Dielectric breakdown of alumina thin ﬁ lms produced by pulsed direct current magnetron sputtering

Alumina ﬁ lms (~2 μ m thick) were deposited with a mixed Cu/Al interlayer onto copper. Direct current (DC)/ Pulsed DC (PDC) magnetron sputtering techniques were independently compared for reactive alumina sputtering. In DC sputtered ﬁ lms, elemental aluminium of 9.2 at.% and nano-crystallites were present within the x- ray amorphous matrix, resulting from target arcing. Defects lead to premature dielectric breakdown/increased current leakage. PDC sputtering improved ﬁ lm quality by removing crystallites, metallic clusters and through thickness cracking. Time dependent dielectric breakdown (TDDB) measurements were carried out using conductive atomic force microscopy identi ﬁ ed an improvement in dielectric strength (166 to 310V μ m − 1 ) when switching from DC to PDC deposition power. TDDB suggested that at high applied ﬁ eld the dominant pre- breakdown conduction mechanism was Fowler-Nordheim tunnelling in DC ﬁ lms. Tensile pull-o ﬀ adhesion ranged from 56 to 72MPa and was highest following incorporation of an Cu/Al blended interfacial layer. Scratch testing indicated various cracking/buckling failures.


Introduction
As the aerospace industry progresses towards "the more electrical aircraft", there is a growing dependence on materials with superior dielectric strengths and high packing densities. Demands on aircraft electrical systems are increasing as the industry moves towards highvoltage direct current power generation/distribution and seeks to design aircraft distribution based on "fly by wire", requiring more powerful actuators [1][2][3]. Weight is a key design constraint in aircraft and automotive components such as electromechanical actuators, which are set to be heavily utilised [4]. Dielectric strength of > 1 kVμm −1 could offer a reduction in insulation thickness (> 3 μm) for enamelled magnet wire which is used in motor windings. Litz wires are also starting to be used for motor windings in aircraft. Whereas even thinner insulation could also be utilised for low voltage applications. Current commercial technology utilises tens of microns of insulation for instance commercial polyimide insulated wire can have an insulation thickness of 33 μm and a breakdown strength of 250 Vμm −1 . The motivation of increased breakdown strength and thermal conductivity has led to the implementation of polymer composite materials, often including filers or multi-layers. Chemically produced Polyimide films with dielectric strengths between 145 and 181 Vμm −1 for thickness between 7 and 21 μm respectively have been produced [5]. In 2016, Carreri et al. used bipolar reactive sputtering, to produce films of 3 μm with dielectric breakdown strengths of up to 1.5 kVμm −1 . Techniques used for ultrathin films such as plasma enhanced atomic layer deposition are capable of producing alumina films of breakdown strengths of up to 840 Vμm −1 whilst spin coating has achieved breakdown strength as high as 544 Vμm −1 with sufficient doping [6][7][8]. However a requirement to maintain low temperature deposition and ability to coat 3D objects such as wires makes sputtering techniques desirable. Thermal stability of the coating is also key in producing better insulation as an increase could potentially facilitate a wiring core size reduction for equivalent current throughput.
Thin-film ceramics such as Alumina have been used extensively in the micro-electronics industry for dielectric layers in capacitors with a dielectric constant of 8.5 whilst Tantalum pentoxide films have emerged in this area due to their improved dielectric constant > 20 [9]. However, the potential transparency of Alumina films provides broad functionality for anti-degradation protective films,  [10,11]. Alumina thin-films have been fabricated by chemical vapour deposition (CVD), plasma enhanced chemical vapour deposition, atomic layer deposition, electrophoretic deposition and sol-gel based methods [12][13][14][15]. Shortcomings of these techniques include high temperature operation leading to degradation of the substrate, inadequate adhesion or lengthy drying/aging stages [16].
The Physical Vapour Deposition (PVD) technique of Magnetron Sputtering has been chosen here due to its potentially low deposition temperatures (< 100°C) and high tune ability, with regards to structure and composition of the resulting film [17]. Deposition by a Radio frequency (RF) potential allows for the utilisation of dielectric target materials. Limitations include the low inherent deposition rates in the order of 1 μmhr −1 as reported by Cremer et al. at a target power of 300 W, which have been mitigated by using reactive Direct Current (DC), for which half of the applied sputtering power increased deposition rates to 3.5 μmhr −1 [18]. Hence RF's practical application for coating of high throughput products such as wires is diminished. As an alternative a DC source can be utilised for preparing dielectric coatings from conducting targets by using reactive gas sputtering, during which sputtered aluminium reacts with a working gas of oxygen to condense an alumina film [17,19,20]. The further development towards a Pulsed DC (PDC) potential neutralises surface charge, this prevents target arcing. This is achieved by periodically reversing the target voltage. PDC has been shown to have benefits over DC with fewer stoichiometric and structural defects due to the increased sputtering stability [21,22]. Using PDC Kelly et al. have made higher quality films (with fewer defects intrinsic to DC sputtering) with a deposition rate higher than that achieved with RF sputtering [20,23,24]. PDC is a good compromise between film quality and deposition rate when considering RF and DC depositions as alternatives [18,25]. Sputtered alumina has been deposited with various pulsing parameters in an attempt to improve structure, composition and deposition rate (regardless of duty cycle %) [6]. PDC sputtering is also being used to produce high quality films with low residual stresses especially when utilising Highly Power Impulse Magnetron Sputtering techniques [21]. Hollow cathode sputtering was recently explored to increase plasma density, ionising a higher amount of neutrals resulting in denser coatings [26].
The aim of this study is to understand the relationship between micro/nano structure, dielectric break down strength and mechanical properties of alumina thin films deposited using reactive DC and PDC magnetron sputtering onto copper substrates, with a goal to produce the next generation of adherent, electrically insulating and thermally stable wire and lamination coating materials to significantly reduce weight and volume in electronic devices, critical to aircraft.

Substrate preparation and magnetron sputtering
Copper disks (20 mm and 10 mm diameter, 2 mm thick, > 99.9%) were polished using sequentially higher grit 8″ silicon carbide paper (P240-P4000) with water as a lubricant, followed by polishing on a MD-Chem polishing pad (Struers®) utilising colloidal silica (particle size 0.06 μm). Disks were then sonicated in acetone and then ethanol for 10 min each. A TEER UDP-650 equipped with an ENI RPG50 power supply with arc handling capabilities was used to deposit Alumina from an Al target (> 99.6% purity, 175 × 380 mm). A Cu target (> 99.6% purity, 47 mm ø) was additionally used for deposition of a graded interlayer for enhanced adhesion between the substrate and the Alumina film. The TEER UDP 650 is configured with between 1 and 4 target materials to enable continuous rotation of substrates past each depositing material as shown in (Fig. 1a). All substrates were fixed at a substrate to target distance of 55 ± 1 mm and were rotated at 5 RPM.
The chamber was pumped down to a base vacuum of < 11 × 10 −3 Pa. All depositions utilised in a working gas mixture of pureshield argon (99.999%) and oxygen (99.999%) acquired from BOC©. Gas flow ratios were controlled using an Optical Emission Monitoring System (OEM) at a set point of 23% of the original plasma intensity, for stoichiometric Alumina deposition. The OEM controls the argon/oxygen gas ratio to obtain a desired plasma intensity at a certain wavelength (396 nm) [24]. Oxygen was added until the plasma reached an intensity of 23% of the original, the intensity associated with stoichiometric Alumina. Prior to deposition all substrates were bias etched with a pulse frequency of 250 kHz and a step time of 500 ns at −150 V for 15 min. Alumina coating variations were produced by three comparative methodologies (See Fig. 1b for schematic) with the following notations; Conventional DC films (DC Films), Conventional DC films with a copper/alumina blended interlayer (BDC films) and Pulsed DC (PDC Films). DC Films were deposited with a primary Al interlayer deposited for 5 min at 6 A, as the OEM stabilised at 100% Ar. The OEM was then set to 23% and oxygen was introduced over 5 min facilitating deposition of the Alumina layer once the set point was reached. Oxygen was added according to a feedback loop in response to the OEM intensity, resulting in an oxygen flow rate which depended on the corresponding OEM response [24]. BDC Films included a Cu/Al blended interlayer which was produced by ramping the Al target current from 0 to 6 A under a DC potential and simultaneously lowering a potential on a Cu target from 50 to 0 W over 5 min. PDC Films contained an Al interlayer deposited for 5 min using a pulsed DC potential of 4 A with a similar OEM reduction as DC Films towards deposition of Alumina. All interlayer deposition used a substrate bias of −50 V using the sample pulse parameters as the previous bias etching. Deposition parameters for all coatings following attainment of the 23% OEM set point are outlined in Table 1. External heating was not applied and substrate temperatures were measured using RS Pro temperature sensitive labels, which indicated temperatures ranging from 110 to 120°C for all samples.

Fourier transform infrared spectroscopy
Fourier Transform Infrared Spectroscopy (FTIR) using a Bruker Tensor FTIR instrument equipped with an Attenuated Total Reflectance (ATR) attachment with a diamond mirror and ZnSe lens was used to measure sample IR absorbance between 500 and 4000 cm −1 wavenumbers with a resolution of 4 cm −1 . Samples were analysed by pressing the coating surface in contact with the ATR diamond.

X-ray photoelectron spectroscopy
X-ray photoelectron spectroscopy (XPS) of coating surface was conducted using a VG scientific Escalab Mark II with an Al Kα nonmonochromatic source. Survey spectra were collected between 0 and 1200 eV using a step size of 1.0 eV and dwell time of 0.2 s and 2 scans followed by high definition scans using a step size of 0.2 eV and dwell time of 0.4 s with 5 scans for O_1S, Al_2P and C_1S photoelectron emissions. Casa XPS was used for elemental quantification, peak deconvolution and oxidation state analysis. All spectra were calibrated to with respect to C_1S photoelectron emission at 284.8 eV.

X-ray diffraction
Films were deposited onto borosilicate glass coverslips (Agar scientific L46R19-5 19 mm ø cover glasses) to avoid signal from the underlying copper substrate and were analysed using X-Ray diffraction (XRD). A Bruker D8 advanced (Cu Kα source at 40 kV and 40 mA) was used for glancing angle scans at 1.2°, scanning between the 2θ values of 15 and 80 o using a step size of 0.04 o with a dwell time of 12 s.

Scanning electron microscopy
Scanning electron microscopy (SEM) was carried out using a JEOL 7100F FEG-SEM microscope, for sample thickness measurements and cross sectional morphology. A working distance of 10 mm, 15 kV acceleration voltage and an aperture number 4 were used for imaging. TEM lift out sample was prepared was using a FEI quanta 200 3D dual beam FIB/SEM equipped with a Ga ion beam (Magnum™ Column) using a standard procedure. A 5 μm Pt top layer to protect the sample was applied prior to milling. Lift outs were performed on BDC and PDC samples. Liftout samples were an average of 15 × 5 μm.

TEM
Transmission electron microscopy (TEM) was carried out using a JEOL JEM 2100+ in order to analyse lift out samples. The TEM was equipped with Gatan US1000XP and JEOL STEM detectors for conventional imaging and bright field and annular dark field scanning TEM respectively with an accelerating voltage of 200 kV was used for imaging. EDX was also carried out on these samples using an Oxford instruments X-max 150.

Time dependant dielectric breakdown
Atomic Force Microscopy (AFM) with time dependant dielectric breakdown (TDDB) was carried using an Oxford Instruments Cypher S equipped with a high voltage module and an Asylum Research ASYELEC.01 Ti/Ir coated tip. Measurements were carried out by first acquiring a 30 × 30 μm topographic image in tapping mode and then a grid of 4 × 4 locations was determined on the image. At each location a voltage ramp of 0-150 V at a rate of 37.5 V/s was applied to the copper substrate until a sharp increase in current was detected through the probe (past 10 nA). Thinned DC and BDC samples were produced for these measurements with an alumina deposition time of 12.5 min hence copper substrate was acting as the bottom electrode and the probe acting as the top electrode.

Pull off adhesion strength
Pull off testing was carried out using a P.A.T. handy (DFD instruments) pull off adhesion testing unit. Dollies of 2.8 mm radius-were glued to the surface using DFD E1100S epoxy, cured at 140°C for 60 min on a heat plate in direct contact with the base of the coated disk. Prior to curing stubs were cleaned with compressed air to remove any particulate, stubs were pressed down once epoxy was applied to remove any air bubbles. Excess glue was removed with a cylindrical cutting tool following curing. Stubs were then removed using the P.A.T handy. Removal sites were analysed using light microscopy using a Nikon LV100ND upright microscope, for failure mechanism.

Scratch testing
Scratches were made using a CETR UMI multiple specimen test system equipped with a Rockwell C indenter. The indenter was cleaned (using isopropyl alcohol) and dried after each scratch. A force ramp of 0.75 Ns −1 was used to apply a 3 mm scratch over a period of 20 s with a max load of 15 N. A Nikon LV100ND light microscope was used to collect scratch images. Image J software was used to analyse scratches for cohesive, adhesive and complete interfacial failures being identified. Failures in this case were defined as the first appearance of specific failure. Test was carried out in accordance to BS EN ISO 20502:2016.

FTIR, XRD and XPS
FTIR showed broad absorbance at < 1000 cm −1 relating to the AleO LO phonon vibrations (Fig. 2a). Peak positions varied for DC, BDC and PDC films and are shown at 875.6, 862.1 and 804.3 cm −1 respectively. Peak broadening within the PDC analysis was accompanied by a decrease in the wavenumber. XRD revealed the X-ray amorphous nature of the films, with peaks appearing in the DC and BDC films attributed to the Aluminium interlayer (Fig. 2b).
As deposited surface compositions and structures of DC, BDC and PDC films on copper substrates were analysed by XPS (Fig. 2c). The Al_2P photoelectron emission was present in all high resolution spectra located at 74.75 and 74.67 eV for DC and BDC films respectively and was attributed to Al 3+ (Al 2 O 3 ) [27]. DC and BDC sputtered films contained a distinct shoulder from the Al_2P emission at 72.00 and 71.85 eV respectively which were associated with Al 0 /metallic aluminium. Fig. 1. a) Plan configuration of the targets and substrates within the TEER UDP 650 magnetron sputtering system. Copper disks were fixed to the substrate holder and continuously rotated at 5 RPM past the Al and Cu plasmas for coating deposition. b) Alumina deposition was compared by three methodologies; (DC Films) produced by DC sputtering, (BDC Films) produced with a Cu/Al blended interlayer for improved adhesion and (PDC Films) which utilised a pulsed potential to prevent target arcing. The proportion of Al 0 in DC and DBC films were found to vary from 5.01 and 9.62% following quantitative peak deconvolution. Additionally Al:O ratios were obtained from the survey spectra for all films and were found to vary from 0.56 to 0.68. Most significantly Al 0 / metallic aluminium peak was absent in the PDC film (Fig. 2c) whilst the Al 3+ component was shifted to a lower binding energy of 74.37 eV ( Table 2).

Scanning and transmission electron microscopy
Cross sectional electron micrographs showed DC and BDC films had been deposited to a thickness of ca. 2 μm following deposition rates of 1330 nmhr −1 . PDC films had a markedly slower deposition rate of 76 nmhr −1 as can be seen by the SEM and TEM images ( Fig. 3 and Fig. 4 respectively). DC and BDC films contained 170 nm thick metallic interlayers whilst the interlayer in PDC was 30 nm as shown in (Fig. 4). Aluminium interlayers were deposited before alumina deposition, by sputtering the target in argon prior to oxygen addition. During this time the OEM response was calibrated to 100% for aluminium sputtering in argon. This was followed by reactive sputtering of alumina, where the flow of reactive gas was controlled by the OEM.
The addition of a blended Cu/Al interlayer and an additional 5 min to reach the 23% OEM set point were used in BDC films, with the view to improving adhesion.
Both DC and BDC films had distinct striation patterns, seen diagonally (Fig. 3d). Voids within the structure of BDC films, in the form of cracks became apparent during TEM analysis (Fig. 3d). Crystallites of ca. 10 nm were found throughout the BDC coating (Fig. 4a) in contrast PDC films did not have any morphological features such as striation/ layered patterns as seen in (Fig. 4a and b). The structure appeared entirely amorphous. EDX analysis confirmed the presence of the Al rich interlayer of 30 nm thickness.

Mechanical properties
Pull off strengths of the coating from the substrate in the DC samples occurred by partial interfacial detachment at 63.0 ± 6.4 MPa. BDC samples failed predominantly through complete interfacial detachment at 72.3 ± 5.6 MPa with some occurrence of partial failure and one instance of failure in the adhesive. PDC samples failed at 55.7 ± 2.9 MPa, with a mixture of partial interfacial failure and failure in the epoxy adhesive (Table 3, Fig. 5).
Scratch testing was used to show adhesive and cohesive failures, failure came in the form of a combination of chevron and conformal cracking (Lc 1+2 ) and buckling (Lc 3 ). DC films exhibited a cohesive failure at 1.00 ± 0.05 N corresponding to combination cracking, adhesive failure became apparent at 2.10 ± 0.18 N seen as buckling. BDC films had cohesive strength of 0.30 ± 0.03 N and adhesive strength of 1.78 ± 0.07 N. PDC films experienced cohesive failure due to combination cracking at 0.87 ± 0.05 N and adhesive failure due to buckling at 1.69 ± 0.16 N. Failure mechanisms for each sample as seen in (Fig. 5).

Electrical characterisation
The thickness of DC and BDC was reduced to allow full breakdown measurements within the maximum voltage to be applied 150 V (Table 4). Electrical breakdown voltage measurements showed that DC films broke down at 128.1 ± 0.8 V with all 16 measurements resulting in complete breakdown and leakage prior to breakdown (< 200 Vμm −1 ) was observed up to ca. 0.20 nA. Breakdown for BDC was seen at 137.4 V with greater error of ± 3.5 V however only 10 of 16 samples broke down before 150.0 V. Due to variance in breakdown the difference in breakdown voltage was not significantly different for the two data sets. Dielectric strength (E BD ) was calculated using the breakdown voltage and specimen thickness. Leakage occurred in an oscillating pattern until breakdown (Fig. 6b and c) where oscillations are more pronounced in the BDC films.
Breakdown in PDC films occurred at 66.7 ± 4.6 V, corresponding to a dielectric strength of 310.1 Vμm −1 with all 16 measurements resulting in breakdown. The maximum leakage current in these films was 0.13 nA. Pre-breakdown oscillations were smaller than in the other films, only clearly seen in 5 of the measurements, all occurring at higher applied field > 380 Vμm −1 .
Further analysis of the IeV data for samples which completely broke down was used to evaluate the conduction mechanism. Fowler-Nordheim tunnelling was evaluated by plotting ln (J/E 2 ) against the reciprocal field strength, where J is the current leakage density (A/ cm 2 ). These plots yielded two linear regions with negative gradients Fig. 2. a) FTIR spectra for Alumina thin films on copper foil, showing AleO LO phonon peaks. b) XRD diffraction patterns for alumina thin films as deposited onto borosilicate glass slides. Peaks indexed are for metallic Aluminium ICDD-PDF-00-004-0787, with an amorphous hump apparent from the substrate. c) XPS deconvolution of Al_2P peaks for DC, BDC and PDC films with main Al 3+ peak and Al 0 shoulders.   ( Fig. 7) at high field strength > 120 Vμm −1 for DC and BDC films.
Other common mechanisms were explored including Schottkey electrode limited and Poole-Frenkel bulk limited emissions [28]. However, further analysis of the data revealed high current variation at low applied fields, making mechanism determination difficult. Analysis of PDC IeV curves at low applied field yielded no clear conduction mechanism. Fowler-Nordheim tunnelling was suggested as the conduction method at high field, with graphical analysis yielding a singular linear response (Fig. 7).

Structural
The use of reactive sputtering to form alumina from metallic targets (such as using aluminium to form alumina films) is well documented and has been reviewed extensively by Kelly et al. among others [17,19,20]. All reactively deposited films (DC, BDC and PDC) in the current study were amorphous in structure as evidenced by the lack of Bragg diffraction peaks in the XRD scans (Fig. 2b). The peaks for crystalline aluminium, which were only observed in DC and BDC films resulted from the aluminium interlayer deposited between the substrate and alumina coating. (Note PDC coatings only had a 30 nm aluminium layer, which was not detected by the glancing angle XRD). In alignment   with this, XRD scans of BDC without a 170 nm interlayer showed no crystallites, resulting in an XRD pattern identical that of PDC. Reactively sputtered Ta 2 O 5 films produced by Sethi et al. were also found to contain crystallites (through TEM) in an apparently amorphous film [29]. The films analysed by XRD were deposited on borosilicate glass substrates -to eliminate copper Bragg diffraction peaks from the spectra-resulting in an amorphous hump region. The FTIR data confirmed the presence of alumina AleO bonding (LO) with absorption bands seen from 800 and 900 cm −1 (Fig. 2a), with FTIR AleO peak positions agreeing with XRD results, suggesting that all films were amorphous due to the lack of a peak at ca. 530 cm −1 [29]. Shifting of the absorption band from 867.3 to 804.8 cm −1 was seen upon the application of PDC power. This could be an indicator of higher levels of intrinsic stress within the PDC films when compared to DC and BDC films [30]. Cracking was seen by TEM (Fig. 3) in the DC and BDC films which may have been caused by the intrinsic stress. FTIR shifting of this kind was seen by Haanappel et al. in CVD produced alumina films where a lower internal stress for films formed at high temperature resulted in a shift to a higher maximum wavenumber, accompanied by an increase in intensity [30].
Alumina films were previously reported by Kelly et al. and showed formation of stoichiometric alumina with an OEM set point of 25% [23]. Film stoichiometry was determined for films produced at an OEM set point of 23% (this reduced value was selected to ensure the films were not Al rich). XPS found the Al:O ratios to be close to stoichiometric alumina (0.66) with values between 0.56 and 0.68 in the top 10 nm of the film, varying partially as a result of the different amounts of elemental aluminium in the films. The shoulder within the Al_2P XPS peaks for DC and BDC films indicated the presence of metallic aluminium which could be conductive pathways (Fig. 2). Metallic Al may have been introduced into the films by arcing caused by the build-up of the dielectric material on the target surface, which caused the ejection of metallic target material [20]. The Al_2P shoulder was shifted to a lower electron emission energy than the peak for Al 3+ by an average of 2.78 eV ( Table 2). XPS of alumina films deposited by Reddy et al. showed the presence of peaks with a binding energy for aluminium oxide and metallic aluminium peaks as seen in this study [27]. This problem was mitigated in PDC samples, showing that the pulsing parameters (150 kHz and a duty cycle of 40%) were adequate to prevent target arcing whilst still producing stoichiometric films, in agreement with other studies on pulsed DC deposition [17,31].
TEM analysis showed crystallites were dispersed within the X-ray amorphous films in the sub 10 nm size range as in Fig. 4. Crystallites in DC deposited films have been noted in alumina systems with the presence of metallic crystallites also noted in other films such as in aluminium nitride and aluminium oxynitride [24,32]. A layered structure to the films was also apparent with EDX identifying layers of oxygen deficient material. These striations increased and became more abundant with increasing sputtering time. Hence this is likely a result of target poisoning cycles resulting in more arc events giving rise to the striations. Micrographs also showed cracking within the cross section, a result of the high stress in the DC deposited films.
In contrast TEM analysis of the PDC films showed a purely amorphous structure absent of crystallites; moreover the film had no striations which were common place in the BDC films. This is directly a result of the pulsing applied and the improved process stability allowing an amorphous film with fewer defects to be formed and an  absence of cracks and striations.

Electrical characterisation (conduction and breakdown)
Films from the current study deposited onto polished copper substrates for 12.5 min using DC power had varying breakdown strength as measured by AFM TDDB methods. DC and BDC films had dielectric strengths of 210 and 160 Vμm −1 respectively. However, breakdown in these films was not instantaneous, a maximum leakage prior to breakdown of ca. 2 nA was seen for DC and BDC films. Pre-breakdown leakage included current oscillations (Fig. 6c) attributed to charge trapping and de-trapping. Crystallites and elemental aluminium in the material (Fig. 2c and Fig. 4a) could have facilitated the trapping within the film.
IeV analysis has been used frequently to generate information about conduction mechanisms in metal insulator metal systems (MIM). Commonly conduction in such systems is a result of Fowler-Nordheim tunnelling, Schottkey emission or Poole-Frenkel emission. Charge transport mechanisms and graphical analysis of current leakage have been studied extensively [28,33].
Conductive AFM measurements have been carried out by Ganesan et al. to assess the electrical properties of ultra-thin alumina films (< 1 nm produced by ALD), reiterating increase in dielectric strength with decreasing thickness and showing breakdown voltages as high as 13 kVμm −1 . Pre breakdown oscillations were seen in these ultra-thin alumina films and attributed to charge trapping and de-trapping in the films, whereby the creation of negatively charged traps oppose the injection of additional charge, reducing the current flow [34,35]. Defects created by applied field contributed to stress induced leakage current within their study. Conduction mechanisms in alumina films has been studied widely and often attributed to multiple conduction mechanisms, such as a combination of ohmic and non-ohmic thermal excitation conduction (Shottkey or Poole-Frenkel) mechanisms as seen by Zhu et al. who explored the effect of a nano-phase dispersed within anodic alumina structures on resistive switching behaviour [36]. Poole-Frenkel conduction was also determined to be the predominating mechanism in alumina films on silicon by Kolodzey et al. [37].
At lower applied electric fields conduction mechanisms in DC and BDC films aren't clear due to high variation in current (Fig. 7) which carried through in further analysis. Despite this, it is likely that thermionic emission plays a part in the conduction. The presence of a crystalline phase seen in TEM imaging (Fig. 4), could facilitate charge trapping, suggesting Poole-Frenkel emission could be a suitable conduction mechanism. The presence of elemental aluminium (Fig. 2c) could form conductive pathways through the material, whereby the mechanism would be determined by the conduction between such pathways. Tunnelling between metallic channels in sputtered AlN x O y films, similar to MIM structures has been documented by Borges et al. [32].
Conduction at high applied field prior to breakdown in DC and BDC films was attributed based upon analysis of E-J curves in Fig. 7. The linear relationships between the reciprocal of the applied field (cmMV −1 ) and ln (JE −2 ) at high fields relative to breakdown (< 120 Vμm −1 ) is consistent with the Fowler-Nordheim mechanism. This kind of conduction at high fields has been seen, by Groner et al. in ALD alumina films thinner than 10 nm, at fields of above 380 Vμm −1 [38]. In the current study two linear regions were observed in both BDC and DC films Fowler-Nordheim plots, indicating that there were multiple barriers being tunnelled through. Multiple linear regions have been seen in Fowler-Nordheim analysis of GaN nanorods by Evtukh et al. resulting from different tunnelling barriers, with a steeper slope occurring at higher applied field [39].
In contrast the PDC films showed a dielectric strength of 310 Vμm −1 meeting the target of 300 Vμm −1 which was also above 200 Vμm −1 of polyimide films deposited by Diaham et al. using chemical methods [40]. Dielectric strength values were calculated from breakdown voltages and thickness data. This is lower than breakdown seen in films produced by Bartzsch et al. using pulsed DC sputtering from a double ring magnetron having a breakdown of 620 Vμm −1 . Dielectric strength was four times less than films produced by Carreri et al., attributed to the use of mid frequency pulsed DC power supplies equipped with special arc handling equipment, producing films which had a breakdown strength of 1.5 kVμm −1 [6,22]. Electrical Measurements made on these reported films were made using probe based methods, opposed to the AFM methods used in this study. Both probe and AFM based methods are capable of producing IV curves which can be used to determine breakdown voltage and breakdown mechanism. However AFM methods were ultimately chosen for their high spatial resolution (with the only size constraint being the tip diameter of 25 nm) and ability to easily form arrays (as demonstrated in this work). Probe based methods were avoided because of the probes inability to avoid extrinsic defects accounting for some of the variation in the reported results [41,42]. The AFM method used here have allowed for consistent local area measurements.
Leakage in PDC films was apparent with an average maximum leakage of 0.13 nA seen in Fig. 6. This clear improvement over leakage seen in DC films prior to breakdown is a result of the improved structure as seen in TEM and XPS. The removal of elemental aluminium and crystallites from the surface and bulk of the materials is expected as a direct result of reduction in arc events due to the implementation of the pulsed power supply. The lower amount of charge trapping sites contributes to improved dielectric strength as well as the reduced leakage current. Current oscillations were seen only for a number of measurements. These oscillations were again attributed to charge trapping and de-trapping. Fowler-Nordheim tunnelling is proposed as a candidate for the conduction mechanism at high fields due to the linear response when plotting E −1 against ln (JE −2 ). High current variation meant that the conduction mechanism at low applied field wasn't clear as for DC and BDC films. It is possible that thermionic mechanisms are responsible for conduction prior to Fowler-Nordhiem conduction.

Mechanical characterisation
Alumina adhesion has been explored to various substrates and with a number of interlayer conformations. Järvinen et al. showed improvement in alumina film (RF deposited) adhesion to copper with the inclusion of Ti (which are known to improve adhesion to steels) and mixed Ti/Ni interlayers, the benefit of interlayers in alumina thin film systems has been widely shown, other metal systems have also shown promise [43][44][45]. Vuoristo et al. have produced thin Al 2 O 3 coatings on copper substrates using magnetron sputtering with varying power supply options focusing on the use of the coatings for electronic applications and adherence was found to improve with the application of a suitable bond layer to protect from delamination during thermal cycling [43].
However, the use of a mixed RF DC blended Cu/Al interlayer onto copper has not been performed. The mechanical adhesion of the films presented in this study appeared to improve with the addition of the bimetallic interlayer with a longer sputtering period during the transition between aluminium and alumina (DC 63.0 ± 6.4 MPa and BDC 72.3 ± 5.6 MPa respectively). The improved adhesion strength is a result of higher amounts of material mixing at the CueAl interface due to RF Cu sputtering [46]. The lack of pull off failure at the Al-Al 2 O 3 interface suggests that the additional time taken to reach the 23% OEM may not be necessary for promoting adhesion. The prevalence of partial interfacial delamination in DC films could be a result non uniform film stress distribution, arising from the lack of Cu in the interlayer. Pull off testing for PVD deposited alumina isn't widely documented but values obtained in this study agree well with results for electron beam evaporated copper deposited onto sputter cleaned polycrystalline alumina by Erck et al., who achieved a maximum adhesion of 88 MPa [47].
Failure modes seen in alumina scales by Bull et al. were similar to those seen in the current study for the alumina films, scratches exhibited cohesive failure through conformal cracking (L C1+2 ) and subsequent adhesive failure through buckling (L C3 ), as seen in scratch test imaging [48]. Scratch testing of DC films exhibited L C1+2 and LC 3 forces 1.0 N and 2.1 N higher than BDC films. Despite the addition of Cu improving pull off adhesive strength and the longer Al, Al 2 O 3 blending in these films, the tests showed lower critical load. Scratch loads were comparable to alumina films deposited by CVD onto stainless steel by Haanappel et al.at varying pressures showed critical loads of 0.1 N for a micron thick film. PDC thin films however, showed a lower adhesion strength in both cases. Pull off testing showed partial interfacial delamination to occur at 55.7 ± 2.9 MPa and scratch analysis yielded the lowest adhesive (L C3 ) values at 1.7 N and these failures were likely a result of a combination of effects. The lower Al sputtering rate, depositing a thinner interlayer, thus forming a more highly stressed interlayer as well as the lowered film thickness resulting in more premature substrate deformation and achievement of critical load. Critical load dependence on film thickness has been demonstrated, with an increase in thickness resulting in a larger critical load [49].

Conclusions
This study investigated the structure, mechanical and dielectric breakdown properties of sputtered alumina films onto copper. Deposition using pulsed and non-pulsed power was compared. Direct current deposition was found unsuitable to produce defect free films, as was shown to produce striations, crystallite defects and metallic aluminium conduction pathways in the structure, observed by TEM and XPS. This led to average breakdown of 183 Vμm −1 and maximum leakage current prior to breakdown of < 2.00 nA. These complications were mitigated by use of pulsed direct current deposition which corresponded to considerably higher dielectric strength and lower leakage, 310 Vμm −1 and 0.13 nA respectively. However, the pulsed method did not increase the adhesion strength of the films, in which the incorporation of a mixed Cu/Al interlayer (which were incorporated into BDC films) was found to have the highest pull off adhesion strength with an average failure strength of 72.3 MPa. This suggests that a combination of a mixed RF Cu and DC Al adhesion layer combined with a pulsed DC dielectric layer would have the potential to offer alumina films with improved pull off adhesion strength and dielectric properties when deposited onto copper, when compared to DC deposited films.