ADVANCES IN LASER SINGULATION OF SILICON Paper #770 Leonard Migliore1, Kang-Soo Lee2, Kim Jeong-Moog2, Choi Byung-Kew2 1Coherent, Inc., Santa Clara, CA, USA 2HBL Corporation, Daejeon 305-811 Korea Abstract Individual devices have been saw-cut ("singulated") from wafers since the invention of the integrated circuit. As circuits get smaller and wafers get thinner, however, the limitations of mechanical processes begin to interfere with productivity. Saws tend to chip thin wafers. Newer, more porous "low-k" dielectrics also are prone to chipping when mechanically cut. At this time, 355nm diode-pumped solid-state lasers are have advantages over mechanical saws when cutting silicon less than 200 microns thick. We have been examining methods of improving laser cutting speed and edge quality to increase the laser's effective cutting capacity. A significant improvement in process rate can be achieved by shaping the focused beam. The fracture strength of the edges, as measured by bend testing, was maintained. Introduction Microelectronic devices are made in large numbers on silicon wafers up to 300 mm in diameter. These devices must be cut out of the wafers prior to packaging them. The standard technology for performing this operation is the diamond saw, which works well for traditional wafers with thicknesses of about 630 having silica (SiO 2 )dielectric coatings. The linear cutting speed for diamond saws on fullthickness wafers is generally below 100 [1]. Two trends in integrated circuits, however, interfere with the diamond sawing process: Thinner wafers (for smaller memory packages) tend to shatter under the forces of the cutting blade, while low-k dielectrics (for higher performance logic circuits) chip at the cut line because of their poor mechanical properties. The blade materials and process settings required to mitigate these problems result in greatly reduced cutting speeds compared to those obtained with full-thickness wafers. Laser cutting offers a great advantage here. Since the material removal process is thermal rather than mechanical, forces on the workpiece are drastically reduced in comparison to those produced by sawing. This essentially eliminates any issues with shattering the wafer and reduces chipping of low-k dielectrics. A secondary advantage of laser dicing is that focused laser beams can be smaller than any practical saw blade, allowing a narrower kerf, which means that more devices can be put on any given wafer. Laser Selection Dicing of silicon requires high laser fluence to vaporize the material and high absorption to localize the volume where the energy is deposited. These constraints lead to the choice of Q-switched frequencydoubled or tripled solid-state lasers using neodymium as the active medium. Such lasers, especially if diodepumped, are capable of high peak powers and excellent beam quality. Choosing between doubled (wavelength 532 or 527 nm) and tripled (wavelength 355 or 351 nm) Nd light involves a compromise between available power and cut quality. The absorptivity of silicon at 532 nm is 6.55 X 103 cm-1, while at 355 nm the absorptivity is 106 cm-1 [2]. These numbers translate to an absorption depth of 1.5 at 532 nm and 10 nm at 355 nm. The more localized absorption at the shorter wavelength concentrates the laser energy in a smaller volume, leading to a smoother surface. The efficiency of the frequency conversion process, however, is less for tripling than it is for doubling, so 355 nm lasers typically have less power than 532 nm lasers. The experiments described here used a 10 watt 355 nm DPSS laser (Coherent Avia X). Process speeds in silicon dicing have been shown to correlate reasonably well with laser power, so it should be possible to extrapolate the results to determine the speeds achievable with higher-powered lasers. Cutting Method High beam quality Q-switched diode-pumped solidstate lasers deliver pulses with energies around 100 µj. This amount of energy is only sufficient to vaporize less than a micron of silicon, so cutting through any
significant thickness of material requires multiple passes. The travel speed must be high to minimize thermal effects. The combination of high travel speed with multiple passes suggests that the beam be manipulated with a galvanometer scanner, although this limits the allowable length of the cut. Experimental Setup As noted above, the source used was a Coherent Avia X 355 nm diode-pumped solid-state Q- switched laser. The output of this laser was directed by a series of plane mirrors to a Scanlabs Hurryscan 10 galvanometer scanner equipped with a 100 mm focal length flat-field scanning lens. Applying Self's equations [3] to the laser's published beam characteristics (waist diameter 3.5 mm, M2 <1.3), the theoretical 1/e2 spot size at focus is 16. A drilling test performed on polyimide produced spots with a measured diameter of 15. This is reasonably good agreement with theory since there is no way to measure the 1/e2 diameter directly and the polyimide does not exactly duplicate the form of the incident energy. Figure 1: Polyimide print of focused beam with normal optical train (measured diameter is 15 ) Experimental Procedure The focal plane of the scan lens was determined by performing drilling tests on polyimide film for both the round and elliptical beams, examining the results under a microscope and selecting the position that produced the smallest spot. Standard 100 and 125 mm diameter bare silicon wafers were thinned to 100 and 200 at a commercial wafer grinding facility. Sample wafers were set on a vacuum chuck that was relieved below the cutting regions to allow through cutting of the wafer. The chuck was then set below the galvanometer scanner so that the top surface of the wafer was in the focal plane of the scan lens. A 6.35 mm X 25.4 mm rectangle was programmed into the scanner controller. Test cuts were made using a variety of scanning speeds, laser pulse repetition rates and numbers of passes. For tests made using the elliptical beam, the major axis of the ellipse was aligned parallel to the long side of the sample. As was expected, extra passes had to be made to cut the short sides because of the directional cutting characteristics of the elliptical beam. These passes were not included in the cutting results. Test cuts were evaluated on the basis of minimum net cutting time (travel speed/ number of passes). Two representative conditions were selected for each material thickness and beam shape and a series of 10 specimens was made for each condition. To generate an elongated focal spot, a cylindrical lens with a focal length of 10 meters was inserted into the optical path 100 mm before the entrance pupil of the scan lens. The astigmatism introduced by this lens caused the focal spot to become an ellipse. Beam prints generated on polyimide were measured at 12 X 80. Figure 3: Wafer test pattern Figure 2: Polyimide print of focused beam with cylinder lens in optical path (measured size is 12 X 80 ) Bend testing The cut samples were subjected to a 4-point bend test. Force was applied by sets of 3.2 mm diameter pins set 5 mm apart. This geometry generates a constant moment over a test length of 10 mm.
For each set of test conditions, 5 specimens were tested with the beam entrance side in tension and 5 specimens were tested with the beam exit side in tension (All failure in this material is expected to be in tension). Figure 4: Test sample in bend test carrier (sample is 6.35 X 25.4 mm in size) Figure 5: Sample and carrier in bend test fixture Experimental Results Table 1: Cutting speeds achieved with round and elliptical beams (R=round beam, E=elliptical beam) No. passes Net speed 102 R 1000 60 16.7 102 R 500 30 16.7 102 E 2000 32 62 102 E 500 8 62 203 R 1000 400 2.5 203 R 500 160 3.1 203 E 2000 300 6.6 203 E 500 100 5 Table 2: Laser conditions for optimum cutting results Rep khz Power W Energy µj 102 R 1000 100 6.65 67 102 R 500 100 6.65 67 102 E 2000 80 7.86 98 102 E 500 80 7.86 98 203 R 1000 80 8.2 103 203 R 500 80 8.2 103 203 E 2000 50 9.3 190 203 E 500 50 9.3 190 Table 3: Bend test results (average of 5 samples for each condition) Surface Strength MPa 102 R 1000 top 150 102 R 1000 bottom 166 102 R 500 top 160 102 R 500 bottom 186 102 E 2000 top 160 102 E 2000 bottom 180 102 E 500 top 154 102 E 500 bottom 160 203 R 1000 top 174 203 R 1000 bottom 125 203 R 500 top 170 203 R 500 bottom 127 203 E 2000 top 200 203 E 2000 bottom 135 203 E 500 top 140 203 E 500 bottom 135 A) Cutting Speed Discussion The elliptical beam allowed for a dramatic improvement in cutting speed for the 100 wafer, the net cutting speed rising from 16.7ond with the round beam to 62ond with the elliptical one. The speed also improved for the 200 wafer, but not by as great a factor. There are several reasons why this occurs. For a pulsed laser to produce a continuous cut, the pulses must overlap. With a spot diameter of only 15, the laser's pulse repetition rate must be very high to allow reasonable travel speeds. This requirement conflicts with the natural behavior of Q-switched lasers to lose energy as the repetition rate rises. In addition, a small spot results in a very high fluence which can exceed the optimum for efficient material processing. Simply increasing the spot diameter mitigates these issues but
increases the amount of material that must be removed, slowing the cut speed even more. An elliptical spot solves all these problems as long as linear cuts are acceptable. Pulse overlap is very high at any repetition rate, allowing the selection of a repetition rate that produces high average power. The fluence can be dropped to a more optimal value. Finally, the kerf is only as wide as the minor axis of the ellipse, minimizing the volume of material to be removed. Table 4 summarizes the effect of the elliptical spot on the processing conditions used in these tests. Table 4: Spot overlap and fluence for silicon cutting Rep khz Overlap % Fluence J/cm2 102 R 1000 100 37 33 102 R 500 100 69 33 102 E 2000 80 69 12 102 E 500 80 92 12 203 R 1000 80 22 51 203 R 500 80 61 51 203 E 2000 50 50 23 203 E 500 50 87 23 In Table 4 it can be seen that overlaps of 50% or more with a 15 diameter beam require relatively high laser pulse repetition rates which lower the laser's average power. In addition, the resulting fluences are too high to achieve the highest material removal rates. Figure 6: Etch depth vs. fluence for silicon at 355nm [4] 1.4 1.2 1.0 0.8 0.6 0.4 0.2 0 0 10 20 30 40 50 60 70 80 90 100 110 120 Fluence (J/cm2) B) Edge quality Figure 7: Edge of 100 sample cut with 15 diameter round beam Figure 8: Edge of 200 sample cut with 15 diameter round beam Figure 6 shows material removal vs. fluence for silicon. There is a minimum fluence below which no material is removed. The etch rate then rises rapidly as the fluence increases to 10 J/cm2, then flattens out for higher fluences. This is the key to the increased removal rate of an elliptical spot. As an example, consider a round spot with a fluence of 40 J/cm2. Each pulse will remove about 0.6 of silicon. If the pulse overlap is 50%, each pass will remove about 1.2 of silicon. If the spot is extended by a factor of 4, the irradiance drops to 10 J/cm2 but, at the same travel speed, the overlap increases to 88 percent. Each pulse removes 0.4 of silicon but eight pulses strike each point for a total removal of 3.2 per pass. Figure 9: Edge of 100 sample cut with 12 X 80 elliptical beam Further increases in travel speed can be obtained by the ability to operate at lower pulse repetition rates that produce higher pulse energies. These effects have combined in this experiment to produce a speed increase of 3.75 X in 100 silicon.
Figure 10: Edge of 200 sample cut with 12 X 80 elliptical beam The stress in the sample may be calculated using Euler-Bernoulli bending theory. For a rectangular beam subjected to a uniform moment M, the stress is zero along the beam centerline and rises to a maximum at the top and bottom. Maximum tensile stress σ as a function of M is: σ = 6M (2) W * T 2 where W = sample width and T = sample thickness Samples cut with a round beam exhibited considerable recast on the exit side of the beam (in Figures 7 to 10, the beam entrance side is at the top). This was greatly minimized with the elliptical beam. Edge quality was also examined by bend testing. A 4- point bend test fixture (see Figure 5) was used. The geometry of the fixture is shown in Figure 11. Figure 11: Schematic of bend test geometry Bend test results (Table 3) were inconclusive, possibly because of the small number of specimens used for each condition. The only significant information obtained from these tests seems to be that the laser beam exit side of the 200 thick wafer has more defects than the entrance side. A statistical analysis of the bend test data was performed in order to determine the probability of fracture as a function of stress for each test condition. The Weibull strength distribution [5] was used. Failure stresses were ranked and probabilities were assigned in accordance with F TEST PIECE P = i N + 1 (3) L F In this fixture, a force F is applied to two pairs of cylindrical pins. The moment arm L for each pair is 5mm while the gauge distance G is 10mm. The moment induced by a force of F newtons is then G where P is the probability of the ith fracture and N is the number of samples. The Weibull modulus β and the characteristic stress α were then estimated from a plot of ln(ln(1/(1-p)) vs. ln (σ) [6]. The characteristic stress α is that at which 63% (1-1/e) of the samples would be expected to fail while the modulus β, which is the slope of the plot, indicates the amount of scatter of the data (higher moduli represent less scatter) A Weibull strength distribution analysis of the results is shown in Table 5. F 2 * L (1) or 0.005 F n-m This moment value is constant over the gauge length G and drops to zero outside the gauge length. This test is then sensitive to any defects within the gauge length. The applied force was measured by a high-sensitivity load cell whose output went to a peak-sensing readout. All specimens exhibited brittle fracture in tension. Force values at fracture were recorded
Table 5: Weibull characteristic strength (α) and shape parameter (β) for silicon bend tests Surface α β MPa 102 R 1000 top 171 5.6 102 R 1000 bottom 171 14.25 102 R 500 top 163 13.4 102 R 500 bottom 201 5.25 102 E 2000 top 169 10.2 102 E 2000 bottom 190 6.6 102 E 500 top 157 2.5 102 E 500 bottom 164 6.3 203 R 1000 top 187 5.8 203 R 1000 bottom 130 13.3 203 R 500 top 180 7.0 203 R 500 bottom 138 5.5 203 E 2000 top 206 9.8 203 E 2000 bottom 140 11.5 203 E 500 top 147 10.6 203 E 500 bottom 140 14.2 This analysis confirms the results of the simple examination of the mean break stress: There is no significant difference in the fracture strength between the round and elliptical beams, and the bottom edge of the 200 thick samples has more defects than the top edge. This variation is not seen for the 100 thick samples. The strength values themselves are lower than typically reported fracture stresses for diced silicon [7], which can be as high as 500 Mpa. This can be attributed to the use of the 4-point test, which is sensitive to fractures along the entire gauge length. The 3-point bend test commonly used on small dies only applies the full moment to the region directly under the center load pin, drastically reducing the area being tested. Also, proprietary processes [1] are available which claim to increase the fracture strength of lasercut edges to that of diamond-sawed ones. Conclusions Significant improvement in the cutting speed of thin silicon wafers can be achieved by shaping of the beam. For wafers with a thickness of 100, laser cutting speeds are greater than those achievable with diamond dicing saws. While cutting with elliptical beams must be in the direction of the spot's major axis, this limitation is shared with mechanical saws. Cut edge quality with an elliptical beam is comparable with that achieved with round beams. Further work There are several directions which may be followed. First, testing was done with a 10 watt laser. Considerably more laser power is available, suggesting that the maximum cutting speed can be increased even more. Second, only one elliptical geometry was used. There is probably an optimum shape for the spot which must be found by experiment. Additionally, the optimum fluence should be determined. Finally, wafer thicknesses other than 100 and 200 must be examined. Very high speed was reached with 100, while the speed for the 200 material was not competitive with mechanical processes. References [1] Toftness, R.F. et al (2005) Laser Technology for Wafer Dicing and Microvia Drilling for Next Generation Wafers, in Photonics West 2005, SPIE, Vol. CDS158:5713-10 pp 54-66 [2] Aspnes, D.E., Optical Properites of c-si, in Hull, R., (ed) Properties of Crystalline Silicon, 1999, INSPEC, p. 679 [3] Self, S.A. (1983) Focusing of spherical Gaussian beams, Appl. Optics 22(5), 658-661 [4] Greuters, J. and Rizvi, N. (2003) UV laser micromachining of silicon, indium phosphide and lithium niobate for telecommunications applications, in Proceedings of the SPIE vol. 4876; Opto-Ireland 2002, pp. 479-486 [5] Cotterell, B. et al (2003) The Strength of the Silicon Die in Flip-Chip Assemblies, Journal of Electronic Packaging Vol 125 pp. 114-119 [6] Dorner, W. (1999) Using Microsoft Excel for Weibull Analysis, Quality Digest Jan. 1999 [7] Perrottet, D. et al (2005) Heat damage-free Laser- Microjet cutting achieves highest die fracture strength, in Photonics West 2005, SPIE, Vol. CDS158:5713-42 Meet the Author Leonard Migliore is a Staff Engineer at Coherent, Inc. In that capacity he examines laser-material interactions in industrial processes and develops enhancements that improve manufacturing productivity. Mr. Migliore, whose background is in materials science, has been working with industrial lasers since 1979 and has participated in the development of many laser welding, cutting, drilling and heat treating installations.