Investigating the Nucleation of AlOx and HfOx ALD on Polyimide: Influence of Plasma Activation

There is an increasing interest in atomic layer deposition (ALD) on polymers for the development of membranes, electronics, (3D) nanostructures and specially for the development of hermetic packaging of the new generation of flexible implantable micro-devices. This evolution demands a better understanding of the ALD nucleation process on polymers, which has not been reported in a visual way. Herein, a visual study of ALD nucleation on polymers is presented, based on the different dry etching speeds between polymers (fast) and metal oxides (slow). An etching process removes the polyimide with the nucleating ALD acting as a mask, making the nucleation features visible through secondary electron microscopy analyses. The nucleation of both Al2O3 and HfO2 on polyimide was investigated. Both materials followed an island-coalescence nucleation. First, local islands formed, progressively coalescing into filaments, which connected and formed meshes. These meshes evolved into porous layers that eventually grew to a full layer, marking the end of the nucleation. Cross-sections were analyzed, observing no sub-surface growth. This approach was used to evaluate the influence of plasma-activating polyimide on the nucleation. Plasma-induced oxygen functionalities provided additional surface reactive sites for the ALD precursors to adsorb and start the nucleation. The presented nucleation study proved to be a straightforward and simple way to evaluate ALD nucleation on polymers.


Introduction
Atomic layer deposition (ALD) is a highly conformal thin film deposition technique. It consists of sequential self-limited reactions between two precursors and offers a precise control of the thickness of the layer [1,2]. A wide range of metals and metal oxides can be deposited by ALD, as compiled in the Atomic Limits database [3]. Thermal ALD processes operate at high temperatures, usually in the range of 200-400 • C [4]. Such high temperatures are not compatible with a large majority of polymers. Although there are strategies to lower the deposition temperature, such as using plasma-enhanced ALD (PEALD), herein the focus is on thermal ALD. The development of ALD is intimately related with the microelectronics industry, which in the past decades has been the main driving force of the advances in ALD technology. As such, ALD was first developed as a coating technology on solid inorganic surfaces [5][6][7], such as metals, glass or silicon wafers. At present, the reaction mechanisms during nucleation and growth of ALD materials on these surfaces are well understood. The recent evolution of the electronics and medical device industry ation stage. Some techniques are very advanced, expensive and not easily available (LEIS), while others are limited in their depth of information, being too large for the analysis of the ultrathin ALD layer in its nucleation stage (FTIR). Others do provide precise information about the surface topography in the early nucleation stage, but they fail to detect the closure of a full layer at the end of the nucleation (AFM and STM).
In recent years, ALD has been widely used in hermetic packaging of medical devices, especially for the encapsulation of flexible devices. Traditional implantable medical devices such as pacemakers, cochlear implants or defibrillators have a rigid and rather large control unit. Nonetheless, the recent development of flexible and stretchable electronics has enabled the fabrication of miniaturized and flexible implantable medical devices [72][73][74][75][76][77]. In contrast to traditional rigid housings (sealed metal, ceramic and glass casings), the material selection for the encapsulation of flexible or soft devices is complex, being almost impossible to find all the requirements (biocompatible, biostable, hermetic, flexible) in a single material. To that end, polymers and ALD ceramic films can be combined, where the polymer provides a flexible support to the hermetic ALD films. The brittle ceramic films should be sufficiently thin (in the order of tens of nanometers) for the packaging to be flexible, which can be obtained by atomic layer deposition (ALD) [54][55][56][57][58]. This strategy has already been proven to provide long-term protection (several years) against water permeation at body temperature [78,79].
Polyimide is a high performance polymer, with excellent mechanical resistance, thermal and chemical stability, flexibility, excellent insulating properties and relatively good moisture barrier properties compared to other polymers [80][81][82][83]. This outstanding set of properties make polyimide suitable for a variety of applications ranging from electronics, medicine, membrane separation to aerospace and military industries [81,84]. Certain types of biocompatible polyimides are widely used in the packaging of implantable electronics and electrodes not only for its barrier performance [85][86][87][88][89][90][91][92][93] but also for its compatibility with microfabrication processes, being used as a structural component of devices [94][95][96][97].
Metal oxides such as Al 2 O 3 and HfO 2 are biocompatible, stable under the high temperatures of microfabrication and sterilization processes and, more importantly, excellent barriers, meeting the hermeticity requirement for encapsulating long-term implantable medical devices [98]. Polyimide and ALD deposited Al 2 O 3 /HfO 2 can be combined to develop an ultra-hermetic package [78,79]. Yet, the adhesion between adjacent layers is critical for the package to retain its long-term hermeticity. If the adhesion between the layers is poor, moisture could accumulate at their interface, leading to fast lateral diffusion and an unsatisfactory performance of the barrier. When the adhesion is good, the moisture that permeates through the defects of a layer stops at the interface, unless the defects of the stacked layers are aligned. Even in such an unlikely case, since the barrier is formed by a multilayer stack, the moisture permeation would be interrupted when reaching the next interface. Thus, the nucleation and early growth of the ALD ceramic are key to the polymer/ALD interface since they determine the type of interaction between both materials. Very thin ceramic layers (5-10 nm) are deposited to obtain a hermetic encapsulation for implantable devices [29][30][31][32][33]78,79]. Although single ALD layers provide a good degree of hermeticity that could be enough for short term implantation, the combination of ALD layers provides a high degree of hermeticity as required for long-term implantable devices [78,79,91,99]. For such an application, it is crucial to ascertain the moment at which the ALD nucleation is finished, in other words, to define the minimum amount of ALD cycles necessary to obtain a fully closed and thus hermetic layer of the ceramic material.
Herein, a qualitative study of ALD nucleation on a polymeric substrate was conducted, using an innovative approach to visualize the nucleation of Al 2 O 3 and HfO 2 on a biocompatible polyimide. Selective reactive ion etching (RIE) of polyimide was applied using the nucleating ALD as a mask to reveal the features of ALD nucleation under microscopy inspection. The nucleation of Al 2 O 3 and HfO 2 was studied by field-emission-gun scanning electron microscopy (FEG-SEM) and by XPS analysis. Cross-sections of thick Al 2 O 3 and HfO 2 ALD layers were analyzed by scanning transmission electron microscopy (STEM) and energy-dispersive X-ray spectroscopy (EDS) to elucidate whether surface or subsurface nucleation and growth occurred. Lastly, the surface of polyimide was modified via nonthermal plasma activation under an air atmosphere to evaluate the influence of introducing reactive groups on the surface of polyimide on Al 2 O 3 and HfO 2 ALD nucleation.

Sample Preparation
Polyimide films were prepared by spin coating PI2611 and PI2610 onto 6.5 cm 2 square borosilicate glass with a Polos SPIN200i (SPS Belgium, Herselt, Belgium) in two consecutive steps: 10 s at 1000 rpm and 60 s at 4000 rpm, at an acceleration rate of 500 rpm/s. The spin coating step was followed by a pre-baking step of 10 min at 200 • C on a hot plate and the samples were subsequently cured for 30 min at 200 • C and 60 min at 300 • C in an N 2 atmosphere.
A selection of polyimide substrates was treated with a non-thermal plasma activation process prior to the ALD deposition to study the influence of the surface functionalization on ALD nucleation. The plasma activation was carried out in a home built dielectric barrier discharge (DBD) reactor, where the discharge occurred between two circular copper electrodes (5.5 cm diameter). The electrodes were covered with glass as dielectric material. The lower electrode was connected to ground through a 10.4 nF capacitor, while the upper electrode was connected to an AC high frequency power source (50 kHz). The substrates were placed on the lower electrode and fixed with double sided tape. The plasma activation of polyimide was carried out in a dry air atmosphere. First, the reactor was pumped down to at least 0.05 kPa. Then, air was introduced in the reactor at a flow rate of 1.0 standard litre per min (slm) and the pressure was raised to 5.0 kPa. Plasma was ignited at this moment, with a discharge power of 3.0 W. The samples were exposed to the air plasma for 1.0 s, a previously optimized parameter for BPDA-PPD polyimide [100].
All ALD depositions were carried out in a commercial reactor (Ultratech Savannah S200 G2, Cambridge NanoTech, Cambridge, MA, USA) at 150 • C and 53 kPa. The Al 2 O 3 deposition process consisted of alternating pulses of TMA (0.03 s and 30 s N 2 purge) and H 2 O (0.03 s and 30 s N 2 purge). HfO 2 was deposited by alternating pulses of TDMAH (0.4 s and 60 s N 2 purge) and H 2 O (0.03 s and 60 s N 2 purge). The GPC of the ALD processes was determined by measuring the thickness of layers deposited on witness Si wafers: 1.20 ± 0.02 Å/cycle for Al 2 O 3 and 1.16 ± 0.01 Å/cycle for HfO 2 . The witness Si samples were analyzed with an FS-1EXs ellipsometer (FilmSense, Lincoln, NE, USA). The results were fitted with a Cauchy dispersion model (on Si) to obtain the thickness of the ALD films.
The reactive ion etching (RIE) process was performed in a Plasma Therm Batchtop VII (Belgium) etcher under a N 2 /O 2 mixture atmosphere. The pressure in the chamber was kept constant at 20 Pa and the mixture flows were 5 sccm (standard cubic centimeter per minute) N 2 and 20 sccm O 2 . The plasma was ignited at a power of 75 W and the samples were exposed to the etching process for 5 min.

Characterization
After the RIE step, the samples were inspected with a JEOL JSM-7000F FEG-SEM (JEOL, Tokyo, Japan) (acceleration voltage 2.7 kV) equipped with a secondary electron detector. Secondary electron images were recorded at five different spots randomly distributed in the center of the sample at 20,000× and 50,000× magnification. Prior to imaging, the samples were coated with a thin layer of gold (approximately 20 nm) using a plasma sputter coater. Two samples per condition were examined. All the images were automatically processed with Image J 1.52a software to estimate the area of polyimide covered by the nucleating ALD through particle counting. The particle analysis calculation was performed for the 20,000× magnification images, considering a minimum particle size threshold of 100 px 2 . The values reported per condition are the average of the images taken at five random spots, and two replicates per condition are investigated. The results of pristine and plasma-activated polyimide were analyzed with paired-samples t-test in SPSS 26.0 software, at a significance level of 0.05.
Cross-sections of polyimide/Al 2 O 3 (40 nm)/polyimide and polyimide/HfO 2 (50 nm)/polyimide stacks were analyzed with high angle annular dark field scanning transmission electron microscopy (HAADF-STEM) and EDS. Thin lamella were prepared on a Cu Omniprobe TEM grid, using a Thermo Fisher Helios focused ion beam secondary electron microscope (FIB-SEM, Thermo Fisher, Waltham, MA, USA). Prior to the FIB preparation the samples were coated with gold. The analyses were performed on a Tecnai Osiris TEM (FEI, Hillsboro, OR, USA), equipped with a super-X detector, operated at an acceleration voltage of 200 kV. One sample of Al 2 O 3 ALD and one sample of HfO 2 ALD were analyzed.
In addition to FEG-SEM inspection, the surface chemical composition was studied by XPS analysis. The measurements were performed using a PHI 5000 Versaprobe II (ULVAC-Physical Electronics, Chigasaki, Japan) spectrometer with an Al Kα X-ray source (hν = 1486.6 eV) operating at 25 W. The photoelectrons were detected with a hemispherical analyzer at 45 • to the normal of the samples' surface and the pressure in the chamber was maintained below 10 −6 Pa. Survey scans were recorded with a pass energy of 187.85 eV (0.8 eV step) while high resolution spectra (C1s, O1s, Al2p, Hf4f) were recorded at 23.5 eV (0.1 eV step). The spectra were analyzed using Multipak software (v 9.6.1), applying an iterated Shirley background and correcting the final peak intensities by the relative sensitivity factors provided by the manufacturer. All the spectra were calibrated (C-C at 285.0 eV) and a Savitzky-Golay smoothing procedure was applied prior to any analysis. The instrumental error is 1%. Two samples per condition were analyzed at four randomly located points in the center of the sample. The values reported are the average and standard deviation of these 8 measurements.

Introduction
Instead of an analytical, quantitative approach, herein we propose a visual approach to study the ALD nucleation on polymers. Polymers are organic materials with much faster dry etching (also known as RIE) speeds compared to metal oxides. The approach in this work is based on this difference in etching speed between polyimide and the ALD metal oxide. A closed metal oxide layer has yet to be formed for the nucleation phase of the ALD deposition on the polyimide, so the local metal oxide sites act as a hard mask for the underlying polymer when subjected to RIE. This means that the polyimide, which is not covered by the nucleating metal oxide, is etched away to a certain degree. The RIE step is key for visualizing the features of the ALD nucleation under SEM. The topographical differences between the polymeric substrate and the ultrathin nucleating ALD material are very small, making it impossible to distinguish the ALD material from the polyimide substrate. The RIE process, etching the polyimide not covered by the metal oxide, increases the topographical difference between the polymer and the metal oxide, enabling the analysis of the ALD nucleation behavior under an SEM. The process is schematically  Figure 1. This approach allows to visualize and observe the progression of the nucleation on polymers; however, it is not possible to discern whether surface or subsurface growth occurs, as only a top-view image is obtained with SEM inspection. Subsurface growth and nucleation could be inspected by analyzing a cross-section of the polymer/ALD layers under STEM combined with EDS analysis. not covered by the nucleating metal oxide, is etched away to a certain degree. The RIE step is key for visualizing the features of the ALD nucleation under SEM. The topographical differences between the polymeric substrate and the ultrathin nucleating ALD material are very small, making it impossible to distinguish the ALD material from the polyimide substrate. The RIE process, etching the polyimide not covered by the metal oxide, increases the topographical difference between the polymer and the metal oxide, enabling the analysis of the ALD nucleation behavior under an SEM. The process is schematically represented in Figure 1. This approach allows to visualize and observe the progression of the nucleation on polymers; however, it is not possible to discern whether surface or subsurface growth occurs, as only a top-view image is obtained with SEM inspection. Subsurface growth and nucleation could be inspected by analyzing a cross-section of the polymer/ALD layers under STEM combined with EDS analysis.

ALD Nucleation Study of Al2O3 and HfO2 on Polyimide
Using an optimized RIE process of 5 min, the evolution of Al2O3 and HfO2 nucleation on polyimide was studied by depositing an increasing amount of ALD cycles on a set of substrates followed by FEG-SEM and XPS analyses. Prior to studying the ALD nucleation on polyimide, the surface of pristine polyimide, polyimide after the RIE process and polyimide with a fully closed (c.a. 100 nm) Al2O3 layer on top after RIE were inspected (see Figure 2). As expected, the surface of pristine spin coated polyimide was smooth. After RIE, the surface topography and morphology of polyimide clearly changed, a roughness commonly referred to as grass-like roughness was created as a consequence of the etching process [101]. The surface topography of polyimide with a closed ALD layer on top (Al2O3 100 nm) was the same as that of pristine polyimide, due to the high conformality of ALD depositions combined with the high resistance of the thick AlOx layer to the RIE process.

ALD Nucleation Study of Al 2 O 3 and HfO 2 on Polyimide
Using an optimized RIE process of 5 min, the evolution of Al 2 O 3 and HfO 2 nucleation on polyimide was studied by depositing an increasing amount of ALD cycles on a set of substrates followed by FEG-SEM and XPS analyses. Prior to studying the ALD nucleation on polyimide, the surface of pristine polyimide, polyimide after the RIE process and polyimide with a fully closed (c.a. 100 nm) Al 2 O 3 layer on top after RIE were inspected (see Figure 2). As expected, the surface of pristine spin coated polyimide was smooth. After RIE, the surface topography and morphology of polyimide clearly changed, a roughness commonly referred to as grass-like roughness was created as a consequence of the etching process [101]. The surface topography of polyimide with a closed ALD layer on top (Al 2 O 3 100 nm) was the same as that of pristine polyimide, due to the high conformality of ALD depositions combined with the high resistance of the thick AlO x layer to the RIE process.
not covered by the nucleating metal oxide, is etched away to a certain degree. The RIE step is key for visualizing the features of the ALD nucleation under SEM. The topographical differences between the polymeric substrate and the ultrathin nucleating ALD material are very small, making it impossible to distinguish the ALD material from the polyimide substrate. The RIE process, etching the polyimide not covered by the metal oxide, increases the topographical difference between the polymer and the metal oxide, enabling the analysis of the ALD nucleation behavior under an SEM. The process is schematically represented in Figure 1. This approach allows to visualize and observe the progression of the nucleation on polymers; however, it is not possible to discern whether surface or subsurface growth occurs, as only a top-view image is obtained with SEM inspection. Subsurface growth and nucleation could be inspected by analyzing a cross-section of the polymer/ALD layers under STEM combined with EDS analysis.

ALD Nucleation Study of Al2O3 and HfO2 on Polyimide
Using an optimized RIE process of 5 min, the evolution of Al2O3 and HfO2 nucleation on polyimide was studied by depositing an increasing amount of ALD cycles on a set of substrates followed by FEG-SEM and XPS analyses. Prior to studying the ALD nucleation on polyimide, the surface of pristine polyimide, polyimide after the RIE process and polyimide with a fully closed (c.a. 100 nm) Al2O3 layer on top after RIE were inspected (see Figure 2). As expected, the surface of pristine spin coated polyimide was smooth. After RIE, the surface topography and morphology of polyimide clearly changed, a roughness commonly referred to as grass-like roughness was created as a consequence of the etching process [101]. The surface topography of polyimide with a closed ALD layer on top (Al2O3 100 nm) was the same as that of pristine polyimide, due to the high conformality of ALD depositions combined with the high resistance of the thick AlOx layer to the RIE process.   FEG-SEM images (magnification ×20,000 (top) and ×50,000 (bottom)) of pristine spin coated polyimide (PI), polyimide after the RIE process and polyimide with a closed ALD layer on top after the RIE process. Figure 3 shows the progress of Al 2 O 3 and HfO 2 nucleation on polyimide. The features of the grass-like roughness of polyimide and of the nucleating ALD were also visibly different, especially at higher magnifications, making it possible to differentiate the underlying polyimide (dark grey) from the nucleating ALD (light grey). As the nucleation progressed, the grass-like roughness of polyimide was no longer visible. The features of the nucleating ALD differed considerably from one ALD material to another, to be expected since the adsorption and reaction of the precursors is different for each ALD process. ferent, especially at higher magnifications, making it possible to differentiate the underlying polyimide (dark grey) from the nucleating ALD (light grey). As the nucleation progressed, the grass-like roughness of polyimide was no longer visible. The features of the nucleating ALD differed considerably from one ALD material to another, to be expected since the adsorption and reaction of the precursors is different for each ALD process. At five cycles, islands were observed for both Al2O3 and HfO2 homogeneously distributed over the sample; however, the islands were considerably smaller for HfO2. In the case of Al2O3, the islands were not only larger, but also presented branches. At seven cycles, Al2O3 continued to nucleate on local sites over the sample, forming more islands. Some of them started to connect one to another, forming some sort of filaments. HfO2 nucleation progressed in the same way, but the features were more easily distinguishable as they did not present as many small branches as for Al2O3. HfO2 nucleated on more sites on polyimide, and neighboring islands merged into very well defined filaments. At 10 cycles, the difference between Al2O3 and HfO2 were even more obvious. For Al2O3, barely any change compared to seven cycles could be appreciated, but the filaments observed at At five cycles, islands were observed for both Al 2 O 3 and HfO 2 homogeneously distributed over the sample; however, the islands were considerably smaller for HfO 2 . In the case of Al 2 O 3 , the islands were not only larger, but also presented branches. At seven cycles, Al 2 O 3 continued to nucleate on local sites over the sample, forming more islands. Some of them started to connect one to another, forming some sort of filaments. HfO 2 nucleation progressed in the same way, but the features were more easily distinguishable as they did not present as many small branches as for Al 2 O 3 . HfO 2 nucleated on more sites on polyimide, and neighboring islands merged into very well defined filaments. At 10 cycles, the difference between Al 2 O 3 and HfO 2 were even more obvious. For Al 2 O 3 , barely any change compared to seven cycles could be appreciated, but the filaments observed at seven cycles in HfO 2 samples had connected with each other, forming meshes. At 13 cycles, the nucleating Al 2 O 3 had formed an incipient mesh structure, while for HfO 2 a porous layer was already formed. From 13 to 15 cycles, Al 2 O 3 nucleation progressed substantially and grew into a porous layer. The HfO 2 film presented visibly smaller pores at 15 cycles. For Al 2 O 3 , the pores were round and homogeneous in shape and size. On the other hand, the pores of HfO 2 had various sizes and shapes, more elongated rather than round. Eventually, at 20 cycles, a fully closed layer was obtained for both ALD processes, marking the end of the nucleation stage.
Despite the differences, the same phases could be identified in the nucleation of both materials. Initially, Al 2 O 3 and HfO 2 ALD processes nucleated on specific sites homogeneously distributed over the polyimide substrate, forming isolated islands. As more cycles were being deposited, the islands grew into filaments. There are two possible routes of formation of these filaments: (i) Precursor molecules adsorbed on reactive sites that were still available forming islands next to the already existing ones until these coalesced transforming into the filamentous shapes; (ii) Lateral growth of the islands occurred through random growth of Al 2 O 3 or HfO 2 on the already grown islands rather than through the adsorption of the precursors on polyimide. Via visual inspection, it is difficult to determine which of the two routes occurred, but most likely both mechanisms contributed to the early stage nucleation. These filaments grew with each ALD cycle and coalesced forming mesh structures, which progressively gained density leading to the formation of porous films. Eventually, the pores closed and a continuous, pin-hole free layer was formed at 20 cycles, both for Al 2 O 3 and HfO 2 . At the last stages of the nucleation, the growth mechanisms were dominated by the second mechanism of lateral growth on the already deposited Al 2 O 3 and HfO 2 , as the likelihood of reactive sites on the polyimide to be vacant or even physically available was rather low. These observations clearly pointed to a nucleating mechanism based on the adsorption of precursors at reactive sites of the polyimide and further lateral growth, commonly referred to as island-coalescence nucleation or growth [34,65,70,71,102,103]. Island growth of Al 2 O 3 on H-terminated Si was already observed by selective etching of SiOx through the "defects" of the nucleating Al 2 O 3 layer [104].
The recorded FEG-SEM images were processed with Image J software to estimate the area of polyimide covered by the nucleating ALD, as a way to evaluate the evolution of the nucleation in a qualitatively but numerical manner. Figure 4 shows how the ALD coverage increased as the nucleation progressed, that is, as more cycles were being deposited. Although the features of the nucleating ALD were different for Al 2 O 3 and HfO 2 , the progress of ALD coverage of polyimide was similar, as depicted in Figure 4. At five cycles, the coverage was estimated to be 24.3 ± 0.4% for Al 2 O 3 and 22.0 ± 2.4% for HfO 2 during the islands stage. At seven cycles, the coverage increased to 34.4 ± 0.6% (Al 2 O 3 ) and 32.3 ± 2.5% (HfO 2 ). At 10 and 13 cycles, the coverage of the nucleating ALD was higher for HfO 2 (42.9 ± 2.3% and 55.7 ± 2.2%, respectively) than for Al 2 O 3 (34.5 ± 1.0% and 48.9 ± 2.2%, respectively), but both presented a similar value after 15 cycles (74.7 ± 6.2% for Al 2 O 3 , 74.1 ± 2.9% for HfO 2 ).  In addition to the visual study of the nucleation, XPS analyses were also performed. The surface concentrations of Al and Hf (Figure 5a) were evaluated over an increasing amount of Al2O3 and HfO2 ALD cycles, respectively. The concentration of both elements progressively increased as more cycles were being deposited, following the same trend as An approximate range of the area of polyimide covered by the nucleating ALD could be derived for each stage of the nucleation based on these results. During the islands phase, the coverage was within 0-30%, increasing to 31-40% when filaments were formed. The coalescence of the filaments into meshes corresponded to a coverage of 41-50%, while the coverage of the porous layers was within a broader range, namely 51-99%.
In addition to the visual study of the nucleation, XPS analyses were also performed. The surface concentrations of Al and Hf (Figure 5a) were evaluated over an increasing amount of Al 2 O 3 and HfO 2 ALD cycles, respectively. The concentration of both elements progressively increased as more cycles were being deposited, following the same trend as the ALD coverage values. For Al 2 O 3 , at the very early nucleation (5-10 cycles) the Al concentration barely changed, and remained within 6-8 at.%, corresponding to the island/filament stages, as observed in Figure 3. Conversely, the concentration of Hf increased after each amount of cycles: 3.0 ± 0.1 at.% at five cycles (island stage), 4.6 ± 0.1 at.% at seven cycles (filament stage) and 6.9 ± 0.1 at.% at 10 cycles (mesh stage). The Al content went up to 10 . The depth of information of the XPS technique at the conditions that the measurements were performed is approximately 5-10 nm for polymer surfaces. Therefore, if a thin ALD layer of 1-2 nm deposited on polyimide is being measured, the signal of the underlying polyimide will still be present and considerably affect the elemental composition that is measured. Once the nucleation is finished and more cycles are deposited, the thicker the Al 2 O 3 or HfO 2 layer becomes, and the lower the signal of the underlying polyimide is, until only the signal of the ALD layers is being detected and a constant Al or Hf concentration is observed (not shown).  In conclusion, it can be stated that the XPS and the FEG-SEM imaging information presented parallel trends and matching results. This agreement confirmed that the RIE process did not significantly damage the deposited ALD, as otherwise there would have been a considerable disagreement between the observations from both techniques. This proved that the visualization approach was a reliable way of studying the nucleation of ALD layers on polymeric substrates. However, as already mentioned in the introduction section, the approach is limited to a top-view imaging of the nucleating process, not being possible to identify whether surface or sub-surface ALD growth is occurring.
For that reason, cross-sections of polyimide/Al 2 O 3 (40 nm)/polyimide and polyimide/HfO 2 (50 nm)/polyimide stacks were analyzed with HAADF-STEM and EDS. Figure 6 shows the analysis of Al 2 O 3 and HfO 2 cross-sections. The HAADF-STEM images show the Al 2 O 3 or HfO 2 ALD layer sandwiched between the thicker polyimide layers. Polyimide is an oxygen-containing polymer; therefore, the oxygen signal (in blue) is present through the whole area of analysis in the EDS maps, but it is much more intense on the area corresponding to the Al 2 O 3 and HfO 2 layers, where the respective Al or Hf are also present. There were no signs of sub-surface growth of the ALD, as a sharp interface between polyimide and the ALD layers was observed. These observations thus imply that the nucleation and growth of both Al 2 O 3 and HfO 2 ALD processes occurred at the surface of polyimide only. The polyimide used in this work (BPDA-PPD polyimide) inherently contains reactive groups (C=O, C-O, N-C=O), which are consequently available at the surface as reactive sites for the nucleation during the ALD process. It seems that the nucleation started by the adsorption of precursor molecules at those reactive sites.

Influence of Plasma Activation on the ALD Nucleation
Previous research has shown that surface functionalization of the substrate prior to ALD deposition provides (additional) reactive sites during ALD nucleation [44][45][46][47][48][49]. The HAADF-STEM cross-section analysis revealed that no sub-surface growth occurred as pristine BPDA-PPD polyimide has reactive groups available for the nucleation of Al2O3 and HfO2. Here, the influence of plasma-activating polyimide prior to the ALD deposition on the nucleation behavior was explored. The additional reactive sites introduced by the plasma activation can increase the formation of covalent bonds between polyimide and Al2O3 or HfO2 during the nucleation. A higher chemical interaction between both layers would increase their adhesion in a multilayer barrier. Polyimide surface was activated under air plasma and an increasing amount of Al2O3 or HfO2 ALD cycles was deposited on it. Figures 7 and 8 show the evolution of Al2O3 and HfO2 nucleation, respectively, on both plasma-activated and pristine polyimide. The surface characterization of the activated polyimide with the plasma process used in this work

Influence of Plasma Activation on the ALD Nucleation
Previous research has shown that surface functionalization of the substrate prior to ALD deposition provides (additional) reactive sites during ALD nucleation [44][45][46][47][48][49]. The HAADF-STEM cross-section analysis revealed that no sub-surface growth occurred as pristine BPDA-PPD polyimide has reactive groups available for the nucleation of Al 2 O 3 and HfO 2 . Here, the influence of plasma-activating polyimide prior to the ALD deposition on the nucleation behavior was explored. The additional reactive sites introduced by the plasma activation can increase the formation of covalent bonds between polyimide and Al 2 O 3 or HfO 2 during the nucleation. A higher chemical interaction between both layers would increase their adhesion in a multilayer barrier.
Polyimide surface was activated under air plasma and an increasing amount of Al 2 O 3 or HfO 2 ALD cycles was deposited on it. Figures 7 and 8 show the evolution of Al 2 O 3 and HfO 2 nucleation, respectively, on both plasma-activated and pristine polyimide. The surface characterization of the activated polyimide with the plasma process used in this work has been described in detail elsewhere [64]. The air plasma activation introduced oxygen-containing groups such as C-O, C=O and O-C=O on the surface of polyimide, increasing the surface oxygen content by approximately 10 at.%. These oxygen functional groups were found to act as nucleating sites for both Al 2 O 3 and HfO 2 ALD processes, as evidenced by the images shown in Figures 7 and 8. 1, x FOR PEER REVIEW 13 of 20 13 cycles on pristine polyimide. Likewise, the coverage of 10 (from 34.5 ± 1.0% to 44.5 ± 0.8%) and 13 cycles (from 48.9 ± 2.2% to 56.9 ± 1.3%) also significantly increased on plasmaactivated polyimide (see Figure 4). After 15 cycles, a porous Al2O3 layer had been formed on both plasma-activated and pristine polyimide. The pores on plasma-activated polyimide were smaller, and of different sizes, as they were being progressively closed. The coverage after 15 cycles was very similar, 70.0 ± 4.5% on plasma-activated polyimide and 74.7 ± 6.2% on pristine polyimide. A fully closed, pin-hole free layer was formed at 20 cycles on either plasma-activated or pristine polyimide. The differences of the nucleation on plasma-activated and pristine polyimide were more obvious in the case of HfO2 ALD. While only HfO2 islands were formed after five cycles on pristine polyimide, filaments were already visible on plasma-activated polyimide (see Figure 8), reaching a significantly higher coverage (27.0 ± 1.1% compared with 22.0 ± 2.4%). After a few more cycles, at seven cycles, HfO2 had grown into a mesh on

Conclusions
In this research, an approach to visualize the nucleation of ALD processes on polymeric substrates was introduced. It is based on the dry etch selectivity between the deposited ALD metal oxide and the polymer substrate under an RIE process.
The approach was successfully applied to study the nucleation of Al2O3 and HfO2 ALD on polyimide, in the framework of developing flexible and hermetic barriers for the packaging of implantable devices. Al2O3 and HfO2 ALD processes produced a full, pinhole free layer on polyimide after 20 cycles. Both processes presented an island-coalescence nucleation model, that is, the nucleation started at reactive sites forming islands and further lateral growth led to coalescence of the islands through the progressive formation of filaments-meshes-porous layers, and eventually pin-hole free layers. This evolution could also be observed in the surface Al and Hf content, and in the O1s, Al2p and Hf4f high resolution XPS spectra.
Cross-sections of thick Al2O3 (40 nm) and HfO2 (50 nm) layers embedded in polyimide were analyzed by HAADF-STEM and EDS, revealing no sign of sub-surface growth. The features of the nucleating ALD on plasma-activated polyimide were visible on a more advanced nucleation stage compared with pristine polyimide. In Figure 4 it can also be observed that Al 2 O 3 and HfO 2 ALD coverage was higher on plasma-activated polyimide, except for 15 cycles of Al 2 O 3 . In the case of Al 2 O 3 , the influence of the plasma activation was not so evident at lower amounts of cycles (see Figure 7). At five and seven cycles, it was difficult to distinguish differences in the FEG-SEM images of plasma-activated and pristine polyimide. The islands were slightly larger and filaments were starting to form on the plasma-activated polyimide. The coverage of Al 2 O 3 ALD was almost the same on plasma-activated polyimide as on pristine polyimide after seven cycles, 34.2 ± 1.2% and 34.5 ± 1.0%, respectively. On the other hand, the coverage after five cycles was significantly higher on plasma-activated polyimide (25.5 ± 1.0% versus 24.3 ± 0.4%). As more cycles were deposited, differences between Al 2 O 3 nucleation on plasma-activated and pristine polyimide became more patent. After 10 and 13 cycles, Al 2 O 3 had already formed a mesh on plasma-activated polyimide while it was only beginning to do so after 13 cycles on pristine polyimide. Likewise, the coverage of 10 (from 34.5 ± 1.0% to 44.5 ± 0.8%) and 13 cycles (from 48.9 ± 2.2% to 56.9 ± 1.3%) also significantly increased on plasma-activated polyimide (see Figure 4). After 15 cycles, a porous Al 2 O 3 layer had been formed on both plasma-activated and pristine polyimide. The pores on plasma-activated polyimide were smaller, and of different sizes, as they were being progressively closed. The coverage after 15 cycles was very similar, 70.0 ± 4.5% on plasma-activated polyimide and 74.7 ± 6.2% on pristine polyimide. A fully closed, pin-hole free layer was formed at 20 cycles on either plasma-activated or pristine polyimide.
The differences of the nucleation on plasma-activated and pristine polyimide were more obvious in the case of HfO 2 ALD. While only HfO 2 islands were formed after five cycles on pristine polyimide, filaments were already visible on plasma-activated polyimide (see Figure 8), reaching a significantly higher coverage (27.0 ± 1.1% compared with 22.0 ± 2.4%). After a few more cycles, at seven cycles, HfO 2 had grown into a mesh on plasma-activated polyimide, while only filaments were observed on pristine polyimide. As more HfO 2 cycles were deposited, the nucleation on plasma-activated polyimide remained in a more advanced stage, showing a denser mesh at 10 cycles, and less porous layers at 13 and 15 cycles. HfO 2 coverage (see Figure 4) was also considerably and significantly higher on plasma-activated polyimide: 45.6 ± 0.9% versus 42.9 ± 2.3% (10 cycles), 68.6 ± 2.8% versus 55.7 ± 2.2% (13 cycles) and 86.6 ± 4.5% versus 74.1 ± 2.9% (15 cycles). After 20 cycles a fully closed HfO 2 layer had been formed on both plasmaactivated and pristine polyimide.
The marked improvement of HfO 2 nucleation on plasma-activated polyimide, in comparison to Al 2 O 3 nucleation on plasma-activated polyimide, denotes a high influence of the surface composition of the substrate on the HfO 2 ALD process. TDMAH nucleation mechanisms and early growth seem more susceptible to the surface composition of the substrate, that is, to the availability of reactive sites. This could be related to the nucleation and early growth mechanisms of both materials. Both Al 2 O 3 and HfO 2 initially nucleate on the reactive sites at the surface of polyimide, but, if after the very first cycles, TMA molecules preferentially react with the already grown clusters of Al 2 O 3 , further growth of Al 2 O 3 would not be considerably influenced by the availability of surface reactive sites. Whereas, if HfO 2 nucleation is mostly driven by the chemisorption of TDMAH molecules on surface reactive sites of polyimide, it would be largely influenced by the availability of these reactive sites.
The coverage of the nucleating ALD was higher on plasma-activated polyimide, and the FEG-SEM imaging revealed a more advanced stage of the nucleation on plasmaactivated polyimide than on pristine polyimide. However, despite these observations, Al 2 O 3 and HfO 2 nucleations were finished at 20 cycles on both plasma-activated and pristine polyimide. This does not necessarily mean that the number of cycles needed to complete the nucleation stage is not affected, but rather that 15-20 cycles is a wide range and the exact number of cycles to obtain a pin-hole free Al 2 O 3 or HfO 2 layer on plasmaactivated polyimide was not determined. Based on the observed results, the availability of functional groups at the surface of polyimide influences the nucleation, not in terms of speed, but in terms of the chemical interaction between the ALD layers and the polymer. A higher concentration of functional groups at the surface of the polymer provides more reactive sites for the nucleation, increasing the amount of covalent bonds formed between the polymer and ALD layers. The oxygen groups introduced by the plasma activation acted as reactive sites for the nucleation, resulting in an increased chemical linkage between the successive ALD layer and the polymeric substrate. This chemical interaction can result in a better adhesion between the multiple layers of hermetic packaging based on polymers and ALD metal oxides [105].

Conclusions
In this research, an approach to visualize the nucleation of ALD processes on polymeric substrates was introduced. It is based on the dry etch selectivity between the deposited ALD metal oxide and the polymer substrate under an RIE process.
The approach was successfully applied to study the nucleation of Al 2 O 3 and HfO 2 ALD on polyimide, in the framework of developing flexible and hermetic barriers for the packaging of implantable devices. Al 2 O 3 and HfO 2 ALD processes produced a full, pinhole free layer on polyimide after 20 cycles. Both processes presented an island-coalescence nucleation model, that is, the nucleation started at reactive sites forming islands and further lateral growth led to coalescence of the islands through the progressive formation of filaments-meshes-porous layers, and eventually pin-hole free layers. This evolution could also be observed in the surface Al and Hf content, and in the O1s, Al2p and Hf4f high resolution XPS spectra.
Cross-sections of thick Al 2 O 3 (40 nm) and HfO 2 (50 nm) layers embedded in polyimide were analyzed by HAADF-STEM and EDS, revealing no sign of sub-surface growth. This confirmed the initial surface nucleation of the ALD processes at the reactive groups inherently present on polyimide.
For the envisioned application of the barriers, the adhesion at the polymer/ALD interface is key. For that reason, the beneficial influence of plasma activating the surface of polyimide prior to the ALD depositions was explored. The oxygen groups introduced by the plasma activation (C-O, C=O, O-C=O) acted as additional reactive sites for the adsorption of the precursor molecules, as evidenced by the FEG-SEM images that revealed a more advanced stage of the nucleation on plasma-activated polyimide than on pristine polyimide, for the same number of cycles. This ensured the formation of additional covalent bonds between the polymer and the successive ALD layers.
The presented results can be considered as one step forward in order to obtain a better insight and understanding of the nucleation behavior of ALD metal oxides on a biocompatible polyimide. However, the exact influence on the improvement of the barrier properties of metal oxide-polymer stacks still needs to be determined.