Low-temperature ALD of metallic cobalt using the CoCOhept precursor: Simulation-assisted process development for deposition on temperature sensitive 3D-structures

Atomic layer deposition (ALD) of metals is an ongoing topic of research. Especially, metallic cobalt is a material of interest for deposition. A thin cobalt layer can be used as a seed layer for electrochemical deposition of copper.¹ It is a suitable material to replace copper interconnects in current technology nodes.² Cobalt layers play an important role in magnetic sensor systems.³ A further application of metallic cobalt thin films is the use as antibacterial coating.⁴ This also involves deposition on temperature sensitive materials, which may not be exposed to temperatures above 150  $°$C.

In former works, we demonstrated the ALD process of CoCOhept for metallic cobalt that operates at temperatures in the range from 50 to 110  $°$C.⁵ Within this work, we show how the simulation of the reaction chemistry supported the work and was confirmed by experiments. Especially for the identification of a suitable precursor for low-temperature processing, the work is accompanied by density functional theory (DFT) calculations. Precursor screening was applied to a set of similar cobalt precursors presented in the work of Georgi et al.⁶ The so selected precursor dicobalt-hexacarbonyl-n-heptyne $Co 2 ( CO ) 6 HC≡ CC 5 H 11$—in short CoCOhept—was used for further process optimization.

Computational fluid dynamic (CFD) simulations accompanied the experimental process optimizations. The motivation for this investigation was to identify the qualitative influence of the chamber height on the precursor transport and, therefore, support the experimental process development. We demonstrate the optimized low-temperature process on silicon deep trench structures and on structured temperature sensitive photoresist structures.

The precursor CoCOhept $[ Co 2 ( CO ) 6 HC≡ CC 5 H 11]$ was synthesized according to Georgi et al.⁶ and Billington et al.⁷ After synthesis, CoCOhept was stored under argon atmosphere in a standard bubbler type stainless steel vessel including a supersonic level sensor. The bubbler was heated to 30  $°$C by an external heating jacket to provide a sufficient vapor pressure. The internal bubbler temperature has been measured using an internal thermocouple. The precursor was bubbled with 100 SCCM, purified argon. This flow was varied for the dosing experiments between 200 and 300 SCCM.

The deposition has been done within a scia Atol 200 machine. As described in previous work,⁵ the scia Atol 200 is a 200 mm single wafer reactor with a variable chamber height. It can be operated with a capacitively coupled plasma at 13.56 MHz. The plasma source was provided by FAP. Each ALD cycle consists of four pulses: precursor dosing (0.2–10 s), precursor purge (1 s), $H 2$ plasma pulse (1–8 s), and $H 2$ purge step (1 s). During the plasma pulse, purified hydrogen is injected into the chamber and the plasma is ignited for the whole pulse duration. The processes were done at a substrate temperature of 85  $°$C.

The process development was done on standardized 200 mm Si (100) wafers with a thermally grown 100 nm silicon oxide. The photoresist testing structures were fabricated using an OiR974-09 resist by Fujifilm with a nominal thickness of 0.9  $μ$m. The pattern was created by i-line projection lithography using an Nikon NSR2205i11D stepper. These photoresist testing structures were then coated with the cobalt film via ALD at 85  $°$C. For this deposition, the precursor pulse was set to 6 s and plasma pulse length to 2 s, respectively.

The HAR testing structures were etched after the lithography steps at an Applied Materials Centura to open the 100 nm $SiO 2$ hard mask layer. The deep trench etching was done on SPTS Omega followed by a polymer and resist removal process with oxygen plasma.

The film thicknesses has been measured within the chamber with the iSE ellipsometer, and externally with a RC2 ellipsometer, both by J.A. Woollam Co. Both systems used an angle of incidence of $70 °$⁠. The film thickness was determined using a combination of a Drude–Lorentzian model for the metallic part and a Tauc–Lorentzian model for possible oxidizations.^5,8,9

The thermogravimetry and mass spectrometry (TG-MS) measurements were performed on an TGA/DSC1 1600 system from Mettler Toledo with MX1 balance coupled with a Pfeiffer Vacuum-Thermostar GSD 301 T2 mass spectrometer. The results were obtained with a heating rate of 5 K/min under a flow of argon 60 ml/min for inert conditions.

The detailed structure analysis has been done with a Zeiss Supra scanning electron microscope (SEM). The images were created by the use of an electron energy of 2 keV on resist samples and 3 keV on silicon samples using the in-lens detector. The Supra is equipped with an x-ray detector to enable energy dispersive x-ray spectroscopy (EDX). The analysis took place using an electron energy of 19 keV.

Surface morphology had been measured using an FRT MicroProf300 tool. The measurement is based on an highly achromatic lens and a spectrometer giving a surface morphology information based on the wavelength with the highest reflection.

The sheet resistance of the deposited cobalt films had been measured using an Polytec automatic four-point probe 280I using a head with tip radius of 500  $μ$m.

The presented experimental work was supported by different modeling approaches to gain further process insights. Reactor scale simulations were used to investigate the precursor distribution insight the reaction chamber. In addition, atomistic simulations were utilized to better understand the chemical processes on the wafer surface.

The used ALD equipment scia Atol 200 allows for a variable reaction chamber height. The height is adjustable in the range of 5–20 mm. To assist in the process optimization of the Co ALD process, reactor scale process simulations based on computational fluid dynamics (CFD) were used. This allowed to investigate and characterize the influence of different chamber heights on the precursor distribution.

The commercial finite volume solver CFD-ACE+ Version 2021.0 from ESI, now Applied Materials, was used.¹⁰ 

For this investigation, we simulated a steady state in the chamber, which is reached quickly after starting the precursor dosing step. This approach is sufficient as it allows to qualitatively characterize the effect of different chamber heights.

The reaction chamber geometry for a chamber height of 20 mm is shown in Fig. 1. The processing gases flow into the chamber using a showerhead, which is implemented as multiple circular inlets here. Due to the showerhead symmetry, only one third of the volume is modeled. The outlet is a circular ring located at the bottom.

Meshes of 3.5–6 $× 10 6$ tetrahedral cells, depending on the chamber height, were used to ensure mesh-independent solutions.

To simulate the dosing step, an initial uniform gas of pure argon was defined. During the simulated dosing step, 200 SCCM of gas containing argon and precursor is flowing in. While the exact amount of precursor is difficult to determine, a reasonable value of 0.04 mol fraction was chosen in this investigation. A process pressure of 50 Pa at the outlet was defined. The wafer temperature was 85  $°$C, while all other wall temperatures and the gas inflow temperature were set to 30  $°$C.

In addition to CFD simulations on the macroscopic scale, the adsorption and surface reactions of Co precursors on the Co surface were studied using DFT at the atomistic scale. All calculations were performed using the Quantum-Espresso code.¹¹ Our DFT calculations incorporate the Perdew–Burke–Ernzerhof (PBE) functional within the generalized gradient approximation for exchange and correlation interactions.¹² Valence electron wave functions and charge densities were expanded using a plane-wave basis set with kinetic energy cut-offs of 35 and 280 Ry, respectively. Electron-ion interactions were described using the Garrity–Bennett–Rabe–Vanderbilt (GBRV) ultrasoft pseudopotential, with valence electron configurations of C: $2 s 22 p 2$⁠, H: $1 s 1$⁠, Si: $3 s 23 p 2$⁠, and Co: $3 s 23 p 63 d 74 s 1$⁠.^13,14 To account for dispersion interactions, we used the spin-polarized van der Waals functional (svdw-DF).¹⁵ Geometric structures were optimized using the BFGS method, with convergence thresholds set to $1× 10 − 4$ Ry for total energy and $1× 10 − 3$ Ry/bohr for forces. The minimum energy pathways and corresponding saddle points were investigated using the CI-NEB method with eight images and a force threshold of 0.05 eV/Å.^16,17 The Co surface was modeled using a four-layer slab with a p(5 $×$ 5) supercell, focusing on the most stable (001) surface. We considered different Co precursors of type $[ Co 2 ( CO ) 6 R 1 C≡ CR 2]$⁠. Among them, precursor with the propyne $[ HC≡ CCH 3]$ has the simplest structure and has been studied considerably.^18,19 Thus, it was used for our subsequent DFT calculations on the surface reactions of Co ALD.

The work of Georgi et al. led to a variety of precursors with varying substituents which have been evaluated for CVD processing.⁶ To develop and optimize an ALD or CVD process, it is important to understand the detailed surface reactions at the microscopic level. Our previous experimental work⁵ on Co ALD has mainly focused on the growth rate and characterization of Co thin films rather than on the chemical mechanisms. In a typical ALD process, the first step involves the chemisorption of a precursor molecule on the substrate surface. An ideal precursor should readily adsorb on the surface and remain intact until the introduction of the coreactant (e.g., hydrogen). However, in practice, the surface chemistry of metal precursors is often more complex, especially with transition metal substrates. Ligands of these precursors may decompose on the surface, introducing impurities into the film.²⁰ In this subsection, DFT calculations are performed to explore exemplarily on the surface chemistry of the precursor $Co 2 ( CO ) 6 HC≡ CCH 3$ for both ALD half cycles. This alkyne-based metal-organic compound is assumed to be suitable as simplified precursor for all the shown Co precursors from Table I. Furthermore, Gibbs free energy for this and other previously reported precursors for the second ALD half cycle were summarized and used for the assessment of their usability in ALD processes.

For an efficient low-temperature ALD process of metallic cobalt, the precursor must demonstrate stability on the surface and high reactivity with the coreactant in the second half-cycle. To assess precursor reactivity, we use the reaction energy between the ligand and adsorbed H atoms as a descriptor. Based on the work of Georgi et al.,⁶ a set of alkynes [ $R 1 C≡ CR 2$] was selected for computational evaluation: [ $HC≡ CCH 3$], $[ ( CH 3 ) 3 SiC≡ CSi ( CH 3 ) 3]$⁠, $[ HC≡ CSi ( CH 3 ) 3]$⁠, $[ C 2 H 5 C≡ CC 2 H 5]$⁠, and $[ HC≡ CC 5 H 11]$⁠. The calculated Gibbs free energies for these reactions are presented in Table I. At 0 K, all precursors show unfavorable (positive) reaction energies, but as the temperature increases, the Gibbs free energies decrease. Among the evaluated alkynes, $( CH 3 ) 3 SiC≡ CSi ( CH 3 ) 3$ demonstrates the lowest reactivity and is not suitable for Co ALD. In contrast, $HC≡ CC 5 H 11$ exhibits the highest reactivity toward $H ∗$⁠, making it a promising candidate for experimental validation. This precursor is referred to as CoCOhept.

As mentioned before, the work of Georgi et al. led to a variety of precursors with varying substituents which have been characterized by thermogravimetry (TG) among others.⁶ Thermogravimetric measurements provided vapor pressure data that allowed to choose the optimal evaporation temperature. Whether the observed temperature of thermal degradation is usable for chemical vapor deposition (CVD) processes was proven and the Co precursor CoCOhept was found to be suitable based on the TG results. More than 20 wt. % residue was found. This weight loss exceeds the theoretical amount of remaining cobalt due to sublimation without decomposition of the precursor as a strong influence of the heating rate on the evaporation of the compound.⁶ With the chosen Co precursor, CoCOhept TG-MS measurements were performed within this work for comparison with simulated reaction chemistry for the equivalent alkyne-based metal-organic precursor $Co 2 ( CO ) 6( HC≡ CCH 3)$⁠. The results of the TG measurement of CoCOhept are shown in Fig. 5 and summarized in Table II. The first mass loss occurs at around 85  $°$C concomitant with the appearance of a peak at m/z 28. This can be attributed either to the loss of carbonyl ligands in accordance to the simulation results or to epsilon-cleavage from the $C 7$ chain by liberation of ethylene. While CO release is commonly observed for carbonyl complexes, a final statement cannot be made here as it requires high-resolution mass spectrometry.

Parallel to the appearance of m/z = 28 between 85 and 220  $°$C also m/z of 29 as well as 12 are observable accounting for $C 2 H 5 +$ and $C +$⁠, respectively, as break-down products from heptyne ligand. $CH 3 +$ with m/z = 15 is observed at temperatures above 250  $°$C and originate from further decomposition of the heptyne ligand. Previously, our saturation studies of this precursor on the wafer surface revealed the layer growth per cycle in dependency of different precursor pulse times. The results show an ALD-like saturation behavior at 85 and 90  $°$C after a certain pulse length. Furthermore, the hydrogen plasma pulse is necessary to generate metallic cobalt, as could be shown with XPS studies within this work. Otherwise, mainly cobalt oxide layers were deposited as shown for a process temperatures of 150  $°$C.⁵ In summary, the continuous mass loss of the precursor is the result of the evaporation of the precursor molecule as well as the thermal decomposition under release of ligands. An ALD process forming uniform metallic cobalt thin films is possible with the precursor CoCOhept combined with a hydrogen plasma step in the second half-cycle and low process temperatures between 50 and 110  $°$C.

Further precursor properties are necessary to enable reactor scale simulations presented in Sec. IV C. To model this ALD process, the precursor properties for the new precursor CoCOhept are required. The thermodynamic data are extracted from quantum chemical model and statistical mechanics.²¹ The parameters for the NASA polynomials²² describing the thermodynamic data of CoCOhept are $a 1=5.93865732$⁠, $a 2=1.80983200× 10 − 1$⁠, $a 3=−2.08512746$ $× 10 − 4$⁠, $a 4=1.41000676× 10 − 7$⁠, $a 5=−4.19807001× 10 − 11$⁠, $a 6=69676.4546529$⁠, and $a 7=9.24374276$⁠. The transport properties are calculated from Lennard–Jones potential in our simulation software. The Lennard–Jones parameters $ϵ/ k b$ and $σ$ are estimated using the method of corresponding states²³ at the boiling point. The Lennard–Jones parameters $ϵ/ k b$ and $σ$ for the new precursor CoCOhept are 470 K and 7.95 Å, respectively.

As the right precursor had been found, the necessary conditions of the system components for an optimum coating result can be considered. Using CFD simulations as described in Sec. III, the influence of chamber height on the precursor distribution inside the chamber is investigated. Figure 6(a) shows the precursor partial pressure on the wafer surface for a chamber height of 5 mm. The showerhead gas outlet positions are marked with small black circles. A local increase in precursor partial pressure below each showerhead outlet is clearly visible. The outlet positions from the showerhead have a high influence on the precursor distribution.

In contrast, the precursor distribution within a chamber with 15 mm height [Fig. 6(b)] shows a completely different distribution. The precursor concentration has a continuous gradient from the wafer center to the edge. There are no local areas with increased precursor partial pressure. The showerhead outlets have no significant influence on the local partial pressure distribution.

These simulation results could be verified with deposition experiments. The homogeneous continuous film deposition could be shown for processes in a chamber with 15 mm height. Figure 6(d) shows a photo of a wafer processed under these conditions. The metallic film is clearly visible on the underlying dark blue colored 100 nm $SiO 2$ film. The film grows continuously and no effect of the showerhead is visible.

In contrast, the wafer processed at a lower chamber height of 10 mm [Fig. 6(c)] shows a clear influence of the showerhead on the film homogeneity and thickness distribution. The photo has been optimized regarding brightness and contrast in order to emphasize the local influences of the showerhead.

The feature size of the local differences is lower than the measurement spot of the ellipsometer. To still get a value of the local distribution of the coated cobalt film, we used the FRT MicroProf300 surface inspection tool with a lateral resolution below 2  $μ$m. Figure 7 shows surface measurement results of the processed wafer shown in Fig. 6(c), the wafer which was processed at a low chamber height of 10 mm. The wafer was measured entirely with measurement point distance of 1 mm.

The measurement provides the surface profile of the wafer according to the most reflective wavelength passing the achromatic sphere at the measurement head. The polished wafer with the thermally grown $SiO 2$ is sufficiently flat that reference wavelength will not change significantly. The second value the measurement provides is the intensity of the reflected light at the measurement wavelength. We use this value in order to get a rough estimation about the amount of deposited cobalt. This can be done as the 100 nm thermally grown $SiO 2$ layer has a low reflectance in a wide wavelength range. In contrast, the metallic cobalt film has a significant higher reflectance.

The plotted wafer map of Fig. 7 verifies the calculations shown in Fig. 6(a). The areas with locally high precursor partial pressure are accompanied by a higher cobalt deposition. We assume that the other areas suffer from an incomplete precursor saturation resulting in a lower growth rate. Further processes were done on increased chamber height of 15 mm.

The optimized precursor flow conditions are the base for further process optimizations. Previous experiments⁵ showed that the precursor dosing pulse requires a substantial time of at least 5 s to achieve a saturated surface while the purging times could be reduced to 1 s or less. The amount of evaporated precursor is—in ideal case—proportional to the carrier gas flow through the bubbler evaporator.²⁵ Using this dependency, we varied the amount of carrier gas in order to investigate the influence of an increased precursor dose to the pulse timing behavior.

Figure 8 shows the influence of the carrier gas flow rate to the saturation behavior. The plot shows the growth per cycle for different dosing times. Each curve represents the experiment on one wafer. The wafer is precoated with 100 cycles, prior the dosing time tests, to suspend the substrate influence. The data points have been fitted to an exponential decay curve of type $r(t)= r 0⋅ ( 1 − e − a ⋅ t )$ adapting the assumption of Tuomo Suntola [Eq. (18), the probability of surface occupation].²⁶ The growth per cycle $r$ in dependence of the precursor dosing time $t$ saturates in the maximum value $r 0$⁠. We define the slope by the arbitrary factor $a$⁠.

The change of precursor dosage has little effect on the saturation curves shown in Fig. 8. For all three test sets, the process saturates at a value of $r 0≈0.10 Å / cycle$⁠. Any variations are lower than the measurement uncertainty. The arbitrary slope factors are calculated as $a 100=1.16/ s$⁠, $a 200=1.24/ s$⁠, and $a 300=0.97/ s$⁠. There is just a slight difference between the three values and these are close to the previously found reference value of 0.92/s.⁵ We assume that these differences are still within the range of measurement uncertainty. In consequence, there could be no further dosing time optimization with the increased amount of precursor. It is likely that the limiting factor for the CoCOhept dosing is not the amount of delivered precursor.

Also, we estimate the transport time through the gas tubes (⁠ $≤$2 m length with inner radius of 4 mm) for 200 SCCM precursor at 30  $°$C to be below 0.1 s. Therefore, the precursor transport from the inlet valve to the reaction chamber is very fast and not a limiting factor. The most likely limiting factor is, thus, the reaction time of the precursor at the low process temperatures.

The main mechanism during the precursor pulse is the precursor adsorption and surface saturation. In case of an unsaturated surface, the growth per cycle will go down. In contrast, the plasma pulse length mainly influences the amount of removed ligand molecules. In the case of a too short plasma pulse, not only the growth per cycle will be influenced, but also the film composition. To take this behavior into account, each plasma pulsing length test was performed on a single wafer running a total of 1500 cycles. The precursor pulse length was set to 6 s to ensure saturation state.

Figure 9 shows the influence of the plasma pulse length on the resulting film thickness and film homogeneity. The violin plot²⁷ in green shows the cobalt film thickness distribution on each wafer. Additional to that, the second violin plot in orange shows the corresponding resistivity distribution of the cobalt film on each wafer.

The sheet resistance was not measurable on the wafer with the 1 s plasma pulse process. Increasing the plasma pulse length to 2 s resulted in measurable values but with a wide range over two orders of magnitude from 632.3  $μΩ cm$ to 86.3  $mΩ cm$⁠. The 8 s plasma pulse length process resulted in a cobalt film with a mean resistivity of 201.1  $μΩ cm$⁠, and a median value of 203.8  $μΩ cm$⁠. These results are in the same range as comparable processes. Park et al. could achieve a resistivity of 90  $μΩ cm$⁠.²⁸ The precursor CCTBA which was used by Park et al. is similar to our CoCOhept but the process temperatures differ significantly: where CoCOhept works from 50 to 110  $°$C, does CCTBA operate from 125 to 200  $°$C.^5,28 Cobalt ALD processes with other classes of precursors like $Co ( EtCp ) 2$ (Ref. 29) or Co(⁠ $tBu2$DAD) $2$ (Ref. 30) can also achieve films with resistivities way below 200  $μΩ cm$⁠, but all of them require deposition temperatures above 125  $°$C and higher. In comparison to other processes, CoCOhept processes provide films with a comparable but higher resistivity but can operate at much lower temperatures.

The previously developed ALD process has been integrated on different test structures to demonstrate its performance and versatility. This included vertical high aspect-ratio structures like deep trenches. Figure 10 shows a merged collection of scanning electron microscopy images of one of those trench structures.

The opening width of the $SiO 2$ hard mask was 2  $μ$m with a slightly structure broadening in the silicon substrate to 2.15  $μ$m. The cobalt film has been deposited on the hard masking layer and the silicon sidewall of the trench. In total, the trench had an aspect ratio of 30:1. With the conditions given, the cobalt coating could be deposited down to a trench depth of 7  $μ$m, resulting in an aspect ratio of 3.5:1.

The ALD window of the CoCOhept process allows us to process at low temperatures in the range of 50 to 110  $°$C.⁵ This temperature range enables the possibility to deposit metallic cobalt on temperature sensitive substrates. To demonstrate this feature, a special test structure had been prepared. The target structures are parallel bars as are typically for interconnect structures or optical gratings.^31,32 We created a dedicated test structure with several parallel bars with a nominally bar width of 500 nm and a gap of 500 nm. This test structure is designed as a common bar chart. The chart shows the average temperature of Germany from 1881 to 2023. [The data are retrieved from the DWD Climate Data Center (CDC) giving the annual regional averages of air temperature—2 m above ground—as of January 2024.] This test structure has been transferred to a photolithography reticle to be used as mask for i-line lithography.

The test pattern has been transferred to the photoresist OiR974-09 on several test fields on 200 mm wafers. The wafers with the created patterns were then coated with metallic cobalt. The ALD process had 1000 cycles at 85  $°$C. The precursor pulse length was set to 6 s in order to ensure saturation state. The plasma pulse length was set to 2 s. With this value, the high deposition rate is achieved, but the plasma impact to the photoresist is as low as possible. This is necessary as hydrogen plasma will etch the photoresist.³³  Figure 11 shows a SEM image of the whole testing structure after the coating. The whole test structure remains thereby intact and no pattern collapse was observed although the ALD process involves 1000 $H 2$ plasma pulses.

Figure 12 shows the cross-sectional view of the resist bar structures. The initial state prior ALD is shown in part (a). The photoresist and the subjacent $SiO 2$ layer are insulating materials. This causes charging effects in the image. The photoresist bar structure has a height of 990 nm and a width of 345 nm on the top side and 475 nm on the base. The state after the Co ALD process is shown in Fig. 12(b). Now, the bar height is reduced to 870 nm. The width has decreased to 320 nm on the top side and 440 nm on the base.

The used inline detector clearly shows the material contrast at the edge of the photoresist structures. A homogeneous film covers the entire surface of the photoresist as well as the $SiO 2$ substrate which is free of photoresist. The image highlights the two measurement points which were used for the EDX analysis: 1—the side edge of the structure and 2—the top of the structure.

In Fig. 13, the EDX spectra of these two positions are plotted. The spectra show the emission lines of cobalt, silicon, oxygen, and carbon.³⁴ The shown cobalt peaks refer to the deposited ALD film. The other elements are expected as the substrate is silicon, the interlayer is silicon oxide, and the photoresist consists mainly of a hydrocarbon polymer. The characteristic line at 2.3 eV can be attributed to sulfur. The origin of this contamination is unclear. The reference measurements of the cobalt ALD film did not show any traces of sulfur contamination.⁵ 

The EDX measurements verify that the low-temperature ALD process resulted in a photoresist bar structure with cobalt coating. The measurement on the structure edge and top show a comparable high signal intensity. This confirms that the edge in Fig. 12(b) consists of deposited cobalt.

Within this work, we demonstrated the implementation of a novel low-temperature ALD process for metallic cobalt. DFT simulations were successfully used to verify the reaction chemistry of the alkyne-based metal-organic Co precursors and to identify the precursor CoCOhept as the most suitable precursor for low-temperature ALD processes. This processes uses hydrogen plasma as reactant in the second half-cycle. CFD simulations were beneficially used to identify optimized parameters for the chamber geometry and precursor delivery to achieve a homogeneous deposition results. These results were verified by the experimental results on the wafer-level.

The analysis of the influence of the precursor pulse time with different precursor dosing rates showed that the precursor adsorption step is most likely reaction limited and not significantly influenced by the amount of dosed precursor. We further investigated the influence of the hydrogen plasma pulse length in order to achieve a conducting film. Plasma pulse lengths of 8 s result in a resistivity of 201.1  $μΩ cm$⁠, a value comparable to other low-temperature cobalt ALD processes.

We applied the optimized cobalt ALD process to several three-dimensional structures on silicon and polymer resists. The deposition in HAR trench structures on silicon was successfully down to aspect ratios of 3.5:1. We could further demonstrate the successful, conformal deposition on temperature sensitive photoresist bar structures without significant changes in polymer topography.

This work was funded by the ERDF fund of the European Commission and by funding of the Free State of Saxony of the Federal Republic of Germany (project ALMET). We wish to acknowledge the support of the whole cleanroom team of the Center of Micro and Nano Technologies of the University of Technology Chemnitz and of the Fraunhofer Institute for Electronic Nano Systems.

The authors have no conflicts to disclose.

Mathias Franz: Conceptualization (lead); Formal analysis (equal); Investigation (equal); Writing – original draft (lead); Writing – review & editing (equal). Linda Jäckel: Conceptualization (supporting); Formal analysis (equal); Investigation (equal); Writing – original draft (supporting); Writing – review & editing (equal). Xiao Hu: Conceptualization (supporting); Formal analysis (equal); Investigation (equal); Writing – original draft (supporting); Writing – review & editing (supporting). Lysann Kaßner: Formal analysis (equal); Writing – original draft (supporting); Writing – review & editing (supporting). Camilla Thurm: Investigation (supporting); Writing – review & editing (supporting). Dirk Rittrich: Investigation (equal); Writing – review & editing (supporting). Christian Helke: Resources (supporting); Writing – review & editing (supporting). Jörg Schuster: Conceptualization (supporting); Funding acquisition (supporting); Writing – review & editing (supporting). Marcus Daniel: Funding acquisition (supporting); Investigation (supporting); Writing – review & editing (supporting). Frank Stahr: Funding acquisition (supporting); Resources (supporting); Writing – review & editing (supporting). Natalia Rüffer: Investigation (equal); Writing – review & editing (supporting). Robert Kretschmer: Investigation (supporting); Resources (supporting); Writing – review & editing (supporting). Stefan E. Schulz: Funding acquisition (lead); Writing – review & editing (supporting).

The data that support the findings of this study are available from the corresponding author upon reasonable request.