Fabrication of a Patterned Interference Filter Array Deposited by an Atmospheric Pressure Vapour Deposition Technique∗

A method is demonstrated of making interference filters fabricated using a layer of metal oxide deposited by a new atmospheric pressure atomic layer deposition (ALD) technique. By depositing different thicknesses of metal oxides between reflecting layers, filters with various spectral characteristics can be made. A method of patterning the ALD layers has been devised using ink-jet and this allows the fabrication of a colour filter arrays of any design. [DOI: 10.1380/ejssnt.2009.284]


I. INTRODUCTION
Colour filter arrays (CFA) are component parts found in many types of display and used in digital cameras in image sensors.In displays, a CFA is placed over white pixels to create coloured images.In sensors, such as those used in cameras, a CFA is used in front of a panchromatic sensor to allow the detection of colour.The CFAs are usually an array of red, green and blue areas laid down in a pattern [1].A common array used in digital cameras is the Bayer [2] pattern array.The resolution of each colour is reduced by as little as possible through the use of a 2×2 cell, and, of the three colours, green is the one chosen to be sensed twice as it is the one to which the eye is most sensitive and contains most of the luminance information of the image to be captured.
Similar arrays can be used to create colour in displays.For example Vollman describes arrays for liquid crystal displays (LCD) [3] and Winters describes an array for an organic light emitting diode, OLED device [4].
Typically, CFAs use band-pass filters, which ideally transmit light over a narrow wavelength range within the red, green or blue regions of the spectrum and absorb light at other wavelengths.Such arrays can be made in many ways, including printing or ink-jetting coloured inks, photographically creating coloured dyes, conventional photolithography, electrodeposition [5] and laser ablation of dyes.An alternative CFA uses dichroic or interference filters, based on a Fabry-Perot etalon [6], which has a cavity with dimensions chosen to transmit a particular wavelength of light.The cavity is formed of a layer of dielectric material sandwiched between two parallel reflecting layers.The reflecting layers are commonly partially transmitting smooth metal films.
The transmission and any wavelength of a Fabry-Perot etalon is caused by interference between the reflections of light between the two reflecting surfaces.Constructive interference occurs if the transmitted beams are in phase, allowing the light to have optimal transmission at that wavelength.If the beams are not in phase, there will be different amounts of interference and transmission will be reduced.The phase difference(δ) depends on the wavelength (λ) of the light (in vacuum), the angle the light travels through the etalon (θ), the thickness of the etalon (l) and the refractive index of the material between the reflecting surfaces (n) and is given by the equation If both surfaces have a reflectance R, the transmittance function of the etalon is given by: A development of this approach replaces the partially transmitting metal mirrors with dielectric thin film mirrors ("Bragg stacks") which have the benefit of greater transmittance at a particular desired pass wavelength of the filter.Thin film reflectors typically consist of alternating layers of two materials with high and low refractive indices respectively [6,7] each with a thickness corresponding to a phase change of 90 • .Highly efficient nonabsorbing filter structures can be constructed using this approach.
In this paper it is assumed that the filter will be viewed normal to the filter and θ = 0 • and the above equations were used to model the filter spectra.The reflectance/wavelength curves of a single metal layer and Bragg stack coated on glass were determined before completing the series of coatings.It was assumed that the reflectance of the top reflector was the same as that coated directly on glass.
Some devices, such as OLEDs, are sensitive to oxygen and water and must therefore be sealed to keep out air and moisture which reduces their efficiency operational lifetime.This is a particular issue when a substrate other than glass is used, since the permeability of most suitable materials is generally much greater than that of a glass substrate.One way to reduce the oxygen and moisture sensitivity is to coat the array with a thin inorganic metal oxide, which is known to have good barrier properties [8] against both oxygen and moisture, provided the metal oxide layer remains pin-hole free.The manufacture of these barrier layers by vacuum ALD has been described by Ghosh [9].
Chemical vapour deposition (CVD) and atomic layer deposition (ALD) are examples of well-known techniques for laying down thin layers of material, especially metal oxides, onto a substrate [10].Chemical vapour deposition is a chemical process used to produce high-purity, highperformance solid materials.The process is often used in the semiconductor industry to produce thin films of dielectrics and semiconductors.In a typical CVD process, the substrate is exposed to one or more volatile precursors, which react and/or decompose on the substrate surface to produce the desired material.The thickness is controlled by diffusion of the reactants to the surface.Atomic layer deposition (ALD) uses multiple selflimiting, sequential surface chemistry reactions to deposit conformal thin films onto various substrates.ALD uses similar chemistry to CVD except that the ALD reaction breaks the CVD reaction into two or more partial reactions, keeping the precursor materials separate during the reaction sequence.It is usually carried out at reduced pressure, contacting the substrate with one reactant to form a monolayer, removing this reactant by evacuation and adding the second reactant to form a single layer of the desired material.The excess of second reactant is removed by evacuation to complete an ALD cycle.The whole sequence is repeated until a layer of the desired thickness is formed.The reaction is self-limiting at each partial reaction step.Hence, dense highly conformal films are created.
Recently, a new method has been developed Levy et al., based on an idea of Suntola [11], to lay down metal oxide layers at room temperature and atmospheric pressure [12].This method uses a flow manifold which spatially separates the precursor gases.The substrate is moved under the head in such a way that each part of the substrate is contacted alternately with a metal precursor and oxidant, with an intervening inert gas purge to 'clean' off excess material.In this way, atomic layers of metal oxide can be deposited with an accurately controlled thickness.The method is relatively rapid, enabling up to four ALD growth cycles per second, and is scalable.Therefore, it could be used to coat large areas and continuous flexible webs with a number of different metal oxides, including those of titanium and aluminium.The oxide layers created in this manner can be patterned by depositing them over a pattern of swellable polymer printed using inkjet.The deposited layers crack where they have been coated over the polymer allowing subsequent penetration of a solvent through the metal oxide layer.The polymer is softened and dissolves in the solvent, releasing the metal oxide attached to it.The metal oxide remains attached to the polymer-free regions of the substrate.The edges of regions are sharp and no polymer residue is seen on the substrate when viewed through an optical microscope at 100x magnification.
Below, we describe a method to make a simple CFA demonstrating the use of the rapid atmospheric ALD process in conjunction with a lift-off patterning process based on an ink-jet method of mask generation.This might be applied to a large area CFA suitable for a display or imaging device.
We also demonstrate the use of the new rapid ALD process to fabricate all-oxide interference filters.

II. EXPERIMENTAL A. Atomic Layer Deposition
In all the examples, thin film deposition was carried out using apparatus very similar to that described by Levy in the examples section in the patent [10].Figure 1 shows a side view of the manifold through which gases flow as indicated by the annotation.The manifold effectively passes over the substrate in a reciprocating motion as indicated by the double headed arrow at a height of about 50 microns.In this way, the substrate is contacted in turn by a metal precursor and oxidant, the excess of each being removed by a flow of inert gas.Each reactant is supplied to the appropriate section of the manifold by blowing inert gas at a regulated flow rate through a bubbler containing the liquid reactant.The gas emerging from the bubbler is then mixed with inert carrier gas in controlled proportions before entering the manifold.The manifold used gave two ALD cycles in each direction of movement giving four ALD cycles per oscillation.In practice, it was found to be simpler to move the substrate on a heated platen below the fixed manifold.
High refractive index layers were either titanium dioxide or zinc oxide, whilst alumina was coated as a lower refractive index material.For titanium dioxide, titanium tetrachloride (Sigma-Aldrich) was in one bubbler and water in the second.For alumina, a 1 molar solution of trimethylaluminium in heptane (Sigma-Aldrich) was in one bubbler and water in the second.For zinc oxide, a 15% solution of diethyl zinc in hexane was in one bubbler and water in the other.All ALD coating was carried out onto 1 mm thick, 63 mm square, Schott AG borosilicate glass, mounted on a moving, heated platen.The glass had been cleaned in a 10% solution of Decon 90 for 10 minutes with gentle agitation at 20 • C followed by rinsing in flowing demineralised water for two minutes.The platen temperature was set to 200 • C for zinc oxide and alumina and 100 • C for titanium dioxide.The speed of the platen was set as 50 mm/s and a travel distance of 50 mm in each direction.For our apparatus this generates 2 complete ALD cycles per second.For the metal oxide coated in this experiment this equates to about 7 nm metal oxide per minute.For all oxides, the flow rate of the carrier gas through the bubblers was 50 ml/min.The flow rate of diluting carrier gas was 300 ml/min for the water reactant and 200 ml/min for the metallic precursor.The flow rate of the inert separator gas was 2000 ml/min.Dry nitrogen was used for the inert carrier gas and inert separator gas in all instances.A calibration was run to determine the film thickness, measured using a Woollam α-SE ellipsometer, as a function of the number of substrate oscillations for the three oxides.
Figure 2 shows the calibration curves for alumina, zinc oxide and titanium dioxide.The coatings are linear in thickness with number of oscillations of the ALD head.Alumina appears to show an induction period when deposited in this way, since there is very little deposition of material below 100 oscillations.The rates of deposition for zinc oxide and titanium dioxide were 0.064 nm/metaloxidant cycle and 0.102 nm/metal/oxidant cycle for alumina each cycle taking approximately 0.5 s.Grigorias [13] reports a similar growth rate of alumina of 0.09 nm/cycle for a 1.5 s sample grown by conventional ALD.Kim [14] reports a rate of 0.202 nm/cycle for a 4 s total cycle for zinc oxide but this included 1 s metal cycle time to allow the metal precursor to saturate the surface which was emphasised to be necessary.Because of the need to have a rapid process, in the experiments described here the metal precursor only had a 20 ms residence time over each portion of the substrate in each cycle which would account for the slow growth rate.Sammelselg et al. [15] report a similar growth rate of 0.04-0.06nm cycle for a 4 s cycle for titanium dioxide.
These curves were used to aid deposition of layers with the correct thickness for the interference filters.As an additional check, the actual thickness and refractive index of each layer was measured by ellipsometry before the next step of the process was carried out.The crystal form of the coatings was determined by X-ray diffraction.
The results are shown in Table I along with references to materials made by conventional ALD at low pressure (about 1 Torr).

B. Fabrication of Interference Filters
The spectral transmittance of all optical filters were recorded using a Hewlett-Packard 8457 spectrophotometer.A simple three layer design was constructed to demonstrate the potential of the atmospheric pressure ALD equipment.A layer of aluminium (Goodfellow) about 10 nm thick was deposited by vacuum evaporation using a Moorfield Minilab.Onto this was deposited a layer of titanium dioxide 180 nm thick.A further layer of aluminium about 10 nm thick was deposited onto this.The thicknesses of these layers were determined by ellipsometry.The measured spectral transmittance of the filter was compared to the expected spectra modelled using conventional thin-film optics theory [6].The results are shown in Fig. 3.
An interference filter with Bragg reflectors was also constructed as follows: 5 alternating layers of zinc oxide and alumina each approximately 80 nm thick, were deposited onto the glass starting with zinc oxide.A 220 nm thick layer of alumina was coated onto this followed by another series of 5 alternating layers of 80 nm zinc oxide and alumina as shown in Fig. 4. The thicknesses of the layers were determined by ellipsometry.Again, the measured spectrum of this filter was compared to the modelled spectra as shown in Fig. 5.
It was decided to demonstrate a colour filter array of the simpler type of filter, using just a single layer metal reflector.In order to estimate the thicknesses required for an interference filter, a series of different thicknesses of titanium dioxide were coated successively onto a layer of aluminium 10 nm thick evaporated onto glass.The spectra of these coatings were recorded as shown in Fig. 6 and were used to determine the appropriate thickness of titanium dioxide for each colour in the CFA.Comparing the local minima in optical density with the non-overlapping visual sensitivity curves suggested by Boynton [18]  and 320 nm should give blue, green and red separation.

C. Patterning Titanium Dioxide Layers
A mixture of 25% w/w Fluoropel P604A (Cytonix Corp) + 75% perfluorodecalin (Sigma Aldrich) was used as a release agent to pattern the surface prior to deposition of the oxide film Its performance was determined by coating over half the width of glass slide using a K Control Coater (RK Print Instruments) with a 6 micron meter-bar, to deposit a wet film thickness of 6 micrometers.The solvent was allowed to dry and the glass was coated with titanium dioxide to a thickness of 200 nm by ALD.The sample was then washed with perfluorodecalin with gentle agitation and mild abrasion to remove the Fluoropel.It was observed that the region of the glass that had been coated with the Fluoropel prior to coating by ALD with titanium dioxide had no titanium dioxide layer on it after washing in solvent when observed by optical microscopy at 100× magnification.The untreated region of the glass had a layer of titanium dioxide which could be measured by ellipsometry and was unchanged by removal of the Fluoropel.Observation of the sample of titanium dioxide deposited on the Fluoropel before treatment with perfluorodecalin using optical microscopy revealed that the ALD coating of titanium dioxide was covered in very fine cracks when deposited on top of the Fluoropel, see Fig. 7.The titanium dioxide on the untreated coating had no cracking.Presumably the cracks allowed the solvent to penetrate the titanium dioxide layer and reach the Fluoropel which dissolved, undermining the titanium dioxide which subsequently became detached and washed off.
This lift-off technique enabled a means of patterning as the Fluoropel could be ink-jetted in a perfluorodecalin solvent.This patterning method could be used to make any pattern in an ALD coated layer.

D. Manufacture of the Colour Filter Array
A simple colour filter array was created by a combination of ALD and inkjet printed P604A, by printing squares of the fluoropolymer to act as a resist for the ALD layers.A glass slide was first coated with a 10 nm layer of aluminium by vacuum evaporation.Next a layer of tita- nium dioxide approximately 220 nm thick was deposited by ALD.
A mixture of 25% w/w Fluoropel P604A + 75% perfluorodecalin was deposited using a Dimatix DMP 2800 ink-jet printer.A pattern of three x three 5 mm squares of P604A was printed using the Dimatix printer, as shown in Fig. 8

A. Interference Filters
The layer thicknesses were determined to be 8.4 nm aluminium on the substrate, 180nm titanium dioxide and 8.0 nm aluminium on the top.Figure 4 shows the spectral characteristic of the filter.This is compared to an a priori model of the structure.The refractive index of the titanium dioxide was measured during thickness measurement by the ellipsometer and found to be 2.35.The filter did show band stop characteristics but differed from the modelled spectrum in that the peaks were displaced, the minima of our sample were greater than the model and the maxima were lower than the model.The model makes no allowance for roughness of the aluminium layers due to their probable discontinuous nature at this thickness and this would be expected to significantly affect the quality of fit.No allowance for the possibility of oxidation of the aluminium to alumina was made in the ellipsom- etry measurements of the whole filter.Nevertheless, the overall shape of the transmittance spectrum is in qualitative agreement with the model.It was found that a small increase in the measured thickness of the titanium dioxide to 193 nm was needed to fit the spectral position of the density minimum at 535nm and this figure was used in the modelled curve in Fig. 4.Although this filter is simple to make, it has poor overall transmittance as would be expected from filters using metal mirrors.
Fig. 5 shows the spectral transmission density of the more complex optical filter that uses a Bragg reflector on both sides of the cavity layer.The curve shows density minima at 470, 510 and 640 nm and density maxima at 570 and 700 nm.This filter appears magenta in transmission as it has a stop band in the green part of the visible spectrum.The filter has a much sharper cut-off than that of the simple one, and has much lower minimum optical density.A model was fitted and the best fit thickness of the central alumina layer was found to be 265 nm.The 11 layers exceeded the capability of the ellipsometer to model the reflectance of multilayer thin film structures and no accurate layer thickness determination could be made after coating was complete.There is an increased optical density in the experimental sample probably due to light absorption and reflection by the glass substrate which was not considered in the model.the ink-jetted masks.

IV. CONCLUSION
We have demonstrated a simple, scalable method for fabricating interference filters using a layer of metal oxide deposited by the atmospheric pressure ALD technique at about 7 nm per minute, between partial reflecting layers of evaporated silver metal.By depositing a different thickness of titanium dioxide, filters with various spectral characteristics can be made.An improvement to the transmission was made by using a Bragg reflector on either side of the half-wavelength filter each consisting of a stack of five alternating layers of low refractive index alumina and high refractive index titanium dioxide.A low cost and scaleable means of patterning the ALD layers using an ink-jetted resist has also been shown.By combination of the inkjet patterning of ALD and evaporative deposition of metal layers, a new route to fabricating colour filter arrays has been demonstrated.This method could be used to prepare any suitable pattern of CFA.The method is not restricted to making a 3 colour array e.g.RGBW filter [19] to give improved light transmission could also be fabricated.Colour filters created by this method should also have good barrier properties against both oxygen and water making them particularly suited to OLED and other environmentally sensitive devices.

FIG. 1 :
FIG. 1: Side view of the ALD manifold showing different gas chambers.

FIG. 2 :
FIG. 2:Graph of calibration curves used to determine metal oxide laydown thickness.

FIG. 4 :FIG. 5 :
FIG. 4: Diagram of the structure of the interference filter with Bragg reflector.

FIG. 6 :
FIG.6:The effect of titanium dioxide thickness on transmitted spectra.

FIG. 8 :
FIG. 8: Areas of the CFA.(a) The pattern for the blue regions, (b) the pattern for the green regions, (c) the predicted thicknesses of titanium dioxide and (d) Schematic design of the colour filter pattern.
(a).The sample was next coated with a layer of titanium dioxide approximately 50 nm thick before printing another pattern of three × three 5 mm squares of fluoropolymer, as shown in Fig. 8(b).After laying down a final layer of titanium dioxide approximately 50 nm thick, the fluoropolymer was removed using perfluorodecalin solvent and mild abrasion.Finally, another 10 nm layer of aluminium was deposited by vacuum evaporation.The selected thicknesses of titanium dioxide are shown in Fig. 8(c) and Fig. 8(d) shows the expected transmitted colours.

FIG. 9 :
FIG. 9: Image of final colour filter array viewed in reflection.

Figure 9
Figure9shows an image of the simple colour filter array viewed by reflection.One can see the complimentary colours of the transmission better using this technique.The yellow squares corresponding to blue transmission, the magenta, green transmission and the cyan, red transmission.The edge of the colour looks sharp but close examination shows gaps caused by poor registration of

TABLE I :
Refrective indices and crystalline forms of the coatings.