HIGH SPEED LASER STRUCTURING OF CRYSTALLINE SILICON SOLAR CELLS

Transcription

1 HIGH SPEED LASER STRUCTURING OF CRYSTALLINE SILICON SOLAR CELLS S. Eidelloth, T. Neubert, T. Brendemühl, S. Hermann, P. Giesel, and R. Brendel Institut ür Solarenergieorschung Hameln (ISFH), Am Ohrberg 1, D Emmerthal, Germany ABSTRACT Fast laser processing is commonly done using laser beams with Gaussian beam proiles in combination with scanning optics. Single laser pulses only aect a limited area beneath the Gaussian intensity bell and result in circular impact regions. Adjacent impact regions have to overlap considerably when continuous processing larger areas. Thus, the processing speed is greatly enhanced by replacing the Gaussian proile with a lat-top intensity proile and by replacing the radial symmetric cross section with a rectangular beam cross section. However, processing with a rectangular lat-top laser beam through a scanner has, to the best o our knowledge, not yet been demonstrated. We report on the successul design and experimental tests o a new laser system that images a rectangular lat-top laser beam through a scanner. Our socalled Simultaneous Scanning and Laser Beam Imaging system (SIMSALABIM) machines a inger pattern that covers 50 % o a (125 x 125) mm 2 crystalline Si solar cell in 14 s. Two parallel slab lasers with increased output power should process the same area in just 2.5 s. INTRODUCTION High eiciency, waer based cell concepts call or well deined geometric patterns. Photolithography is expensive. Thereore alternative processes like inkjet printing or laser structuring need to be pushed to a level capable o meeting industrial needs. Laser machining tents to result in higher investment, but lower cost o operation compared to printing. How to urther reduce laser process times and thereore increase waer throughput is the topic o this paper. We compare the structuring index (distance between the centres o two single impact regions) o round Gaussian beams and rectangular lat-top beams. We also discuss structuring times or inger shaped geometries. The geometry ater the scanning mirrors would lie on a curved plane without a so called -theta lens. We need the -theta lens to homogeneously machine the lat waers. When a Gaussian laser beam propagates through an -theta lens it approximately stays Gaussian, because the Fourier transorm o a Gaussian proile again is a Gaussian proile. Thus, the -theta lens o the scanner has no crucial impact on the Gaussian intensity proile o the laser beam. However, a rectangular lat-top intensity proile is not invariant to Fourier transormation. We describe a new laser system that Fourier transorms the shaped laser beam beore the scanner. The -theta lens then perorms a reverse transormation. Thus, we get the wanted intensity proile at the waer surace. Our SIMSALABIM system paves the way or joining high speed scanning with the beneits o beam shaping. STRUCTURING INDEX AND SINGLE PULSE ABLATION AREA Laser structuring, like edge isolation, is commonly perormed by a beam with Gaussian-like intensity proile with and x + y 2 2 w( z) I( x, y, z) I ( z) e, (1) w0 I0( z) = Imax (2) w( z) 2 2 w( z) z = 1. (3) w z R The ablation area A(z) in a simplistic model (see Fig. 1) only depends on the ablation threshold intensity I th, the maximum Intensity I 0(z), and the beam radius w(z). Fig. 1: Schematic intensity distribution o a Gaussian laser beam. - Threshold intensity I th, maximum intensity I 0(z) and beam radius w(z) deine the ablation area A at a deinite position z d in propagation direction z. The maximum ablation area A max results at the optimum position z max. 1/6

2 I the beam geometry is ixed (installed optics deine the Rayleigh length z R o the Gaussian beam and t r: Ramping & return time or scanning mirror thereore the hyperbolic shape in propagation direction z is acceleration/deceleration, depending on the ixed), the maximum ablation area A max is obtained (see number o structuring lines N, scanning Fig. 1) at a corresponding position velocity v s, mirror weight, etc. z max I ± max zr 1 Imax > e Ith = e Ith. (4) 0 Imax e Ith This optimum position generally depends on the maximum Intensity I max in the beam, the Rayleigh length z R and the ablation threshold intensity I th o the material. The maximum size A max o the round ablation area as unction o laser pulse energy E p and pulse duration t p can be estimated or ideal beam geometry with A max 1 Ep =. (5) e t I For ablating a continuous line, the round spots are projected with the spot overlap δ circle onto the waer surace. The according structuring Index d gauss (that is the distance between the centers o two adjacent single pulse impact regions) becomes [1] : d gauss p th 4 Ep = ( 1 δ circle ) (6) eπ t I The volume beneath the Gaussian bell in Fig. 1 corresponds to the power P o the laser pulse. The cap and the surrounding part o the bell are lost, because only the inner power-cylinder is utilized or the ablation process. Thereore, structuring with a lat-top cuboid at same pulse energy and square beam cross section results in a larger structuring index d lat-top: p th The laser machining o a waer in practice starts with waer handling and image recognition. Additionally, the total process time includes idle jumping time or the mirrors to jump between separate parts o the scanned geometry (this is negligible in our case). Pure structuring time There are two limiting scenarios or the pure structuring time. In the irst scenario the minimum time is limited by the velocity o the scanning mirrors (scanner limited). In the second scenario it is limited by the size o the structuring index in combination with the laser pulse repetition rate R (laser limited). We discuss both scenarios in detail and consider a inger pattern which is shown in Fig. 2. This pattern includes a busbar or a ull-squared waer. The preerred scanning direction or the busbar is perpendicular to the direction or inger scanning. In Fig. 3 the whole pattern is scanned in one direction. This structuring mode is avored or more complex pseudosquare patterns (depending on scanning parameters, pattern geometry and spot size). We use the overlap actor δ to generally compare the structuring indices or circular and square shaped single pulse impact regions above. The ollowing discussion speciically applies to rectangular ablation areas A = b h (we have a rectangular lat-top intensity proile) and assumes a constant overlapping distance d o. d lat-top Ep = ( 1 δ square) (7) t I Rectangles need less spot overlap than circles and with spot overlaps δ circle 0.3 and δ square 0.15 a theoretical time yield actor o d lat-top/d gauss = 1.8 or line ablation is achieved, comparing round Gaussian proiles and lat-top proiles with a square cross section. Possible disadvantages o shaped beams are power losses through the beam shaping optics, shorter depth o ocus and lower pattern resolution. p th PROCESS TIME We primarily concern two components o the total process time: t s: Pure structuring time, depending on geometry o pattern, structuring indices (d x, d y), laser pulse repetition rate R and scanning velocity v s respectively. Fig. 2: Finger geometry with busbar. - In practice, it is not possible to utilize the entire ablation area A = b h o a single laser pulse. The areas need to overlap by a distance d o in order to reliably achieve a complete coverage. A scanning process ills the inger geometry with lines, having a distance d x to each other and an internal structuring index d y. Line distance and structuring index are interchanged or busbar scanning in perpendicular direction to that o the ingers. 2/6

3 The line internal structuring index thus becomes d x = b - d o or busbar scanning and d y = h - d o or inger scanning in the perpendicular direction. We calculate the corresponding scanning velocity v = d (8) s,y R y in inger direction or a given laser pulse repetition rate R. When optimizing R, one has to keep in mind, that the line index d x and the number o lines per inger n have to it the inger width b, using a small overlap d o and ollowing the equation The pure structuring time n dx = b do. (9) h t N N b d o s = = (10) vs,y dx R dy applies or the total N lines o the N ingers (width b, length h, without busbar height h b). Assuming the single pulse ablated area A to be constant or dierent aspect ratios, the time = b d h o ts N d x A R dx + do h do (11) is written as a unction o the line index d x. It has a minimum or an optimum index N h b d o s ( R ) = (12) R A( R ) do t dx = dy = A do (13) and a corresponding optimum velocity v = ( A d ). (14) s, opt R o However, the real output o lasers might deinitely deviate rom this behavior. Pulse duration t p( R) and threshold intensity I th(t p) are taken constant in this approach. We assume A to be approximately proportional to the pulse energy E p( R). Thus, the ablation area A( R) usually decreases with increasing laser pulse repetition rate R. In our irst scenario the optimum ablation velocity v s,opt is too high or the scanner. It is not possible to get discrete spots onto the waer and the process is scanner limited. We assume a scanning velocity limit o 15 m/s. For a laser system with a high repetition rate o R = 1 MHz an ablation area larger than (15 x 15) µm 2 is required to reach this limit. The next generation o picosecond lasers (or even some currently available lasers, see e.g. [2]) should be able to achieve this speciication. With small ablation areas A the number o lines N and thereore the ramping and return time t r (see below) would be large, even rising the demand in scanning speed. I one chooses such a scanner limited process it might be advantageous to reduce the line number N by shaping the beam rectangular (short side o rectangle would parallel the scanning direction). We put the second scenario into practice or our SIMSALABIM system. Our process is laser limited and works with large single pulse ablation areas at a low laser pulse repetition rate o 50 khz. Even with next generation slab lasers this kind o process is likely to stay laser limited or the near uture. I we consider or example a structuring index o d x = d y = 150 µm the laser system would have to deliver a laser pulse repetition rate o R = 100 khz to reach the scanner limit. A lot o pulse energy has to be stored in the laser active media to ablate an area larger than (150 x 150) µm 2. Furthermore the laser wavelength and the pulse duration have to be low, granting low laser induced silicon damage [3]. These speciications are hard to achieve. Ramping & return time Our machining programs include acceleration and deceleration ramps or each line to ensure constant structuring index d y. Within a single ramping & return time t r1 the scanning mirrors delect the laser beam to the next line s starting point and velocity (see Fig. 3). Thus, the aspect ratio should be 1 (the cross section o the laser spot should be a square). Furthermore, the minimum structuring time t s,min is reached at an optimum laser repetition rate R,opt, depending on the unction A( R). This unction is characteristic or a speciic machining process and laser system. The area A generally depends on E p( R), t p( R) and I th(t p). The pulse energy E p stored in the active media o a laser usually decreases with increasing R. It sometimes can be described like an unloading curve o a capacitor : 1 R Ep( R ) Emax 1 e τ = (15) Fig. 3: Scanning path, including structuring path, ramping & return path and idle jumping path. 3/6

4 With constant angular acceleration o the mirrors the ramping & return times are a unction o the scanning velocity. Then or short (busbar) lines, it could be helpul to reduce structuring velocity. The structuring time would increase, but ramping & return time would decrease, decreasing total process time. However, our scanner controller does not use a constant acceleration. The controller automatically adjusts the acceleration as unction o scanning velocity: or low line structuring speed low ramp accelerations apply. The acceleration time o our scanner (hurryscanii 14, SCANLAB AG) is ms or angular speeds 50 rad/s and the acceleration is not constant within the acceleration time. Our single ramping & return time t r1 stays constant at about 1 ms, independent o structuring speed. Thereore, with our current scanner controller we are not able to optimize the scanner acceleration and speed as unction o the line length. Let us now estimate the process time or industrial machining with a next generation slab laser (P av 200 W) o a (125 x 125) mm 2 pseudo square waer. A laser limited process, scanning 256 inger lines with a length o h = 125 mm at v s = 10 m/s theoretically takes t s = 2.3 s pure structuring time. Adding another 253 busbar lines with an average length o h b = 3 mm results in a total pure structuring time o t s = 2.38 s. Ramping & return with t r1 = 1 ms and 509 lines would inally increase the process time to t = 2.9 s. Thus, with two parallel slab lasers the process time should be below 2.5 s, including waer handling, image recognition and idle jumping. For the scanner limited scenario with high repetition rate lasers and low line distances the ramping & return time gets more important. As an example we increase the line number by a actor o 10 to increase the pattern resolution. Then the scanning velocity v s would have to be increased and the single ramping & return time t r1 would have to be decreased by the same actor to keep the process time constant. This high line number kind o process could be realized with the application o high repetition rate lasers and polygon scanners. Polygon scanners should allow or scanning speeds > 50 m/s and the return time would correspond to the small gate time or blending out mirror edges. However, polygon scanners are not commercially available or our applications yet. two additional lenses l 2, l 3) images the intensity proile down to the substrate. The task o the irst lens l 1 is to supply the telescope with the already shaped laser beam at required object position. This optical setup was proposed by Edgewave GmbH. We use an -theta lens with a ocal length large enough ( = 0.25 m) to machine (156 x 156) mm 2 waers. This results in a total optical path length o approximately 1.5 m. Fig. 4: Components o our home build SIMSALABIM system. Figure 5b shows the principle o our SIMSALABIM system. The let telescope lens represents the optics l 2, l 3 and results in a Fourier transormation o the squared lattop proile. The right telescope lens o the principle igure represents the -theta lens o the scanner and thus the reverse transormation. Behind the telescope an image o the object is projected onto the waer surace and results in square like ablation areas (see Fig 9). a) SIMSALABIM SYSTEM Figure 4 shows the components o our home-built Simultaneous Scanning and Laser Beam Imaging - system (SIMSALABIM). One o the advantages o our INNOSLAB laser (λ = 532 nm, t p 10 ns, P av 50 R = 50 khz, Edgewave GmbH) is its resonator shape. The output beam already has a lat-top intensity distribution in unstable resonator direction [4]. The Gaussian shape in stable resonator direction is transormed to a lat-top intensity shape by means o a beam shaping module, directly attached to the slab laser. Lenses (l 1, l 2, l 3) and mirrors (m 1, m 2) guide the shaped laser beam to the scanner. The schematic o our optical assembly is shown in Fig. 5a. A telescope (consisting o the -theta lens l 4 and b) Fig. 5: Optical assembly a) and principle b) o the SIMSALABIM system. We place scanning mirrors inside the telescope to move the image and structure the geometry pattern. (Since the intensity inside the telescope is concentrated one has to keep in mind not to exceed the damage threshold o the scanning mirrors.) Thus our process 4/6

5 simultaneously delects and images the laser beam. To the best o our knowledge this is the irst time a lat-top intensity distribution was guided through a scanner or high speed laser structuring o solar cells. Our SIMSALABIM system thus paves the way or joining high speed scanning with the beneits o beam shaping. EXPERIMENTAL ESTIMATION OF OPTIMUM LASER REPETITION RATE We measure the average output power P av( R) o our SIMSALABIM system ater the scanner and determine the pulse energy E p = P av / R (Fig. 6a). The average output power and pulse energy o our picosecond laser system (utilizing a SUPER RAPID laser rom LUMERA LASER GmbH) is shown in Fig. 6b or comparison. a) The pulse energy in Fig. 6b is itted with a doted line and approximately behaves like an unloading curve (15). The pulse energy E p o the SIMSALABIM system irst increases with increasing laser repetition rate R and then decreases. The optimum repetition rate R,opt is not obvious at a irst glance. We assume the pulse energy E p to be proportional to the single pulse ablation area A. Using this approximation and neglecting the inluence o d o we can simpliy equation (12). The pure structuring time t s ( R) is approximately inversely proportional to the product ( ) Pav pv = R Ep R = (16) 3 R o repetition rate and square root o pulse energy. This characteristic product p v is shown in Fig. 7. The product p v o our SIMSALABIM slab laser system (INNOSLAB) has its optimum laser repetition rate at R,opt = 50 khz. The maximum o the characteristic product o our slab laser system is larger than that o our picosecond laser system. b) Fig. 7: Characteristic product p v or our SIMSALABIM slab laser system (INNOSLAB) and our picosecond laser system (SUPER RAPID). - The maximum o the product p v (laser repetition rate times square root o pulse energy) results in minimum process time, i the pulse energy is proportional to the single pulse ablation area and the machining process is not scanner limited. RAMP LENGTH Fig. 6: Average output power P av and pulse energy E p. - a) Our SIMALABIM system includes an INNOSLAB laser. b) The output o another laser system with a SUPER RAPID picosecond laser is shown or comparison. Its pulse energy behaves like an unloading curve. The lengths o acceleration and deceleration ramps o our SIMSALABIM slab laser system are shown in Fig. 8. We use 1.52 mm ramp length or acceleration and 1.25 mm or deceleration at a scanning velocity o v s = 7 m/s. The reason or dierent ramp lengths or acceleration and deceleration is not yet clariied, but may originate rom the scanner controller or a dierence in the laser on and laser o delay. The ramps are required to grand constant spot distances at the start and at the end o the structuring lines. The ramp lengths increase with increasing 5/6

6 structuring velocity. As explained above, the time or In order to structure a geometric pattern we ablate ramping & return stays constant at t r1 1 ms. overlapping lines with a speed o v s,opt = 7 m/s. The photograph o such a pattern on a ull-squared waer is shown in Fig. 10. We also machine more complex patterns on pseudo-square waers. A inger pattern (including busbar) that covers 50 % o a (125 x 125) mm 2 pseudosquare waer takes 14 s (without handling and image recognition). CONCLUSIONS & OUTLOOK Fig. 8: Length o acceleration and deceleration ramps d r as unction o scanning velocity v s. EXPERIMENTAL PROCESS TIME We machine oxidized Si waers using the above described SIMSALABIM system at the optimum laser repetition rate R,opt = 50 khz. Figure 9 shows single pulse ablation areas. The approximately squared patterns do not show any remaining silicon oxide islands inside the opened areas. Simultaneous Scanning and Laser Beam Imaging (SIMSALABIM) allows or high-speed laser machining with rectangular lat-top intensity proiles. Our SIMSALABIM process is not limited to waer substrates and could be applied to all processes calling or a special intensity proile. Edge isolation, marking, cutting and laser doping are a ew examples. Machining with a squared lat-top intensity proile instead o a round Gaussian proile theoretically increases the line processing velocity by a actor o 1.8. There are two possibilities how to signiicantly reduce our current process time o 14 s. One is to use a (currently available) high repetition rate laser system ( R > 1 MHz). Polygon scanners would have to be developed or this application because they can oer the required high scanning velocities and low return times. Our current structuring process time o 14 s with a single slab laser system is limited by the output power and repetition rate (P av 50 W, R = 50 khz) o our slab laser. With 2 parallel SIMSALABIM systems and next generation slab lasers (P av 200 W) industrial waer structuring will be possible within 2.5 s. ACKNOWLEDGEMENTS Fig. 9: Optical microscope image o slab laser ablation areas on an oxidized silicon waer. The authors would like to thank Dr. Du, Edgewave GmbH, or custom product development and help with optical design. We also thank Dr. Harder, ISFH or ruitul discussions. The inancial support by the German ministry o BMU or the project LaserInvest (No A) is grateully acknowledged. REFERENCES Fig. 10: Photograph and optical microscope image o a inger pattern. - Machined with a quasi square lat-top intensity proile through a scanner on an oxidized silicon waer. [1] P. Engelhart, Lasermaterialbearbeitung als Schlüsseltechnologie zum Herstellen rückseitenkontaktierter Siliciumsolarzellen, PhD thesis, 2007, pp , Leibnitz University Hannover. [2] Data sheet o HYPER RAPID 50, LUMERA LASER GmbH, 2009 [3] S. Eidelloth, Untersuchung von laserinduzierten Schädigungen bei der Strukturierung von Siliziumwaern ür Hocheizienz-Solarzellen, Diploma thesis, 2006, University o Applied Sciences Münster [4] Product inormation sheets o Edgewave GmbH, /6