Investigations of the adsorbed layer of polysulfone: Influence of the thickness of the adsorbed layer on the glass transition of thin films

The glass transition temperature (T_g) is one key parameter used to characterize the behavior of amorphous materials, such as polymers, as it defines the temperature range used in final applications. The glass transition occurs over a range of temperatures and is characterized by a change in the underlying molecular dynamics of a material resulting in significant changes in shear modulus, heat capacity, thermal expansion, and volume, to name a few properties.¹ Although the glass transition has a high importance for applications, its molecular origin is still under debate. Investigations on thin polymer films may help understand this phenomenon further. Moreover, thin polymer films can be considered as model systems for studying confinement effects on the properties of polymers.

For thin polymer films, specifically with film thicknesses less than 200 nm, T_g can deviate from the bulk value. Various trends, such as an increase, decrease, or constant values of T_g, have been previously reported as the film thickness is reduced. Since the early studies of Keddie et al.,^2,3 there have been many investigations that studied the thickness dependence of T_g for thin films employing a variety of techniques such as ellipsometry,^2–15 Brillouin light scattering,^4,16 x-ray reflectivity,^17–21 fluorescence spectroscopy,^13,22–24 dielectric spectroscopy,^25–33 and AC-chip calorimetry.^{31,32,34–36} It is worth mentioning that in these studies one must differentiate between the thermal and the dynamic glass transition temperature. On the one side, a thermal glass transition temperature is estimated using so-called static experiments such as ellipsometry, where the polymer changes from an equilibrium to a nonequilibrium state during cooling at T_g. On the other hand, a dynamic glass transition temperature is obtained from experiments carried out in equilibrium, such as dielectric spectroscopy in the frame of the linear response theory. As discussed in Ref. 31 the thermal and the dynamic T_g can have different thickness dependencies. However, only the thermal T_g is further considered here.

From these investigations, it was revealed that confinement effects and the polymer–substrate interactions play a pivotal role in the thickness dependence of the thermal T_g. At the polymer–air interface, both the number of polymer segments and the density are reduced compared to the bulk due to missing segment–segment interactions. These effects result in an increased surface mobility that causes a reduction in the thermal T_g. However, for thin films supported on a substrate with non-repulsive interactions, the polymer segments adopt a conformation at this interface with a decreased mobility and cause an increase in the thermal T_g compared to the bulk value. Mortazavian et al.³⁷ measured in fundamentals three regions or “layers” with different thermal activities. First, segments adsorb at the substrate interface, forming a tightly bounded layer called the irreversibly adsorbed layer, which is stabilized by the connectivity of the polymer chain.^38–40 Second, a layer comprised of segments located further away from the substrate is formed, where the polymer segments have bulk-like properties. Third, the region at the polymer–air interface was referred to as the mobile surface layer. The competing contributions of these idealized layers result in a complicated thickness dependence of the average thermal glass transition temperature of the entire thin film.

The adsorbed layer has attracted significant attention in research and applications, arising from its importance in material properties. Therefore, the properties of the adsorbed layer have been directly investigated. As discussed above, it is considered as an irreversibly adsorbed layer due to the unlikelihood of a simultaneous detachment of many polymer segments as high activation energy is required for such a process. Nevertheless, it is worth noting that desorption of this layer has been reported recently employing fast scanning calorimetry.⁴¹ Guiselin et al.⁴² proposed for the first time in the literature the idea of preparing an irreversibly adsorbed layer from the melt. This idea can be transferred to spin coating a solution onto an attractive substrate, annealing the prepared film to form the adsorbed layer, and subsequently washing away the non-adsorbed chains using a good solvent. The obtained layer is then annealed at an elevated temperature to allow the chains to adopt more stable conformations. This procedure has been repeated by other scientists in further studies,^14,43–51 where the thickness of the adsorbed layer was measured concerning the adsorption time. The growth kinetics of the adsorbed layer of polystyrene (PS) was studied by Housmans et al.³⁹ by measuring the thickness at different annealing times at temperatures above the T_g of the bulk polymer after leaching. The adsorption process displayed a two-step growth mechanism. First, at short times, the thickness of the adsorbed layer grows linearly with time, and the segments directly pin to the substrate interface forming a flattened conformation referred to as “trains.”^52–54 After a critical time, the substrate surface becomes saturated with segments, and the growth of the adsorbed layer continues by diffusion of segments through the already formed layer. This diffusion process leads to a logarithmic time dependence of the growth kinetics. During this process, the chains are stretched at a cost of entropy and form “loops,” anchored at two points and dangling ends called “tails.”^39,53 The crossover between the linear regime (growth by pinning) and the logarithmic regime (growth by diffusion) is indicated by a crossover time that depends, for instance, on the annealing conditions and molecular weight. A review of the adsorption process and solvent leaching techniques can be found elsewhere.^55,56

Recently, in addition to this two-step growth mechanism, a further growth step was identified for poly(bisphenol A carbonate) (PBAC) at shorter times than characteristic for the linear growth.¹⁴ The molecular origin of this pre-growth regime was assigned to an orientation of the phenyl rings in the main chain forming stacks, which required an enthalpic input to rearrange. At long annealing times, desorption was observed where the segments simultaneously detach from the substrate by dewetting. A desorption process was also discussed to occur for poly(2-vinyl pyridine) (P2VP).⁵¹ 

In this work, thin films of polysulfone (PSU) prepared on Si substrates were used as a model system for main chain polymers with high thermal stability. Both the growth kinetics of the adsorbed layer and the influence of the adsorbed layer on the thickness dependence of the glass transition temperature were investigated. The results obtained for PSU are compared to those found for PBAC since PSU also contains phenyl rings in the backbone. Several results previously obtained for PSU^11,15,23,24 suggest a decrease in T_g with reducing the film thickness. Labahn et al., on the contrary, reported an increase in the ideal glass transition temperature T₀ with decreasing film thickness employing dielectric spectroscopy.²⁸  T₀ is found 30–50 K below T_g and is, therefore, related to the thermal T_g. Therefore, this work aims to determine the glass transition of thin PSU films with the investigation of the adsorbed layer using spectroscopic ellipsometry (SE) and atomic force microscopy (AFM).

PSU with a molecular weight (M_w) of 35 000 g/mol and PDI of 2.19 were purchased from Sigma Aldrich (Taufkirchen, Germany). The chemical structure of PSU is shown in the inset of Fig. 1 (right side). The structure of polysulfone is amorphous. Due to the sterically complex backbone, it does not crystallize. The glass transition of PSU was found to be T_g = 457 K, where the estimation of the glass transition temperature is discussed in the following.

The samples were prepared by dissolving the PSU pellets in chloroform to obtain a master solution, which was then subsequently diluted again with chloroform to obtain solutions with different PSU concentrations.

Conventional DSC measurements were performed using a Perkin Elmer DSC 8500 instrument (Perkin Elmer, Waltham, MA, USA) with a heating/cooling rate of 10 K/min in the temperature range of 293–523 K. Nitrogen was used as a purge gas with a flow rate of 20 l/min. Baseline corrections were performed before each sample run by measuring an empty pan and subtracting the measured heat flow from the data obtained for the sample. The second heating run was used for analysis. The samples were prepared by film casting a PSU solution into a disk-shaped mold. A closed box saturated with chloroform was used to allow for slow evaporation of the solvent. For complete evaporation of the solvent, the film was then annealed for 120 h at 493 K in an oil-free vacuum.

The silica substrates were obtained from CrysTec (Berlin, Germany). The substrates have a high electrical resistivity as they are only slightly doped. The silica substrates were cleaned in the first step by rinsing them in acetone. To clean it further down to the microscale, supercritical carbon dioxide (CO₂) obtained by a snow jet gun was applied to the surface of the substrate. As the last step, the substrate was placed in plasma cleaner with an oxygen (O₂) atmosphere for 10 min at 60 W to activate the OH groups at the surface and to increase the surface energy, allowing for better interaction between the substrate and the polymer.

The diluted solutions with different concentrations were then spin-coated onto the substrate. The detailed spin coating procedure is described elsewhere.¹⁴ After spin coating, the samples were placed in an oil-free vacuum oven. Three different sets of thin film samples were prepared. The temperature of the samples was increased slowly from room temperature in steps of 10 K to two different annealing temperatures of T = 467 K (T_g + 10 K) and T = 487 K (T_g + 30 K). At T_g + 10 K, the samples were annealed for 5 and 72 h, where at T_g + 30 K, the thin films were annealed for 120 h. It is aimed to investigate the effect of different annealing conditions on the thermal T_g by preparing thin films under three different annealing conditions.

To prepare the adsorbed layer, the solvent leaching approach using the process of Guiselin brushes was applied.⁴² Thin films with a thickness of 120 nm were spin coated on an Si substrate and then annealed in an oil-free vacuum oven at three different annealing temperatures above the bulk T_g as T_g + 10 K (467 K), T_g + 30 K (487 K), and T_g + 60 K (517 K) for various annealing times. Afterward, they were solvent leached by washing away the non-adsorbed or loosely bounded chains using chloroform as a good solvent for PSU. The thin films were soaked for 20 min in a chloroform bath and then further rinsed again with chloroform before the samples were placed into an oil-free vacuum oven at 493 K for 20 min to remove the solvent.

AFM measurements were conducted using an MFP-3D origin atomic force microscope (Oxford Instruments, Abingdon, UK) in tapping mode with non-contact silicon SPM sensors (Nanosensors, Neuchatel, Switzerland) having a resonance frequency of 209–497 kHz. Scan rates of 1 Hz were used and, 512 points and lines per image were obtained. The film thickness and the thickness of the adsorbed layer were estimated by scratching through the film down to the Si substrate using a blade. The film thickness was taken as the difference between the surface of the film and the substrate.

Image processing and analysis were performed using the freely available software Gwyddion.⁵⁷ Any possible discontinuous defects measured during the scan were fixed by the software as the initial image analysis step. After that, the large defects due to the scratching of the surface were masked, and the rows were aligned to make any hidden topography visible during the measurements. Additional image processing was done by repeating these steps and adjusting the scale to restrict the color bar for ease of viewing.

Table S1 in the supplementary material presents the estimated film thickness by AFM and the concentration of the diluted solution used for spin coating. Furthermore, no signs of dewetting visible in the topography images were detected, and the surface roughness of each thin film was low.

An example of an AFM measurement of an adsorbed layer obtained from a solvent leached PSU film annealed at T_g + 30 K for 120 h is shown in Fig. 1. The figure shows a profile taken across the surface of the adsorbed layer (left side) and the height profile (right side).

The ellipsometric investigations were carried out by an M-2000DI ellipsometer (J.A. Wollam, Lincoln, NE, USA). To confirm the film thickness estimated by AFM, the ellipsometer was used in a multi-angle configuration with three angles of incidence (65°, 70°, and 75°). Apart from establishing a base value for the layer thickness, this measurement is also used to determine the dielectric function/optical constants of the polymer material. This procedure is carried out for every sample as the optical properties may vary for different film thicknesses. Wavelengths in the range from λ = 192–1697 nm (709 discrete wavelength points were used for analysis. The temperature dependence of the thickness for each sample was measured with the angle of incidence of 70°. The temperature of the sample was controlled by a Linkam heating stage THMSEL 600 adapted to the Wollam ellipsometer. For these measurements, the sample was first heated to 100 °C (373 K). Then, temperature scans from 100 °C (373 K) to 230 °C (503 K) were carried out with a heating rate of 2 K/min. The stability of the temperature was 0.1 K. To check the stability of the whole system, the sample was held at T = 503 K for 5 min. After that, the sample is cooled down to room temperature in 5 min. The data during the cool-down period were not used for analysis. The two ellipsometric quantities that were obtained from the measurements are the amplitude ratio of the polarized light (ψ) and the phase difference between the incoming and outcoming light (Δ).^58,59 The whole measurement was controlled by Complete-EASE-software, v. 6.57.

The spectra of the ellipsometric transfer data were analyzed using a model for the supported film, including the Si substrate with a native oxide layer with a fixed thickness of 2 nm. The dielectric functions for this Si substrate were given by Herzinger et al. elsewhere.⁶⁰ A temperature-dependent variant for the substrate data was considered for the temperature scans. The analysis was done in a two-step procedure. In the first step, a pre-fit was carried out. Here, a B-spline function was used to estimate the start parameter for the dielectric function.⁶¹ A structured absorption edge was observed for PSU in the UV range of the light. For the description of the absorption edge, a general oscillator model was used to model the dielectric function of PSU.^58,62–64 In the fitting procedure, instead of the spline function now, the selected oscillator model was employed for the analysis of the temperature scans as the spline function cannot be used. The model included the sample temperature for every point on the temperature scan measurement to account for the temperature-dependent dielectric function of the Si substrate. Using multi-angle fits, the parameters were optimized at room temperature. For analysis of the temperature scan, in addition to the thickness of the film and ε_∞ the value of the dielectric function at infinite frequencies, all parameters of the oscillator model were kept constant. By this procedure, the film thickness d was obtained as a function of time or temperature. For analysis of the measurements, the program Complete-EASE software, v. 6.70 is utilized. An example of the fitting of the ellipsometric data is given in the supplementary material, Fig. S2.

Contact angle measurements were carried out for PSU and Si substrates using an automated contact angle system (Krüss, Hamburg, Germany) using the static sessile drop method. Diiodo-methane, ethylene glycol, glycerol, and water were used as test liquids.

The growth kinetics of the adsorbed layer of PSU was investigated by measuring the thickness of the solvent leached film as a function of annealing time for the selected annealing temperatures. Three annealing temperatures (T_g + 10 K, T_g + 30 K, and T_g + 60 K) and annealing times between 0 h (unannealed) and 336 h were used.

Figure 2 shows the thickness of the adsorbed layer as a function of the annealing time for the three selected annealing temperatures. Deviations from the reported two-step growth mechanism, observed for other polymers with a simpler backbone,^39,45 were obvious. It was observed that the thickness of the adsorbed layer increases by a pre-growth step at the shortest annealing times. This pre-growth step was also reported previously for PBAC by some of the authors.¹⁴ However, for PSU, this step requires a longer time compared to PBAC, which allows for a more detailed analysis. For the annealing temperature of T_g + 10 K, the pre-growth regime needs almost 25 h to be completed before a change in growth mechanism is observed. Different from PBAC, the pre-growth step is also observed for the highest annealing temperature T_g + 60 K with a duration of 8 h. For comparison, for PBAC annealed at T_g + 30 K, the pre-growth step is completed after 1 h. The increased time scale of the pre-growth step of PSU in comparison to PBAC might be due to the higher number of phenyl groups in the repeating unit of PSU (four for PSU and two for PBAC). As discussed in Ref. 14, the bulky phenyl rings in both PBAC and PSU adopt a preferred parallel orientation with respect to the surface of the Si substrate during spin coating. Myers et al.⁶⁵ studied the effect of the solvent on the orientation of phenyl rings at the polymer–substrate interface for thin polymer films. It was found that for aromatic solvents, the phenyl rings had a perpendicular orientation regarding to the normal of the substrate surface and for non-aromatic solvents, a parallel orientation was detected. For a 2 wt. % solution of PSU in chloroform, it was revealed by sum frequency generation spectroscopy (SFG) that the phenyl rings of PSU orient mostly in a preferred parallel orientation at the polymer–substrate interface.

The thickness of the unannealed adsorbed layer is found to be less than 1 nm (0.8–0.9 nm), as shown in Fig. 2. A similar value was found for PBAC.¹⁴ Quantum chemical calculation for the case of dimers of benzene rings reveals a slip-parallel orientation of the benzene rings most stable configuration due to π–π interaction.⁶⁶ This stable slip-parallel configuration has a minimal interaction energy of −2.48 kcal/mol, where the distance of the benzene rings in the dimer is ∼0.37 nm. Furthermore, it is further assumed that the discussed model calculation also has some relevance for the phenyl groups in the PSU main chain. The spin coating process thereby leads to the formation of a thin adsorbed layer having a thickness of 0.7–0.8 nm in which the phenyl rings of PSU form dimers preferentially parallel oriented at the substrate surface. Then, these preferentially oriented dimers can act as a kind of nucleus on which further phenyl rings can stack on top of each other during the pre-growth step. Therefore, the pre-growth step is assigned to the formation of stacks in the early stages of the formation of the adsorbed layer. The thickness of the adsorbed layer formed during the pre-growth step depends on the annealing temperature (see Fig. 2).

Figure 3 shows the thickness of the adsorbed layer at the end of the pre-growth step vs the annealing temperature. These results reveal that the stack formation is most effective in the temperature range between T_g + 10 K and T_g + 30 K and saturates for T_g + 60 K. For the annealing time of T_g + 10 K, the adsorbed layer at the end of the pre-growth process mainly consists of dimers of phenyl rings as the thickness of the adsorbed layer is ∼1 nm. However, for the annealing temperature of T_g + 30 K, the thickness of the adsorbed layer at the end of the pre-growth is ∼2.5 nm. This means that for this annealing temperature, ∼eight phenyl rings stack on top of each other, whereas for the annealing temperature of T_g + 60 K, the stacks are formed by ten phenyl rings in slip-parallel configuration.

E_a denotes the activation energy, τ_∞ symbolizes the time constant at infinite temperatures, and R is the general gas constant. A fit of the Arrhenius equation to the data results in an activation energy of 75 kJ/mol.

Figure 2 shows that at times longer than characteristic for the pre-growth step, a linear growth regime is also observed for PSU, such as for polymers with a flexible backbone. For the two highest annealing times, a change in the linear growth kinetics to a logarithmic one is observed. Unfortunately, no higher annealing temperatures than T_g + 60 K can be employed as the PSU might thermally degrade or dewet. Therefore, the saturation of the growth process cannot be proven as Eq. (5) could not be tested.

The linear growth rate ν was plotted as a function of inverse temperature in the Arrhenius plot (see Fig. 5). From the fit of the Arrhenius equation to the data of the linear growth rate, the activation energy was calculated to be 62 kJ/mol, which matches the values reported in the literature for the change from linear to logarithmic.^39,51 The activation energy for localized fluctuations of PSU measured by BDS was 45.5 kJ/mol,²⁸ which confirms that the adsorption process is not localized as the activation energy is higher than expected for a localized process and instead points to a process involving at least dimers. In Ref. 68, Song et al. reported a fast equilibration mechanism in disordered materials, including polymers called the slow Arrhenius process (SAP). It was also shown that this process is also related to the adsorption kinetics of the adsorbed layer. For PSU for the activation energy of the SAP, a value of 62 kJ/mol was found.⁶⁹ This value agrees with the activation energy found for the linear growth regime. Therefore, in might be concluded that the adsorption process of PSU is related to its SAP.

From the analysis of the data shown in Fig. 5, the linear growth rate at infinite temperature υ₀ could be estimated to be 13 nm/s +/− 6 nm/s. In Ref. 70, a relationship between the activation energy of the SAP process and the growth rate at infinite temperatures was derived. Using this relationship, a value of ∼4.4 nm/s can be calculated using the activation energy for the SAP of 62 kJ/mol. This value is a bit smaller than experimentally estimated. However, it is worth mentioning that υ₀ was estimated from only three data points that involve larger errors. Nevertheless, the estimated value is found within the range of 2–15 nm/s for a variety of polymers.

For PBAC, evidence for desorption of the adsorbed layer was reported for higher annealing temperatures and times.¹⁴ This is different from PSU, where in Fig. 2 no sign of desorption is observed. This might be due to the stronger interaction of the PSU segments with the substrate in comparison to that of PBAC.

The surface roughness values for the adsorbed layer of PSU were obtained from the AFM measurements by measuring the average roughness value for a smaller image size of 5 × 5 µm² using Gwyddion software. Figure 6 shows the surface roughness as a function of the annealing time for the selected annealing temperatures. The values obtained for PSU are significantly lower compared to those measured for PBAC. For PSU, the maximum value of the surface roughness is found to be 1.8 nm for the highest annealing temperature at the longest annealing time, where the observed value for PBAC for the same annealing conditions was 3.5 nm. This result indicates that a more stable adsorbed layer is formed for PSU compared to PBAC. This result is also in agreement with the observation that no desorption of the adsorbed layer is detected for PSU. For all annealing temperatures, the surface roughness increases with the annealing time. For both T_g + 30 K and T_g + 60 K, a step-like increase in the surface roughness is observed, whereas the step shifts to shorter times with increasing annealing temperature. For T_g + 10 K, no inflection pointing to saturation is detected, as in growth kinetics, no logarithmic growth regime was reached in the covered experimental time scale. Generally, the surface roughness of the adsorbed layer increases with increasing annealing temperature, where for T_g + 10 K, its time dependence starts with the value estimated for an unannealed sample.

The thermal glass transition temperature for bulk PSU was determined by measuring a film-casted sample by DSC. The heat flow curves were fit with a sigmoidal function and subsequently differentiated concerning temperature. The peak position of the differentiated fit was assigned as T_g. Multiple film-casted samples were measured with DSC confirming T_g = 457 K, which agrees with the values reported for PSU in the literature.^{11,15,23,24,28,71}

The estimated T_g values were plotted as a function of film thickness in Fig. 8 for two different annealing conditions. The value for the bulk T_g is included in the figure. The plot shows an increase in T_g with decreasing film thickness. This result points to an increasing contribution of the adsorbed layer with decreasing film thickness to the T_g of the whole film. The restricted segmental motions at the substrate–polymer interface require elevated temperatures to allow for a glass transition. For the different annealing conditions, different dependencies of the thickness dependence of the thermal glass transition are obtained. For the samples annealed at T_g + 30 K for 120 h, the increase in T_g begins at around 100 nm with an increase of 15 K for a film with a thickness of 26 nm compared to the bulk sample. However, for the samples annealed at T_g + 10 K for 72 h, the increase in T_g begins at a much lower film thickness, around 45 nm, with an increase of 11 K for the thinnest measured film (17 nm) compared to the glass transition temperature for the bulk. So, the difference in the onset thickness for both annealing conditions is ∼55 nm. Moreover, for the films annealed at T_g + 10 K for 72 h, the increase in T_g is sharper with decreasing the film thickness than that for the films annealed at T_g + 30 K for 120 h. For the films annealed at T_g + 10 K for 5 h, no increase in the glass transition temperature with decreasing thickness could be detected down to thicknesses of 30 nm.

The different dependence of T_g on the film thickness obtained for the different employed annealing conditions might be related to the different thicknesses of the adsorbed layer formed under these conditions, although the difference in the thickness of the adsorbed layer obtained from leaching experiments is only ∼3 nm.

In Fig. 9, the width parameter w obtained from the fit of Eq. (6) to the temperature dependence of the thickness measured by ellipsometry is plotted vs the film thickness. Similar to the dependence of the glass transition temperature on the film thickness, the dependence of the width of the ellipsometric glass transition on the film thickness is quite different for three annealing conditions. For annealing at T_g + 10 K for 72 h, the width of the glass transition is constant down to film thicknesses of ∼40 nm and then increases to a relatively narrow maximum at ∼25 nm, followed by a strong decrease. For the annealing condition T_g + 30 K for 120 h, the increases in the width of the glass transition start at ∼100 nm and then display a broad peak with decreasing film thickness. It is worth noting that the onset thicknesses for the width of the glass transition are the same as the onset thicknesses for the glass transition temperature. For the samples annealed at T_g + 10 K for 5 h, only an increase in the width of the glass transition is detected for film thickness of around 30 nm.

As a preliminary approach, the width of the glass transition can be considered as an expression of the spatial heterogeneity of the glass transition. This approach agrees with the idealized three-layer model of thin films, which assumes a dynamical heterogeneity of it. The broader peak obtained for the annealing condition at T_g + 30 K for 120 h points to a more heterogeneous film structure compared to that obtained at T_g + 10 K for 72 h. This conclusion might agree with the higher surface roughness of the adsorbed layer obtained from the solvent leaching experiments. For the annealing condition T_g + 10 K for 5 h, only the pre-grow step is observed during the formation of the adsorbed layer with a thickness of ∼1 nm.

As mentioned before, Labahn et al.²⁸ also reported an increase in the Vogel temperature T₀, which is related to T_g for PSU thin films measured using dielectric spectroscopy. The Vogel temperature is obtained by fitting the Vogel/Fulcher/Tammann equation^73–75 to the dielectric relaxation rates, showed an increase beginning at ∼100 nm. Here, for films below 17 nm, T_g could not be determined as the thickness changes in the thin film become similar to the uncertainties of the measurements. In the literature, a strong reduction in T_g for PSU thin films measured with fluorescence spectroscopy^23,24 and ellipsometry^11,15 has been reported. However, for these investigations, different substrates, and sample preparation steps as well as annealing conditions were used. Evans et al.²⁴ measured PSU samples spin-coated onto quartz glass slides annealed at T_g + 20 K for 2 h and found a reduction in T_g of about 20 K. Moreover, it was argued that polymers with a higher fragility have the largest reductions in T_g. Ma et al.¹⁵ prepared thin PSU films with a thickness of 200 nm where no annealing was applied and only the surface T_g was measured to discuss the reduction when compared to bulk T_g. In this respect, it should also be noted that fluorescence-based measurements were found to overestimate the contribution of the free-surface effect in thin films where ellipsometry underestimates it.¹³ In addition to the consideration, these studies classified the interaction between Si substrate and PSU as non-attractive, which was not the case in this work. The Si substrates had, as discussed above, a high resistivity due to a low doping level, which results in an increased surface energy compared to highly doped Si substrates. The effect of plasma treatment with oxygen increased the surface energy too.

In order to calculate the interfacial energy between Si and PSU, the surface tension must be first determined for both compounds. Therefore, contact angles were measured using common test liquids. The measured contact angles are presented in Table I for each test liquid. 2–6 contact angle measurements were taken to estimate the arithmetically averaged values with reasonable errors.

The subscripts S and P refer to the substrate and polymer, respectively. The estimated interfacial energy between Si and PSU was 3.46 mN/m. This value is higher than the critical value of 2 mN/m reported by Fryer et al.⁶ Above this value, an increase in T_g was reported for both PS and PMMA films (thickness ∼20 nm) on Si wafers modified with octadecyl trichlorosilane (OTS) to vary the surface tension of the silicon substrate. The estimated interfacial energy between Si and PSU gives evidence for attractive interactions between the substrate and polymer. This is also in agreement with the formation of an irreversibly adsorbed layer that leads to an increased T_g compared to the bulk sample with the decreasing film thickness, as discussed above.

Madkour et al.³⁶ partly summarized the difference between the bulk T_g and T_g for a thin film vs the interfacial energy for various polymer–substrate systems. It is important to mention that the reported plot contains thermal and dynamic T_g values. In Fig. 10, data are included that were obtained from previous studies in the literature, and it contains the data points for PSU on Si annealed at T_g + 30 K for 120 h and at T_g + 10 K for 72 h from this investigation. A trend is observed that with increasing interaction energy between the polymer and substrate, T_g of the thin film increases as the interaction energy between the substrate and polymer controls the formation of an adsorbed layer. Nevertheless, for a few systems, no changes in the T_g values were reported even with strong interfacial interactions, namely, PVAC on both aluminum and Si substrates measured with a surface sensitive AFM technique. Based on these results shown in Fig. 10, it might be concluded that there is a rough correlation between the interfacial energy and the change in T_g compared to the bulk value. However, the interfacial energy is not the only parameter that affects the thickness dependence of T_g values of thin films. In addition, the annealing conditions have a direct effect on the thickness and conformation of the segments of the irreversibly adsorbed layer. This is confirmed here for the data of PSU having the same interfacial energy but different thicknesses of the adsorbed layers (see Fig. 10).

The annealing conditions used in the previous studies of PSU (see Refs. 11, 15, 23, and 24) were not sufficient enough to allow for the growth of an adsorbed layer at the polymer–substrate interface. Annealing the sample after spin coating allows for further growth of the adsorbed layer. Figure 10 shows that the samples annealed at higher temperatures and longer times than those previously reported in the literature^11,15,23,24 displayed not only a larger increase in ΔT (T_{g, film} − T_{g, bulk}) but also a larger onset thickness. As discussed above, particularly for T_g + 30 K at 120 h, the onset thickness is around 100 nm, and for T_g + 10 K at 72 h, it is near 45 nm. Napolitano et al. reported, in Ref. 44, a proportionality between an increase in the adsorbed layer of PS and the increase in T_g. For poly(4-tert-butylstyrene) (PTBS), Perez-de-Eulate et al.⁴⁷ also showed that the large confinement effects that result in large reductions in T_g can be erased with prolonged annealing. In Fig. 8, the difference in ΔT for the annealing conditions clearly shows the same dependence as described for PS and PTBS. Therefore, combining both a non-attractive interaction between the polymer and substrate and short annealing times might lead to the conclusion that the segments at the polymer–air interface dominate the contribution to the overall T_g value of the thin film. However, for conditions that allow for the growth of an adsorbed layer, T_g increases with a reduction in film thickness due to the dominating contribution of the adsorbed layer. This effect was not only seen for PSU but also for other polymers, such as PS, PBAC, P2VP, PMMA, poly(vinyl chloride) (PVC), and PET.^{6,14,18,44,51,80}

The general aim of the presented investigation is to study the influence of the adsorbed layer formed at the polymer/substrate interface on the glass transition of thin films of polysulfone. Therefore, in the first step, the adsorbed layer of PSU was prepared by solvent leaching approach. For this procedure, thin films with a thickness of 120 nm were prepared. The films were then annealed at three different annealing temperatures (T_g + 10 K, T_g + 30 K, and T_g + 60 K) between 0 h (unannealed) and 336 h and then leached using the good solvent chloroform for PSU. Subsequently, the thickness of the obtained layer was measured using AFM. At the shortest annealing times, a pre-growth step was found. The origin of this step was assigned to the stacking of phenyl rings at the substrate. After a given time, which depends on the annealing temperature, the growth kinetics changes from the pre-growth to the linear regime. During this growth step, the adsorbed layer grows by pinning segments at the substrate. The crossover from the linear to the logarithmic regime was observed only for T_g +30 K and T_g + 60 K. For T_g + 10 K, annealing times longer than at least 240 h would be required for a crossover from the linear to the logarithmic regime. The growth in the logarithmic regime now occurs due to crowding by segments at the substrate by a diffusion of segments through the tightly bounded layer by stretching. Furthermore, no chain desorption was detected for any solvent leached sample, which was assigned to the increased stability of the adsorbed layer for PSU compared to PBAC. The surface roughness of each sample also confirmed this temperature-dependent behavior of the adsorbed layer. Only the samples annealed at T_g + 10 K did not show a plateau at long annealing times.

The glass transition temperature of thin films was secondly determined, employing spectroscopic ellipsometry using the samples annealed either at T_g + 10 K for 5 and 72 h or at T_g + 30 K for 120 h. For the annealing condition T_g + 10 K for 5 h, no real adsorbed layer is formed and only the pre-growth step due to stacking of phenyl rings is observed. The condition T_g +10 K for 72 h represented films with adsorbed layers formed in the linear regime, and the condition T_g +30 K for 120 h, was for thin films where the adsorbed layer is also formed in the logarithmic regime. The glass transition temperature increased for both the latter conditions compared to the bulk value, with a reduction in film thickness down to 17 nm. However, the samples annealed at T_g + 30 K for 120 h, the increase in T_g with decreasing film thickness had a higher larger onset thickness of ∼100 nm, compared to the samples annealed at T_g + 10 K for 72 h with an onset thickness of ∼45 nm. The thicknesses of the adsorbed layer obtained by the solvent leaching experiments were 1.4 nm for the samples annealed at the lowest annealing temperature of T_g + 10 K for 72 h and 4.1 nm for the samples annealed at T_g + 30 K for 120 h. This result evidenced a strong correlation between the adsorbed layer and the increase in T_g with decreasing film thickness compared to the bulk value.

The supplementary material provides the relation between the concentration of the spin coating solution and the thickness of the thickness of the film; analysis of the pre-growth step; and analysis of the ellipsometric measurements.

BAM is acknowledged for financial support in the frame of the Ph.D program.

The authors have no conflicts to disclose.

Hassan Omar: Data curation (equal); Writing – original draft (equal). Shayan Ahamadi: Data curation (equal). Deniz Hülagü: Data curation (equal). Gundula Hidde: Data curation (equal). Andreas Hertwig: Data curation (equal); Methodology (equal). Paulina Szymoniak: Conceptualization (equal); Data curation (equal); Supervision (equal); Writing – review & editing (equal). Andreas Schönhals: Conceptualization (lead); Project administration (lead); Supervision (lead); Writing – review & editing (equal).

The data are part of a running research project. The data that support the findings of this study are available from the corresponding author upon reasonable request.