6 s) and Tc (∼40 °C) and different Ts and tprg conditions: (a) Ts ∼ 100 °C, tprg ∼ 7 s and (b) Ts ∼ 70 °C, tprg ∼ 15 s.

images of the samples indicated by arrows in Figure 4c, which have similar growth rates despite their different growth conditions. The SEM and AFM images show that the film grown at the lower Ts (70 °C) region had a much smaller grain size and a higher nucleation density compared with that grown at a higher Ts region (100 °C), whereas the two films had almost identical layer densities. This suggests that the film growth at the lower Ts region is dominated by the nucleation of the GST films on the SiO2 surface, whereas that at the higher Ts region is dominated by the coagulation of the GST grains from the sparsely formed nuclei. Therefore, the estimated smaller Ed,mono at a low Ts range represents the desorption energy of the precursor molecules from the SiO2 surface, and the higher Ed,mono at the high Ts range represents the desorption energy of the precursor molecules from the GST nuclei. The generally low-nucleation density of the GST layer on the SiO 2 surface7,8,13 supports these experimental results. The transition of activation energy played important roles for not only growth rate but also roughness of the film. The AFM measurement results show that root-mean-square roughness of films were 2.96 and 7.29 nm for the lower and higher Ts region, respectively. A smoother film was obtained in the low Ts region because of the simultaneous formation and uniform growth of nuclei.

(19)

Because of the different activation energies, θeq,mono,Ge(OC2H5)4 and the purge function were described in the two cases according to Ts, as shown in eq 20. The θ for the Ge(OC2H5)4 precursor can be described under different Ts values as eq 20.

θmono,Ge(OC2H5)4

⎧f ⎪ Ge(OC2H5)4 θeq,mono,Ge(OC2H5)4| ⎪ gmono,Ge(OC H ) | , Ts < 84 °C 2 54 ⎪ Ts < 84 ° C ⎪ Ts< 84 ° C =⎨ ⎪f θ | ⎪ Ge(OC2H5)4 eq,mono,Ge(OC2H5)4 ⎪ | g , T ≥ 84 °C ⎪ Ts≥ 84 ° C mono,Ge(OC2H5)4 T ≥ 84 ° C s ⎩ s

(20)

According to eqs 17−20, an equivalent equation of θ for the Ge(OC2H5)4 precursor as a function of Ts, Tc, tinj, and tprg can be derived, and the growth rate can be estimated. It should be noted that the theoretical monolayer growth rate is valid when tprg ≥ 5 s. Figure 4e shows the influence of Ts on the growth rate of the GeTe2 component in the (GeTe2)(1−x)(Sb2Te3)x film. Here, 6 s tinj, 5 s tprg, and 40 °C Tc were adopted for deposition. The theoretical coverage was calculated using eqs 17−20, assuming the same deposition conditions as those in the deposition experiments. The theoretical growth rates were estimated on the basis of the product of the theoretical coverage and the ideal monolayer growth rate for Ge(OC2H5)4 (202 (ng cm−2) cy−1). The experimental growth rates of GeTe2 showed a gradual decrease with increasing Ts up to 80 °C and showed an abrupt decrease at the higher Ts region. Eventually, almost no film was grown at Ts ≥ 130 °C because o fthe excessive desorption of the precursors at this high Ts region. It should be noted that the theoretical growth rates were well-corresponded with the experiment results, showing that the change of the trend is caused by the change in the activation energies in eqs 18−20. Figure 4f shows the influence of Tc on the growth rate of the GeTe2 component in the (GeTe2)(1−x)(Sb2Te3)x film. Here, 6 s tinj, 5 s tprg, and 70 °C Ts were adopted for deposition. The theoretical coverage was calculated in the same way as in Figure 4e. With increasing Tc, the experimental and theoretical growth rates increased. It should be noted, however, that the theoretical model certainly underestimated the actual growth rate as Tc (or P) increased. This could have been due to the fact that the physisorbed molecules formed more of a multilayer configuration as P increased for the given tprg of 5 s, whereas the theoretical model still assumed a monolayer configuration.

IV. CONCLUSIONS The incorporation mechanisms of the GeTe2 layer into the (GeTe2)(1−x)(Sb2Te3)x film during ALD were examined in detail for two different Ge precursors: Ge(OCH3)4 and Ge(OC2H5)4. Although the incorporation mechanism of the Sb2Te3 layer is the genuine ALD type, the weak chemical interaction between the Ge precursors and the Sb−Teterminated surface resulted in the physisorption of both Ge precursors and thus was shown not to be a genuine ALD mechanism. As a result, the growth rate eventually became almost zero when the Ge precursor purge time increased to ∼20 s. The dynamic balance between the adsorption and desorption kinetics of the Ge precursors, however, allowed an ALD-like saturation behavior regarding the precursor injection time to be achieved even when multilayer formation of the physisorbed Ge(OC2H5)4 molecules occurred. The Brunauer− Emmett−Teller-(BET)-type analysis of the growth behavior revealed that the maximum physisorption density of the precursor molecules even with a multilayer configuration is still far lower than the full coverage of the monolayer, which is the fundamental reason that ALD-like saturation can be achieved even for the multilayer physisorbed Ge(OC2H5)4 molecules. The behavior of the growth rate with the variations in the substrate temperature and precursor bubbling temper1590

dx.doi.org/10.1021/cm4034885 | Chem. Mater. 2014, 26, 1583−1591

Chemistry of Materials

Article

ature can also be quantitatively predicted using the physisorption model based on BET theory. This study sheds new light on the ALD mechanism of compositionally complicated thin films.



AUTHOR INFORMATION

Corresponding Author

*E-mail: [email protected]. Notes

The authors declare no competing financial interest.



ACKNOWLEDGMENTS This study was supported by the Converging Research Center Program through a National Research Foundation of Korea (NRF) grant funded by the Ministry of Education, Science, and Technology (2012K001299).



REFERENCES

(1) Suntola, T. Mater. Sci. Rep. 1989, 4, 261. (2) Yamada, N.; Ohno, E.; Akahira, N.; Nishiuchi, K. i.; Nagata, K. i.; Takao, M. Jpn. J. Appl. Phys. 1987, 26S4, 61. (3) Pellizzer, F.; Pirovano, A.; Ottogalli, F.; Magistretti, M.; Scaravaggi, M.; Zuliani, P.; Tosi, M.; Benvenuti, A.; Besana, P.; Cadeo, S.; Marangon, T.; Morandi, R.; Piva, R.; Spandre, A.; Zonca, R.; Modelli, A.; Varesi, E.; Lowrey, T.; Lacaita, A.; Casagrande, G.; Cappelletti, P.; Bez, R. Symp. VLSI Technol., Dig. Tech. Pap. 2004, 18. (4) Kim, Y.-T.; Hwang, Y.-N.; Lee, K.-H.; Lee, S.-H.; Jeong, C.-W.; Ahn, S.-J.; Yeung, F.; Koh, G.-H.; Jeong, H.-S.; Chung, W.-Y.; Kim, T.K.; Park, Y.-K.; Kim, K.-N.; Kong, J.-T. Jpn. J. Appl. Phys. 2005, 44, 2701. (5) Process Integration, Devices, and Structures. International Technology Roadmap for Semiconductors, 2011. (6) Choi, B. J.; Choi, S.; Shin, Y. C.; Hwang, C. S.; Lee, J. W.; Jeong, J.; Kim, Y. J.; Hwang, S.-Y.; Hong, S. K. J. Electrochem. Soc. 2007, 154, H318. (7) Choi, B. J.; Choi, S.; Shin, Y. C.; Kim, K. M.; Hwang, C. S.; Kim, Y. J.; Son, Y. J.; Hong, S. K. Chem. Mater. 2007, 19, 4387. (8) Choi, B. J.; Choi, S.; Eom, T.; Ryu, S. W.; Cho, D.-Y.; Heo, J.; Kim, H. J.; Hwang, C. S.; Kim, Y. J.; Hong, S. K. Chem. Mater. 2009, 21, 2386. (9) Knapas, K.; Hatanpäa,̈ T.; Ritala, M.; Leskelä, M. Chem. Mater. 2009, 22, 1386. (10) Pore, V.; Hatanpäa,̈ T.; Ritala, M.; Leskelä, M. J. Am. Chem. Soc. 2009, 131, 3478. (11) Ritala, M.; Pore, V.; Hatanpäa,̈ T.; Heikkilä, M.; Leskelä, M.; Mizohata, K.; Schrott, A.; Raoux, S.; Rossnagel, S. M. Microelectron. Eng. 2009, 86, 1946. (12) Sarnet, T.; Pore, V.; Hatanpäa,̈ T.; Ritala, M.; Leskelä, M.; Schrott, A.; Zhu, Y.; Raoux, S.; Cheng, H.-Y. J. Electrochem. Soc. 2011, 158, D694. (13) Eom, T.; Choi, S.; Choi, B. J.; Lee, M. H.; Gwon, T.; Rha, S. H.; Lee, W.; Kim, M.-S.; Xiao, M.; Buchanan, I.; Cho, D.-Y.; Hwang, C. S. Chem. Mater. 2012, 24, 2099. (14) Brunauer, S.; Emmett, P. H.; Teller, E. J. Am. Chem. Soc. 1938, 60, 309. (15) Langmuir, I. J. Am. Chem. Soc. 1916, 38, 2221. (16) Gorte, R.; Schmidt, L. D. Surf. Sci. 1978, 76, 559. (17) Ranke, W.; Joseph, Y. Phys. Chem. Chem. Phys. 2002, 4, 2483. (18) Christmann, K. Surf. Sci. Rep. 1988, 9, 1.

1591

dx.doi.org/10.1021/cm4034885 | Chem. Mater. 2014, 26, 1583−1591