Localized thin film damage sourced and monitored via pump-probe modulated thermoreflectance

Mitigating the potentially destructive effects of thermal buildup in materials with reduced dimensionalities is crucial to creating functional devices on the nanoscale. Major concerns in the micro- and nano-electronics communities are device efficiency limits, reliability, and the onset of mechanical failure of integrated components and metallic interconnects subjected to increasingly large thermal loads.^1,2 A common mechanism of thermally induced failure is thin film damage via dewetting and delamination.³ 

Common means of characterizing thin film damage consist largely of inspecting a material system after a damage event has occurred, for example, heating or ion-irradiating a film and reviewing the subsequent morphological changes in a scanning electron microscope (SEM).^4–6 This post-mortem analysis is often time consuming and lacks the ability of real-time, direct probing of the damage event that an in situ detection method would enable.

Alternative techniques to characterize film damage include the measurement of electrical resistance,⁷ x-ray diffraction,⁴ and optical detection of dewetting via major changes in the transmission of light through a film^8,9 or acoustic reflection off film interfaces.¹⁰ In particular, optical techniques have the advantage of providing in situ probes of real-time damage dynamics, while also offering the possibility of utilizing the energy deposition rates achieved by short-pulsed lasers to study material damage.^11–15 On very high energy scales, ablation of thin films has been seen to be of particular interest to researchers both computationally^16–20 and experimentally.^21–25 

In this work, we demonstrate the use of the modulated thermoreflectivity signal as a means to monitor thermally induced damage of a thin metal film on a substrate. Thermoreflectance is the reflectivity change of a material due to a perturbation in temperature.^26–28 This quantity is described by the electronic dielectric functions in the metal and, hence, is intimately related to electronic excitations, scattering, and structure.^29,30 At sub-ablation energy scales, using ultra-fast pulsed lasers in a pump-probe experimental setup, we demonstrate the detection of film damage via thermoreflectance of the metal film. We show, through the measurement of the reflected probe, sensitive detection of the degree of damage following the absorption of a train of pump pulses. The advantage of this approach lies in the use of time-domain thermoreflectance (TDTR),^31,32 an experimental setup that exhibits enhanced sensitivity to changes in material and microstructural properties of the material being interrogated. As typical TDTR measurements rely on an intact thin metal film, this study also serves to inform TDTR, and other thermoreflectance-based technique, users of the thresholds and signatures of thin film damage that may inhibit the use of such measurements.

Time-domain thermoreflectance is a measurement technique during which a train of laser pulses is split into two beams: an amplitude-varying pump beam, which contains a modulated train of laser pulses creating a frequency-dependent heating event at the sample surface, and a time-delayed probe beam, which samples the heating event as it decays between pulses. The component of the reflected probe beam at the heating frequency of the pump is isolated using a lock-in amplifier and the thermoreflectance can be monitored as a function of pump-probe delay time. By changing the delay of the probe pulses to coincide precisely with the arrival of the pump pulse at the sample surface, we maximize the signal-to-noise ratio of our reflected probe. The thermoreflectance coefficient of a material is extremely sensitive to changes in morphology, defects, and carrier concentrations, which can provide indication of the extent and time evolution of induced damage. Additionally, by utilizing all optical heating and sensing events, TDTR can provide information regarding localized damage on the spatial length scale of focused beam diameters.

It is well known for thin films that temperatures needed to induce damage, such as dewetting, decrease with the film thickness. For example, 20 nm thick gold films have been shown to begin island formation, and thus visible damage characteristics via a SEM, at temperatures as low as 350 °C.³³ Using focused lasers and a low thermal conductivity substrate, these temperatures can easily be achieved in a local area with relatively low power, nJ/pulse sub-picosecond oscillators, which are commonly used in TDTR experiments. This allows for in situ detection of localized damage caused by such temperature rises.

To demonstrate the use of TDTR as an efficient means of detecting film damage, we use a 20 nm film of electron-beam evaporated Au on a glass substrate deposited at 6 $×10−6$ Torr using planetary rotation to ensure uniform layer deposition. The glass substrates are treated using a series of alcohol cleans (methanol, acetone, and isopropyl alcohol) in addition to an O₂ plasma clean for 30 min. The use of a thin metal film on a low thermal conductivity substrate allows us to achieve sufficiently high localized temperatures and simultaneously monitor thermoreflectance, thus achieving an in situ measurement of thin film damage.

Our TDTR system is based around the output of a Spectra Physics Tsunami oscillator with a bandwidth of 10.5 nm centered around 800 nm with a repetition rate of 80 MHz, leading to 12.5 ns between pulses. The output of the laser is split into pump and probe paths which are further manipulated in a manner described below. The wavelength of the pump is doubled to 400 nm and modulated using an electro-optical modulator (EOM) with an 8.8 MHz square wave. In order to create a significant heating event, the pump pulses were focused down on the surface of the Au film. The 800 nm probe pulses are focused to an area less than that of the pump pulse. To further maximize the signal, the path lengths between the pump and probe are matched such that both pulses arrive at the surface of interest within 500 fs of each other. The cross correlation of the pump and probe pulses yields a FWHM of 700 fs. Using a lock-in amplifier, we are able to isolate this modulated component of the reflected probe beam and achieve a signal-to-noise ratio of well over 1000:1. The magnitude of the thermoreflectance signal is recorded every 300 ms and the signal is monitored for 5 min continuously.

The power of the pump pulse train, measured by an optical power meter, is varied to determine the onset of damage in the film. The induced damage is further inspected using a scanning electron microscope to demonstrate the use of the TDTR signal as a means of structural characterization; any change in the thermoreflectance signal can then be correlated to the clear microstructural changes from laser interactions.

As previously mentioned, laser pulses in our TDTR experiment emanate from the Ti:Sapphire oscillator once every 12.5 ns. This time between pulses leads to the accumulation of the energy from the various pulses; in other words, full dissipation of the generated heat into the substrate from the average power absorbed by the pump and probe pulses occurs on a longer time scale than 12.5 ns, especially for thermal insulators like glass. This results in a steady state temperature rise from the average power absorbed by the Au film, which is the local heating event that drives film damage. We can approximate this temperature rise, $ΔT$⁠, from the multilayer solution to the radially symmetric heat equation with a periodic Gaussian heat source absorbed at the surface of a thin film on a semi-infinite substrate, given by³¹ 

where P₀ is the absorbed power, r₀ is the probe radius (5.0 $μm$⁠), r₁ is the pump radius (13.2 $μm$⁠), and G(k) is the Hankel transform of the radially symmetric heat equation on the surface of a multilayer semi-infinite substrate. This function takes into account the thermal conductivity and volumetric heat capacity of each layer (in this case, a gold thin film and a glass substrate) and is explained in greater detail by Cahill.³¹ The assumed thermal conductivity of the glass substrate is 1.35 W m⁻¹ K⁻¹ and the assumed volumetric heat capacity is 1.62 MJ m⁻³ K⁻¹.³⁴ The thermal conductivity of the gold film is measured using electrical resistivity measurements and the Wiedemann-Franz law³⁵ for each film thickness (corresponding to 60 W m⁻¹ K⁻¹ for the 10 nm film, 130 W m⁻¹ K⁻¹ for the 20 nm film, and 190 W m⁻¹ K⁻¹ for the 30 nm film) and the volumetric heat capacity is assumed to be 2.49 MJ m⁻³ K⁻¹.³⁶ The absorbed power is measured by the analysis of transmission and reflection of incident power, and the pump radius is measured using a scanning slit optical beam profiler. Error in the calculated steady state temperature arises from variation in the spot radius (±1 μm).

As shown in Fig. 1(a), at low power fluxes near the measured power threshold, the thermoreflectance signal is constant throughout the entire scan, signifying a stable, unmodified Au film; this is typical for a standard TDTR measurement at a fixed time delay. Fig. 1(c) shows that higher power fluxes well above this threshold, however, exhibit a nearly immediate, within the 300 ms time constant of our lock-in amplifier, decay in the thermoreflectance signal once the pump beam impinges upon the surface. The decay of the thermoreflectance signal is observed in Fig. 1(b), at an intermediate power density, where the signal drifts significantly but does not diminish as rapidly or reach as low in magnitude as the higher intensity case.

These samples were analyzed further via an SEM. Figure 2 shows that the severity of damage correlates to the intensity of the heating (discussed in more detail below), as well as the degree of the thermoreflectance decay, as shown in Fig. 3. In the intermediate intensity case (Fig. 1(b)), the damage is subtle and the thermoreflectance signal has changed due to very small changes in the microstructure. In this case, it can be very challenging to locate regions of damage via a SEM using post-mortem procedures, whereas our TDTR signal has given a real-time indication of the initiation of film damage. Thus thermoreflectance decay is a good indicator not only of the onset of film damage but also of the extent of the damage occurring. In the extreme case, the gold has receded and left very large open areas in the film. By the thermoreflectance signal (Fig. 1(c)), we can see that this process occurred relatively quickly compared to the case of Fig. 2(b), and then an equilibrium was reached where no further decay in the signal was observed.

We quantify the extent of the damage by the percent of uncovered area in the SEM images. In Fig. 3, we compare the percentage of uncovered area to the decay in thermoreflectance for multiple times prior to the measured time constant at a given power density for our 20 nm films. The change in the thermoreflectance signal is comparable to the amount of area that has been uncovered by damage, especially at degraded signal intensities less than 60%. Although the trend matches, the difference in the detected change in the film at higher signal degradation can be attributed to the sensitivity of thermoreflectance to factors such as film thickness and microstructure, which are likely to change significantly as Au agglomerates into the bridges and islands shown in Fig. 2 and in higher magnification in Fig. 5. It is important to note that the time scale over which damage initiates, due to heating from pulse accumulation, is likely less than the 300 ms time constant of our lock-in amplifier. Even so, we are able to detect this damage and comment on the time evolution of this damage over longer time regimes. Thus, processing dynamics occurring at earlier times are outside the scope of this article.

The degradation in the thermoreflectance signal can be used to quantify a damage threshold for a given thin film system. We employ a damage metric of the percent change between the peak thermoreflectance signal and the signal after 150 s (well after the degradation plateau for the highest power density case). We plot the film damage based on TDTR signal degradation as a function of absorbed power, recast as induced temperature rise based on Eq. (1). To demonstrate this thermoreflectance metric, we fabricated three Au samples with nominal film thicknesses of 10 nm, 20 nm, and 30 nm. We can see from Fig. 4(a) that the relative damage threshold of each sample is distinct and increases with increasing film thickness. We use 10% signal degradation as a metric for the onset of damage, a criteria which mimics that used elsewhere in the literature³³ and ensures a signal well outside of the noise of our measurement (over 3 standard deviations). By this metric, the onset of damage occurs at equivalent temperature rises of $16±26$ K, $275±45$ K, and $383±25$ K for the 10 nm, 20 nm, and 30 nm films, respectively. Thresholds are determined by the interpolation between points where the temperature rise and 10% damage criteria are not explicitly measured. Compared to a similar thin film dewetting analysis on thin Au films using bulk anneals and post-mortem SEM review, we find that the onset of localized damage sourced and detected by TDTR occurs at roughly equivalent temperature rises to those found in the literature where island formation on 20 nm of Au on glass occurs at 344 K.³³ Furthermore, these findings show a direct correlation of film thickness to the ability to withstand incident energy without damage, also consistent with findings from other methods in the literature.³ The error in the measurement of the signal degradation comes from the standard deviation of average measurement data from 10% of the total scan time. We fit the decay in the thermoreflectance signal to an exponential decay in time, $Ae(−t/τ)$⁠, from the onset of heating to the 150 s point and plot the resulting time constants, τ, in Fig. 4(b). This provides a metric for the rate of damage in the films which is normalized to the absolute value of the thermoreflectance signal.

The mechanism of film damage observed in these measurements is likely driven primarily by thin film dewetting. As seen in the higher magnification SEM image in Fig. 5, the microstructural signature of the damage includes faceted islands around the uncovered substrate. This signature is consistent with the well known characteristics of solid thin film dewetting.^3,33 While this demonstrates the final microstructure after the laser induced heating, the exact evolution of the thin film microstructure during the signal degradation may include other thermally induced mechanisms that alter the thermoreflectance of the film such as burning of contamination, strain induced delamination, and coarsening of processing defects. This measurement provides insight into the degree of film damage, but further information into the dynamics of the evolution of the thin film structure during the damage process is beyond the scope of this study.

It is important to note that the films used in this study were deposited using a relatively low energy technique. This could very well result in pinholes, micro-cracks, and other non-idealities in the as-deposited films that may lead to discrepancies in the absolute value of damage thresholds with other findings. A major strength in the use of thermoreflectance as a monitor for film damage, however, is that we are able to detect relative changes in the film structure upon heating via changes in the reflected signal. The detection of relative changes enables us to remain sensitive to alterations in the film structure regardless of the initial defect state or film quality as long as we can obtain a baseline measurement, in this case using low enough power where damage does not occur, against which one can compare measurements using higher powers.

This measurement technique for the damage threshold can easily be extended to other material systems. The process of depositing a blanket thin film on a substrate and monitoring changes in thermoreflectance provides a real-time measurement of local damage. The real-time detection allows for immediate feedback, without the necessity of subsequent microscopy. In addition, the highly local damage allows for large statistical sampling and high-throughput testing of damage thresholds since the damaged spot is generally very small compared to the dimensions of a given substrate. In other thermally induced film damage measurements, one would have to sacrifice an entire substrate for each data point and nucleation of damage, if not widespread, could be much harder to detect than with local heating and detection. Depending on the application, the local heating used in TDTR is likely small enough to even be considered virtually non-destructive. Lastly, these findings can be extended to ultra-fast pump-probe measurement techniques in general to verify that the laser intensities being used are below thresholds for material damage.

In conclusion, thermoreflectance can be used as a convenient, rapid, high-throughput metric to judge changes in film microstructure and can therefore lend insight into thin film damage. Information such as the degree of film damage can be determined in extremely controlled systems and on very small scales. In the findings used to demonstrate this technique, we show that significant changes in the local film microstructure occur very close to previously reported temperature thresholds for Au thin films and the degree of damage scales with increasing power, confirming the validity of thermoreflectance as a means to determine film damage characteristics.

This material is based upon work supported by the Air Force Office of Scientific Research under Award No. FA9550-15-1-0079.