1. Introduction

Laser material processing entered industrial production in the 1970s where in particular laser welding machines were introduced in the automotive sector. Since then, an ever growing number of laser processing applications has been developed ranging from large-scale machining to micro-structuring and medical uses. Depending on the special requirements of each application, the use of wavelengths from UV to mid-IR, and interaction times from cw illumination to ultrashort pulses in the picosecond range are meanwhile established techniques, and the fundamental mechanisms are mostly well understood. Femtosecond lasers are just crossing the barrier for large-scale technological usage, mainly because femtosecond laser systems have only recently achieved a sufficiently high performance and reliability level required for integration in mass production processes. From a scientific point of view, however, the general advantages of material structuring in the highly nonlinear range of energy input during fs laser ablation or local material modification are well-known, such as very low collateral thermal damage in the processed region [1,2] or the possibility of local micro-cutting inside transparent materials [3,4]. A lot of very interesting processes in this regime have been developed and studied in detail by many scientists in the last years (see, e.g [5–10].).

In this work, we focus on the process of selective thin film ablation from absorbing, especially semiconducting substrates. This delamination is of special interest in the field of photovoltaics, e.g., for mask-less opening of passivation layers for contacts on wafer-based Si solar cells. The process in general, however, is transferrable to vastly different optical elements, where micro-structuring of a thin film coating is wanted. Very recently, we were able to show that, for the example of SiO₂ on Silicon, the use of fs laser pulses is beneficial in terms of low ablation threshold, minimum substrate damage and wide processing window [11]. In the following, we present an extended study of femtosecond laser selective delamination of the transparent dielectric layers SiO₂, Si_xN_y or Al₂O₃ on Si wafers, using pulse durations from 50 fs to 2000 fs, and covering the laser wavelength range from 400 to 1030 nm. We have studied ablation thresholds and quality of the opened surfaces by optical and atomic force microscopy in dependence of pulse duration, laser wavelength and thin film material. In addition we present examples for micro-patterns prepared by this process and discuss the potential for production of optical elements.

2. Experimental

The samples used in this investigation were three different types of thin dielectric films on planar silicon substrates: (i) thermally grown SiO₂, thickness 100 nm; (ii) Si_xN_y film made by PECVD, thickness 110 nm; and (iii) Al₂O₃ film, thickness 50 nm, prepared by atomic layer deposition. The local removal of these layers has been examined by way of ablation experiments using three different laser systems: the Ti:sapphire laser systems (1) Legend (Coherent) with pulse duration of 700, 1000 and 2000 fs at a wavelength of 800 nm and (2) Spitfire (Spectra Physics) with pulse duration of 50 fs at λ = 800 nm and λ = 400 nm, and (3) the Yb:KGW laser system Pharos (Light Conversion) with a pulse duration of 280 fs at wavelengths of 1030 nm and 515 nm. All laser beams had been focused on the sample surface by help of plano-convex lenses; the characterization of beam size and shape in the focal range was done by a uEye camera having a pixel size of 6 µm. The pulse energy was measured with a pyroelectric sensor. Fine tuning of the pulse energy was done in two ways: (1) with a computer controlled half-wave plate and a thin-film polarizer using the Ti:sapphire laser systems and (2) by software controlled change of the regenerative amplifier current combined with a photodiode monitored readjustment (Pharos).

Each sample was individually fixed on top of a motorized x-y-table by help of a vacuum-chuck; the laser pulse sequence as well as the required precise, synchronized movement of the positioning stage was controlled with home-made software. For studying the delamination as a function of locally applied laser energy density on a reliable statistical basis, lines of 100 areas with a spot-to-spot distance of 100 µm and constant pulse energy have been prepared for each combination of sample and laser parameters. Each spot has been irradiated by a single pulse; lines differing only by pulse energy were also separated by a distance of 100 µm, enabling an unambiguous ex post mapping of the applied laser energy (each line number refers to a special, constant pulse energy). After irradiating the samples, the machined areas were characterized by light microscopy (using a Zeiss Axioplan 2 imaging microscope) as well as atomic force microscopy (Witec alpha 300). The optical microscopy was used to determine the diameters of visibly changed regions, in particular ablation craters. In general, there is always a considerable modification of the optical appearance of laser-irradiated spots already at fluences below the occurrence of ablation craters, and in most cases opened regions are still surrounded by such an optically changed region (‘corona’). Therefore we use, in accordance with previous work [11,12], two different threshold fluences: (i) The breaking threshold fluence Φ_th,b defined as the minimum local fluence needed for an ablation of the dielectric layer from the silicon substrate, and (ii) the melting threshold fluence Φ_th,m defined as minimum fluence where visible colour (reflectivity) changes can be seen. The typical location of these thresholds within a Gaussian beam is shown schematically for three different maximum fluences Φ₀ in Fig. 1 , where also the related diameters D_m (between vertical, red dotted lines) and D_b (between vertical, black dotted lines) of modified regions on the sample surface are given.

 

Fig. 1 Fluence profile of a Gaussian beam with beam radius ω₀ at Φ₀/e². At local fluence above the melting threshold fluence only a change of colour (reflectivity) can be seen (a); fluences above the breaking threshold lead to delamination of the dielectric layer (b), at sufficiently high excess energy finally to ablation of the transparent layer (c).

Download Full Size | PDF

It is obvious from Fig. 1 that further increase of the laser fluence beyond the situation of Fig. 1(c) will produce ablation craters and coronas of increasing diameter. This paves the route to determining melting and breaking thresholds ex post by analyzing the sizes of ablated areas as a function of laser fluence. For a spatially Gaussian laser beam the diameter of an irreversibly changed round area is related to focus radius ω₀, threshold fluence Φ_th and maximum fluence Φ₀ of the applied pulses by [13]:

This relation holds in our case for both melting and breaking [11,13], so that a fit of the experimentally obtained sizes based on Eq. (1) provides a very reliable method to derive the threshold fluences for any individual combination of materials, laser wavelength and pulse duration.

Finally, atomic force microscopy has been conducted on selected spots in order to assess the quality of the Si surface left after delamination of the dielectric layer at different energy densities. The topological information is then used to recognize a third, ‘damage’ threshold Φ_th,d as upper limit of a fluence range of constantly very smooth surfaces in the opened area (typical rms roughness: 2−3 nm). Beyond this threshold, often marked by a narrow hole in the center of the ablated area, any further increase of fluence leads to a quickly increasing area of much higher roughness (rms ≥10 nm). This apparent damage of the Si surface, which is probably due to melting and re-crystallization, is normally not visible by optical microscopy and could, as such, easily be overlooked.

3. Results

We start the presentation of experimental results with an example for a region showing ablation after irradiation by a 515 nm, 280 fs laser pulse. In the microscope image given on the left-hand side of Fig. 2 , one can clearly recognize the central round area – obviously a blank Si surface – surrounded by a (slightly elliptical) ring of changed color, apparently the corona with the Si_xN_y still on top. The AFM cross section presented on the right-hand side proves this interpretation: the crater is almost exactly 110 nm deep, corresponding perfectly to the thickness of the dielectric layer, and the crater diameter of approximately 33 µm matches the size of the central light region. Due to the very flat crater bottom and the lack of significant Si ablation, we call this state damage-free delamination. Moving towards higher laser fluence, at some point a significant corruption of the Si opened surface is observed (mostly a narrow hole of ~100 nm depth in the center); we mark this change as damage threshold, giving the upper limit for a selective, damage-free ablation of the transparent layer.

 

Fig. 2 Microscope image (left) and AFM cross section (right) of ablation crater after one 515 nm, 270 fs pulse.

Download Full Size | PDF

Systematic studies varying materials and laser parameters, as specified above, have been conducted. The fluence-dependent results have been analyzed separately for each combination of material, laser wavelength and pulse duration, yielding the individual melting and breaking thresholds. As demonstrated exemplarily in Fig. 3 , different dielectrics may exhibit significantly different reaction to the same laser parameters, here again for the λ = 515 nm, 280 fs pulses: Fig. 3(a) shows diameters (average over 3 craters for each laser pulse energy) of crater and corona for SiO₂, whereas the same information is presented for Si_xN_y in Fig. 3(b). The solid lines are in each case least mean squares fit curves using Eq. (1). The very good agreement with the experimental data provides evidence that reliable values of melting and breaking threshold can be obtained from that data evaluation. Comparing in particular ablation diameters at identical pulse energies, one recognizes that silicon oxide is more easily delaminated from the substrate than silicon nitride. Possible reasons for this observation will be discussed below.

 

Fig. 3 Dependence of ablation and corona diameters versus laser pulse energy for (a) SiO₂ and (b) Si_xN_y thin films on Si; solid lines represent least mean squares fits to the data based on Eq. (1).

Download Full Size | PDF

While the melting and breaking thresholds can be obtained by the systematic, nicely comparable method detailed above, the determination of the damage threshold can only be done ‘manually’, by inspecting spot by spot by atomic force microscopy. In addition, there is no simple parameter like diameter to be averaged for higher accuracy. That is why the damage thresholds are being determined less precisely. An overview of all three thresholds and further parameters derived from the variety of the experimental series conducted, is given in Table 1 .

Table 1. Overview of Melting, Ablation and Damage Thresholds Obtained for Irradiation of Three Different Transparent Dielectrics with Different Laser Parameters

View Table

There are some interesting trends or individual differences in the results comprised in Table 1, which need special attention. First of all, as we have already reported recently [11], the results obtained with 800 nm laser wavelength exhibit a considerable dependence of Φ_th,m and Φ_th,b on the pulse duration, while the damage threshold is obviously only related to the total amount of laser energy. On the other hand, the ablation thresholds for 50 fs and 700 fs differ by less than 10%, which lets us expect less than 5% change of thresholds comparing the results obtained using 50 fs or 280 fs pulses, respectively. Thus it appears reasonable to compare the ablation results for 800 nm, 50 fs pulses with the other wavelengths in disregard of pulse duration. Doing this for the SiO₂-covered samples, a monotonous decrease of Φ_th,b by a factor of two upon decrease of the wavelength from 1030 to 400 nm is recognized. An even stronger decrease by almost a factor of 4 is found for the pertinent melting thresholds; also the damage thresholds follow in general the same trend, however with the maximum value of Φ_th,d not at the longest wavelength, but at 800nm. From a technical point of view, these results somehow discourage the use of shorter (visible) wavelengths for the damage-free delamination, because then the processing window (energy interval between breaking and damage threshold) is very narrow. Also the collateral damage (corona) is relatively largest because of the low melting threshold, as can be seen in Fig. 2, where quite obviously the total area of corona (color change without ablation) exceeds that of the actually opened spot. A final noteworthy point is the higher ablation threshold of silicon nitride compared to SiO₂. The implications of all those trends and differences for the physical mechanisms will be discussed below (chapter 4).

One important question in context with the selective ablation mechanism is: what happens with the ablated material? If the process is really a selective ablation, the delaminated disk of the ablated material should simply be blasted off its original place and then fall down again to the sample, at least in our experimental setup, where the wafers are positioned horizontally. Figure 4 shows that in fact such complete disks are ablated: the two light circles at the bottom are two out of 100 spots ablated at constant pulse energy, while the next line of 100 spots is outside the image area. Thus, clearly the gray area in the upper part of the image is an ablated disk of Al₂O₃, which may have been ejected from one of the two opened spots in the lower part of the image, or any further ablated spot at larger distance. This is on one hand a nice confirmation of the idea that no considerable energy input in the dielectric layer is involved in the ablation mechanism, at least for the laser wavelength 800 nm used here. On the other hand, this debris has to be regarded when the technique shall be used for preparation of optical microstructures, because without further precaution some ablated slices might, just by chance, fall to a place where one of the next laser pulses is then prevented from making a clean ablation there.

 

Fig. 4 Microscope image of Al₂O₃ coated Si wafer, showing two ablation craters (close to bottom of image) and an almost undamaged, ablated disk of the delaminated thin film (marked by orange arrow).

Download Full Size | PDF

Due to the transparency of the material, however, this seems not to be a very dramatic problem, as can be seen in Fig. 5 . Just to demonstrate the flexibility of the method, as well as the good optical contrast which can be prepared by it, we have written the double signet of our university by pixel-wise delamination of the 50 nm thick Al₂O₃ layer using the 800 nm, 50 fs laser pulses. There are a few white spots in the dark areas of the actual image, which are recognized in higher magnification as delaminated slices. There is, however, no indication that the bitmap itself has been corrupted by the debris during image processing. Nonetheless a vertical positioning or a constant air flow during the procedure should be beneficial if very sensitive optical microstructures are to be produced by this method.

 

Fig. 5 Double signet of Martin-Luther-University Halle-Wittenberg, prepared by selective delamination of Al₂O₃ from Si wafer.

Download Full Size | PDF

4. Discussion and conclusion

The results reported above give several important hints to understand the delamination mechanism in more detail. For the discussion of the physical processes which are most probably responsible for our observed trends, we resume and extend the ideas of our previous publications on this topic [11,14]. The scenario derived for the dynamical laser-matter interaction leading to selective delamination is as follows: linear and, increasing with the rising intensity of the laser pulse, non-linear absorption in the silicon causes an increasingly hot electron system [15], until – still within the rising edge of the pulse – dense electron-hole plasma [16] close to the silicon-dielectric interface dominates the interaction. Due to this quasi-metallic absorption, the remaining energy of the laser pulse is then deposited within a very low penetration depth of the order of 10 nm, while the deeper region of the Si wafer experiences no further energy input directly by the laser. Since shorter pulses of equal total energy provide higher peak intensity, a concomitantly higher amount of nonlinear absorption will then achieve the required electron-hole plasma density even faster. This appears to be the reason why the ablation threshold decreases towards shorter fs pulses. In any case, only a very small, surface-near volume of the silicon will undergo a phase change created by the laser pulse, leading finally to the quasi selective ablation of the dielectric coating.

Now we can test if this picture for the selective delamination is suitable to explain also the new results presented in this work, obtained at different laser wavelengths and for different dielectric layers. One main finding considering the values compiled in Table 1 was the decrease of melting and breaking thresholds with decreasing laser wavelength. This is well compatible with the linear absorption coefficient of Si, which increases from 30 cm⁻¹ at 1030 nm to 9.5⋅10⁻⁴ cm⁻¹ (corresponding to 1/e penetration depth of 105 nm) at 400 nm [17]: for larger absorption coefficients, generation of charge carriers starts already at low intensities (in the linear interaction regime), so that the electron-hole plasma is achieved earlier within the pulse with the consequence of lower ablation thresholds. On the other hand, the energy input by linear absorption only is compressed to a much smaller depth, leading to higher temperatures in that volume, before the quasi-metallic absorption limits energy input to the very uppermost Si layer. For instance, if we calculate for λ = 400 nm the temperature increase only due to linear absorption directly at the silicon surface by ΔT = Φ_th,m⋅α / (c⋅ρ) [11] for the melting threshold of 60 mJ/cm², we end up at ΔT = 3.500 K. Vice versa, considering heating of the solid as well as the latent heat of fusion for silicon, the energy density of Φ_th,m would be sufficient to melt a Si layer of ~90 nm thickness. So it is obvious that in this wavelength range the strong linear extinction of Si leads to a deposition of the total pulse energy in a thin layer of the order of 100 nm, which qualitatively explains the large corona as well as the low damage thresholds observed at 515 nm and, in particular, 400 nm laser wavelength. If the individual differences of SiO₂ and Si_xN_y are only due to different thickness and optical reflection factors (which have not been considered here) of the two materials, or if possibly two-photon absorption in the dielectric layer already takes away an appreciable amount of laser energy, cannot be distinguished from our experiments.

No matter which wavelength we are using, to break the surface and eject a piece from the dielectric layer requires significantly more energy than melting; apparently vaporization of a thin silicon layer is necessary. We want to get upper and lower limits for the thickness of such an evaporated layer. For this, we first use the same approach as above and add up the energy required to heat silicon to its vaporization temperature (3540 K) with heat of fusion and heat of vaporization, and then calculate which volume can be evaporated with the energy density at the breaking threshold. For instance, for the 50 fs pulses at 800 nm and the pertinent value of Φ_th,b = 286 mJ/cm² a maximum layer of 70 nm depth could be evaporated. However, we have to regard that the linear absorption coefficient leads to a penetration depth of ~11 µm, so that initially the main part of the pulse energy heats up the much larger volume rather than evaporating the top layer. Also, the energy density is not yet corrected for the sample reflectivity, and we have to consider that breaking of the bonds at the crater edge as well as the mechanical ejection of a disk of the dielectric material takes up another portion of energy. Over all, it is obvious that only a much smaller layer of the order of 10 nm is converted to the gas phase by the laser pulse energy. Remembering the AFM results, we can get another clue: as there was a considerable time between laser processing and AFM scanning, the opened silicon surface should be covered again by the well-known ‘native’ oxide layer of typically 5 nm thickness. On the other hand, we found in each case that the crater depth was almost perfectly identical with the original dielectric layer thickness. So it is very plausible to assume that 3 to 5 nm of Si are removed and then replaced by the native oxide. We are currently performing Raman and electron microscopy of ablated areas in order to learn more about the structural details of the Si below the irradiated spot. Preliminary results suggest that the crater bottom is still monocrystalline silicon coated by ~4 nm native silicon dioxide. The complete study will be published in a forthcoming paper.

In general, the technique demonstrated in this work has a large potential for two different branches of application: (i) damage-free removal of passivation layers on semiconductor solar cells, e.g. for electrical contacts, and (ii) mask-less preparation of gratings and more sophisticated refractive optics based on transparent thin films on semiconductors. For the latter, more detailed studies will be necessary to explore technical limits like minimum pattern dimension, maximum dielectric layer thickness, feasible material combinations etc.. The main scientific finding is that an electron-hole plasma generated by the ablating pulse itself is apparently the key to the ‘smooth’ delamination, which is counteracted by too strong linear absorption. Thus, an indirect semiconductor like Si, and laser pulses as short as possible and at a wavelength only moderately above the band gap, appear to be the best choice for the process.