The impact of substrate bias on a remote plasma sputter coating process for conformal coverage of trenches and 3D structures

With the progression towards higher aspect ratios and finer topographical dimensions in many micro- and nano-systems, it is of technological importance to be able to conformally deposit thin films onto such structures. Sputtering techniques have been developed to provide such conformal coverage through a combination of coating re-sputtering and ionised physical vapour deposition (IPVD), the latter by use of a secondary plasma source or a pulsed high target power (HiPIMS). This paper reports on the use of an alternate remote plasma sputtering technique in which a high density (>1013 cm−3) magnetised plasma is used for sputter deposition, and additionally is shown to provide IPVD and a re-sputtering capability. From the substrate I–V characteristics and optical emission spectroscopy (OES) data, it is shown that remote plasma sputtering is an inherently continuous IPVD process (without the need of a secondary discharge). Through the reactive deposition of Al2O3 onto complex structures, scanning electron microscopy (SEM) and energy dispersive x-ray spectroscopy (EDX) results demonstrate that applying a negative substrate bias during film growth can result in re-sputtering of deposited material and film growth on surfaces obscured from the initial sputter flux. Using 5 : 1 (height : width) aspect ratio trenches, the substrate bias was set to 0,−245 and  −334 V. At 0 V substrate bias, the alumina coating is predominantly deposited on the horizontal surfaces; at  −344 V, it is predominantly deposited onto the side walls and at  −245 V a more uniform layer thickness is obtained over the trench. The process was optimised further by alternating the substrate bias between  −222 and  −267 V, with a 50% residence time at each voltage, yielding a more uniform conformal coverage of the 5 : 1 aspect ratio structures over large areas.


Introduction
The conformal coating of trenches, vias and other 3D structures is a critical requirement in many modern technologies including, for example, micro-electromechanical systems (MEMS) [1] and inter-connect (IC) fabrication [2]. Although chemical vapour deposition (CVD) and derivative techniques can accomplish good conformal coverage [3], they are not always suitable due to the high process temperatures required or the absence of a suitable pre-cursor material. Whilst the common physical vapour deposition (PVD) techniques (e.g. electron beam evaporation, magnetron sputtering) are largely free of these limitations, they suffer from inherent poor step coverage of non-planar surfaces due to shadowing effects caused by the essentially line-of-sight transit of material from the target material source to the substrate. This can lead to coating build-up on the top edges of trenches and vias, resulting in void formation [4].
For many PVD processes, a degree of conformal coverage can be achieved through substrate motion during coating deposition or through increasing the process pressure, the latter reduces the mean free path of scattering collisions between the sputtered flux and process gas (usually Ar) increasing scattering frequency. Both techniques increase the angular distribution of arriving sputter material at the substrate [5], allowing some coating of shallow trench and via side walls, but with limited ability to coat more demanding 3D structures, especially where 'overhangs' are present.
Further improvement of PVD conformal coverage can be achieved using ionised physical vapour deposition (IPVD) in which an electrical substrate bias is used to influence the directionality and energy of arriving ionised material [6,7]. For most PVD processes significant ionisation of the physical vapour is only achieved if a secondary plasma discharge is used, usually either an inductively coupled plasma (ICP) [8] or an electron cyclotron resonance (ECR) microwave plasma [9,10]. Alternatively, high energy pulsed magnetron sputtering techniques can directly produce highly ionised species [11], though at the expense of deposition rate.
Substrate bias is used to modify the arrival characteristics of these ions. Their direction and kinetic energy can be controlled [4], resulting in greater penetration into high aspect ratio structures such as trenches or vias [10,12].
In a high density plasma sputter process, a much higher ion contribution to the process is provided by the plasma gas (usually Ar) ions. In addition to the usual densification improvement resulting from low energy ion bombardment [13], by increasing the bias the Ar ions will eventually have sufficient energy to cause re-sputtering of the deposited material. The re-sputtered material greatly increases the angular distribution of the overall sputtered material flux, allowing surfaces that are 'shadowed' from the target material source to be coated and providing a mechanism to conformally cover 3D surfaces.
This paper investigates the sputter deposition of thin films onto biased substrates using a remote plasma sputtering technique. The remote plasma sputter system has been described in detail by Thwaites [14] and numerous publications describing the technology to deposit thin films materials such as ITO [15], IZO [16], ZnO 2 [17], HfO 2 [18], ZnS : Mn [19] have been reported in the literature.
The main aims of this work are (i) to show the capability of remote plasma sputtering to generate an ionised sputter flux; (ii) to characterise the effect of substrate bias on sputter rate and the re-sputtering process; (iii) to demonstrate that this technology through re-sputtering can conformally coat 3D substrates.
Previous preliminary experiments indicated that a substantial proportion of the sputter flux might be continuously ionised without the need of a secondary discharge. In this investigation, the effect of substrate biasing on the sputtered material flux has been examined by measuring the substrate I-V characteristics and by optical emission spectroscopy (OES). The influence of substrate bias on the deposition rate is then examined for the deposition of Al and Al 2 O 3 thin films. Through the reactive deposition of alumina, this paper describes the use of substrate bias to improve the conformal coating of 3D structures and 5 : 1 (height : width) aspect ratio trenches. Scanning electron microscopy (SEM) and energy dispersive x-ray spectroscopy (EDX) were employed to monitor the thin film coverage and uniformity, and the crystal structure of the deposited thin films was determined by x-ray diffraction (XRD).

Plasma characteristics
OES studies were performed to confirm the presence of ionised sputtered material. For this a remote plasma sputter unit was equipped with an internal fibre-optic system which can be positioned to allow localised monitoring of the plasma. The integration time was 15 ms and the data averaged over 250 scans. A collimator approximately 10 cm long and with similar diameter to the head on the spectrometer detector (approx. 4 cm) was used to minimise deposition onto the lens.
The OES measurements were made when sputtering a Ti target. The process parameters employed were as those used for the deposition of Al thin films, described in section 2.2.
To determine the I-V characteristics at the substrate under different bias conditions a dc power supply was connected to the substrate table in conjunction with a multi-meter to measure current.

Film deposition
All films were deposited at room temperature using a remote plasma system developed by Plasma Quest Ltd [14]. A schematic diagram of the remote plasma system used in this work is given in figure 1. The technology is based upon the use of a remote, high density (>10 13 cm −3 ions) but low energy (<10 eV) Ar gas plasma generated in a side arm plasma source adjacent and connected to the vacuum coating chamber. The visible glow region of the remotely generated plasma is confined and guided to the target by the magnetic field produced by two electromagnets. Within this glow region the ions are of uniform high density but have insufficient energy to cause sputtering.
Application of a target bias accelerates the ions to the target resulting in uniform sputtering and therefore uniform erosion across the full target surface. This is particularly beneficial during reactive sputtering as the high rate uniform target erosion (3.9 × 10 18 atoms s −1 for Al using a current of 0.5 A over a 4″ target) enables a stable reactive process to be achieved at constant oxygen flow without the need for feedback control. As a result there was no need for O 2 flow feedback systems typical required for other sputter techniques.
Although the remote plasma system appears to operate in a manner similar to an ion beam source, it is important to note that the plasma ions have no particular directionality and only low (thermalised) energy until accelerated in the target sheath by the target bias. Therefore the sputtering process shows no directional effects from incoming ions, other than those usually observed in diode and magnetron sputtering.
For this study, Al and Al 2 O 3 films were grown using a high purity (99.999%) Al metal 4″ target. Argon of 99.999% purity was used as the sputter gas and oxygen of 99.999% purity was used as the reactive gas for the deposition of Al 2 O 3 .
The base pressure was 8.0 × 10 −6 mbar with a process pressure consistently maintained at 3.1 × 10 −3 mbar. The RF plasma source power and DC target voltage were kept constant at 600 W and 600 V respectively. The target to substrate spacing was 20 cm. An RF substrate bias, V b, was applied and increased in discrete steps between 0 and −600 V. Systematic RF powers were selected to assist in the later determination of sputter rate/power behaviour (0, 110, 120, 150, 200 W), but are also shown as equivalent bias voltages (V b = 0, −245, −267, −334, −445 V) for comparison with other published data.

Film characterisation
Thickness measurements of all films deposited on glass were conducted using a Taylor Hobson Talystep profilometer.
SEM and EDX analysis was performed using a Jeol JSM-7100F instrument, employing a Schottky field emission gun, operated at 20 keV. A Thermo Scientific UltraDry Energy Dispersive Spectrometer (EDS) detector with a NORAN System 7 x-ray Microanalysis System was used to record the EDX elemental maps. Glancing angle x-ray diffraction (GAXRD) using a 1° incident angle was undertaken on a Panalytical X'pert Pro diffractometer using Cu Kα radiation (λ = 1.5406, 40 kV, 30 mA) and a 2Θ range of 15-75°.
X-ray photoelectron spectroscopy (XPS) was undertaken using a VG Thermo Scientific Theta Probe employing a monochromated Al Kα X-radiation operated at 15 kV and 20 mA. The hemispherical analyser was run at a pass energy of 20 eV and a step of 0.1 eV. The binding energies of the photoelectron peaks were referenced to the adventitious hydrocarbon C 1 s peak at 285.0 eV. Quantification was performed using the Thermo Scientific Avantage software which employs instrument modified Wagner sensitivity factors after a Shirley background subtraction to determine atomic concentrations.

Conformal coverage studies
To test the conformal coverage of remote plasma sputtered Al 2 O 3 thin films, 3D structures were fabricated. The structures were fabricated by anisotropically etching Si 3 N 4 in a Reactive Ion Etcher (RIE) at 40 mT (CF 4 and Ar) at 40 W for 50 min, followed by an isotropic wet etch of the Si for 3 min using HF : Nitric Acid : DI Water (2 : 25 : 25) to produce 1 μm undercut features. Cross-sectional SEM images of these structures are shown in figure 2.

Results and discussion
3.1. Plasma 3.1.1. Remote plasma system characteristics. The remotely generated plasma used in this work is a spatially extensive, variable strength magnetised plasma in which the ion and electron motion is significantly modified from that normally associated with other sputter techniques, with consequent impact  on plasma generation, sputtering and deposition behaviour. Of particular relevance to this study, for the remotely generated plasma, the motion of electrons and ions perpendicular to the magnetic field direction is disproportionately restricted such that the usually higher electron drift mobility becomes less than the ion drift mobility. This inversion is most obviously demonstrated through an electrically isolated (floating) substrate charging positively, as opposed to negatively as reported with other similar systems [20][21][22]. The application of substrate bias in magnetron sputtering to permit re-sputtering is well documented [23]. However, it is unclear what the effect of the directionally and spatially variant electromagnetic field and magnetised plasma in the remotely generated plasma will have on re-sputtering when a substrate bias is applied.

Substrate bias measurements.
Initial work to quantify the impact of the magnetised plasma involved measuring the I-V characteristics of the substrate stage when applying a dc bias voltage to an electrically conductive substrate, but no target bias. As is the normal convention for a plasma I-V curve the electron current into the probe is displayed on the positive y axis. The I-V curve obtained by varying DC substrate bias V b , varied between −80 and +60 V, is given in figure 3. This data is similar to a typical I-V response from a Langmuir probe in a magnetron plasma with the exception of a positive voltage zero current cross over point, not the usual negative voltage. This confirms that the ion drift mobility is higher until a positive voltage of 2.8 V is applied, at which point a dynamic equilibrium between electron and ion flux is reached. Critically for this work, for bias voltages above 2.8 V, increasingly higher electron currents are drawn from the magnetised plasma, confirming that the electrons remain sufficiently mobile to prevent charge build up when an RF substrate bias is applied (required for non-conducting metal oxide thin films).

IPVD.
The presence of an ionised sputter flux in proximity to the substrate was determined by the focused OES system earlier described. For the deposition of Ti used in our initial studies, the plasma emission spectra obtained with and without target bias are shown in figure 4. For no target bias, only Ar peaks are present as expected. Additional peaks appear with an application of target bias. Comparison with published data [24] shows these additional peaks are due to both excited Ti atoms and ions. The four peaks at 308.02, 323.55, 334.88 and 375.96 nm correspond to singly ionised Ti, showing that remote plasma sputtering can offer the benefits of an IPVD process.
The ionised flux behaviour can be influenced through the use of an applied substrate bias [6]. Figure 5 shows the progressive increase in the intensity of the four Ti + OES peaks close to the substrate with increasing negative substrate bias. As would be expected, the ionised flux is attracted by the negative bias, directing more Ti ions to the substrate.

Deposition rate.
The effect of substrate bias (V b = 0 V to −222 V) on deposition rate is presented in figure 7. Application of −90 V to the substrate leads to the deposition rate being increased by 2 nm min −1 compared to zero bias, consistent with the attraction of additional Al ions towards the substrate (XPS analysis of the Al thin film confirmed that the increase in deposition rate is not due to the implantation of Ar). Hence, both the deposition rate and OES data show that increasing the substrate bias results in an increase of ionised material flux to the substrate. At a V b > −90 V, the deposition rate decreases. This is due to the increasing re-sputter rate at higher voltages. Hence, any increase in the deposition rate from more Al ions arriving at the substrate is negated by the stronger re-sputtering process. and −134 V being applied. However, the preferential (1 1 1) orientation at 0 V is not preserved as the substrate bias is increased to −134 V and a more random grain orientation develops. For the deposition at 0 V bias, the Al (f.c.c.) exhibits pronounced preferential growth on the lower energy (1 1 1) plane. We believe that the more random grain orientation observed with increasing substrate bias (−90 V and −134 V)   is due to the increased adatom mobility induced by ion bombardment [25], though further experimentation would be required to confirm this.

Deposition rate.
To examine the effect of substrate bias on a reactive process, Al 2 O 3 thin films were grown. For the deposition of these thin films, the O 2 flow remained constant (3.4 sccm) and the substrate bias was varied from 0 to −550 V. The deposition rate as a function of substrate bias for the reactive deposition of Al 2 O 3 ( figure 9) displays a similar trend to the Al process (figure 7). No variation in stoichiometry was observed with an increase in substrate bias. However, the V b corresponding to maximum deposition rate increased from −90 V for Al to −300 V for Al 2 O 3 . It is well known that the sputter rate of metal oxides is lower than metals and hence a shift of the V b maximum to higher voltages would be expected. Similar trends in the deposition rate have also been reported when varying the substrate bias for Cu 2 O [23] and TiN x O y [26].

Coating of complicated architectures.
From section 3.3.1, it is evident that remote plasma sputtering has the capability to deliver both an IPVD process and, with sufficient substrate bias, that coating re-sputtering occurs. Thus, in principle, remote plasma sputtering has the potential to coat complicated architectures with the deposited film acting as a sputter source for surfaces obscured from the depositing flux.
To test the conformal coverage, 3D structures were fabricated as described in section 2.4 and cross-sectional SEM images of these structures are shown in figure 2. For this initial study, based on the results given in figure 9, Al 2 O 3 has been deposited onto the 3D structures at substrate biases of 0,−267 and −445 V. Following thin film deposition, the 3 coated structures were mounted in Bakelite, ground and polished to provide the cross sectional SEM images and EDX elemental maps shown in figure 11.
At substrate bias values of 0 and −267 V, the SEM images and EDX elemental maps in figure 11 are consistent with primarily line-of-sight deposition for the Al 2 O 3 layer with no discernible deposition of the film on the underside of the overhanging structure. Increasing the substrate bias to -445 V promotes re-sputtering of Al and O; it is evident both from the SEM images and EDX maps that a significant amount of Al 2 O 3 has been deposited on the surface of the undercut feature.

Coating of trenches.
A more comprehensive set of trials was conducted to test the combined IPVD and re-sputtering capability of the remote plasma sputtering technique to deliver conformal coating of high aspect ratio trenches. 5 : 1 aspect ratio structures fabricated from a multi-component metallic alloy with metal coated trenches were used for with consequent non-uniformity when applied to the full system substrate (100 mm dia).
A more controllable method was therefore investigated, in which the substrate bias was alternately switched between −222 and −267 V, with a 50% deposition time at each voltage. V b = −222 V will cause preferential coverage of the horizontal surfaces and −267 V preferential coverage of the vertical surfaces. The result is shown in figure 15. Although the conformal coverage does not display a uniform thickness on both the horizontal and vertical surfaces, the film deposited on the sidewalls exhibits a more uniform thickness over the whole structure. Optimisation of the process parameters will enable further improvements to be made in the total uniformity of the conformal coating on such high aspect ratio structures.
Other work has shown that delicate (polymeric) substrates can be protected from ion bombardment through the deposition of a thin initial layer with no substrate bias applied. Hence, on application of the bias, ion bombardment and resputtering occurs from the initial Al 2 O 3 layer and the substrate is protected, allowing conformal coverage of plastic and organic material based structures.

Conclusion
It has been shown that remote plasma sputtering is inherently an IPVD process (that does not require an extra plasma source), and together with Ar ion bombardment the use of substrate bias can be used to influence the growing film. It is evident from OES and deposition rate data, that for remote plasma sputtering, a bias can be applied to the substrate in order to increase the energy and control the direction of the incident ions at the substrate. For lower substrate bias voltages, an increase in deposition rate is observed for both Al and Al 2 O 3 due to the attraction of sputter flux ions. At a substrate bias of −90 V for Al and −300 V for Al 2 O 3 the deposition rate reaches a maximum. At higher substrate bias voltages the deposition rate decreases. This is due to the increasing resputter rate at higher voltages.
Initial trials have shown that through the use of substrate bias in remote plasma sputtering conformal coating of complicated architectures can be achieved by the coating itself acting as a sputter source for overhang structures.
For the deposition of Al 2 O 3 onto 5 : 1 aspect ratio structures it has been shown that through control of the substrate bias the coating thickness on the vertical and horizontal planes can be varied. For low or zero substrate bias the majority of the coating will be deposited on the horizontal planes. But, employing a substrate bias >−267 V results in the majority of the coating being deposited on the side walls. This occurs through the attraction of ionised sputter flux and re-sputtering of the thin film on the horizontal surfaces due to ion bombardment. Finally, it has been shown that by alternating the substrate bias between −222 and −267 V, with a 50% residence time at each voltage, conformal coverage of 5 : 1 aspect ratio structures can be achieved. It is expected that further optimisation of the process parameters will lead to a uniform film thickness on all surfaces of high aspect ratio structures.