Pulsed laser processing plays a crucial role in additive manufacturing and nanomaterial processing. However, probing the transient temperature field during the pulsed laser interaction with the processed materials is challenging as it requires both high spatial and temporal resolution. Previous transient thermometry studies have measured neither sub-100 µm spatial resolution nor sub-10 ns temporal resolution. The temperature field induced by Gaussian laser beam profiles has also not been accounted for. Here, we demonstrate a 9 ns rise time, 50 µm sized Pt thin-film sensor for probing the temperature field generated by a nanosecond pulsed laser on a semiconductor thin film. The measurement error sources and associated improvements in the thin film fabrication, sensor patterning, and electrical circuitry are discussed. We carried out the first experimental and theoretical analysis of spatial resolution and accuracy for measuring a Gaussian pulse on the serpentine structure. Transparent silica and sapphire substrates, as well as 7–45 nm insulation layer thicknesses, are compared for sensing accuracy and temporal resolution. Finally, the measured absolute temperature magnitude is validated through the laser-induced melting of the 40 nm thick amorphous silicon film. Preliminary study shows its potential application for probing heat conduction among ultrathin films.

Pulsed laser processing has been widely applied in manufacturing, including machining,1 marking, welding, and annealing of semiconductors.2 Recently, it has been incorporated into the additive manufacturing of materials of high melting point and thermal conductivity, including tungsten.3 Furthermore, pulsed laser processing of nanomaterials has been demonstrated recently, including the sintering of nanoparticles4–6 and nanowires7–9 as well as the directed assembly of photonic structures.10–14 For a wide spectrum of applications, precise characterization of the involved physical and chemical mechanisms will play a significant role in the understanding of light-matter interaction15 and improving the manufacturing quality control.16–18In situ characterization at microscale spatial resolution and at nanosecond time scale is therefore desired. Although it provides the most fundamental information for characterizing these processes, temperature measurements at such a resolution are still missing, hindering the wide study and application of pulsed laser interaction with materials. The main challenges lie on the 1010 K/s heating and cooling rates, the microscale minimum resolution, and the spatial nonuniformity induced by the spatially varying laser beam intensity profiles. It is noteworthy that a Gaussian distribution is the most widely used laser beam shape due to its capability for high numerical aperture focusing and stability in long-distance light delivery.

The resistive thermometer is an ideal candidate for probing pulsed laser-induced temperature fields on various materials. In general, the microscale temperature probe can be categorized into three types:19 electrical, optical, and physical contact. Physical contact methods such as scanning probe thermal microscopy or thermocouples typically have slow temporal response due to the large thermal mass of probe. On the other hand, electrical resistance,20,21 thermal radiation,22,23 and thermoreflectance method feature nanosecond temporal resolution. Though the nonintrusive nature of radiation and thermoreflectance offers unique strength, mounting a bulky optical apparatus onto the laser processing setup is both time- and resource-consuming for R&D. Another limit of the optical method is that it is not capable of operating with broadband energy sources such as lamps and plasma, commonly used in semiconductor processing, because they will interfere with the probing optical wavelengths. Furthermore, surface modifications including ablation, melting, or oxidation24 introduced by the laser will also affect the optical response, lowering the accuracy of the temperature measurement. Electrical resistance is ideal for studying the laser material interaction as it does not require a dedicated optical setup, nor prior optical property characterization and analysis of the sample surface condition.

High spatial and temporal resolution temperature measurements of Gaussian laser pulse irradiation are lacking in the literature. Brunco et al.20,21 developed a Pt thin film resistive thermometer to probe the nanosecond laser heating of thin silicon films. A Pt film was embedded underneath the absorbing layer with nitride as the insulation layer. The sensor identified the melting temperature for pure Si and its alloy assisted by heat transfer simulation. However, the rise time was found to be 100 ns, and the spatial resolution is also limited to 1 mm. A dynamic thin-film based microscale thermocouple25 with 28 ns rise time and 25 µm size has been fabricated. However, the sensor does not probe the temperature field of the laser irradiated area. Instead, it probed the region proximal to the laser irradiated area. Furthermore, none of the sensors demonstrate the capability to probe temperature differences caused by the subtle variation of film thicknesses. Finally, all the above measurements did not address Gaussian shaped beam profiles associated with focused laser beams. A systematic analysis of the required sensor pattern and algorithm adjustment based on the thermal analysis is required.

In the current study, we show the design and validation of a Pt based thermometer of 9 ns rise time and 50 µm active area. We demonstrate the capability to detect temperature distributions induced by Gaussian laser beam profiles, discuss the error sources and corresponding optimization steps before establishing its spatial, temporal resolution, accuracy, and precision. Coupled thermal and electrical simulation is employed for comprehensive theoretical analysis. Furthermore, we confirm the absolute level of sensor measurements against nanosecond laser-induced melting of an amorphous silicon film. Finally, we discuss the sensor’s applications for probing the heat transfer of multilayer ultrathin films.

The overall sensing instrument consists of the sensor, electrical circuits, and data processing unit. The fabrication and calibration of a 50 µm sized 50 nm thick Pt sensor on transparent substrates has been described in previous work,26 which is designed for probing transient temperature fields induced by pulsed lasers. The Pt thin film is adhered to transparent substrates through the adhesion layer, and then it is covered by an insulation layer on the top. Through wet-etching, we remove the oxide and expose the contact pad region of the Pt film. Then, laser machining patterns both oxide and Pt in the center region of the sensor to have the serpentine structure. The patterned and insulated sensor structure is illustrated in Fig. 1(a). The same sensors and Temperature Coefficients of Resistance (TCR) from previous work26 are used in the current study. The sensor’s electrical contacts are directly connected to the lead wire with a silver paste [Fig. 1(a)]. For laser processing, an absorbing layer will be deposited on top of the sensor and receive laser energy. Our previously reported work26 showed that the sensor capped with an additional germanium absorbing layer can provide repeatable TCR performance over several cycles of heating and cooling. For characterization of the sensor performance, both sensors with and without top layers are studied.

FIG. 1.

Schematics of the sensor, its working principle, and error analysis. (a) The schematics of the sensor with an inset showing its cross section. (b) The working principles with the upper row showing the data flow and the lower row showing the correlations, and (c) the error analysis.

FIG. 1.

Schematics of the sensor, its working principle, and error analysis. (a) The schematics of the sensor with an inset showing its cross section. (b) The working principles with the upper row showing the data flow and the lower row showing the correlations, and (c) the error analysis.

Close modal

Figure 1(b) shows the schematics of the ultrafast thermometer’s working principle. The sensor temperature after laser irradiation forms a T(x, y, t) field because the Pt layer is thermally thin in the z-direction. The temperature distribution T(x, y, t) leads to the electrical conductance distribution σ(x, y, t) due to σ(T) correlation. Different from the Brunco et al.20 flat beam irradiation, the current sensor has a serpentine structure for probing Gaussian beam irradiation. As a result, the detailed temperature distribution T(x, y, t) and associated σ(x, y, t) will be further discussed in Sec. III A. Note that the relevant RF signal characteristic wavelength can be estimated to be 3–30 m, based on λ=cfVF and a velocity factor (VF) of 0.5 as well as a frequency f of 10–100 MHz. Since the sensor size is 2–5 orders smaller than the RF signal wavelength, we infer that the local variation σ(x, y, t) will not affect the RF signals beyond a lumped resistance R(t). The R(t) then affects the measured V(t) through the customized circuit. Based on the transient V(t) and other static voltage measurement, we can formulate the probed resistance Rprob (t) into algorithms. Finally, the resistance is translated to the probed temperature Tprob(t) based on calibrated TCR from Ref. 26.

As depicted in Fig. 1(c), the sources of errors are categorized into three fields, i.e., electrical, thermal, and material. Electrical errors are associated with the contact resistance and circuit response. Material errors comprise the sample resistance stability against different thermal annealing effects from fabrication and the degradation effects associated with laser machining. Thermal errors are associated with laser spot size and the serpentine sensor structure. Based on these analyses, we illustrate the efforts on sensor design and fabrication toward reducing error and improving sensitivities in Secs. II A–II C.

The substrate substantially affects the laser-induced spatial and temporal temperature evolution. As mentioned in our recent work,26 the transparent substrate allows for heating from top or bottom. Sapphire and silica are two chemically inert and thermally stable substrates with melting points reaching 1617 °C and 2050 °C, respectively, higher than most metals or semiconductors. From the classic pulse energy heating of semi-infinite solids,27 both the thermal conductivity and the product of the specific heat and density will affect the transient temperature evolution at a given depth. These substrates have distinctly different thermal conductivities (silica is 1.5 W/mK and sapphire is 25 W/mK) and relatively close specific heat and density, i.e., both their specific heat values are very close to 700 J/kgK, while silica has a density of 2650 kg/m3 and sapphire 3980 kg/m3. The selection of the substrate offered two typical cases for real substrates at the time scale we are interested in. For ultrafast laser processing, most thermal processes conclude within 100 ns. The thermal penetration depth in both aforementioned materials is within 1 µm (sapphire and silica’s thermal diffusivities are 8.97 × 10−6 m2/s and 8.09 × 10−7 m2/s). Consequently, silica substrates resemble applications where the Pt layer is backed by thermally insulating materials like silicon oxide. On the other hand, sapphire mimics the thermal behavior of substrates having high thermal conductivities, such as crystalline Si, Ge, and III-V at elevated temperatures. To improve the adhesion of the Pt film to silica, we deposited 7 nm Al2O3 through atomic layer deposition (ALD) as an adhesion layer.26 

Depositing an insulation layer between the Pt film and the probed specimen is critical as it prevents electrical shortage, chemical diffusion, or mechanical damage to the Pt layer. The sensor film quality is affected by the thickness of the insulation layer. In Table I, the Pt film is insulated with 45 nm of PECVD oxide (350 °C, 15 min) or 7 nm thick ALD alumina (300 °C 40 min). Sheet resistances are reduced to a similar level after depositing a different insulation layer. The main cause is the high-temperature annealing accompanying the deposition processes. Insulation can also improve the mechanical properties of the film from brittle to ductile with the enhancement of adhesion (see the supplementary material, Fig. S1). The sensor’s TCR performance is adapted from the cooling parts reported in Ref. 26. The thickness of the insulation layer will further affect the sensor’s sensitivity and its potential application is discussed later.

TABLE I.

Sheet resistance measurements of Pt film on different substrates before and after deposition of different insulation layers.

BeforeInsulationInsulation
Adhesioninsulationwith oxidewith alumina
No.Substrateslayers(Ω/sq)(Ω/sq)(Ω/sq)
Silica … 3.81 2.58 2.57 
Silica Al2O3 3.46 2.67 2.54 
Sapphire … 3.37 2.70 2.71 
BeforeInsulationInsulation
Adhesioninsulationwith oxidewith alumina
No.Substrateslayers(Ω/sq)(Ω/sq)(Ω/sq)
Silica … 3.81 2.58 2.57 
Silica Al2O3 3.46 2.67 2.54 
Sapphire … 3.37 2.70 2.71 

Different from the “transmission line” setup by Brunco et al.,21 a clear DC and RF separated path is realized through 2 identical bias tees (Fig. 2), which eliminated the tuning and matching of the capacitor and inductor. The DC route biases the sensor and constitutes the charge conservation to formulate the original resistance R0. The RF route is dedicated to isolating the transient voltage caused by the resistance change. The bandwidth of the RF is greater than 4 GHz based on the specifications of bias tee. Four channels of signals are terminated in a 500 MHz 4G sample/s oscilloscope (Rigol, DS4000). A LabVIEW program is developed to simultaneously record the signals and interpret the temperature in situ with laser processing.

FIG. 2.

Schematics for the electrical circuits and experimental setup. Channel 1–4 records transient voltage, laser triggered photodetector signal, DC bias voltage, and DC load voltage.

FIG. 2.

Schematics for the electrical circuits and experimental setup. Channel 1–4 records transient voltage, laser triggered photodetector signal, DC bias voltage, and DC load voltage.

Close modal

The error associated with the circuit can be recognized mainly in the waveform distortion and the response time limit. The waveform distortion is caused by impedance mismatch and is discussed in this section, while response time limit will be discussed in Sec. III B. All the electrical connections in the RF route are made with coaxial cables that match the impedance of 50 Ω. The coaxial optimized probe station cannot fit into the chambers where laser processing happens. Therefore, the parasitic capacitance and inductance of our lead wires will serve as sources for impedance mismatch. The analysis of the captured waveform entails a ringing frequency of 50 MHz [Fig. 3(a-iii)], whose effective wavelength is estimated as 6 m based on VF×cf and the velocity factor (VF) of 0.65. The clip and leads are 1 m in total length, two orders larger than the dimension of the sensor (<1 cm), and less than one order smaller than the effective wavelength. Therefore, we confirm that the clip and leads are the main sources of impedance mismatch and the resulting ringing. The mismatch can be reduced by registering the sensor to an impedance matched printed circuit board (PCB) through wire-bonding, which is not pursued in this study.

FIG. 3.

Ringing effect and comparison of signal processing options. Signals of nanosecond laser irradiation on the sample with 50 nm Ge deposited on (a) 7 nm alumina as an insulation layer or (b) 45 nm oxide as an insulation layer. [(i)–(ii)] Comparison of low pass filter with the S-G filter with (i) a large number of elements or (ii) a small number of elements; (iii) the FFT based frequency analysis of signal after filtering. In the legends, the filter type “0” stands for no filter, “1” is Savitzky-Golay filter with an element size of 73 or 53 and order of 3, “2” is low pass filter with an element size of 88, cut-off frequency 100 MHz, and “3” is the time-corrected “2.”

FIG. 3.

Ringing effect and comparison of signal processing options. Signals of nanosecond laser irradiation on the sample with 50 nm Ge deposited on (a) 7 nm alumina as an insulation layer or (b) 45 nm oxide as an insulation layer. [(i)–(ii)] Comparison of low pass filter with the S-G filter with (i) a large number of elements or (ii) a small number of elements; (iii) the FFT based frequency analysis of signal after filtering. In the legends, the filter type “0” stands for no filter, “1” is Savitzky-Golay filter with an element size of 73 or 53 and order of 3, “2” is low pass filter with an element size of 88, cut-off frequency 100 MHz, and “3” is the time-corrected “2.”

Close modal

Finite Impulse Response (FIR) digital filters are compared for ripple signal removal. In Fig. 3, we show that the frequency domain optimized low-pass filter can preserve better peak features than the time-domain optimized Savitzky-Golay (S-G) filter. It is shown that the S-G filter can be tuned toward the performance of a low pass filter at a certain time scale [Fig. 3(b-i)] but fails at another [Fig. 3(b-ii)]. It is mainly because the ringing signals have a characteristic frequency at 50 MHz which can be easily removed by a low-pass filter. In Fig. 3(c-ii), the frequency ripple has been removed for filter type = 2 at a frequency beyond 50 MHz. However, even with an optimized filter, we still lost the granular peak information for the signals of Fig. 3(a-i). Larger error on the maximum temperature is generated on a slower heating and cooling process [Fig. 3(b) compared to Fig. 3(a)]. As a consequence, no digital signal processing is adapted for comparing measurement with simulations.

The serpentine structure was suggested but not discussed in any detail in Brunco’s paper.21 In general, the serpentine structure can effectively increase the Pt sensing unit resistance and reduce the contact resistance error while keeping a compact footprint. However, it generates thermal error under Gaussian beam irradiation, which will be discussed in Sec. III A.

Illustrated in Fig. 4, probe resistance and contact pad resistance are bundled series resistances in the circuit. Note that the RTC here includes the contact resistance between the probe and the contact pad, termed RC, as well as the contact pad intrinsic resistance RCP. Through silver paste bonding, we reduced the contact resistance RC to the minimum value of below 0.5 Ω. The four-point probe method is not applied here due to the circuit design illustrated in Sec. II B. The contact pad intrinsic resistance RCP is defined by the space required to implement silver paste bonding, which contributed to the majority of RTC. Therefore, the probe resistance RT should be engineered significantly higher than RCP with a given lateral dimension. According to Ohm’s law R=ρLA=ρLW*d, the ratio of LW determines the resistivity in a thin film pattern, where d is the thickness, and L and W are the total length and width of the pattern. The silver paste is controlled to be 4 mm wide (WCP = 4 mm) and the distance to the sensor pattern is controlled to be 1 mm (LCP = 1 mm). As LCPWCP is 0.25, the LTWT ratio is then required to be at least 20. From the densely packed serpentine structure, we found that an N fold serpentine structure has LtotWtot=Lind*NWind, where L and W are the length and width, the subscript “ind” and “tot” stand for an individual stripe or the total pattern inside the serpentine structure. The definitions can be further referred to in Fig. 4. The larger Wtot is preferred to reduce the thermal error. The lower bound of Wtot is set to avoid laser machining induced defects and surface agglomeration in the vicinity of the edge, requiring Wtot larger than twice the femtosecond laser beam diameter used for machining. The problem is then transformed to find maximum WT with constrains as follows: 2DWtotLind/N, LtotWtot>20, where D is the diameter of the laser machined spot, which is 1 µm. It is easy to see that the optimal solutions lies on the boundary where Wtot = Lind/N, and hence, LtotWtot=N2. Therefore, the maximum Wtot is achieved at minimum N = 5. The typical resistance of a fivefold serpentine pattern on 50 nm Pt film after annealing is measured to be 65 ± 5 Ω. The residual resistance is measured to be 1.5 ± 0.5 Ω. Therefore, the contact resistance error is controlled below 3.3%.

FIG. 4.

Laser machined sensor geometries and electrical resistance analysis. (a) The image of a 5-fold serpentine structure with 4 mm active size. (b) Microscopic image of a sensor with a 50 µm active area. The dimensions of Lind, Ltot and Wind, Wtot are labeled with respect to the sensor. (c) The breakdown schematics of the resistances of the sensor. RS is the sensor resistance, and RT is the probe resistance. RTC is the total contact resistance including RCP, the contact pad resistance, and RC, the contact resistance between the probe and contact pad.

FIG. 4.

Laser machined sensor geometries and electrical resistance analysis. (a) The image of a 5-fold serpentine structure with 4 mm active size. (b) Microscopic image of a sensor with a 50 µm active area. The dimensions of Lind, Ltot and Wind, Wtot are labeled with respect to the sensor. (c) The breakdown schematics of the resistances of the sensor. RS is the sensor resistance, and RT is the probe resistance. RTC is the total contact resistance including RCP, the contact pad resistance, and RC, the contact resistance between the probe and contact pad.

Close modal

We analyze and describe the three performance indicators, i.e., resolution, accuracy, and precision. Our signal is the transient temperature evolution curve that we obtained during the capturing process. Our analysis will examine both the maximum temperature and normalized temporal variations. For laser-induced damage, and melting or deformation threshold studies, the maximum temperature is of interest. For thermal conductivity or frequency domain measurement, the temporal evolution is more important than the maximum temperature. For probing laser-induced chemical reaction processes, both the maximum temperature and temporal evolution should be considered.

Before our discussion, we first analyze the heat transfer problem involved in the sensing process. The vertical thermal diffusion length Ldiff=αt is below 500 nm for 10 ns and 2.5 µm for 250 ns for both oxide and sapphire substrates. The full-width half maximum (FWHM) of the laser-irradiated area is 50 µm. Therefore, the vertical temperature gradient Tpeak/Ldiff is by 1–2 orders larger than the lateral gradient Tpeak/(1/2FWHM). Hence, the 1D vertical heat dissipation to the substrate dominates the temperature evolution. For the actual sensing processes, however, the measured temperature represents the lumped thermal resistive effects over the entire sensor. Additional details are revealed from coupled thermal and electrical simulations.

Temperature measurements aim to estimate the location and time dependent temperature field, i.e., T(x, t). For a Gaussian beam laser heating, the spatial resolution and accuracy is of great interest. We first apply comprehensive simulation to analyze the performance. Experimentally, we will rely on changing input source’s spatial distribution to study the spatial performance of the sensor. In Fig. 5, coupled heat transfer and electrical conductance simulation is carried out in COMSOL with a simplified 3D transient model. Insulated with a 7 nm ALD alumina layer, a 50 µm wide 50 nm thick Pt sensor on sapphire is irradiated with a 13 nanosecond laser pulse with different laser beam diameters. The laser fluence is set to generate a temperature change of 500 K–600 K, which is the temperature range of interest. For small beam sizes [FWHM = 25 µm, Fig. 5(a-i), we note that the temperature field on the sensor shares a similar Gaussian distribution with the incident laser beam. The spatially averaged temperature increase is only half of the center temperature [Fig. 5(b-i)]. On the other hand, the large laser beam (FWHM = 100 µm) irradiation induces a uniform temperature distribution on the sensor pattern. As a result, the average temperature is only 50 K lower than that from center temperature (500 K). The 10% difference between the average and the center temperature is caused by the cooling of the sensor’s edges. In the zoomed-in image and cross section temperature plot from the supplementary material, Fig. S2, the edge temperature has dropped to half of the center temperature. The cross-sectional average temperature is consistent with the overall average temperature. Here, we define a “resistive T” representing the calculated probed temperature TP based on RPR0=TCR*(TPT0), where RP is the simulated overall resistance change. From the plot in Fig. 5(b), the “resistive T” is identical to the average T regardless of the temperature distribution, which indicates that the actual probed temperature will be the spatial average temperature. Further explanation on this identity is not pursued due to the scope of current work. In summary, the spatial resolution of the sensor is 100 µm and the system error is 10% due to the effect of serpentine structure and Gaussian irradiation.

FIG. 5.

Experiments and simulation for the spatial resolution and associated thermal error for 50 µm size sensor on sapphire. (a) Simulated temperature contour of the sensor exposed to a 13 ns laser pulses with beam sizes of (i) 25 µm diameter and (ii) 100 µm diameter. Here, d stands for the width of the sensor, which is 50 µm. The plot is selected at the 22 ns when the peak temperature is achieved. (b) The time evolution of the simulated center temperature (“Center T”), average temperature (“Average T”), and resistance back-calculated temperature (“Resistive T”) under a laser beam of (i) 25 µm diameter and (ii) 100 µm diameter. (c) (i) The experimental temperature evolution history under two sizes of laser beams. (ii) The simulated center T and resistive T evolution under two sizes of laser beams. All the samples are insulated with 7 nm ALD alumina without a top absorbing layer.

FIG. 5.

Experiments and simulation for the spatial resolution and associated thermal error for 50 µm size sensor on sapphire. (a) Simulated temperature contour of the sensor exposed to a 13 ns laser pulses with beam sizes of (i) 25 µm diameter and (ii) 100 µm diameter. Here, d stands for the width of the sensor, which is 50 µm. The plot is selected at the 22 ns when the peak temperature is achieved. (b) The time evolution of the simulated center temperature (“Center T”), average temperature (“Average T”), and resistance back-calculated temperature (“Resistive T”) under a laser beam of (i) 25 µm diameter and (ii) 100 µm diameter. (c) (i) The experimental temperature evolution history under two sizes of laser beams. (ii) The simulated center T and resistive T evolution under two sizes of laser beams. All the samples are insulated with 7 nm ALD alumina without a top absorbing layer.

Close modal

The edge-induced error can be reduced or properly accounted for through postprocessing. Expanding the stripe width can reduce the ratio of the cooled edge width to the stripe width. However, for a given 50 µm sensor size, the stripe width has a maximum value of 7.5 µm for a five-fold serpentine pattern, limiting the minimum error. Alternatively, the oxide substrate can reduce the edge cooling as it has one order lower thermal diffusivity compared to sapphire. It is noteworthy that the edge cooling induced error is a systematic one that can be properly estimated with COMSOL simulation and corrected as a ratio regardless of laser fluences. Postprocessing correction is a viable route for removing the error.

Experimentally, we further show that the temperatures normalized with respect to the peak signal value are dominated by one-dimensional (1D) vertical heat dissipation to the substrate. These normalized transients agree well with the 1D simulated temperature evolution in Fig. 5(c-i). Furthermore, simulations presented in Fig. 5(c-ii) show that the heat dissipation from the edge of Pt stripe only affects the temperature evolution at the transition from the initial fast drop to the long-tail cooling.

Besides the spatial resolution, we will also experimentally analyze the temporal resolution and accuracy. The temporal resolution is defined by the response time, which is characterized by evaluating the impulse response of the sensor. Picosecond laser pulses are by two orders faster than the thermal relaxation time (∼1 ns) as well as the oscilloscope response time (1 ns), and they are considered as impulse inputs into the system. The sensor’s response is listed in Fig. 6(a). We found that the signal first increased instantly, then gradually reached a peak within 12 ns, and started dropping afterward. From basic 1D heat transfer analysis, the temperature increase inside the Pt layer should conclude within 1 ns. However, it takes 12 ns to reach the peak, which invalidated the measurement of the peak temperature. We think the main cause of such a long response time (9 ns, 1/e) is on the nonideal DC power supply.21 The charge cannot be restored within 1 ns. Therefore, the voltage first increases to a medium level, then continues to increase gradually while the charge is being restored. Asymptotic curve fitting [dashed line in Fig. 6(a)] may help to estimate the peak temperature; however, the error will be significant as the curvature near the peak is high.

FIG. 6.

Temporal evolution upon picosecond and nanosecond pulsed laser irradiation. (a) Experimental laser signal and sensor response upon picosecond pulse irradiation on an oxide substrate. (b) Both experimental and simulated laser signals and sensor responses upon nanosecond laser irradiation on sapphire and oxide substrates. All the samples are insulated with 7 nm ALD alumina.

FIG. 6.

Temporal evolution upon picosecond and nanosecond pulsed laser irradiation. (a) Experimental laser signal and sensor response upon picosecond pulse irradiation on an oxide substrate. (b) Both experimental and simulated laser signals and sensor responses upon nanosecond laser irradiation on sapphire and oxide substrates. All the samples are insulated with 7 nm ALD alumina.

Close modal

For the temporal accuracy, we compared the experimental and simulated sensor’s normalized temperature evolution on different substrates. In Fig. 6(b), sensor response signals on both sapphire and oxide matched well with the simulation. Since the sensor’s maximum temperature is reached beyond the response time [shadowed area in Fig. 6(b)], no significant distortion shall exist for the probed peak temperature and following cooling processes.

Since the sensor has a fixed location, the spatial precision of measurements only depend on the imposed laser beam intensity distribution. The laser energy spatial distribution fluctuation induced precision loss is lumped into the temporal precision of the magnitude. Temporal precision is defined as the variation of T(xo, t), where xo stands for the typical location. Intuitively, it is quantified by measuring the variance of a time dependent temperature measurement at a given location. The T(xo, t) is essentially T(t), a one-dimensional array of signals. This one-dimensional signal has overall two characteristics, the absolute magnitude and the relative trend. Ten independent measurements are carried out through shining a continuous train of 13 ns laser pulses onto a Pt sensor with 45 nm oxide insulation and backed by a silica substrate. We first confirmed the temporal alignment of the temperature evolution through cross correlation evaluation of all the signals. Then, we note that the measured transient peak temperature increase has an rms (Root Mean Square) level of 197.27 K and an rms error of 7.24 [Fig. 7(a)]. The error percentage is then 3.6%, which is close to the laser fluence fluctuation (4%). After normalization of the transient history with respect to the peak temperature and filtering, we plotted the normalized error as a function of time and found it to be on the order of 1% [Fig. 7(b)].

FIG. 7.

Overlay of multiple measurements and average error for raw temperature and normalized temperature history. (a) The raw temperature evolution and error, and (b) the normalized and filtered temperature history and error. The sensor is on silica and insulated with 7 nm ALD alumina.

FIG. 7.

Overlay of multiple measurements and average error for raw temperature and normalized temperature history. (a) The raw temperature evolution and error, and (b) the normalized and filtered temperature history and error. The sensor is on silica and insulated with 7 nm ALD alumina.

Close modal

Below, we describe the experimental verification of the sensor’s measurement through a nanosecond transient process with a known reference temperature. The spatial and temporal accuracy as well as the system accuracy are therefore confirmed. We utilize the melting points of the amorphous and crystalline silicon film as known reference temperatures. On the grounds of prior laser crystallization studies,22,28 the melting points have been verified as essentially constant in the nanosecond time regime. 40 nm of amorphous silicon thin film is deposited through the PECVD method on top of 35 nm silicon nitride insulation layer due to its high mechanical strength against laser-induced mechanical shock. Finally, the sample is irradiated through a 13 ns pulsed laser (New Wave Polaris II, 532 nm). Though the film thickness is smaller than three times the penetration depth, accounting for multilayer thin film interference,29 we calculated the absorption of the laser energy that is confined in mainly in a-Si film layer [Fig. 8(b-ii)].

FIG. 8.

Experimental validation of the temperature measurement magnitude and temporal evolution. (a) Probed peak temperature against the incident fluence. A linear fit is applied before phase transformation and ablation. The inset is the tested sample geometry. (b) (i) Bright field images of laser irradiated 40 nm amorphous silicon film on silicon nitride insulating layer. (ii) The multilayer thin film analysis of the reflection and absorption. Top stands for a-Si film and sub stands for the Pt layer. (c) (i) Experimental and simulated temperature evolution of the Pt layer under insulation and silicon films with a fluence of 2 J/cm2. (ii) Simulated temperature evolution of the silicon top surface and Pt layer under high and low fluence threshold indicated in (a).

FIG. 8.

Experimental validation of the temperature measurement magnitude and temporal evolution. (a) Probed peak temperature against the incident fluence. A linear fit is applied before phase transformation and ablation. The inset is the tested sample geometry. (b) (i) Bright field images of laser irradiated 40 nm amorphous silicon film on silicon nitride insulating layer. (ii) The multilayer thin film analysis of the reflection and absorption. Top stands for a-Si film and sub stands for the Pt layer. (c) (i) Experimental and simulated temperature evolution of the Pt layer under insulation and silicon films with a fluence of 2 J/cm2. (ii) Simulated temperature evolution of the silicon top surface and Pt layer under high and low fluence threshold indicated in (a).

Close modal

The laser-induced melting is clearly captured in the plot of peak temperature against the laser pulse fluence [Fig. 8(a)]. For a given sample, we gradually increase the fluence and simultaneously measure the transient temperature evolution. The record peak temperature is the averaged value of the 10 pulses at given fluence. The probed peak temperature first increased linearly with the incident fluence, corresponding to the stage where no phase change is involved. Consequently, the probed temperature rise follows the relation, ΔT = Q/Cp, where Q is the deposited pulse energy on the sensor and Cp is the specific heat. However, at 1.5 J/cm2 the measured temperature deviated from the linear correlation, and the deviation was enlarged with increased incident energy. As the laser profiles are Gaussian, we assume that the gradual deviation indicates the initiation of melting in the center region and subsequent expansion across the irradiated area. Hence, 1.5 J/cm2 is considered to be the onset of the phase transformation, where the latent heat used for melting will stall the temperature increase. After reaching 2.5 J/cm2, the temperature plateaued and soon started to drop. This stage was interpreted as the completion of full melting and crystallization. In the experiments, we started to reduce the fluence, and the temperature dropped to a level lower than before, exhibiting a hysteresis in the temperature vs fluence curve, which is visually guided by the blue arrows. The hysteresis is caused by the fact that an increased amount of crystalline silicon in the laser-irradiated area reduced the absorption of the incident laser irradiation, leading to the drop in temperature increase. In support of this argument, a multilayer thin film interference model29 was implemented and the absorption was predicted to drop with increased crystallinity [Fig. 8(b-ii)]. In the optical image [Fig. 8(b)], the color of the irradiated area has changed, indicating a local refractive index change caused by the phase transformation. Furthermore, we measured the reflectivity at different locations. Pristine regions exhibited lower reflection compared to the crystallized region, lending credit to the optical simulation and proposed interpretation. From the multilayer thin film interference model, we further infer that absorption in the substrate (Pt layer) is negligible compare to the amorphous silicon layer. After the initiation of crystallization, we infer a total energy absorption drop of 5% (50 K/1000 K) through measuring the drop of maximum temperature [Fig. 8(a)]. Comparing it to Fig. 8(b-ii), we conclude that the crystallization level only changes by approximately 20%, indicating that the substrate absorption is still negligible.

The maximum temperature and temporal evolution were confirmed in conjunction with 1D heat transfer simulation. We first verified the agreement of the temporal evolution of the sensor temperature with our modeling. Moreover, we set the modeled incident fluence, F, to ensure that the modeled Pt layer maximum temperature matched with the probed maximum temperature at both low F (onset of amorphous silicon crystallization) and high F (crystalline silicon crystallization). Note that the spatial heterogeneity will cause reduced probe temperature, and the difference for the current sample is estimated less than 50 K. Hence, we intentionally compensated 50 K to the probed temperature as the input for modeling. The matched model was further used to predict the top layer temperatures under two incident fluences, which is plotted together with the sensor temperature at Fig. 8(c-ii). At low F, the top layer reached an amorphous melting point (1485 K) and at high F, the crystalline silicon melting point (1685 K), perfectly aligned with the observations and physical interpretations of Fig. 8(a). Hence, we experimentally confirmed the maximum temperature and temporal evolution of the sensor given the proper accounting of the effects of spatial nonhomogeneity. The uncertainty due to nonhomogeneity effects is within 25 K.

The peak of transient temperature evolution will be significantly affected by the insulation layer or top layer, which implies that the nanosecond-resolved sensing can be used for the thin film thermal conductivity study. As a proof-of-concept experiment, we measured the initial part of the laser-induced temperature for samples with 7 nm ALD alumina or 45 nm PECVD oxide insulation layers. E-beam evaporation is used for depositing 50 nm of amorphous Ge, which is over three times of the optical penetration depth. With a thinner insulation layer, the temperature rise time is shorter and the maximum temperature is higher due to the lower thermal resistance (Fig. 9). The thermal conductivity of PECVD oxide30 and ALD alumina31 is set as 1 W/mK as suggested by references. The laser fluence is the only tuning parameter, and the same fluence is applied to both simulations. Slight differences in the maximum temperature can be caused by the laser energy fluctuation as well as the thin-film interference effect on the reflectivity of the sensor. Similar to the Laser Flash Analysis (LFA), the sensor can be readily used to probe the thermal conductivity of unknown material or the thickness of a known material. Different from LFA that only measures a homogeneous bulk sample, the proposed method provides a laterally microscale and vertically nanoscale measurement on heterogeneous multilayer samples. It is also conceivable that layouts of multiple sensors embedded at different depths in microstructured target specimens could provide 3D transient temperature maps under varying thermal loading configurations.

FIG. 9.

Experimental and simulated temperature evolution on samples with different insulation layers. All the samples are fabricated on silica substrates. The legends indicate the thickness of the insulation layer.

FIG. 9.

Experimental and simulated temperature evolution on samples with different insulation layers. All the samples are fabricated on silica substrates. The legends indicate the thickness of the insulation layer.

Close modal

In the current study, we presented the systematic design, optimization, and implementation of nanosecond resolved microscale Pt thin-film sensors. The error sources were analyzed and optimized during the development of the sensor. Coupled thermal and electrical simulations quantified the spatial resolution and accuracy imposed by the Gaussian incident laser beam profile and the sensing serpentine structure. Regarding the temporal resolution, we characterized the sensor rise time as 9 ns via launching picosecond laser irradiation as a heat impulse input. The temporal evolution of the sensor measurements agree well with simulations on different film structures and substrates. The absolute temperature magnitude was evaluated against amorphous silicon film melting, and the error was shown to be below 25 K. After correcting for the laser energy fluctuation, the temperature measurement attained fluctuation less than 0.5%. Finally, we demonstrated the sensor’s potential applications for thin film thermal conduction with nanometer sensitivity.

See the supplementary material for the additional information regarding sensor fabrication and sensor thermal analysis.

The authors would like to thank Yang Xia from the Department of Applied Science and Technology and Luya Zhang from the Department of Electrical Engineering and Computer Science for helpful discussion on the electrical circuit. The nanofabrication was carried out at the Marvell Nanofabrication Laboratory and the California Institute of Quantitative Bioscience (QB3) of UC Berkeley. The authors thank Chris Zhao from Novellution Technologies, Inc., for helpful discussion on the fabrications. This work was financially supported by the Lam Research Corporation and NSF CMMI funding (Grant No. 1363392).

1.
J.
Long
,
P.
Fan
,
D.
Gong
,
D.
Jiang
,
H.
Zhang
,
L.
Li
, and
M.
Zhong
,
ACS Appl. Mater. Interfaces
7
,
9858
(
2015
).
2.
K.
Sera
,
F.
Okumura
,
H.
Uchida
,
S.
Itoh
,
S.
Kaneko
, and
K.
Hotta
,
IEEE Trans. Electron Devices
36
,
2868
(
1989
).
3.
B.
Nie
,
L.
Yang
,
H.
Huang
,
S.
Bai
,
P.
Wan
, and
J.
Liu
,
Appl. Phys. A
119
,
1075
(
2015
).
4.
H.
Pan
,
N.
Misra
,
S. H.
Ko
,
C. P.
Grigoropoulos
,
N.
Miller
,
E. E.
Haller
, and
O.
Dubon
,
Appl. Phys. A: Mater. Sci. Process.
94
,
111
(
2009
).
5.
H.
Pan
,
S. H.
Ko
, and
C. P.
Grigoropoulos
,
Appl. Phys. A
90
,
247
(
2008
).
6.
W.
Shou
,
B. K.
Mahajan
,
B.
Ludwig
,
X.
Yu
,
J.
Staggs
,
X.
Huang
, and
H.
Pan
,
Adv. Mater.
29
,
1700172
(
2017
).
7.
E. C.
Garnett
,
W.
Cai
,
J. J.
Cha
,
F.
Mahmood
,
S. T.
Connor
,
M.
Greyson Christoforo
,
Y.
Cui
,
M. D.
McGehee
, and
M. L.
Brongersma
,
Nat. Mater.
11
,
241
(
2012
).
8.
S.
Hong
,
H.
Lee
,
J.
Yeo
, and
S. H.
Ko
,
Nano Today
11
,
547
(
2016
).
9.
S.
Han
,
S.
Hong
,
J.
Ham
,
J.
Yeo
,
J.
Lee
,
B.
Kang
,
P.
Lee
,
J.
Kwon
,
S. S.
Lee
, and
M.
Yang
,
Adv. Mater.
26
,
5808
(
2014
).
10.
L.
Wang
,
Y.
Rho
,
W.
Shou
,
S.
Hong
,
K.
Kato
,
M.
Eliceiri
,
M.
Shi
,
C. P.
Grigoropoulos
,
H.
Pan
,
C.
Carraro
, and
D.
Qi
,
ACS Nano
12
,
2231
(
2018
).
11.
D.
Qi
,
S.
Tang
,
L.
Wang
,
S.
Dai
,
X.
Shen
,
C.
Wang
, and
S.
Chen
,
Nanoscale
10
,
8133
(
2018
).
12.
D.
Qi
,
D.
Paeng
,
J.
Yeo
,
E.
Kim
,
L.
Wang
,
S.
Chen
, and
C. P.
Grigoropoulos
,
Appl. Phys. Lett.
108
,
211602
(
2016
).
13.
J.
Long
,
P.
Fan
,
M.
Zhong
,
H.
Zhang
,
Y.
Xie
, and
C.
Lin
,
Appl. Surf. Sci.
311
,
461
(
2014
).
14.
X.
Wang
,
A.
Kuchmizhak
,
D.
Storozhenko
,
S.
Makarov
, and
S.
Juodkazis
,
ACS Appl. Mater. Interfaces
10
,
1422
(
2018
).
15.
P.
Elahi
,
C.
Kerse
,
H.
Hoogland
,
Ö.
Akçaalan
,
D. K.
Kesim
,
B.
Çetin
,
R.
Holzwarth
,
F. Ö.
Ilday
,
B.
Öktem
,
M. D.
k
ı,
S.
Yavaş
, and
H.
Kalayc
oğlu
ı,
Nature
537
,
84
(
2016
).
16.
M.
Islam
,
T.
Purtonen
,
H.
Piili
,
A.
Salminen
, and
O.
Nyrhilä
,
Phys. Procedia
41
,
835
(
2013
).
17.
D.
Hu
and
R.
Kovacevic
,
Proc. Inst. Mech. Eng., Part B
217
,
441
(
2003
).
18.
B.
Yuan
,
G. M.
Guss
,
A. C.
Wilson
,
S. P.
Hau-Riege
,
P. J.
DePond
,
S.
McMains
,
M. J.
Matthews
, and
B.
Giera
,
Adv. Mater. Technol.
3
,
1800136
(
2018
).
19.
D. L.
Blackburn
, in
Twentieth Annual IEEE Semiconductor Thermal Measurement and Management Symposium (IEEE Cat. No. 04CH37545)
(
IEEE
,
2004
), Vol. 20, p.
70
.
20.
J. A.
Kittl
,
R.
Reitano
,
M. J.
Aziz
,
D. P.
Brunco
, and
M. O.
Thompson
,
J. Appl. Phys.
73
,
3725
(
1993
).
21.
D. P.
Brunco
,
J. A.
Kittl
, and
M.
Thompson
,
Rev. Sci. Instrum.
64
,
2615
(
1993
).
22.
M.
Hatano
,
S.
Moon
,
M.
Lee
,
K.
Suzuki
, and
C. P.
Grigoropoulos
,
J. Appl. Phys.
87
,
36
(
2000
).
23.
M.
Hatano
,
S.
Moon
,
M.
Lee
,
K.
Suzuki
, and
C. P.
Grigoropoulos
,
J. Non-Cryst. Solids
266-269
,
654
(
2000
).
24.
Y.
Yue
and
X.
Wang
,
Nano Rev.
3
,
11586
(
2012
).
25.
H.
Choi
and
X.
Li
,
Sens. Actuators, A
136
,
118
(
2007
).
26.
L.
Wang
,
Z.
Jin
,
D.
Paeng
,
Y.
Rho
,
J.
Long
,
M.
Eliceiri
,
Y. S.
Kim
, and
C. P.
Grigoropoulos
,
Sens. Actuators A
100
,
111657
(
2019
).
27.
J. P.
Holman
,
Heat Transfer
(
McGraw-Hill
,
New York
,
1986
).
28.
E. P.
Donovan
,
F.
Spaepen
,
D.
Turnbull
,
J. M.
Poate
, and
D. C.
Jacobson
,
Appl. Phys. Lett.
42
,
698
(
1983
).
29.
J. R.
Howell
,
M. P.
Menguc
, and
R.
Siegel
,
Thermal Radiation Heat Transfer
(
CRC Press
,
2015
).
30.
S.-M.
Lee
and
D. G.
Cahill
,
J. Appl. Phys.
81
,
2590
(
1997
).
31.
C. S.
Gorham
,
J. T.
Gaskins
,
G. N.
Parsons
,
M. D.
Losego
, and
P. E.
Hopkins
,
Appl. Phys. Lett.
104
,
253107
(
2014
).

Supplementary Material