Long-range correlated dynamics in ultra-thin molecular glass films

Nanometer-sized thin films of small organic molecules are widely used in applications ranging from organic photovoltaics¹ and light emitting diodes^2,3 to protective coatings⁴ and high resolution nano-imprint lithography.⁵ It is advantageous to use amorphous films because compared to crystals, they do not have grain boundaries to hinder charge transport, generate cracks and defects, or disrupt the writing processes. Physical vapor deposition (PVD), the common method used to manufacture these films, is usually performed at substrate temperatures below T_g to produce films in the glassy state. However, if the properties at nanoscale deviate significantly from bulk properties, the resulting films can have reduced kinetic and thermal stability. Recent experiments suggest that diffusion at the free surface of organic glasses can be several orders of magnitude faster,^6,7 with weaker temperature dependence compared to bulk diffusion.

The free surface has been shown to result in enhanced and weakly temperature-dependent dynamics on the surface of polymeric glasses^8,9 and significantly affects the properties of ultra-thin polymer films.^9–17 In polymeric systems, the molecular weight of the polymer¹⁴ and the temperature range of the measurement^8,9,14 seem to also affect the observed properties, resulting in ambiguity in the relationship between enhanced dynamics at the free surface and properties of ultra-thin glass films. As such, these results cannot be easily extrapolated to molecular and atomic glass systems. Direct measurements of dynamics in ultra-thin films of molecular glasses can help resolve some of the outstanding questions about enhanced dynamics in polymeric glass systems.

Thick PVD films have been shown to form exceptionally stable glasses upon deposition at temperatures just below T_g.^18–21 While the detailed mechanisms of the formation of stable PVD glasses are still under investigation, most studies^18,19,22,23 indicate that surface-mediated equilibration (SME) is critical to their production. As such, understanding the length scales of mobility gradients can help elucidate the mechanisms of stable glass formation. Such studies may also help predict the performance of thin film devices, which can in turn help design better thin glassy layers for applications.

A systematic study of the dynamics of ultra-thin organic glasses over a wide thickness and temperature range can also help estimate length scales of mobility gradients which can be compared with fundamental length scales of glass transitions as proposed by various theories.^24–27 There are very few studies that measure the dynamics of ultra-thin films of organic glasses with thicknesses less than 100 nm.^28,29 There is some evidence suggesting that the properties of these films may be strongly thickness-dependent.³⁰ Many other experimental efforts have focused on direct characterization of the dynamical heterogeneity and the length scales of the cooperative motions in glassy systems, most of which were performed near or above T_g. In these studies, the sizes of the correlated domains were reported to be a few nanometers.^31–37 In addition to the direct characterizations of the cooperative rearranging regions, these length scales have also been probed by investigating the interfacial effects on glass properties under confinement. Measurements of glass transition temperature of organic liquids in small pores suggest modified dynamics in pores that are a few nanometers in diameter,^38,39 while measurements in thin polymer films widely vary depending on experimental techniques as well as the temperature ranges performed and they measure dynamical variations in thicknesses that range from a few nanometers to several tens nanometers.^{9,12,16,17,38–41} One challenge in understanding these results of polymer thin film studies in the context of correlation lengths of glass transition is the effect of molecular weight on the experimental results.^14,42

In this article, we use dewetting kinetics and cooling-rate dependent T_g (CR-T_g) experiments to measure the dynamics of ultra-thin films of the molecular organic glass, N,N′-Bis(3-methylphenyl)-N,N′-diphenylbenzidine (TPD). While a direct measure of absolute viscosity of thin films cannot be obtained due to potential mobility gradients induced by the free interface, by relating the characteristic dewetting times with CR-T_g measurements,^14,43 we are able to measure the “effective viscosity” of ultra-thin films as function of film thickness and temperature. In the absence of gradients in the dynamics, the effective viscosity equals the film viscosity. We show that ultra-thin films remain mobile far below bulk T_g, and the apparent activation energy for rearrangements decreases sharply for film thicknesses below 30 nm. The sharp decrease in the activation energy indicates that the dynamics in films thinner than 30 nm are strongly correlated and enhanced due to the enhanced surface dynamics.

Small organic molecule N,N′-Bis(3-methylphenyl)-N,N′-diphenylbenzidine (TPD) was purchased from Sigma-Aldrich and used without further purification. Bulk differential scanning calorimetry (DSC) and rheology characterization can be found in the supplementary material, Figure S1 for DSC and ellipsometry, Figure S2 for viscosity characterization. All films in this study were prepared using physical vapor deposition (PVD) in a custom built high vacuum (HV) chamber with a base pressure of ∼10⁻⁷ torr. The details of the chamber are described elsewhere.²¹ Silicon (one side polished, 100 plates from Virginia Semiconductor) with 1–2 nm native oxide layer was used as substrates for all films. For films used for dewetting studies, the substrate temperature during deposition was held at bulk T_g to ensure that the film was at equilibrium during deposition and it did not form a stable glass. A deposition rate of 0.02 nm/s was used for all films. An in-situ quartz crystal micro-balance (QCM) was used to monitor film thickness during deposition. After deposition of a desired average film thickness, the substrate temperature was brought back to room temperature before the sample was removed from the chamber. The total time between the end of deposition to the first atomic force microscopy (AFM) measurement was typically 10-15 min. The thickness of films was also verified by measuring the depth of holes formed during dewetting using atomic force microscopy (Agilent 5420).

For ultra-thin films (h = 8–30 nm) dewetting experiments were performed under isothermal conditions using a custom thermoelectric heating-stage equipped with a thermoelectric module (Custom Thermoelectric TEC) and a thermistor (Oven industries TR91-170) to measure the value of the temperature, along with a DC power supply to control the temperature (Mastech HY3010E). The morphology of the films was simultaneously imaged using AFM. The temperature of the heating-stage was calibrated with melting-point standards (Sigma-Aldrich), the temperature fluctuations were within ±2 K throughout the experiments for the temperature range used. Non-contact AFM tips (Budget Sensors, Tap-300G, resonance frequency 300 kHz, tip radius of curvature <10 nm, force constant 40 N/m) were used for these measurements. To ensure the isothermal conditions, the stage was preheated and equilibrated. As-deposited films were transferred onto the preheated AFM stage immediately after removal from the chamber and were scanned continuously. Dewetting experiments for thick films (h ≥ 100 nm) were performed using optical microscopy (OM) (Olympus BV51) with an in-situ heating stage (Linkam THMS 350 V). Image analysis softwares Gwyddion⁴⁴ and ImageJ⁴⁵ were used to calculate the surface coverage of each sample at various points of time for further analysis. For AFM images a height threshold was set to evaluate the surface coverage, while for OM images an intensity threshold was set for surface coverage evaluations. Optical microscopy measurements were performed with and without nitrogen purge as a control and no significant difference in the dewetting dynamics were observed. All the AFM measurements were performed at atmospheric conditions in an isolation box. No evidence of the atmosphere effect, moisture sorption, or any sign of oxidation of the material was observed at the temperature ranges applied. All measurements were performed above room temperature to avoid water condensation on the surface. Water condensation is typically observed in AFM imaging as increased noise, and no evidence of this was found during imaging above room temperature.

Cooling-rate dependent T_g (CR-T_g) measurements were performed using spectroscopic ellipsometry (M-2000 V J. A. Woollam). Details of the technique can be found in our earlier publications.^14,44,46 For CR-T_g measurements, films were deposited at a substrate temperature of 0.85T_g (279 K) to ensure that the substrate was fully covered with no holes. These as-deposited films were then heated above T_g to transform into ordinary glass for CR-T_g measurements as shown in Figure S9. The samples were then cooled at various cooling rates between 1 K/min and 150 K/min and heated at 150 K/min to a set temperature of 348 K as shown in Figure S10. An isotropic Cauchy model was used to fit raw ellipsometry data as detailed in the supplementary material and previous publications.^14,44,46 AFM imaging was performed before and after measurements (Figure S11) to ensure that the films did not dewet during the process. All ellipsometry measurements were carried out under dry nitrogen purge conditions and at temperatures above room temperature to prevent water condensation on the surface.

Thin films of TPD were prepared by PVD under high vacuum (HV) conditions; HV conditions ensure that the silicon substrates have uniform interfacial energies, so that dewetting is caused only by homogeneous nucleation and hole growth induced by thermal capillary and density fluctuations (Figure S3). Figure 1 shows the root mean square (RMS) roughness of as-deposited PVD films produced at a deposition rate of 0.02 nm/s, while the substrate temperature was held at bulk T_g. The insets show representative AFM images of the observed as-deposited morphologies at various film thicknesses. Measurements were typically performed within 15 min after deposition.

Figure 1 shows that during vapor deposition at T_g, ultra-thin films become rough and have morphology of the same height scale as the film thickness. The film morphologies at thicknesses below 12 nm resemble semi-continuous morphologies typically observed in spinodal dewetting experiments,⁴⁷ with similar spectral distribution functions. A uniform layer starts forming at thicknesses above 20 nm, and the film gradually becomes smoother. For films thicker than 30 nm, the morphology flattens with both time and film thickness. This evolution in the morphology implies that during the deposition there is significant reconfiguration and motion of the molecules due to interfacial interactions, which allow for the formation of spinodal-like features. As the film thickness is increased, surface diffusion becomes more prominent and interfacial tension acts to smoothen the film.

Because these films exhibit a spinodal morphology only minutes after deposition and before a complete film is ever formed, the deposition rate and the size of the spinodal features can be used to provide an estimate of the average diffusion of the molecules during PVD. Based on the spectral distribution shown in Figure S4, the average lateral size of these features in an 8 nm film deposited at T_g is 350 nm. Given the deposition rate of 0.02 nm/s, it takes 600 s to deposit this film. Thus, the average diffusion coefficient of the 8 nm film is in the order of 3 × 10⁻¹⁶ m²/s. As a point of comparison, the bulk diffusion coefficient for most organic molecules at T_g is about 10⁻²⁰ m²/s.⁴⁸ This simple estimation implies that the average dynamics in 8 nm films, measured at T_g, are several orders of magnitude faster than the bulk dynamics, even at a temperature where the bulk material is still at equilibrium. Similar morphological features have been observed in PVD films even when the substrate temperature during deposition was below T_g.⁴⁹ Since organic films used in applications are usually less than 100 nm thick, the morphologies caused by dewetting in thin films must be accounted for, as they may have a strong influence on the performance and stability of the device.

The rough structures of as-deposited ultra-thin films were used as templates for further isothermal dewetting experiments. AFM or optical microscopy (OM) showed that the film morphology continued to evolve with time (examples shown in Fig. 2 and Figure S5, and SI videos). Dewetting of ultra-thin films (h < 30 nm) progressed both through the growth of existing holes and bi-continuous features, as well as the nucleation and growth of new holes due to thermal capillary fluctuations. Isothermal dewetting was observed at temperatures as low as T_g − 32 K, where the bulk viscosity is not measurable and any reasonable extrapolation of the values of viscosity would predict a dewetting time longer than the age of the universe for a bulk film. As the holes continued to grow, material from the holes accumulated in rims, resulting in an increase in the local film thickness outside the holes, which slowed down and eventually stopped the process. In contrast, 100 nm films only dewetted well above T_g (T > T_g + 18 K), where the bulk viscosity is orders of magnitude lower.

Due to the strong apparent film thickness-dependence of the dynamics and non-uniform initial film morphologies, it is not possible to use models based on uniform viscosity and uniform film thickness⁵⁰ to model the kinetics of dewetting in these films. Furthermore, the pre-existing morphology can make the dewetting process appear faster, as the growth of existing holes in thin films is typically faster than the spontaneous nucleation of new holes in thick films. Besides, the substrate interaction potentials in these models^47,50 are poorly understood and slip condition at substrate interface is not explicitly included.

Despite these difficulties, the effective viscosity can be indirectly measured by investigating the temperature dependence of the characteristic dewetting time, τ_dewetting.⁵¹ This is because substrate interactions, surface tension, and the film’s initial morphology are all weak functions of temperature, leaving the film’s effective viscosity as the only temperature-dependent parameter driving dewetting (more details are in the supplementary material). As such, τ_dewetting should be proportional to the effective viscosity of the film. Since the viscosity in molecular glasses has been shown to follow the structural relaxation time (τ_α), τ_dewetting should also be proportional to τ_α.^52,53

Samples prepared under the same condition have the same initial dewetted area, A(0), which is dictated by the initial morphology of the as-deposited films. τ_dewetting was measured by tracking the time evolution of the total dewetted area, A(t), indicated by green color in Figure 3(a). Figure 3(b) shows the change of A(t) as a function of time for a 20 nm film at five different isothermal annealing temperatures. More examples are provided in Fig. S6. A single exponential function can be used to evaluate the characteristic time scale of dewetting,^51,54 A(t) = A(∞) + (A(0) − A(∞))exp(−t/τ_dewetting), where τ_dewetting is the time scale of dewetting. As seen in Figure 3(b), A(t) increases with time in a manner that depends both on the thickness of the film and the annealing temperature. At high enough temperatures, dewetting stops when the equilibrium contact angle has been reached. At lower annealing temperatures, the dewetting process can take longer than the experimental time. Furthermore, dewetting may slow down due to the fact that as the film continues to dewet, it thickens and its viscosity may increase dramatically. To ensure the accuracy of the fit for films held at low temperatures, A(∞) was defined as the free fit value from the data at the highest temperature for the same thickness. It is observed that for films of all thicknesses, within the experimental errors, a single exponential function fits the evolution of the dewetted area, shown as the green dashed line in Figure 3(c). In Figure 3(c), the data of various thicknesses and annealing temperatures collapsed nicely onto the universal curve y = 1 − exp(−t/τ) after scaling with τ_dewetting, indicating the validity of exponential description. We note that variations in choosing A(∞) which can have a large error due to variability of AFM images do not significantly affect τ_dewetting or the final calculated activation energies, as these values change more strongly with temperature than small errors in the fit values. It is also important to note that to the best of our ability to fit the data, all data follow a single exponential decay, and a stretched exponential fit to the data results in stretched exponents between 0.9 and 1 but does not improve fitting accuracy. More details can be found in the supplementary material and Figure S6.

If dynamics of thin films were identical to those of bulk, one would expect the temperature dependence of τ_dewetting to be the same regardless of film thickness, even if absolute dewetting times depend on the film thickness due to different initial morphologies and non-trivial thickness-dependence of the driving forces of dewetting. Figure 4 shows an Arrhenius plot of $−logτdewetting$ versus inverse temperature, 1/T. It is evident that in this temperature range, the slopes of the curves, which represent the apparent thermal activation barriers for rearrangement, E_a, have strong thickness dependence. For an 8 nm film τ_dewetting, and therefore the effective viscosity, changes less than half a decade over the temperature range of 298 K–321 K. τ_dewetting for 30 nm films shows a much stronger temperature dependence, changing more than two decades over the same temperature range. It is important to note that all of these temperatures are well below bulk T_g. The low apparent activation energy of ultra-thin films is consistent with previous studies of dynamics on polymeric thin films.^14,15,43

To extend the range of available viscosities of bulk films to lower temperatures and to determine the vertical shift factors for thin film dewetting data, cooling-rate dependent T_g (CR-T_g) measurements were also performed using ellipsometry. For most bulk organic glasses when the system is cooled at 10 K/min, the supercooled liquid falls out of equilibrium when the average structural relaxation time of the system is equal to τ_α ≃100 s or the viscosity reaches η ≃ 10¹⁰ Pa s. If the system is cooled more slowly, it falls out of equilibrium at a larger value of viscosity. As such cooling-rate CR is inversely proportional to both η and τ_α with measurements at 10 K/min used as a point of reference.

Figure 5(a) shows a typical graph of normalized thickness vs. temperature for a 100 nm (bulk) film for various cooling rates. T_g values for films of all thicknesses and at all cooling rates were evaluated as the intersection of two lines fits in the temperature range of 303 K to 314 K for the glassy region and 339 K to 348 K for the super-cooled liquid region. To ensure accuracy in thin film data fitting, common values of the apparent expansion coefficient were used for the glassy and supercooled liquid regions, obtained from fits to the data for the thickest film (more details are described in our earlier publications^14,46). These values were α_G = (2.10 ± 0.02) × 10⁻⁴ K⁻¹ and α_SCL = (6.57 ± 0.03) × 10⁻⁴ K⁻¹, respectively.

We note that for films with a thickness thinner than h < 25 nm, T_g assignments are subject to some error at low cooling rates, which may result in overestimating the value of T_g. As shown in Figure S12, thin films at low cooling rates show very broad transition, which has also been previously observed in polymeric thin films.¹⁴ As such, the measurements need to be extended to lower temperatures to observe complete transition into glassy state. Unfortunately, in our current setup we cannot monitor the full transition at lower temperatures due to possible water adsorption, despite nitrogen purge used to dry the samples. Thus large experimental error in the values of T_g is expected at these cooling rates. However, the measured T_g can be considered as an upper bound to the real value of the average T_g in these films, with discrepancy increasing with decreasing cooling rates. As a result, the values of the apparent activation energy from CR-T_g (the slope of lines in Figure 5(b)) are also measured to be larger than the actual values and these measurements provide an upper bound for the value of activation energy. These data points with large uncertainties are indicated by open symbols in Figure 5(b) and affect activation energies for films with thicknesses less than 25 nm.

Once the T_g is evaluated at all cooling rates, the data can be plotted as Arrhenius plots of log (cooling rate (CR)) vs. 1/T_g for various film thickness surveyed as shown in Figure 5(b). The CR-T_g for the bulk film of 100 nm can be directly compared with rheology experiments by the assumption that at the temperature where the system falls out of equilibrium (T_g) at a cooling rate of 10 K/min, the relaxation time for most organic glasses are measured to be τ_α = 100 s. As shown in Figure S13, this corresponds to a viscosity of η = 10¹⁰ Pa s for TPD, where the rheology data match dielectric relaxation measurements.

In order to relate τ_dewetting to viscosity, the vertical shift factors were obtained by comparing the data for bulk films (100 nm) with rheology measurements of bulk TPD as shown in Figure 6(a). CR-T_g measurements were used to extend the dynamical range of the bulk measurements (100 nm film) to temperatures close to bulk T_g. More details about the shift factors used can be found in Figures S13-S15.

Figure 6(a) shows the combined data of dewetting and CR-T_g measurements and provides a direct comparison between the bulk and the effective thin film viscosities. In the temperature range of these experiments, there is excellent agreement between the data obtained for 100 nm films and the viscosity and structural relaxation time measurements of bulk TPD.⁵⁵ This strongly indicates that first, 100 nm films behave bulk-like and their effective viscosity matches that of bulk viscosity in the entire dynamical range of these measurements for both types of experiments, and second, other related parameters for the dewetting process, such as substrate interaction energy and surface tension, did not have strong temperature dependences.

As detailed above, the initial morphology of films of various thicknesses is different. As such, the shift factor used to match the 100 nm dewetting data to bulk viscosity is not applicable to other films. Instead, for films of 25-30 nm, the CR-T_g experiments were used to calculate the appropriate shift factors (Figure S13). It is important to note that CR-T_g could not be reliably used for ultra-thin films (h ≤ 20 nm) due to early onset of dewetting. As such, the exact shift factors for these data sets are unknown and the data presented here for these films only reflect the temperature dependence of the effective viscosity and not their exact values. However, based on the simple analysis of diffusion presented above, the effective viscosity is at least about a factor of four faster than that of bulk at T_g, which is consistent with data presented in Figure 6(a).

Figure 6(a) provides a clear picture of the extent by which the dynamics are enhanced in ultra-thin films. While at bulk T_g the dynamics are enhanced by only two to four orders of magnitude, the difference between thin film and bulk dynamics continues to diverge as the temperature is decreased below T_g. For example, at a temperature of T_g − 35 K, the bulk viscosity becomes unmeasurable, while the effective thin film viscosity only changes by less than two orders of magnitude from the value at T_g.

Figure 6(b) shows the apparent activation energy, E_a (slope of log viscosity vs. 1/T) as a function of film thickness as determined via both dewetting and CR-T_g experiments. We note that E_a may vary outside the range of temperatures presented here, for example, closer to or above T_g. As shown in Figure 6(b), in films with h < 20 nm, the E_a is much lower than that of the bulk and has a weak thickness-dependence. The low activation barrier for dewetting confirms that the rough morphologies observed in ultra-thin PVD films are due to fast dewetting during the deposition. E_a increases sharply in films with thicknesses between 20 nm <h < 30 nm and becomes very similar to bulk at h > 40 nm. In this regime, the dynamics of the film are bulk-like during PVD, and surface diffusion acts to smoothen the film. Interestingly, in ultra-thin films, once the local film thickness around the rims reaches 40 nm, the dewetting process also slows down significantly and appears to stop in films with thicknesses ranging between 8 nm and 25 nm as shown in Figure S8 of the supplementary material.

This remarkably sharp transition in the dynamics suggests that the gradient of dynamics induced by the interfacial effects is not the same in films of different thicknesses, as schematically shown on the right of Figure 6(b). In ultra-thin films, the dynamics are enhanced in the entire film, showing little thickness dependence, while in films with thicknesses of 40 nm or more, the dynamics in the entire film is bulk-like except for perhaps a few liquid-like mono-layers near the free surface. If the interfacial effects propagated over the same length scale and with the same strength in films of all thicknesses, then the overall dynamics can be simply described by a weighted average function of faster interfacial dynamics due to the existence of a mobile surface layer with a constant thickness and slower bulk dynamics. One would then expect a gradual decrease in E_a with decreasing film thickness. This behavior is shown in Figure 6(b) as black dashed-line by assuming a mobile layer with a thickness of 8 nm, with identical dynamics as that of the 8 nm film measured in this study. Details of this prediction can be found in the supplementary material and Figures S16 and S17. While assuming a different thickness for the mobile layer or different dynamics may change the onset of the changes in E_a or its final value, the shape of the curve would qualitatively remain the same, showing a gradually decreasing E_a with film thickness. Instead, in experiments, a sharp transition is observed around 30 nm (Figure 6(b)) suggesting strongly correlated dynamics in thin films that result in enhancement in the overall dynamics of the film and not just the top 8 nm.

It is important to note that the temperature dependence of diffusion coefficients measured on the surface of bulk films^6,7,56 is stronger than the temperature dependence for dewetting measured in ultra-thin films in this study, further confirming that the dynamics at the surface of a bulk film are also slower than those measured in ultra-thin films. These observations suggest that the dynamics of the glassy material are correlated over a length scale of ∼10-30 nm and the dynamics of thin films are influenced both by the interfacial dynamics and the glassy dynamics in the layers closer to the center of the film. As such, surface diffusion measurements on the surface of bulk-like films alone are inadequate in predicting the activation barrier for the dynamics in ultra-thin films and the length scale of the interfacial effects. Direct measurements of properties as a function of film thickness are required for determining the correlation length for the dynamics.

While few studies probe dynamics of ultra-thin molecular glass films, observations in polymeric glasses⁴³ show a similar non-linear transition in E_a as a function of film thickness, with the midpoint of transition in the 20-30 nm film thickness region. Earlier studies by Ellison and Torkelson also suggest correlated dynamics in the top and bottom layers of a polymeric film as the film thickness is reduced below 20 nm.¹² However, the transition appears to be much sharper in organic thin films with much lower activation barriers in ultra-thin film regime. This may imply that chain effects are also important in facilitating the dynamics in polymeric thin films. Regardless, the general similar trend and length scale of enhanced dynamics in both small organic molecular and polymeric glasses suggest that long range facilitation of the dynamics may be a characteristic feature of glassy systems. Future studies on more glassy systems are needed to confirm whether these observations are ubiquitous in organic glassy systems.

Our results are difficult to rationalize with existing theories and simulations of the glass transition. Models based on a constant length scale of interfacial effects⁵⁷ are unlikely to capture the strong, almost sigmoidal transition in apparent activation barriers observed here. Instead, models based on long range elastic response^25,58 or which use growing cooperative length scales^26,27 may be able to predict such strong correlated dynamics.

Simulations have observed a growing mobile surface layer that correlates strongly with the size of cooperative rearrangements,^59,60 but these simulations have not observed the sharp transition in apparent activation energy for film thicknesses of 20-30 nm. The sharp transition in dynamical behavior with increasing film thickness is reminiscent of the non-equilibrium phase transitions to immobile phases that have been observed in simulations that were biased to sample low-mobility trajectories.^61–63 However, it is not obvious at all how to map film thickness, enhancing the mobility of the glass, to the biasing potential or pinning procedures used to observe the non-equilibrium transitions.

The long length scales that we observe would be difficult to rationalize with the typical sizes of dynamic heterogeneity that have been measured either experimentally or computationally, which are commonly cited to be on the order of 2-3 nm near T_g. Certainly it is known that surface effects can propagate over large length scales in polymer glasses, where the mobile surface layer associated with T_g confinement effects has been found to approach 10 nm, so it is possible that the same mechanism could be responsible in these molecular glass films. Simulations of polymer glasses have connected the size of the mobile surface layer to the size of cooperatively rearranging regions in the glass,^59,60 so it is possible that an Adam-Gibbs picture could explain the large length scale component of our results. An alternative possibility is the theory of Mirrigian and Schweizer, which captures long-range elastic effects on monomer rearrangements and predicts changes in properties over long length scales away from free surfaces.^25,58

In summary, we have presented that the temperature dependence of the effective viscosity, and thus the structural relaxation time, of thick and ultra-thin films of molecular glasses can be measured via a combination of isothermal dewetting and CR-T_g measurements. We have demonstrated that the rough initial morphology of vapor-deposited thin films is closely related to the enhanced dynamics in ultra-thin films. Films as thick as 30 nm dewet spontaneously well below bulk T_g, indicative of greatly enhanced dynamics in these films. An examination of the thickness dependence of the apparent activation barrier in these films reveals a sharp, sigmoidal transition in the dynamics as the thickness varies between 20 and 40 nm indicating a strong correlation between the dynamics of the free surface and the bulk of the film. This implies an interplay between the facilitation of the dynamics by the interface and the bulk glass, with a considerably large length scale of about 30 nm.

See the supplementary material for detailed material characterization Figs. S1 and S2, discussion about thin film equation, substrate effect, extra dewetting data Figs. S3-S8, more details about CR-T_g experiments Figs. S9-S12, details about shift factors Figs. S13-S15, and discussion about no correlation assumption Figs. S16 and 17.

We thank Dr. Wei-Shao Tung and Professor Karen Winey for help with rheology measurement, M. Reza Rahimi Tabar, Andrea Liu, Mark Ediger, Ramalingam Kailasham, Amit Shavit, and Jack Douglas for helpful discussions. Z.F. acknowledge support from NSF-Career Grant (No. NSF DMR-1350044). Z.F. and R.R. acknowledge seed funding from MRSEC Grant (No. NSF DMR-1120901). K.W. was an undergraduate student enrolled in the Research Experiences for Undergraduates (REU) program during 2014 summer. He is from the Department of Physics, University of Texas Rio Grande Valley. He acknowledges funding from an NSF MRSEC Grant (No. DMR-1120901).