Applied Surface Science 242 (2005) 162167 www.elsevier.

com/locate/apsusc

Femtosecond versus nanosecond laser machining: comparison of induced stresses and structural changes in silicon wafers
M.S. Amera,*, M.A. El-Ashrya, L.R. Dosserb, K.E. Hixb, J.F. Maguirec, Bryan Irwind
a

Department of Mechanical and Materials Engineering, Wright State University, 3640 Colonel Glenn HWY, Dayton, OH 45435, USA b Mound Laser and Photonics Center Inc., Miamisburg, OH 45342, USA c AFRL, Materials Directorate, WPAFB, OH 45433, USA d Sciperio Inc., Stillwater, OK 74075, USA Received in revised form 12 July 2004; accepted 7 August 2004 Available online 12 October 2004

Abstract Laser micromachining has proven to be a very successful tool for precision machining and microfabrication with applications in microelectronics, MEMS, medical device, aerospace, biomedical, and defense applications. Femtosecond (FS) laser micromachining is usually thought to be of minimal heat-affected zone (HAZ) local to the micromachined feature. The assumption of reduced HAZ is attributed to the absence of direct coupling of the laser energy into the thermal modes of the material during irradiation. However, a substantial HAZ is thought to exist when machining with lasers having pulse durations in the nanosecond (NS) regime. In this paper, we compare the results of micromachining a single crystal silicon wafer using a 150-femtosecond and a 30-nanosecond lasers. Induced stress and amorphization of the silicon single crystal were monitored using micro-Raman spectroscopy as a function of the uence and pulse duration of the incident laser. The onset of average induced stress occurs at lower uence when machining with the femtosecond pulse laser. Induced stresses were found to maximize at uence of 44 J cm2 and 8 J cm2 for nanosecond and femtosecond pulsed lasers, respectively. In both laser pulse regimes, a maximum induced stress is observed at which point the induced stress begins to decrease as the uence is increased. The maximum induced stress was comparable at 2.0 GPa and 1.5 GPa for the two lasers. For the nanosecond pulse laser, the induced amorphization reached a plateau of approximately 20% for uence exceeding 22 J cm2. For the femtosecond pulse laser, however, induced amorphization was approximately 17% independent of the laser uence within the experimental range. These two values can be considered nominally the same within experimental error. For femtosecond laser machining, some effect of the laser polarization on the amount of induced stress and amorphization was also observed. # 2004 Elsevier B.V. All rights reserved.
PACS: 61.18j Other methods of structural determination; 61.82 fk Radiation effect of semiconductors; 81.65 b Surface treatment Keywords: Raman spectroscopy; Laser machining; Micromachining; Silicon amorphization; Induced stress; Femtosecond laser machining

* Corresponding author. Tel.: +1 937 775 5095; fax: +1 937 775 5009. E-mail address: maher.amer@wright.edu (M.S. Amer). 0169-4332/$ see front matter # 2004 Elsevier B.V. All rights reserved. doi:10.1016/j.apsusc.2004.08.029

M.S. Amer et al. / Applied Surface Science 242 (2005) 162167

163

1. Introduction Lasers have been widely utilized in metallic materials machining since the early 1970s [1]. More recently, lasers have been utilized in machining nonmetallic materials such as ceramics, plastics, various composites, and semiconductors (e.g., silicon, silicon carbide, etc.) for a number of industrial applications [2,3]. The ability of lasers, especially pulsed lasers, to precisely machine micron and sub-micron features in otherwise hard to machine materials such as ceramics and semiconductors has created a rapidly growing interest in understanding the parameters controlling the limits and the capabilities of this process [4]. A large number of studies have been devoted to investigate laser-based micromachining that covered the different aspects of the machining process [1,48] and the physics of laser/material interaction [9,10]. The development of femtosecond (FS) lasers and their initial application to the machining of a variety of materials has created huge interest in their micromachining potential. Current reasoning suggests that the pulse duration of a femtosecond laser is so short such that there is not sufcient time for any of the pulse energy to be distributed to the substrate in the form of heat. Thus, particularly for low pulse energies, there should be no heat-affected zone (HAZ) resulting from the processing. This is in direct contrast to nanosecond (NS) machining, which has an associated HAZ. The magnitude of the HAZ is a direct result of the machining parameters and can be minimized for nanosecond micromachining. However, the HAZ is only one aspect of how a material is altered during the machining process. It is important to also look at the stresses induced as a function of laser parameters. Previously, the authors utilized micro-Raman spectroscopy to investigate induced stresses and structural changes in single crystal silicon as a result of nanosecond pulsed laser machining [11]. This paper reports the results of an initial study to compare the magnitude of the stresses induced by femtosecond and nanosecond micromachining of single crystal silicon. A variation in both the temporal and spatial distributions of the laser energy exists in the experiments performed using the FS and NS laser systems. The temporal distribution is attributed to the

difference in the laser pulse duration while the spatial distribution is attributed to different laser focal spot sizes. Studying the effects of the temporal distribution is the primary goal of this paper. However, the variation in the spatial distribution of the laser energy can be accounted for by normalizing the pulse energy with respect to the area over which it is delivered. Thus, the data should be interpreted in terms of energy density or uence (J cm2) to understand the machining effects of variations in the pulse duration.

2. Experimental procedure 2.1. Micromachining Laser micromachining experiments were performed using solid-state laser systems having pulse durations in the nanosecond and femtosecond regimes. The nanosecond machining was performed using the Spectra Physics frequency tripled YHP40-355 Nd:YVO4 laser. The wavelength was 355 nm and the pulse duration was nominally 3050 ns depending on the pulse repetition rate. The linearly polarized laser beam was scanned across the stationary <1 1 1> single crystal Si substrate using the ScanLab HurrySCAN10-355 galvanometer laser beam scanner. The theoretical focal spot size was 13 mm. A single pass of the laser beam was used to machine one groove for each of the experimental conditions. The incident laser pulse energy was varied from 3.0 mJ to 193.3 mJ. The pulse repetition rate and scanning speed were varied to maintain a constant pulse overlap. The femtosecond laser micromachining was performed using the Clark MXR laser system. The wavelength was 775 nm and the pulse duration was nominally 150 fs. The laser beam was focused to a 50mm spot. The machining was performed by translating the Si wafer at 1 mm/s through the focal region for all trials. The pulse repetition rate was held constant at 1 kHz, and the pulse energy was varied from 17.5 mJ to 950 mJ. Experiments were also performed using a 1/ 4-wave plate to circularly polarize the beam. Under these conditions, the pulse energy was reduced slightly and experiments were performed from 16.5 mJ to 930 mJ. All grooves were machined using a single pass of the laser beam. Beam cross sections were measured using beam prolers. No attempt was

164

M.S. Amer et al. / Applied Surface Science 242 (2005) 162167

made to change the beam cross sections during machining. 2.2. Micro-Raman spectroscopy Micro-Raman measurements were conducted using a Renishaw1 2000 model. The excitation laser used was Ar+ ion laser with a wavelength of 514.5 nm. Laser power at the sample was kept at 2.5 mW to avoid local heating of the wafer. An area of 100 mm 20 mm around each laser-machined groove was mapped using a step of 1 mm to produce the reported stress and amorphization maps. The Raman machine was calibrated using an untreated area of the same (1 1 1) silicon wafer to determine the stress free peak position at 520 cm1 [12]. Hence, the stresses measured and reported represent the stresses directly resulting from laser machining. Induced stress was calculated from the shift in the silicon peak position using the experimentally measured value for the Raman shift/stress relationship of 1.55 cm1 GPa1 and not the value of 2 cm1 GPa1 obtained from theoretical calculations [13]. The zero stress value for silicon peak position was set to the value obtained from the same silicon wafer away from the laser machined region (about 10 mms). The local induced stress was calculated as follows: Local stress (GPa) = (local peak position520)/ 1.55. Details of such calculations are explained elsewhere [11]. Also, the amorphization induced

was calculated from the relative integrated intensity of the amorphous silicon (a-Si) peak that occurs at 490 cm1 [14].

3. Results and discussions Raman spectra obtained from the laser-machined silicon wafer are best tted by three peaks. The very well characterized peak around 520 cm1 resulting from the triply degenerate optical mode in single crystal cubic diamond phase of silicon (Si-I) [15], a lower intensity peak around 510 cm1, and a broad peak around 490 cm1 that is related to amorphous silicon [16]. The origin of the peak at 510 cm1 is not clear. It was previously assigned to either hexagonal structure of bulk silicon (Si-IV) or nano-crystals of silicon [17]. A typical Raman spectrum along with the lorentzian tting used to determine the characteristics of each peak is shown in Fig. 1. Fig. 2 shows typical color maps of the stress distribution measured within the laser-machined region superimposed on optical micrographs of the area mapped. It is important to note that due to the complexity of the relationship between Raman frequency and individual components of the stress tensor, a direct estimation of individual components of the stress tensor (even under the assumption of plane stress situation) is practically impossible [18]. However, due to the fact that no splitting in the triply degenerate peak around

Fig. 1. Typical Raman spectrum obtained from the machined area. The Raman spectrum was best tted to three lorentzian peaks.

M.S. Amer et al. / Applied Surface Science 242 (2005) 162167

165

Fig. 2. Colored map of the stress distribution in and around the machined groove and an optical micrograph of the mapped area.

520 cm1 was observed, the shift in the Raman peak position can be assumed to be due to uniaxial stress normal to the groove direction [19]. The focus spot size of the two lasers was considerably different (13 mm for nanosecond and 50 mm for femtosecond), and therefore the data were plotted as a function of uence (J cm2) to normalize the results to the same focal area. The resulting data are shown in Figs. 3ac and 4ac. Fig. 3ac shows that the stress induced by the FS laser reaches a maximum before that of the NS laser. This is reasonable since the laser energy delivered by the femtosecond laser occurs over a time period nearly ve orders of magnitude shorter than the nanosecond laser, which results in a very high peak power at the substrate surface (TWcm2). At uence levels that are only slightly above the damage threshold for the substrate, the femtosecond lasermaterial interaction is purely ablative. It is interesting that the induced stress in this region is greater than for the nanosecond laser and increases more rapidly to a maximum value of $1.11.4 GPa before diminishing to a value of $1.0 GPa. It is believed that the initial increase in stress for the FS

Fig. 3. Average induced stress as a function of laser uence for (a) femtosecond linearly polarized, (b) femtosecond circularly polarized, and (c) nanosecond laser. Lines are added to guide the eye.

166

M.S. Amer et al. / Applied Surface Science 242 (2005) 162167

Fig. 4. Average induced amorphization as a function of laser uence for (a) femtosecond linearly polarized, (b) femtosecond circularly polarized, and (c) nanosecond laser. Lines are added to guide the eye.

laser machining may be due to laser induced shock and that the decrease in stress at higher uences may be the result of plasma heating providing a stress-relief or annealing mechanism. It is interesting to note that although the femtosecond interaction induces stress more rapidly, the nanosecond interaction ultimately induces more stress at the higher uences. It is believed that this is due to the thermal component (excessive thermal loading of the substrate) of the nanosecond laser/matter interaction. The experimental results showed also that lower stresses were induced when the FS laser was circularly polarized. This is very interesting result, however, the exact relationship between laser polarization and induced stresses in the machined substrate is not clear at this stage and will be addressed in future investigations. From the experimental data it is clear that it is more difcult to avoid induced stresses when machining with the femtosecond laser. Thus, it is very important to machine the substrate near the threshold for ablation. The data also suggest that more experiments should be performed at lower uences. The optimal processing window for low inducedstress machining requires relatively low uence. Although nanosecond lasers can induce more stress, they can be readily controlled to maintain the induced stress level at or below that of the femtosecond laser. Ultimately, the selection of the appropriate laser for a particular machining application will be made based upon the signicance of induced stress in the nal product and time required to complete the machining operations. Fig. 4ac shows the dependence of the induced amorphization as a function of laser uence. Amorphization is constant with respect to uence for the femtosecond laser over the experimental range studied in this paper. Likewise, the induced amorphization is also constant for the nanosecond interaction except for uence less than 20 J cm2, in which case it decreases sharply. Knowing that the 514.5 nm laser optical penetration depth in silicon is around 770 nm and assuming the Raman cross section for both crystalline and amorphous silicon are equal, the 20 5% amorphization observed can be translated into amorphous layer thickness of 150 40 nm. The thickness of such amorphous layer will denitely affect the induced stress in the silicon substrate.

M.S. Amer et al. / Applied Surface Science 242 (2005) 162167

167

4. Conclusions Micro-Raman spectroscopy was utilized to investigate induced stresses and amorphization in laser machined silicon wafers and to compare such induced changes for femtosecond and nanosecond laser machining. Contrary to expectations, femtosecond laser was found to induce signicant stress and amorphization in single crystal silicon. Results showed that induced stress depends on the laser uence and reaches a maximum around 50 J cm2 and 25 J cm2 for nanosecond and femtosecond lasers, respectively. The maximum stress observed for nanosecond laser was higher than that observed for femtosecond lasers. It was also observed that circularly polarized femtosecond laser induced less stresses that the linearly polarized laser. Amorphization induced ranged around 20 5%. There could be a polarization effect for femtosecond laser machining on induced changes in the substrate. Such effects require further investigation.

References
[1] S.J. Ahn, D.W. Kim, H.S. Kim, K.H. Cho, S.S. Cho, Appl. Phys. A 69 (1999) 527530. [2] J.F. Ready, Industrial Applications of Lasers, Academic Press, London, 1997, p. 394.

[3] ICALEO, Laser Micro-fabrication Session, October 25, 2000. [4] P. Simon, J. Ihlemann, Appl. Phys. A 63 (1996) 505 508. [5] S. Dauer, A. Ehlert, S. Buttgenbach, Sens. Actuators 76 (1999) 381385. [6] J. Li, G.K. Anathausuresh, J. Micromech. Microeng. 11 (2001) 3847. [7] J. Jandeleit, A. Horn, R. Weichenhain, E.W. Kreutz, R. Poprawe, Appl. Surf. Sci. 127129 (1998) 885891. [8] H.W. Bergmann, Appl. Surf. Sci. 9698 (1996) 287295. [9] W. Xu, J. Phys. Condens. Matter 10 (1998) 61056120. [10] G. Marowsky, C.K. Rhodes, Appl. Phys. B 66 (1998) 475 478. [11] M.S. Amer, L. Dosser, S. LeClair, J.F. Maguire, Appl. Surf. Sci. 187 (2002) 291296. [12] I. Dewolf, G. Pozzat, K. Pinardi, D.J. Howard, M. Ignat, S.C. Jain, H.E. Maes, Microelectron. Reliab. 36 (11/12) (1996) 17511754. [13] S.T. Amimoto, D.J. Chang, A.D. Birkitt, SPIE, vol. 3512, Santa Clara, CA, pp. 123129. [14] Y. Kimura, T. Katoda, Appl. Surf. Sci. (117/118) (1997) 790 793. [15] V. Domnich, Y. Gogotsi, S. Dub, Appl. Phys. Lett. 76 (16) (2000) 22142216. [16] M. Marinov, N. Zotov, Phys. Rev. B 55 (1997) 2939. [17] Y. Gogotsi, C. Baek, F. Kirscht, Semicond. Sci. Technol. 14 (1999) 936944. [18] I. De Wolf, H. Norstrom, H.E. Maes, J. Appl. Phys. 74 (7) (1993) 44904500. [19] I. Dewolf, G. Pozzat, K. Pinardi, D.J. Howard, M. Ignat, S.C. Jain, H.E. Maes, Microelectron. Reliab. 36 (11/12) (1996) 17511754.