An elevated temperature study of a Ti adhesion layer on polyimide.

Titanium layers are used to promote adhesion between polymer substrates for flexible electronics and the Cu or Au conducting lines. Good adhesion of conducting lines in flexible circuits is critical in improving circuit performance and increasingcircuit lifetime. Nominally 50 nm thick Ti films on polyimide (PI) are investigated by fragmentation testing under uniaxial tensile load in the as-deposited state, at 350 °C, and after annealing. The cracking and buckling of the films show clear differences between the as-deposited and the thermally treated samples, cracks are much straighter and buckles are smaller following heat treatment. These changes are correlated to a drop in adhesion of the samples following heat treatment. Adhesion values are determined from the buckle dimensions using a total energy approach as described in the work of Cordill et al. (Acta Mater. 2010). Cross-sectional transmission electron microscopy of the Ti/PI interface found evidence of a ~ 5 nm thick interlayer between the largely columnar Ti and the amorphous PI. This interlayer is amorphous in the as-deposited state but nano-crystalline in those coatings tested at elevated temperature or annealed. It is put forward that this alteration of the interfacial structure causes the reduced adhesion.

Introduction

The realisation of electronic circuitry on flexible and stretchable substrates [1-3] requires the patterning of conducting lines onto flexible substrates like polyimide (PI) and polydimethylsiloxane (also known as PDMA). Cr and Ti adhesion layers are used to improve the adherence of the conducting line material [4-6], usually Cu or Au. The exact mechanisms by which materials such as Ti and Cr function as effective adhesion layers are not yet fully understood [7,8].

The mechanical properties of adhesion layers are critical for device performance and lifetime. The failure of the adhesion layer or any of the interfaces leads to reduced performance, if not total failure, of the flexible circuit. Much work is going on in the characterisation of adhesion layer materials and substrate/adhesion layer/conducting material systems [9-12]. A common means of investigating the mechanical properties of film and of the film/substrate interface is the fragmentation test [12-15]. In this experiment, samples of coated substrate are strained uniaxially and the evolution of damage in the coating (cracks, buckles and coating delamination) are tracked with increasing applied strain. With this method the film failure stress and statistical distribution of strength can be assessed [15,17], the shear stress at the film/substrate interface can be characterised [14,18] and the film substrate adhesion can be measured [19,20].

In addition to the characterisation of these systems in their as-deposited state, it is important to consider the effects of external stimuli. One such stimulus is the temperature due to further processing steps and use of the circuit. In a previous study [16] it was found that annealing a Cu/Ti/PI system led to the formation of intermetallic phases and the adhesion of the polymer interface was reduced compared to as-deposited samples. Further investigation of the Ti/PI interface was required to understand the effect of temperature on adhesion. In this study the Ti/PI interface was exposed to 350 °C, a rather high temperature for a metal/polymer system, which should accelerate interfacial ageing conditions. In order to resolve the microstructural processes taking place at the interface, transmission electron microscopy (TEM) investigation of these Ti/PI samples is employed.

Experimental details

A Ti film of nominal 50 nm thickness was deposited onto a cleaned PI substrate (50 μm thick UBE Industries UPILEX-R) by electron-beam evaporation (Balzers BAK 500) at a pressure of 3 ⋅ 10− 7 mbar. The cleaning procedure involved a 24 h, room temperature soak in a 10% aqueous solution of RBS 50 (a laboratory cleaning concentrate with high pH). This soak was followed by a deionised water rinse and a 30 kHz ultrasonic cleaning process. A deposition rate of approximately 0.5 nm s− 1 was maintained during the process. TEM investigation revealed the film to be (60 ± 3) nm in thickness where the error corresponds to the standard deviation of the measurements.

The samples were tested with a mechanical testing frame (Zwick) fitted with a non-commercial vacuum oven. Room temperature tests were carried out in air whilst tests at elevated temperature were conducted at a vacuum of 10− 4 mbar and a temperature of 350 °C. This setup typically required 90 min to reach 350 °C and took 60 min to cool down to a temperature at which the sample could be exchanged. All tests were performed at a strain rate of 4 ⋅ 10− 4 s− 1 on samples with a width of 7 mm and a 24 mm gauge length. The tensile tests took between 1 and 8 min to perform, depending on the level of strain, so the total time for each experiment was 150–160 min. Sample load data could not be recorded during testing as the low loads developed in these samples at elevated temperature could not be reliably resolved above the noise of the system. As no means of assessing the film crack spacing in situ was available in the current setup, crack development profiles were produced by straining multiple samples to different levels of strain. 8–10 samples must be tested to different levels of strain to create a profile of the crack density versus strain equivalent to that produced by an in situ test. In addition to the samples strained at room temperature and 350 °C, several samples were heated to 350 °C in the vacuum oven and held at this temperature for 90 min, these samples were then tested at room temperature.

The microstructure and thickness of the films and the structure of the film/substrate interface were examined with TEM (120 kV Philips CM 12 and 200 kV JEOL 2100F). Cross-sectional TEM samples were produced by a focussed ion beam (FIB, FEI Helios Nanolab) lift-out process. The Ti coating was protected from the ion beam during lift-out by first depositing ~ 2 μm of Pt using the electron beam and a gas injection system. These samples were finished on both sides with a 1 kV, 30 pA Ga+ polish inclined 7° to the surface to minimise the effects of ion damage and Ga-implantation. In addition to the cross sections, plan view TEM samples were produced by a film removal technique outlined in reference [21]. Grain size analysis of the Ti was carried out on these plan view samples by the line intercept method. TEM samples of the films were produced from the as-deposited, annealed and 350 °C tested material.

Examination of the samples after mechanical testing was carried out by optical microscopy (Olympus BX51), by scanning electron microscopy (SEM, Leo 1525) and by atomic force microscopy (AFM, Digital Instruments Dimension 3100) in tapping mode. Several in situ tests of samples in the as-deposited and annealed states were performed inside the SEM using a miniaturised testing frame (Kammrath and Weiss). These tests were carried out at the same nominal strain rate as the ex situ experiments but were strained step-wise such that secondary electron images of the film cracking were recorded at progressively higher levels of strain. AFM measurements were used to determine the dimensions of buckles formed by the Ti film after straining, the measured dimensions were then used to calculate film adhesion energies for the three sample conditions.

Results

Microstructure and cracking

The plan view TEM of the as-deposited, 350 °C tested, and annealed samples revealed the grain size of the samples to be unaffected by the heat treatments, Fig. 1. The grain sizes of these samples were measured to be (32 ± 8), (28 ± 8) and (24 ± 9) nm for the as-deposited, 350 °C tested and annealed samples, respectively. The error presented in these measurements is the standard deviation of the data, there is no statistically significant difference between these grain size estimates. The moiré fringes [22] in all the plan view TEM micrographs, Fig. 1, indicate that overlapping grains are present through the film thickness, i.e. the films do not have a purely columnar structure. This was confirmed from the TEM cross-sections. Finally, selected area diffraction (SAD) of the samples (not presented) confirmed the Ti films to be hexagonal close packed α-Ti.

Fig. 1

Plane view TEM micrographs of the A) as-deposited, B) 350 °C tested and C) annealed Ti films. No significant difference in the film microstructure or grain size was observed.

The fragmentation testing of the samples revealed a stark difference in fracture and buckling behaviour, Fig. 2. The as-deposited sample has cracks with rough edges and uncracked broad buckles with a triangular footprint, Fig. 2A. The 350 °C tested and annealed samples have cracks with very straight edges and cracked narrow buckles with an approximately rectangular footprint, Fig. 2B and c. Fig. 2D presents the evolution of the crack density in the coatings with applied strain, a strong difference in the crack density is observed between the three sample types. The crack density of the annealed sample at crack saturation is highest at 0.40 μm− 1 whilst the as-deposited sample has a density of 0.24 μm− 1 and the 350 °C tested sample 0.14 μm− 1. The crack density data for the sample tested at 350 °C was obtained from ex situ experiments, this causes the greater scatter between data points compared to the crack density data of the as-deposited and annealed samples. A difference in the strain at which cracking is first observed is also found. The annealed sample has the lowest crack onset strain with 1.6% compared to 5%for the as-deposited sample. An accurate crack onset strain could not be determined for the 350 °C tested sample, due to the larger strain intervals between ex situ tests, but the data indicates that cracking begins between 2% and 5% strain.

Fig. 2

SEM micrographs of the samples after cracking have saturated. Panel A) is as-deposited, panel B) 350 °C tested and panel C) is the annealed film. Shown in panel D) are the crack development profiles for the three samples, the lines are purely to guide the eye. Note the different cracks and buckle morphologies of the as-deposited sample compared to the heated samples.

Film adhesion

The interfacial adhesion energy is quantitatively measured with the buckles that form between the cracks. The model developed by Fischer et al. [20] takes into account the strain energy between buckles, the debonding energy and the strain energy of the buckled material. This model was developed for brittle films on polymer substrates and the buckle formation therein caused by the compressive stress evolved in the film perpendicular to the tensile straining axis [20].

To calculate the interfacial adhesion, the dimensions of the buckles are needed, namely, the buckle height, δ, and half buckle width, b. When the buckles' dimensions are plotted as as a function of (b/h), where h is the film thickness, the data can be described by:where α is a fitting parameter. The α parameter is used to calculated the adhesion energy, Γ, using:

In Eq. (2), h is again the film thickness and E′ is the modified elastic modulus ( = 129 GPa). With:where E and ν are the Young's modulus (E = 116 GPa) and Poisson's ratio (ν = 0.32) of the Ti film, respectively.

AFM was used to image and measure the buckle dimensions. As has been previously discussed [16,20,23], buckles which do not travel across the whole crack fragment (partial buckles) better describe the calculated adhesion energy. In this region of the buckle, there is no cracking of the buckle apex (top) or at the base of buckle. For the Ti buckles here only measurements from the end of partial buckles (Fig. 3) were used to determine adhesion. As shown from the different height scales in Fig. 3, the average buckle size of the film strained at room temperature is larger than at 350 °C.

Fig. 3

AFM height images of A) the as-deposited Ti film and B) the 350 °C Ti film. Note the different shapes of the buckles in the samples and larger height range for the as-deposited scan. Circles highlight the tips of partial buckles appropriate for adhesion measurement. Arrows indicate the direction of straining.

The adhesion energy using Eqs. (1) and (2) is calculated by fitting α on a diagram like that shown in Fig. 4. For the as-deposited film a minimum α of α = 3.0 ⋅ 10− 4 is fitted and the minimum is α = 1. 2 ⋅ 10− 4 for the 350 °C tested sample. From the α values the adhesion energy was calculated as (3.5 ± 1.2) J m− 2 for the as-deposited film and (1.4 ± 0.5) J m− 2 for the film tested at 350 °C. The degree of error was determined by changing the minimum α value by approximately ± 0.00005 to better account for the lower data points. A summary of the data for the three types of sample is given in Table 1.

Fig. 4

Plots of vs. b/h for the as-deposited Ti, A), and the 350 °C tested Ti, B), as proposed by the Fischer model [20]. The lower limit represents buckles as they first form and hence gives the most reliable measure of interfacial adhesion.

Table 1

Summary of the film properties determined. ∈ is the observed fracture strain (estimated for the 350 °C sample). The value of adhesion is calculated according to the Fischer model [20] for the as-deposited and 350 °C samples, or Hutchinson and Suo model [24] for the annealed sample.

Condition	Grain size/nm	∈f/%	Crack density/μm− 1	Adhesion/J m− 2
As-deposited	32 ± 8	5	0.24	4.7 ± 0.6
350 °C tested	28 ± 8	2–5	0.14	1.4 ± 0.5
Annealed	24 ± 9	1.6	0.40	2.6 ± 0.8

The annealed film adhesion energy was measured using the Hutchinson and Suo model [24] instead of the Fischer model due to the fact that partial buckles were not found, even at the initial buckling strain. Only buckles which propagated across the whole crack fragment were observed. However, spontaneous telephone cord buckles did form on an annealed sample, Fig. 5, and were used with the Hutchinson and Suo adhesion model. This model also uses the buckle dimensions of buckle height, δ, and half buckle width, b, to calculate the driving stress, σ, and buckle stress, σ, for the telephone cord buckle shape. Buckles were imaged using AFM and cross-sections were measured at the point of inflection of the buckles (shown in Fig. 5, dashed line) and not through the pivot point of the buckles [25-27]. The residual and buckle driving stresses are used in Eq. (4) along with the film thickness, h, and modified elastic modulus, E′, to calculate the mixed mode interfacial fracture energy, Γ(Ψ):where Ψ is the phase angle of loading. For more information on the Hutchinson and Suo model please see reference [24]. This analysis gives a value of (2.6 ± 0.8) J m− 2 for the annealed samples, the error being the standard deviation of the buckle measurements.

Fig. 5

AFM height image of spontaneous buckles found on an unstrained annealed sample. The classic telephone cord shape is observed. Buckle measurements were taken across the point of inflection of the buckles (dashed line) rather than at the pivot point of the buckle.

Cross-sectional TEM

To further investigate the Ti/PI interface, cross-sections of all three sample types were investigated with TEM. Images from an as-deposited sample and a 350 °C tested sample are presented in Fig. 6, of particular note is the different interfacial structure. Cross-sectional TEM of the annealed material was indiscernible from that of the 350 °C tested material and is hence not presented here. The lower magnification images, Fig. 6a and b, show that the basic structure of the coating does not change despite the different testing conditions. Both samples exhibit an approximately 5 nm thick surface layer (indicated to be titanium oxide by energy dispersive X-ray spectroscopy, EDS), crystalline Ti of around 45 nm thickness and an interlayer of 5 nm thickness between the Ti and the PI. However, the higher magnification images, Fig. 6c and d, indicate a difference in structure of the interlayer, amorphous in the as-deposited sample and nano-crystalline in the 350 °C and annealed samples. Cross-sections of areas of PI from which the Ti layer has been delaminated, not presented here, show that the coating consistently delaminates at the interlayer/PI interface, i.e. no Ti remains on the surface.

Fig. 6

Cross-sectional TEM micrographs of the Ti/PI interface of an as-deposited sample, A) and C), and a 350 °C tested sample, B) and D). The white markers on the left of each image indicate the location of the interfaces. Note the crystalline fringes in the interlayer in D) that are not present in C).

Discussion

The observed change in cracking and buckling behaviour of the annealed and 350 °C tested Ti compared to the as-deposited films is a strong indicator of a change in interfacial adhesion. The change in buckle morphology between the 350 °C tested and annealed Ti compared to the as-deposited material is due to the reduced adhesion in the heated samples. The decrease in adhesion, more than a factor of two from 3.5 to 1.4 J m− 2, causes the observed switch from triangular footprint buckles in the as-deposited samples to rectangular footprints in the 350 °C tested and annealed Ti. This occurs as the film stress perpendicular to the loading axis is a function of the film stress parallel to the axis:where ϵ is the applied strain, E is the stiffness of the substrate and ν is the Poisson's ratio of the film. ν is a composite term describing the perpendicular contraction of the substrate as a fraction of the stress parallel to the loading direction, it has elastic and plastic components. The dependence of the parallel, σ, and perpendicular, σ, film stresses as a function of location along the x-axis is shown schematically in Fig. 7. This stress state results in a compressive stress gradient from fragment edge to fragment centre. The release of strain energy driving cracking will therefore also decrease towards the fragment centre. For a system with low film/substrate adhesion, rectangular footprint buckles will form as a rectangle gives the greatest release of strain energy for a given area of film delamination. If adhesion is sufficiently high then triangular buckles will form. This is a result of the compressive strain energy in the centre of the crack fragment being insufficient to drive the formation of new surfaces. For further details of the biaxial stress state in a cracked film on a substrate the reader is directed to reference [12].

Fig. 7

Schematic of the stresses in a film fragment. A) A film fragment on a substrate with the load direction and principal axes indicated. B) The variation in tensile stress parallel to the loading axis, x-direction, across a film fragment during loading. C) The variation in perpendicular, y-direction, film stress in the x-direction. For a fragment that is infinitely broad in the y-direction the perpendicular stress is constant at any given position along the x-axis. After Frank et al. [12].

The reasons for the change in cracking behaviour are less clear. TEM observations, Figs. 1 and 6, indicate that no change in film microstructure takes place during elevated temperature testing or annealing and that the thickness of the surface oxide is also invariant. In addition, the similar crack morphology of the 350 °C tested and annealed samples demonstrates that the effect is permanent following heating at 350 °C rather than being due to the elevated temperature properties of Ti. As all heating was performed in moderate vacuum and titanium's native oxide is highly protective [28,29] it can also be assumed that the change in crack morphology is not due to the infiltration of environmental impurities into the film, causing embrittlement. A final possibility is hydrogen embrittlement of the Ti film. It has been shown that hydrogen environments can cause embrittlement of Ti and its alloys [30-32], it is possible that reactions within the PI at 350 °C lead to the release of hydrogen into the Ti film. No evidence of titanium hydride (TiH2) formation [31] was found in selected area diffraction of any of the coatings but the interaction of hydrogen with grain boundaries and dislocation cores can also lead to embrittlement [30].

A final observation relating to the cracking of the Ti films is the variation in crack density between the samples. The samples tested at 350 °C have the lowest crack density. This is consistent with the findings of Leterrier et al. [18] and Yanaka et al. [33] that saturation crack densities of brittle coatings correlate with substrate mechanical properties, i.e. that the reduced substrate stress at 350 °C leads to a reduced crack density. This difference is most clearly seen by comparing Fig. 2b and c, the crack density of the annealed sample tested at room temperature is more than a factor of two greater than that of the sample tested at 350 °C. It is less clear why the crack density of the annealed sample is higher than that of the as-deposited Ti. Several groups [13,34,35] have related the average film crack spacing at saturation to the adhesion of the film/substrate interface, on the principle that an interface which can maintain a higher shear stress, leading to more closely spaced cracks, possesses greater adhesion. In the present case, the annealed Ti/PI has been shown to have a significantly reduced interfacial adhesion compared to as-deposited Ti/PI and yet the annealed sample has a greater crack density. This effect is also a result of the Ti becoming embrittled by heat treatment. The annealed sample has a significantly lower fracture strain, 1.6% versus 5%, than that seen for the as-deposited Ti, Table 1, this strongly indicates a lower film fracture stress following annealing. A reduction in film fracture stress is also expected to increase the crack density at saturation [14,18,15]. These results show that relating crack spacing and interfacial shear stress to film/substrate adhesion should be done with caution, ensuring all other experimental factors remain constant.

The quantitative adhesion energy results derived from buckle measurements found a factor three reduction for the samples tested at 350 °C. In addition to the quantitative measure of adhesion supplied by the buckle dimension measurement, a qualitative tape test was also carried out on the samples. In this test an adhesive tape was applied to the samples following testing, the removal of the tape leads to local film delamination. By examining the degree of film removal a simple indicator of film/substrate adhesion is achieved. This purely qualitative test was not carried out according to any defined standards. The results from this test, Fig. 8, support the adhesion measurements made from tensile strain-induced buckles, i.e. that the 350 °C tested sample has significantly lower adhesion. Tape-testing of the 350 °C tested and annealed samples always removed the majority of the Ti coating, limited film removal was only achieved for highly strained, > 10% strain, as-deposited samples.

Fig. 8

Optical micrographs of the A) as-deposited Ti and B) 350 °C tested Ti are presented following the application and removal of adhesive tape to the film surface. Significantly more of the 350 °C tested Ti were removed by the tape, indicative of lower adhesion at the Ti/PI interface in this sample. The dashed lines mark the lower edge of the tape.

In order to make quantitative adhesion measurements of the samples in the as-deposited, 350 °C tested and annealed conditions it was necessary to utilise two different models for film adhesion; the Fischer model for uniaxially strained brittle films [20] and the Hutchinson and Suo model for spontaneous buckling [24]. The adhesion values measured for the 350 °C tested and annealed samples are very similar, (1.4 ± 0.5) versus (2.6 ± 0.8) J m− 2, despite the different models used. As the cracking and buckling behaviour of the 350 °C tested and annealed samples is very similar and the tape test also yielded similar results, it is encouraging that the two models give equivalent results for the interfacial adhesion. All the adhesion measurements indicate that the adhesion of the Ti film decreases following heat treatment and that this decrease is irreversible.

A similar study of the adhesion energy of Cu/Ti films on PI also found a decrease in adhesion energy after annealing for 1 h at 350 °C [16]. For the Ti/PI system investigated in this study, the interfacial TEM observations suggest that this change in adhesion is related to the structure of a Ti/PI interlayer at the interface. A ~ 5 nm thick interlayer was observed between the Ti and PI in all samples investigated. However, crystalline fringes consistent with the presence of nano-crystallites of Ti were observed in the interlayers of the 350 °C tested and annealed samples but not of the as-deposited material. This indicates that either the deposition process or the inherent chemistry of Ti leads to the formation of an amorphous layer during the initial stages of deposition. The heating experiment presented here indicates that this amorphous interlayer is thermodynamically unstable. The heat treatment applied here is extreme and the authors do not claim it to be representative of the applications for such adhesion layers but it does highlight a concern warranting further investigation. From the present results it is not possible to say if the change in the interlayer is simply from amorphous Ti to crystalline Ti or if the initial state is a more complex mixing of Ti atoms with the polymer chains of the substrate. Determining changes in chemistry with such high spatial resolution requires highly specialist instrumentation and is beyond the scope of this work, such information would however be very valuable in understanding the processes at work at the interface.

The observation of reduced film adhesion following heat treatment is important for the emerging field of flexible electronics, regardless of the cause of this reduction. Elevated temperatures, > 100°, are routinely experienced by electronic circuitry during deposition, processing and use, if this can lead to reductions in interfacial adhesion of components then circuit lifetime will drop. The 350 °C testing and annealing carried out is not representative of the temperatures experienced by flexible circuitry but as an accelerated test it does illustrate that circuit thermal history is an important factor in circuit lifetime. If the adhesion drop observed is due to the thermodynamic instability of an interlayer then this process will also occur at lower temperatures but more slowly. Understanding and learning to counteract this effect should then lead to improved circuit lifetimes and performance.

Conclusions

Ti adhesion layers on polyimide (PI) flexible substrates exhibit very different cracking and buckling when uniaxially strained in the as-deposited state and at 350 °C. The shape of the cracks and buckles changes and analysis of the buckles shows a significant decrease in interfacial adhesion when tested at 350 °C. TEM of the films in the two states and room temperature testing of samples annealed at 350 °C shows that these changes are due to the heat treatment of the material.

A cross-sectional TEM study infers that the difference in adhesion caused by heat treatment of these Ti films is linked to the crystallisation of an interlayer between the Ti and PI. This change in Ti/PI interface adhesion has enormous implications for thermal management during processing and service and thus for the lifetime of flexible electronic circuits. Additionally, this greater understanding of how Ti acts as an effective adhesion layer between polymers and metallisation materials such as Au and Cu could lead to more adherent, and hence failure-resistant, flexible circuitry.