Parameter Space of Atomic Layer Deposition of Ultrathin Oxides on Graphene.

Atomic layer deposition (ALD) of ultrathin aluminum oxide (AlOx) films was systematically studied on supported chemical vapor deposition (CVD) graphene. We show that by extending the precursor residence time, using either a multiple-pulse sequence or a soaking period, ultrathin continuous AlOx films can be achieved directly on graphene using standard H2O and trimethylaluminum (TMA) precursors even at a high deposition temperature of 200 °C, without the use of surfactants or other additional graphene surface modifications. To obtain conformal nucleation, a precursor residence time of >2s is needed, which is not prohibitively long but sufficient to account for the slow adsorption kinetics of the graphene surface. In contrast, a shorter residence time results in heterogeneous nucleation that is preferential to defect/selective sites on the graphene. These findings demonstrate that careful control of the ALD parameter space is imperative in governing the nucleation behavior of AlOx on CVD graphene. We consider our results to have model system character for rational two-dimensional (2D)/non-2D material process integration, relevant also to the interfacing and device integration of the many other emerging 2D materials.

Introduction

Two-dimensional (2D) materials, such as graphene, offer new and improved functionalities for a wide range of applications ranging from electronics and photonics to energy conversion and storage devices.[1] The effective properties of 2D materials are, however, extremely dependent on their environment, and hence their route to applications critically requires precise control of interfacing and integration in particular with established non-2D materials including metals, metal oxides, and polymers. Characteristics for 2D materials are their strong, predominantly covalent, intralayer bonding, contrasted by their weak out-of-plane interactions dominated by van der Waals forces. Because of these weak out-of-plane interactions, it remains extremely challenging to grow ultrathin continuous layers of such standard materials on top of 2D materials, be it as dielectric, barrier, dopant, contact, light emitter/absorber, carrier recombinator/separator, catalyst, or structural support.[2−10] The properties of a 2D material interfaced with a conventional thin film are thereby not merely dictated by the quality of the components. A significant challenge is to provide an optimum interface between the 2D and 3D structure, which requires a detailed understanding of the various growth modes and of 2D/non-2D material interfacing. Almost all 2D-based electrical devices, for instance, require not only metallic contacts but also interfacing to a common dielectric. While progress has been made in the scalable process integration of chemical vapor deposition (CVD) of 2D materials with atomic layer deposition (ALD) of ultrathin metal oxides,[7,11,12] a fundamental understanding of such interfacing remains in its infancy, hindering the rational process and device design for 2D/non-2D integration.

Here, we focus on the nucleation behavior of ALD aluminum oxide (AlO) on supported CVD graphene, systematically exploring the ALD growth modes and the governing conditions for achieving either selective or conformal AlO deposition on graphene that is either supported by its original growth catalyst or transferred with various levels of defects, wrinkles, and contamination. To date, the most common approaches to enhancing wetting for graphene and, hence, achieving a high AlO nucleation density and more conformal coverage employ either lower deposition temperatures (Tdep) or a surface modification of the graphene using seed layers, functional groups, and a more reactive oxidant to uniformly activate the graphene surface.[2,12−20] However, such approaches can not only degrade the AlO film properties and/or the graphene but also introduce additional elements/states at the interface that can be deleterious to the device functionality. Hence, here we do not employ any additional graphene surface modification but rather focus on the details of the ALD parameter space. Because ALD depends heavily on surface saturation to achieve the self-limiting sequential reactions, the nucleation behavior is mainly governed by three parameters: the available amount of oxidant/precursor for reaction, their mass transport to the surface, and the surface reaction kinetics.[21,22] We address the choice of these parameters in detail to control AlO deposition on CVD graphene. We show that by extending the precursor residence time, by either optimizing the pulse sequences or introducing a soaking period, we are able to overcome the otherwise heterogeneous nucleation that is limited to defect/selective sites and highly dependent on support such as layer numbers and the underlying metal. As demonstrated herein, sub-2-nm thin continuous AlO films can be achieved directly on graphene using standard water (H2O) and trimethylaluminum (TMA) precursors even at a high Tdep of 200 °C. Such a capability to directly integrate a thin continuous AlO film, an archetypical high-κ dielectric, with graphene would allow the further development of a wide range of applications that utilize graphene as the channel material.[23]

Materials and Methods

AlO films were deposited directly by ALD on four different sets of samples: graphene grown on copper metal catalysts (G/Cu), graphene grown on germanium substrates (G/Ge), graphene transferred on SiO2 substrates (G/SiO2), and highly oriented pyrolytic graphite (HOPG; Agar Scientific, 3.5 ± 1.5 mosaic spread). These samples were selected to represent different types of supported graphene because it has been previously shown that the AlO nucleation behavior is strongly affected by the underlying support.[17] The G/Cu samples were grown by CVD using a H2-diluted CH4 (0.1% in argon) precursor at a partial pressure of ∼10–3 mbar and a temperature of 950–1000 °C on polycrystalline copper foils (Alfa Aesar, 25 μm thickness, 99.8% purity), which have been electrochemically polished prior to CVD using diluted H3PO4 (85% aqueous solution, further diluted in H2O with a 7:3 ratio) under a constant voltage of 2.7 V for 7–15 min.[24] The G/Ge samples were grown by CVD on a monocrystalline germanium wafer (110) using a H2-diluted CH4 (CH4/H2 ratio of 1:52) precursor at a partial pressure of ∼1 mbar and a temperature of 920 °C. The G/SiO2 samples were fabricated by transferring the graphene layer from G/Cu to SiO2 substrates (silicon wafer with a 300 nm native oxide) using a polymer support (Microchem 950PMMA A4) and wet chemical etching (0.5 M FeCl3 and 37% HCl), followed by a cleaning process in acetone and H2 annealing at a partial pressure of ∼1 mbar and a temperature of 200 °C, as described in detail elsewhere.[25,26] All CVD and transferred graphene samples used herein were predominantly monolayer graphene (MLG) with complete coverage over the substrates with a size of >1 × 1 cm2. To ensure that the findings in this study were consistent and not skewed by changes in the sample wettability due to adventitious carbon contamination from ambient air,[24] ALD was performed within 7 days after CVD or a transfer process for G/Cu, G/Ge, and G/SiO2 and within 15 min subsequent to mechanical cleavage for HOPG.

AlO films were deposited on all samples by ALD (Cambridge Nanotech Savannah S100 G1) using trimethylaluminum (TMA; purity >98%, Strem Chemicals 93-1360) as the precursor, and unless stated otherwise, the vapor of deionized water (H2O) as the oxidant was delivered alternatingly into the reaction chamber by 20 sccm of a N2 flow. During ALD, TMA and H2O were volatized at a temperature of 40 °C, and when ozone (O3) was used as the oxidant in place of H2O, it was supplied by an ozone generator (DELOzone LG-7, ∼90% power output) at room temperature. The deposition temperature (Tdep) was varied between 80 and 200 °C. All samples were loaded and unloaded while the chamber was at Tdep without bringing the temperature down to room temperature. Prior to ALD, the chamber was pumped until it reached a base pressure (Pbase) of ∼4.5 × 10–1 Torr, while being purged with 20 sccm of a N2 flow for at least 10 min (tpurin). To prevent premature or CVD-like reactions, the chamber was purged after each delivery of an oxidant/precursor with 20 sccm of a N2 flow and a purging time (tpur) that varied depending on Tdep: 60 s purge for 80 °C, 45 s purge for 120 °C, 30 s purge for 150 °C, 20 s purge for 180 °C, and 12 s purge for 200 °C. Unless stated otherwise, the total number of ALD cycles was always limited to only 12 cycles to highlight the nucleation process because a higher number of cycles usually results in a more uniform deposition. For consistency, the oxidant/precursor dose is always approximated by the product of the delivery pressure (Pdos) and residence time (tdos), which are determined by the maximum and full-width at half-maximum (fwhm) values of the chamber pressure profile when the dose is delivered. We avoid the use of pulse time (tpul) as a measurement metric because the same tpul may result in different Pdos and tdos if the carrier gas flow rate, pumping speed, and amount of oxidant/precursor available for volatilization are varied.

To elucidate the effect of the ALD parameters on the AlO nucleation behavior on graphene, ALD was performed under four distinct processes: continuous-flow mode (CM), pretreatment mode (PM), multipulse mode (MM), and stopped-flow mode (SM). Schematic representations of these processes are shown in Figure . CM (Figure a) is an ALD mode commonly used in previous literature,[12−14,18] where H2O and TMA are dosed alternatingly into the reaction chamber and separated by the purging periods. The effect of the oxidant/precursor doses was investigated by varying the H2O/TMA doses between ∼0.14 and ∼2.1 Torr·s, obtained by pulsing H2O (tpulA) between 15 and 300 ms and TMA (tpulB) between 15 and 100 ms. In CM, the doses for both H2O and TMA are always set equally, while the dose for O3, when it is used as the oxidant, is always set at a constant value of ∼30 Torr·s. PM (Figure b) was used here to introduce a surface modification to the sample without the addition of seed layers but rather by exposure to a series of H2O or O3 pulses for a certain period of pretreatment time (tpretreat) prior to AlO deposition. Here, tpretreat is varied between 10 and 300 min for H2O pretreatment and between 2 and 15 min for O3 pretreatment. The oxidant dose and purging time in the pretreatment period are the same as those in the subsequent deposition period, which is performed under the same conditions as those in CM. The extended oxidant/precursor residence time is introduced herein by the use of a sequence of multiple pulses in MM and soaking periods in SM. In MM (Figure c), each reactant/precursor dose is delivered by a sequence of two consecutive pulses in quick succession. The time interval (tintv) between these pulses is adjusted in such a way that tdos becomes the sum of the fwhm of both pulses. In SM (Figure d), the oxidant/precursor soaking period is introduced by stopping the flow to create a static atmosphere in the process chamber for several seconds (thold) right after the oxidant/precursor is dosed. Therefore, the dose in SM is controlled by two independent parameters, tpul and thold. Before the subsequent dose, the flow is continued and the chamber is purged. The effect of the oxidant/precursor residence time in MM and SM was investigated by varying the H2O/TMA tdos between ∼2.5 and ∼3.5 s while keeping all of the other ALD conditions the same as those in CM. Further details of the ALD parameters are described in the Supporting Information (section SI1).

Figure 1

Schematic of the ALD process in (a) CM, (b) PM, (c) MM, and (d) SM. A denotes the oxidant, here H2O vapor or O3, and B denotes the metal precursor, here TMA. The oxidant/precursor dose is calculated from the product of the delivery pressure (Pdos) and the residence time (tdos), which in CM and PM are both governed by a single-parameter ALD pulse time (tpul). All samples are loaded while the chamber is at the preset deposition temperature (Tdep), and the process chamber is purged with N2 for more than 10 min (tpurin) before the ALD process is started. The purge time between the oxidant/precursor pulses (tpur) is varied between 10 and 60 s depending on Tdep. In PM, the samples are exposed to a series of oxidant pulses prior to the ALD process, where the pretreatment time (tpretreat) is determined by the total number of pulses. In MM, each oxidant/precursor is delivered twice in quick succession with a very short time interval (tintv). Thus, tdos in MM can be twice as long as that in CM for the same Pdos. In SM, the flow in the process chamber is stopped for several seconds (thold) to allow the samples to be soaked in the oxidant/precursor. Therefore, tdos in SM can be adjusted independently from Pdos.

The AlO nucleation was characterized by scanning electron microscopy (SEM; Carl Zeiss SIGMA VP) at an acceleration voltage of 2 kV and atomic force microscopy (AFM; Digital Instruments Dimension 3100) under tapping mode at a scanning frequency of 1 Hz. The AlO surface coverage (θ) was calculated based on the contrast observed in SEM images, with bright regions indicating areas of the graphene surface that are covered by AlO films/clusters and dark regions indicating the absence of AlO. Further details of the surface coverage calculation are described in the Supporting Information (section SI2).

Results

Figure shows the typical surface topography of CVD graphene on G/Cu prior and subsequent to ALD AlO using CM (Figure a). Because of the nature of the CVD method used for the growth, the surface topography of G/Cu is dominated by uniaxial graphene wrinkles and Cu surface reconstructions with an average height of 10–25 nm and an interspacing of 200–600 nm,[27−30] which is equivalent to an average feature aspect ratio of much less than unity and a root-mean-square (rms) surface roughness of ∼5 nm (Figure a). When ALD is performed in CM (Figure a) under typical conditions of Tdep of 200 °C and a TMA/H2O dose of ∼0.14 Torr·s, which is obtained by the commonly used tpulA and tpulB settings of 15–30 ms,[17,31,32] the nucleation behavior on G/Cu is highly influenced by the presence of G/Cu surface features. For a low number of ALD cycles, in this case 12 cycles, AlO is observed to nucleate preferentially on the ridges of these features, while the troughs are still relatively, although not entirely, free from AlO (Figure b).[32] Under these ALD conditions, AlO deposition in the troughs occurs subsequently when G/Cu is subjected to further ALD cycles, and a high number of ALD cycles will eventually lead to complete coverage of the G/Cu surface. This behavior was observed after 100 ALD cycles, at which point the AlO layer almost completely encapsulates the G/Cu surface, including the troughs (Figure c). Note that the topography of the deposited AlO layer resembles islandlike clusters, rather than a smooth film, implying a Volmer–Weber-type nucleation mode.[33]

Figure 2

SSEM and AFM images of typical CVD G/Cu before (a) and after (b and c) ALD of AlO in CM at Tdep of 200 °C. (a) Surface topography of the as-grown CVD G/Cu is dominated by graphene wrinkles and Cu surface reconstructions with average heights of 10−25 nm and interspacings of 200−600 nm. (b) AlO deposition on G/Cu with only 12 ALD cycles, demonstrating the nature of the nucleation process that is highly preferential to the ridges. (c) AlO deposition on G/Cu with 100 ALD cycles, demonstrating the eventual complete surface coverage due to deposition on the troughs as the number of ALD cycles increases. In parts b and c, the dark regions indicate the uncovered graphene surface, while the bright regions indicate the presence of AlO clusters/films on the graphene surface. All scale bars represent 500 nm, and the red parallel lines indicate the ridges of G/Cu surface features.

The highly selective AlO nucleation behavior on G/Cu at Tdep of 200 °C leads to the assumption that a lower Tdep is a necessary condition for achieving a more homogeneous nucleation with H2O/TMA.[16,17,33−35] Indeed, a significant change in the AlO nucleation behavior could be achieved by simply altering Tdep while keeping the other deposition parameters constant. As shown in Figure a, a significantly higher nucleation density in the troughs is observed when Tdep is decreased to 150 °C while maintaining a constant TMA/H2O dose of ∼0.14 Torr·s. When Tdep is lowered further to 80 °C, AlO nucleation becomes completely nonpreferential, nucleating almost everywhere on the G/Cu surface and yielding a surface coverage (θ) of ∼98%. Note that the nonpreferential nucleation is not due to the effect of insufficient purging because a too short tpur will result in premature hydrolysis of TMA, which impedes AlO nucleation on graphene (see also the Supporting Information, section SI3). Instead, the very smooth surface topography of the AlO-covered graphene with a rms surface roughness of <1 nm and barely visible G/Cu surface features indicates that the troughs are covered by AlO more than the ridges. This strongly suggests the occurrence of H2O/TMA condensation when ALD is performed at 80 °C. The use of noncondensing O3 replacing H2O as the oxidant at a Tdep of 80 °C is shown in Figure b. In contrast to the nucleation obtained using H2O at a Tdep of 80 °C, that using O3 yields a moderately preferential nucleation on the ridges with θ of ∼65%. The correlation between θ and Tdep in CM is shown in Figure d. In general, a relatively constant θ at an average value of 79–82% can be achieved in CM with Tdep of 120–180 °C using H2O/TMA. A condensing condition occurs at a Tdep of 80 °C, resulting in almost complete coverage of AlO. On the other side of the spectrum, a Tdep of 200 °C is always observed to yield the lowest AlO coverage with θ ∼ 43%, although PM (Figure b) can be employed to improve θ as discussed below.

Figure 3

(a) AlO nucleation by ALD in CM using H2O/TMA at Tdep of 150 and 80 °C. At a Tdep of 80 °C, the AlO coverage (θ) on the G/Cu surface is almost perfectly complete with a considerably smooth surface topography, suggesting condensation of the oxidant/precursor during the ALD process. (b) AlO nucleation by ALD in CM using O3/TMA at a Tdep of 80 °C. AlO nucleation by ALD in PM using (c) H2O/TMA with pretreatment times (tpretreat) of 60 and 300 min at a Tdep of 200 °C and using (d) O3/TMA with tpretreat of 2 and 15 min at a Tdep of 80 °C. The use of pretreatment significantly changes the selective nature of AlO nucleation into a more homogeneous nucleation. The green dotted lines in parts c and d indicate the boundaries between MLG and BLG, where the regions enclosed by the lines represent BLG. (e) Plot of θ on G/Cu by ALD in CM as a function of Tdep based on parts a and b. The red/blue arrows in part e indicate the improvement in θ when ALD is performed in PM, as observed in parts c and d, at the same Tdep. (f) Plot of θ on the G/Cu surface by ALD in PM as a function of tpretreat as observed in parts c and d, where a tpretreat of 0 min corresponds to CM. All scale bars in parts a–d represent 500 nm, and the red parallel lines indicate the ridges of G/Cu surface features. The error bars in parts e and f indicate the standard deviation from the mean. The doses for H2O and TMA in both CM and PM are maintained at ∼0.14 Torr·s, while that for O3 is set at ∼28.65 Torr·s. All AlO depositions are performed with 12 ALD cycles total.

The fact that CM at a Tdep of 200 °C yields the lowest AlO coverage gives rise to the assumption that the graphene surface needs to be uniformly activated by surface modification to obtain a more homogeneous nucleation if a H2O/TMA combination is to be used at a high Tdep.[2,12,13,16,18] Because the use of an additional seed layer is undesirable because of its potential deleterious effect to the device functionality, surface modification is introduced in this study by the use of PM (Figure b), which is essentially an exposure to a series of H2O or O3 pulses for a certain period of pretreatment time (tpretreat), followed immediately by a AlO deposition that is similar to CM without breaking the vacuum. Aside from the additional pretreatment step, the deposition parameters in PM are set to be the same as those in the aforementioned CM, i.e., using a TMA/H2O dose of ∼0.14 Torr·s at a Tdep of 200 °C. Figure c shows that a substantial shift in the AlO nucleation behavior is observed when tpretreat is set at 60 min, with the nucleation is no longer preferential to the graphene ridges but rather distributed evenly between both ridges and troughs. The change in the nucleation behavior is more pronounced when tpretreat is prolonged further to 300 min, at which point the nucleation is significantly more homogeneous throughout the G/Cu surface. Similarly, the switch from CM to PM when O3 is used as the oxidant not only results in a significant improvement to the nucleation density but also completely changes the nucleation behavior, as shown in Figure d. With a tpretreat of just 2 min, the nucleation becomes completely nonpreferential and a highly conformal AlO layer is observed throughout the G/Cu surface. It is also important to note that the nucleation in the troughs is always found to be much more homogeneous than that on the ridges whenever O3 pretreatment is used. The correlation between θ and tpretreat in PM is shown in Figure f. In general, the use of PM improves the AlO coverage on G/Cu, where θ increases proportionally with an increase of tpretreat. While improvement is observed regardless of whether H2O vapor or O3 is used as the oxidant, the effect is much more pronounced for the latter for a short tpretreat. Using O3/TMA, a significant improvement in θ to ∼96% can be observed for a tpretreat of just 2 min, although a further increase of tpretreat to 15 min only increases θ slightly to ∼97%. In contrast, using H2O/TMA, a significant improvement in θ to ∼89% can only be observed when tpretreat is set to 300 min. Because all deposition parameters in PM are exactly the same as those in CM, the observed changes in the otherwise preferential AlO nucleation can all be attributed to the addition of the pretreatment step.

Previous literature has already highlighted that ALD AlO nucleation on MLG can be highly dependent on the underlying graphene support/substrate.[17] Here we observe that, for few-layer graphene, it is also dependent on the number of graphene layers. As shown in Figure d,f, the nucleation density in the troughs of MLG is considerably higher than that in the troughs of bilayer graphene (BLG). While the use of PM, either with H2O or O3, results in a more homogeneous AlO nucleation on MLG, the nucleation on BLG is still highly selective. When PM is performed using H2O with a tpretreat of 60 min, AlO shows very poor nucleation in the troughs of BLG, resulting in an extremely low θ of ∼33%, approximately half that on MLG (Figure f). A significant improvement to the AlO nucleation on BLG can be achieved by extending tpretreat to 300 min. This results not only in an increase of θ on BLG to ∼79% (Figure f) but also in a shift of the AlO nucleation behavior to a more homogeneous nucleation on both ridges and troughs. While a higher θ can be, in general, achieved using O3 with a tpretreat of 2 min, the AlO nucleation density in the troughs of BLG is still much lower than that in the troughs of MLG. Note that the nucleation behavior on the ridges is unaffected by the number of graphene layers, as observed from the constant nucleation density on the ridges across the MLG–BLG boundaries.

The failure of PM to achieve conformal AlO nucleation on G/Cu using H2O/TMA at a Tdep of 200 °C motivates us here to investigate in more detail the limiting parameters at such a high Tdep. Because ALD depends heavily on surface saturation to achieve the self-limiting reactions, there is a possibility that the aforementioned selective AlO nucleation on graphene is due to unsatisfied saturation conditions, and it is unclear in the literature whether these conditions are always satisfied. Thus, we explore the use of higher H2O/TMA doses than the commonly used dose, with the aim of achieving surface saturation to obtain conformal AlO nucleation on graphene. The improvement in the nucleation density under CM at a Tdep of 200 °C due to the use of higher H2O/TMA doses is shown in Figure a. While increasing the H2O/TMA dose from ∼0.14 to ∼0.3 and ∼0.56 Torr·s substantially increases the nucleation density in the troughs, the nucleation behavior itself is relatively unaltered, i.e., is still highly preferential to the ridges, suggesting that the nucleation behavior cannot be easily altered by exclusively changing the H2O/TMA dose. Note that the AlO nucleation on the troughs at such a higher dose always results in a crisscrossed pattern. A transition in the nucleation behavior toward nonpreferential nucleation can be observed once the H2O/TMA dose is increased further to ∼1.31 Torr·s, and consequently an even higher H2O/TMA dose of ∼2.1 Torr·s results in a conformal nucleation of AlO. This finding suggests that conformal nucleation on graphene at high Tdep is attainable if the H2O/TMA dose is sufficient to achieve surface saturation.

Figure 4

(a) AlO nucleation by ALD in CM at a Tdep of 200 °C using increasing doses of H2O/TMA. Although the nucleation is still highly preferential to the ridges, an increase in the H2O/TMA dose significantly improves the AlO nucleation especially on the troughs of G/Cu, which leads to a higher θ. Full AlO coverage is obtained using a H2O/TMA dose of ∼2.1 Torr·s (Pdos = ∼1.05 Torr; tdos = ∼2 s). The typical H2O/TMA dose used in Figures and 3 is ∼0.14 Torr·s (Pdos = ∼0.2 Torr; tdos = ∼0.7 s). A homogeneous AlO nucleation on G/Cu using H2O/TMA at a Tdep of 200 °C can also be achieved by performing ALD either in MM (b) or in SM (c). Under either one of these modes, the H2O/TMA residence time (tdos) could be extended to reach complete AlO coverage without necessarily increasing the H2O/TMA dose pressure (Pdos). Full AlO coverage can be observed in part b when the H2O/TMA dose is at ∼1.65 Torr·s (Pdos = ∼0.55 Torr; tdos = ∼3 s) in MM and in part c at ∼0.7 Torr·s (Pdos = ∼0.2 Torr; tdos = ∼3.5 s) in SM. The AlO surface topography in part c is similar to that of as-grown G/Cu, suggesting a conformal deposition. Plot of θ on the G/Cu surface by ALD in CM, MM, and SM as a function of Pdos (d) and tdos (e). In part d, the color of the marker indicates tdos, while in part e, it indicates Pdos. In general, the relationship between θ and tdos is linear, i.e., θ ∝ tdos, instead of the square root, θ ∝ tdos1/2, until a saturation is reached at tdos ≥ ∼3 s. In parts a–c, all scale bars represent 500 nm and the red parallel lines indicate the ridges of G/Cu surface features, and the error bars in parts d and e indicate the standard deviation from the mean. All AlO depositions are performed with 12 ALD cycles total.

Given that the oxidant/precursor dose is essentially a product of the delivery pressure (Pdos) and residence time (tdos), a sufficiently high dose for conformal nucleation can be obtained by a higher Pdos and/or a longer tdos. Because it is not trivial to explore the effect of each parameter in CM because of the interdependence of Pdos and tdos, i.e., both are controlled by a single-parameter oxidant/precursor pulse time (tpul), we introduce modifications to the ALD process, denoted herein as MM and SM, which allow us to decouple tdos from Pdos. In MM (Figure c), each H2O/TMA dose is delivered by a sequence of two consecutive pulses in quick succession such that tdos is now controlled by the interval time between pulses (tinterv) rather than by tpul. Thus, MM allows tdos to be extended to about twice as long as that in CM without changing Pdos. In SM (Figure d), the sample is soaked in a H2O/TMA dose for several seconds (thold) before being purged, allowing tdos to be controlled by thold rather than by tpul. Thus, SM allows tdos to be completely independent from Pdos and extended virtually indefinitely. The use of MM and SM ALD to obtain a conformal AlO nucleation on G/Cu at a Tdep of 200 °C is shown in Figure b,c. A completely nonpreferential nucleation can be easily obtained with a H2O/TMA dose of ∼1.12 Torr·s, and a further increase in the H2O/TMA dose to ∼1.65 Torr·s results in a highly homogeneous AlO nucleation with complete surface coverage. Similarly, a highly homogeneous nucleation can be achieved by performing ALD in SM with a H2O/TMA dose of just ∼0.7 Torr·s. The similarity in the surface topography between AlO deposited under SM and bare G/Cu suggests that the deposition is highly conformal.

The correlation between θ and Pdos for CM, MM, and SM is shown in Figure d, while the correlation between θ and tdos is shown in Figure e. Although the relationship between θ and Pdos is observed to be approximately linear for just CM, because an increase in the dose from ∼0.14 Torr·s (typical dose) to ∼2.1 Torr·s results in an increase of θ from ∼44% to ∼99%, the overall correlation becomes extremely poor once the nucleation under MM and SM is taken into account. In contrast, a strong linear correlation between θ and tdos can be observed for all ALD modes because a higher tdos results in a higher θ until saturation is achieved at tdos ≥ ∼2 s. It is important to note that a conformal AlO nucleation is obtained with just 12 ALD cycles in MM and SM with a H2O/TMA dose of <1.3 Torr·s, whereas the same dose in CM results in a nucleation behavior that is still preferential with a θ of only ∼82%.

Here we also explore the use of SM to achieve conformal AlO nucleation at a Tdep of 200 °C on HOPG, G/Ge, and G/SiO2. These graphitic surfaces are known to be much less wettable by H2O than G/Cu.[17,24] Comparisons in the nucleation behavior between CM and SM at the same Tdep on these surfaces are shown in Figure a–c. AlO nucleation on HOPG under CM at a Tdep of 200 °C and a H2O/TMA dose of ∼0.3 Torr·s (Pdos = ∼0.3 Torr; tdos = ∼1 s) results in incomplete surface coverage with a relatively low θ of ∼68%. Despite the low θ, the nucleation on HOPG appears to be random and nonselective to only specific sites (Figure a). On the other hand, when CM is performed on G/Ge and G/SiO2 under the same conditions, AlO nucleates selectively on specific, more highly reactive locations, resulting in an extremely low θ of just ∼47% (Figure b) and ∼38% (Figure c), respectively. Although it is more spatially irregular than that on G/Cu, AlO nucleation on G/SiO2 is observed to be highly selective to the randomly oriented graphene folding and defect sites (Figure c). Currently, the most common transfer method used leads to the removal of uniaxial surface features that occur ubiquitously on G/Cu but at the expense of introducing new reactive sites, including folding sites, defects, and contamination, to the graphene. As a result, AlO appears to nucleate preferentially on these newly introduced reactive sites. Similarly, the absence of graphene wrinkles and folding sites on G/Ge suggests that the nucleation is now preferential to domain boundaries and defect sites (Figure b). On the other hand, the nucleation under SM at a Tdep of 200 °C and a H2O/TMA dose of ∼0.7 Torr·s (Pdos = ∼0.2 Torr; tdos = ∼3.5 s) is much more homogeneous across the entire surface, resulting in AlO coverage with θ > 97% on all samples (Figure a–c). Such a homogeneous nucleation allows the formation of sub-2-nm thin continuous AlO films, as measured by AFM (see also the Supporting Information, section SI4), with just 12 ALD cycles. In terms of the dielectric quality, these continuous AlO films exhibit capacitance values of 1.6 and 0.7 μF/cm2 and leakage currents of lower than 1 nA at 0.7 and 2.2 V when ALD is performed in SM with a H2O/TMA dose of ∼0.7 Torr·s for 20 and 50 ALD cycles, respectively (see also the Supporting Information, section SI5). The agreement between these values and those of AlO formation on graphene found in the literature strongly suggests that the AlO films deposited under SM are indeed continuous and have the potential to act as an efficient high-κ dielectric in graphene electronics with EOT < 1.3 nm.[23,36] The fact that the difference between CM and SM used here is only in tdos, i.e., tdos in SM, more than 3 times as long as that in CM accentuates the importance of a longer tdos for obtaining homogeneous AlO nucleation.

Figure 5

AlO nucleation on HOPG (a), G/Ge (b), and G/SiO2 (c) by ALD in CM at a Tdep of 200 °C using a H2O/TMA dose of ∼0.3 Torr·s (Pdos = ∼0.3 Torr; tdos = ∼1 s) for 48 cycles total and under SM at a Tdep of 200 °C using a H2O/TMA dose of ∼0.8 Torr·s (Pdos = ∼0.2 Torr; tdos = ∼4 s) for 12 cycles total. The use of CM yields a relatively low surface coverage of ∼57% on HOPG (a), ∼47% on G/Ge (b), and ∼38% on G/SiO2 (c). In contrast to the nucleation behavior on HOPG, which is relatively nonpreferential, that on G/Ge and G/SiO2 is preferential to the more active locations, e.g., domain boundaries, folding sites, and contaminations, introduced by the transfer process. The use of SM results in an almost perfectly conformal AlO nucleation with surface coverage of >97% on all samples. All scale bars in parts a–c represent 500 nm. (d) Raman spectroscopy analysis of G/SiO2 samples before and after ALD using a photon excitation of 532 nm. The analysis is represented by a plot of the 2D and G peak intensity ratio (I2D/IG) against the D and G peak intensity ratio (ID/IG), a plot of the 2D peak position (ω2D) against the G peak position (ωG) including an indication of the relative strain and doping contributions, and a plot of the 2D peak line width (Γ2D) against the G peak line width (ΓG).

Figure d shows the effect of AlO film deposition on graphene analyzed by Raman spectroscopy on G/SiO2 prior and subsequent to ALD using 532 nm excitation (see also the Supporting Information, section SI6, for individual representative Raman spectra). The peak intensity ratio of the 2D and G bands (I2D/IG) is found at ∼2.88 for as-transferred G/SiO2 and shifts toward a higher value of ∼3.39 after AlO deposition (AlO/G/SiO2) for both CM and SM. Note that here CM is performed using a H2O/TMA dose of 0.3 Torr·s at a Tdep of 80 °C, while SM is performed using a H2O/TMA dose of 0.7 Torr·s at a Tdep of 200 °C, and both yield almost complete AlO coverage with θ > 98% on G/SiO2. The Raman peak intensity ratio between the D and G bands (ID/IG) is ∼0.04 for the as-transferred G/SiO2 samples and remains the same for AlO/G/SiO2 regardless of the ALD mode used. For the as-transferred G/SiO2, the peak frequencies of the 2D (ω2D) and G (ωG) bands are found at ∼2679 and ∼1588 cm–1, respectively, with a ω2D/ωG slope of ∼0.7. When ALD is performed in CM, ω2D and ωG are found at ∼2677 and ∼1585 cm–1, respectively, while when ALD is performed in SM, they are found at ∼2676 and ∼1584 cm–1, respectively. Note that the ω2D/ωG slope shifts to ∼2.2 for AlO/G/SiO2 regardless of the ALD mode used. The line widths of the 2D (Γ2D) and G (ΓG) bands are found at 29.5 (±5.3) and 12.8 (±1.5) cm–1, respectively, for the as-transferred G/SiO2 and shift toward higher values after AlO deposition. When ALD is performed in CM, Γ2D and ΓG are found to be broadened to 32.3 (±6.9) and 16.4 (±1.8) cm–1, respectively, while when ALD is performed in SM, they are further broadened to 33.1 (±7.7) and 17.2 (±2.1) cm–1, respectively. The Γ2D/ ΓG slope is ∼2.2 for all G/SiO2 samples, with or without ALD AlO.

Discussion

Our data show that the deposition of AlO on G/Cu using a typical ALD process, i.e., CM at a Tdep of 200 °C and a TMA/H2O dose of ∼0.14 Torr·s, is strongly affected by the presence of uniaxial G/Cu surface features, where the ridges form preferential AlO nucleation sites. These ridges are the topographically highest points on the G/Cu surface, making them more readily available sites for adsorption of the oxidant/precursor. More importantly, the high curvature of the ridges is known to present the most active sites on supported graphene because of the high strain in the C–C bonds.[27,37,38] Similar to the nucleation on line defects and step edges, the nucleation on these ridges has long been thought to be energetically preferable to the release strain and ultimately relaxes the graphene.[34,37] AlO deposition in the troughs themselves does not take place within the first few ALD cycles but rather starts to occur several tens of cycles later once the ridges, i.e., the most reactive sites, have been fully occupied and passivated by AlO clusters (Figure ).[14,34] The highly selective nucleation behavior at such a high Tdep has led the hitherto conclusion in the literature that conformal AlO nucleation on a graphitic surface using the standard H2O/TMA precursor is notoriously difficult to achieve, and thus a lower Tdep or a surface modification that promotes uniform wetting is required.[13−17,19,20,34,39]

In terms of Tdep, it is widely known that an ideal ALD process can only occur in a very specific Tdep window.[39] A higher Tdep provides sufficient thermal energy to drive the surface reaction to reach completion, although it may also lead to a higher desorption rate of oxidants/precursors from the G/Cu surface, which results in a highly selective nucleation to only the reactive sites with lower θ. On the other hand, a lower Tdep often results in not only incomplete oxidant/precursor reactions[33,35] but also the condensation of oxidants/precursors across the sample.[39] As measured by spectroscopic ellipsometry (see the Supporting Information, section SI7), the refractive index of AlO films deposited at a Tdep of 80 °C is consistently lower, albeit only slightly, than that deposited at 200 °C, suggesting that a lower Tdep results in a lower density in the AlO films.[16,40−42] In addition, the lower desorption rate at lower Tdep corresponds to a longer ALD process time because of a longer purge time needed between pulses.[40] Our data show that, in general, θ increases with a decrease of Tdep, where Tdep of 120–180 °C yield an average θ of 79–75% and a Tdep of 80 °C yields almost complete coverage with θ ∼ 98% (Figure a). Thus, a lower Tdep is definitely favorable if the goal is to alter the AlO deposition behavior so that deposition occurs everywhere across the G/Cu surface.[16,17] However, the fact that the resulting AlO layer is topographically very flat yet porous implies that the deposition is far from the ideal conformal deposition and is instead due to H2O condensation that takes place mostly in the troughs. The presence of H2O condensation at 80 °C can be confirmed by replacing it with O3 because O3 will still be gaseous and not condense at this temperature (Figure a). In contrast to the AlO nucleation using H2O/TMA, ALD with O3/TMA at the same Tdep results in a much lower nucleation density with a θ of only ∼76% (Figure c,d). The absence of condensation is implied by the similarity in the AlO nucleation behavior between O3/TMA at a Tdep of 80 °C and H2O/TMA at a higher Tdep, i.e., preferential nucleation on the ridges. This implies that, as long as the noncondensing conditions are satisfied at low Tdep, the AlO nucleation behavior on G/Cu under CM will always be selective to the most active sites, i.e., the ridges.

A modification to the graphitic surface is often introduced to make it more wettable, either by adding seed layers and functional groups, e.g., Al and PTCA,[2,13] or by using a more reactive oxidant, e.g., O3 and NO2.[14,15,18] We here introduce a surface modification to the G/Cu surface by performing ALD in PM to avoid the use of an undesirable additional seed layer and without the need to use a lower Tdep. When PM is performed using H2O/TMA, it has been suggested that H2O molecules are physically adsorbed onto the graphene surface by van der Waals forces during the pretreatment, which then act as nucleation sites for the subsequent ALD process.[16,43] A higher density of nucleation sites can be, in principle, achieved with a longer tpretreat because it leads to a higher concentration of adsorbed H2O molecules on the G/Cu surface. However, the intermolecular attraction between the H2O molecules may become increasingly dominant and exceed the van der Waals forces, resulting in island-like nucleation sites (Figure d).[16] Our data indeed show that, at a Tdep of 200 °C, θ increases significantly with an increase of tpretreat, despite the fact that the entire process becomes prohibitively long, taking about 300 min of pretreatment to reach a θ of ∼89% (Figure f). An even more effective surface modification can be introduced by performing PM using O3. Because of its reactivity, O3 is commonly used to modify the graphene surface, either by cleaning the graphene surface or by functionalizing it with epoxide groups,[14,15,44−46] to ultimately change the nucleation behavior into a highly homogeneous one.[14,17,47] Indeed, a relatively short tpretreat of 2 min is sufficient to completely alter the AlO nucleation behavior completely nonselective (Figure f). Nevertheless, the use of O3/TMA is less desirable because O3 is known to have a detrimental effect on graphene, especially at a high Tdep.[15] To minimize damage to the graphene, Tdep is always set at 80 °C whenever O3/TMA is used in this study. Nevertheless, even at such a low Tdep, the detrimental effects of O3 to the graphene structure could still be observed (see also the Supporting Information, section SI8). Therefore, a prolonged O3 pretreatment of more than 2 min should be avoided because it not only does not significantly improve the AlO nucleation density but also damages the graphene. In addition, the imposed upper Tdep limit often results in a higher carbon concentration in the deposited AlO layer due to incomplete decomposition of the formate or other carboxylate species,[48] which ultimately leads to a lower AlO density (see also the Supporting Information, section SI7).

While the use of PM allows a much more homogeneous AlO nucleation to be attained on monolayer G/Cu (MLG), it struggles to achieve the same nucleation density on bilayer G/Cu (BLG). Our data show that while AlO nucleation on the ridges of the BLG is very similar to that on the ridges of the MLG, the nucleation density in the troughs of BLG is significantly lower than that of MLG. Interestingly, this behavior is always observed whether H2O or O3 is used as the oxidant, and although our observation is limited to only MLG and BLG, it suggests that AlO always nucleates preferentially on the ridges regardless of the number of graphene layers. The big difference in terms of the nucleation density in the troughs may originate from the difference in polarity between MLG and BLG, where a higher number of graphene layers corresponds to a lower surface polarity.[17,24,49] It is important to note that the effect of the number of graphene layers is stronger when O3 is used as the oxidant rather than when H2O is used, although the difference between θ of MLG and BLG can be minimized by increasing tpretreat. As shown by our data, such a difference can be minimized to <10% after 300 min of pretreatment using H2O and to <30% after 15 min of pretreatment using O3.

As in any gas-adsorption processes, the ALD process is known to be limited by the total amount of oxidant/precursor available for the reaction, quantified by the delivery pressure (Pdos), as well as their mass transport to the surface and the surface reaction kinetics, both quantified by the residence time (tdos).[21,22] Thus, we hypothesize that a conformal AlO deposition can be, in principle, obtained using H2O/TMA at a Tdep of 200 °C by increasing Pdos to compensate for a high desorption rate from the surface and/or by extending tdos to account for mass transport onto the imperfectly flat surface and slow adsorption kinetics of the relatively nonreactive graphitic surface. Our data indeed show that a higher H2O/TMA dose in CM always results in a higher AlO nucleation density, especially on the troughs, as reflected by an increase in θ from ∼44% to ∼82% when the dose is increased by an order of magnitude from ∼0.14 to ∼1.31 Torr·s (Figure d). Despite the significant increase in the nucleation density on the troughs due to the use of a remarkably high H2O/TMA dose, the nucleation behavior remains largely the same, i.e., preferential nucleation on the ridges. It is also important to note that the AlO nucleation in the troughs at a higher dose always results in a crisscrossed pattern (Figure a). While the origin of such a crisscrossed pattern is still unclear, we observe that one of the crisscrossed pattern axes is always aligned to the direction of the flow but independent of the direction of the graphene wrinkles and Cu surface reconstructions. This implies that the flow plays an important role in the nucleation behavior and may strongly affect oxidant/precursor mass transport to the G/Cu surface. While an increase in tpul in CM always yields a higher dose due to a simultaneous increase of both Pdos and tdos, care must be taken because the relationship between them is not linear and is highly dependent on secondary ALD parameters including the carrier gas flow rate and pumping speed.

The use of MM and SM allows us here to decouple tdos from Pdos such that a prolonged tdos could be achieved without necessarily increasing tpul, and consequently Pdos. Typically, a prolonged tdos is employed to obtain conformal deposition on a high-aspect-ratio structure because a longer tdos is required for the oxidant/precursor molecules to fully diffuse into the structures.[50] In fact, it has been estimated that the required tdos would be proportional to the square of the aspect ratio.[22] Given that the aspect ratio of G/Cu is much less than unity, we could argue that the diffusion of oxidant/precursor molecules onto the surface should not be a limiting factor. On the other hand, the long tdos may indeed be needed to account for the slow adsorption kinetics due to the inertness of the graphene surface. Our data show that, for the same Pdos, a longer ,tdos results in a higher θ, while for the same ,tdos, a higher ,Pdos does not necessarily result in a higher θ. In fact, when all data from CM, MM, and SM are combined, θ can only be correlated to tdos but not to Pdos. A strong correlation between θ and tdos is observed when tdos is less than a critical value of ∼2 s, with θ varying linearly with tdos, i.e., θ ∝ tdos, instead of with the square root of tdos, i.e., θ ∝ tdos1/2, suggesting that the ALD AlO on G/Cu is surface-reaction-limited instead of diffusion-limited (Figure e).[50] On the other hand, a saturation is reached, i.e., θ ≈ 100%, when tdos ≥ ∼2 s regardless of the ALD mode used. In addition, the use of SM using H2O/TMA with a tdos of ∼3.5 s allows a much more homogeneous nucleation with θ > 97% to be obtained with just 12 ALD cycles on HOPG, G/Ge, and G/SiO2 (Figure a–c), negating the difficulties in introducing conformal nucleation on the notoriously difficult-to-wet graphitic surfaces. It is important to note that the value of critical tdos may be different from one ALD system to another. It is also worth mentioning that the supporting substrates by themselves, e.g., bare Cu or SiO2 without graphene, are not difficult-to-wet surfaces, and thus homogeneous AlO nucleation could be consistently obtained with the typical parameters in CM (see also the Supporting Information, section SI9). While a conformal nucleation on these graphitic surfaces could still possibly be obtained by CM, a prohibitively high amount of H2O/TMA would probably be required. This finding strongly suggests that the tdos of H2O/TMA needed to obtain conformal nucleation at a Tdep of 200 °C on graphitic surfaces is not excessively long.[50] More importantly, this confirms our hypothesis that tdos is the key parameter to account for the slow adsorption kinetics of H2O/TMA on the relatively nonreactive graphitic surfaces; as such, the use of a lower Tdep and the introduction of a surface modification are not a necessity for conformal AlO nucleation.

Raman analysis of G/SiO2 before and after ALD AlO shows that the ALD process, in either CM or SM, does not introduce additional damage to the graphene structure, as reflected from their identical ID/IG ratios. Thus, unlike the use of O3 as the oxidant,[15] the use of H2O is relatively harmless for the graphene for a range of Tdep values from 80 to 200 °C. We also show here that tdos could be extended by up to 3.5 s in SM without introducing a detrimental effect to the graphene even at a high Tdep. Nevertheless, care must be taken when an extremely long tdos is used because TMA is highly reactive and may result in the undesirable formation of defects on the graphene (see also the Supporting Information, section SI10).

Although nucleation on the ridges has long been thought to be energetically preferable to release the strains and ultimately relax the graphene,[34,37] the effect of AlO nucleation on the mechanical strain is observed to be much less pronounced compared to its effect on charge doping of the graphene. The decrease in ω2D and ωG modes toward lower wavenumbers indicates a decrease in the graphene doping level from ∼3 × 1012 to ∼1012 cm–2 when AlO is introduced under CM at 80 °C, while the mechanical strain level remains similar in magnitude between −0.1 and −0.2% (Figure d).[51,52] On the other hand, when AlO is deposited under SM at 200 °C, the doping level decreases further to <1012 cm–2 and the mechanical strain level decreases slightly to between −0.05 and −0.15%, although the broadening in Γ2D and ΓG indicates that the variation in the nanometer-scale strain is actually increased (Figure d).[53] It has been known that the presence of hydroxyl species on the SiO2 surface induces the formation of charge trap sites that contribute to the doping level and the buckling behavior of G/SiO2. During ALD, the concentration of hydroxyl species on the SiO2 surface is strongly reduced because of induced desorption by thermal treatments.[51,54] In addition, surface saturation by H2O during ALD drives the O2/H2O redox reaction on SiO2 toward H+, which results in the depletion of reactive hydroxyl and peroxide species and leads to the further removal of charge trap sites.[54,55] Thus, the difference in the doping and mechanical strain levels between CM and SM may actually be attributed to the difference in Tdep, where a higher Tdep leads to a higher removal rate of charge trap sites and thus results in lower doping and strain levels. Note that the level of doping and mechanical strain of graphene is strongly influenced by its substrate. Thus, the changes in the doping and strain levels observed here may occur differently if the graphene is supported by substrates other than SiO2. Nevertheless, this strongly suggests that the 12 ALD cycles in SM at 200 °C is a sufficient condition not only for obtaining a homogeneous AlO film but also for decreasing the doping and mechanical strain levels of G/SiO2. As mentioned earlier, the ability to homogeneously deposit ultrathin oxide films on graphene is considered critical for device integration because, for instance, it allows a strong current saturation and a significant gain in voltage and transconductance in high-frequency graphene devices.[23] While we show that a conformal deposition on graphene is possible, its use as a barrier is yet to be investigated and its quality in terms of, for instance, the leakage current, capacitance, or gas permeation remains to be thoroughly quantified. Nevertheless, future work related to ALD on graphitic surfaces should consider extending the residence time if a conformal nucleation is to be achieved.

Conclusions

Our results show that ALD of AlO directly on graphene using the standard H2O/TMA precursors results in nucleation behavior that can be either highly selective or completely homogeneous across the entire surface depending on the deposition conditions. When ALD is performed in CM under a wide range of deposition temperatures, the deposition is highly preferential to the most active sites, i.e., ridges of the graphene wrinkles and Cu surface reconstructions, as long as a noncondensing condition is satisfied. For a condensing condition, the nucleation results in a continuous yet porous AlO film with complete coverage of the surface. A more homogeneous AlO nucleation can be achieved without relying on H2O/TMA condensation by performing ALD in PM, which exposes the graphene surface to H2O prior to the actual ALD process. At a typical deposition temperature of 200 °C, the use of PM allows for a more homogeneous nucleation behavior because the nucleation density in the troughs increases proportionally with an increase of the pretreatment time. Nevertheless, this is not a necessary condition because the key to obtaining a conformal nucleation lies in the H2O/TMA residence time because an extended residence time is needed to account for the slow adsorption kinetics of the relatively inert graphene surface. Here a prolonged residence time is introduced by optimization to the ALD pulse sequence and a soaking period, in the form of MM and SM, respectively. Regardless of the method used, be it CM, MM, or SM, when ALD is performed at 200 °C, there exists a critical residence time below which the nucleation is selective and above which it is much more, if not completely, homogeneous across the entire graphene surface. By extending the precursor residence time, we are able to overcome the otherwise heterogeneous nucleation such that sub-2-nm thin continuous AlO films can be achieved directly on graphene using standard H2O/TMA precursors even at a high Tdep of 200 °C. Because these results could be generally extended to ALD of any other oxides, particularly if homogeneous deposition is required, the work presented here should be considered as a model system for rational 2D/non-2D material process integration, which is relevant to the interfacing and device integration of other emerging 2D materials, including hBN and transition-metal dichalcogenides, and many other difficult-to-wet materials.