- Email: [email protected]
2D Mapping of Chemical and Field Effect Passivation of Al2O3 on Silicon Substrates
Available online at www.sciencedirect.com
ScienceDirect Energy Procedia 77 (2015) 91 – 98
5th International Conference on Silicon Photovoltaics, SiliconPV 2015
2D mapping of chemical and field effect passivation of Al2O3 on silicon substrates Paul M. Jordana,*, Daniel K. Simona, Franz P.G. Fenglera, Thomas Mikolajicka,b, Ingo Dirnstorfera b
a NaMLab gGmbH, Nöthnitzerstr. 64, 01187 Dresden, Germany Technische Universität Dresden, Institut für Halbleiter- und Mikrosystemtechnik, 01062 Dresden, Germany
Abstract Fixed charge and interface defect densities are the critical material parameters for silicon surface passivation. These parameters are measured with high spatial resolution on 150 mm wafers using a novel method called ‘BiasMDP’. In this method the effective carrier lifetime is determined by means of microwave detected photoconductivity while a bias voltage is applied to an electrode on top of the passivation layer. The measured carrier lifetime strongly reduces when the external bias voltage compensates the electric field of the fixed charges. Based on this minimum lifetime fixed charge and interface defect densities are calculated. The sensitivity of BiasMDP is demonstrated on a silicon wafer with Al2O3 passivation, which is processed with thickness gradient. A continuous shift of flat band voltage is measured across the wafer. This shift linearly correlates with oxide thickness as predicted by theory. Furthermore, an Al2O3 passivation stack with local HfO2 interface layer is characterized. The 2D map of fixed charge density shows a variation between 0.5 x 1012 cm-2 and 4.0 x 1012 cm-2 of negative polarity. This inhomogeneity is attributed to the local presence or not presence of the HfO2 interface layer. Finally, a wafer with local surface damage and inhomogeneous carrier lifetime distribution is measured with BiasMDP. The measurement shows that the inhomogeneity is caused by degradation of chemical passivation. BiasMDP is shown to be a powerful method for detecting inhomogeneities of passivation layers. © 2015 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license © 2015 The Authors. Published by Elsevier Ltd. (http://creativecommons.org/licenses/by-nc-nd/4.0/). Peerreview reviewbybythethe scientific conference committee of SiliconPV 2015responsibility under responsibility of PSE AG. Peer scientific conference committee of SiliconPV 2015 under of PSE AG Keywords: passivation; Al2O3; HfO2; carrier lifetime; fixed charges; interface defects
* Corresponding author. Tel.: +49-351-2124990-38; fax: +49-351-47583-900. E-mail address: [email protected]
1876-6102 © 2015 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/). Peer review by the scientific conference committee of SiliconPV 2015 under responsibility of PSE AG doi:10.1016/j.egypro.2015.07.014
92
Paul M. Jordan et al. / Energy Procedia 77 (2015) 91 – 98
1. Introduction High efficiency solar cells require excellent surface passivation. The best passivation for silicon surfaces is currently achieved with dielectric passivation layers, such as Al2O3, SiO2 or SiNx:H [1]. Dielectric layers provide chemical and field effect passivation. The chemical passivation results from low density of open bonds and interface traps (Dit) at the silicon-dielectric interface and the diffusion of hydrogen towards these open bonds during the post deposition anneal. The low density of recombination centers results in a reduced surface recombination rate. The field effect passivation is caused by a high density of fixed charges (Qf), which are formed within the dielectric layer. These fixed charges change the carrier statistics at the silicon surface and suppress recombination [2,3]. For understanding and improving the passivation performance accurate characterization of Qf and Dit is crucial. The common characterization methods are capacitance-voltage [C(V)] [4,5] and corona charge technique [3,6]. In both methods, the silicon surface potential is modified by an applied electrical field. For C(V)-characterization, the passivation layer is sandwiched in a metal-insulator-semiconductor (MIS) structure with metal dot and silicon substrate. The parameters Qf and Dit are extracted from the voltage dependent capacitance curve as flat band voltage (Vfb) and frequency dependent stretch-out, respectively. In the corona charge measurement, a high electric field is applied to a tungsten tip, which ionizes air particles and molecules in a corona chamber. These ions are deposited on a passivated silicon wafer and change the silicon surface potential. Simultaneously, the effective carrier lifetime (ʏeff) is monitored. When the external charges compensate the field effect passivation, ʏeff drops to a minimum. This minimum is used to determine the Qf and Dit values. The increasing importance of dielectric passivation layers for solar cell manufacturing creates the demand for determining material homogeneity and process signatures with high resolution and high throughput. However, 2D mapping is difficult with state of the art characterization methods. Impedance measurements could be done on an array of local MIS structures, which are realized by metal evaporation with a shadow mask. The sequential measurement of local MIS structures is possible in automatic probe stations and is a standard routine in microelectronics. However, this method is extremely time consuming. The corona charge technique combined with highly spatially resolved lifetime measurements also allows fixed charge measurements; however, a reliable data mapping requires a homogeneous and time stable distribution of corona charges on the substrate. Though this technique is commercially available, reports about successful application on solar cells are difficult to find. Several alternative methods were suggested for fast 2D mapping of dielectrics. In the electrolytical metal oxide semiconductor method [7,8] the wafer is contacted with conductive liquids on both sides. A laser beam scans the front surface and generates charge carriers diffusing to the back side of the wafer. This diffusion current is measured as a function of a voltage applied to the dielectric layer on the front side. Modeling these data reveals the local surface recombination rate and flat band voltage. Haug et al. introduced another method based on photoluminescence (PL) imaging of a MIS structure with applied voltage [9]. This method is extremely fast since the camera system detects large areas. However, the surface recombination is deduced from the PL signal intensity and a careful calibration is required to achieve accurate results [10]. This work is based on the ‘BiasMDP’ method, which was recently published [11]. In this method, the lifetime is detected by spatially resolved microwave detected photoconductivity (MDP) while a bias voltage is applied at a back electrode. The parameters Qf and Dit are mapped with high spatial resolution on 150 mm wafers with different passivation stacks. The purpose of this work is to demonstrate the potential of BiasMDP for detecting inhomogeneities of passivation layers. 2. Experimental procedure All measurements were carried out on 150 mm float zone silicon wafers (2-3 cm) with a thickness of 250 μm. After a megasonic clean, 20 nm thick symmetrical Al2O3 passivation layers were deposited by thermal atomic layer deposition (ALD) at 150 °C. All samples were subsequently annealed in forming gas atmosphere (10 % H2, 90 % N2) at 350 °C [12]. Aluminium contacts were deposited in a thermal evaporator system to prepare MIS structures. The reference values for Qf and Dit were determined in C(V) measurements with an impedance analyzer. For BiasMDP, the external voltage was applied between a full area rear contact and the Si substrate, which was locally contacted at the wafer edge in order to minimize photocarrier diffusion towards the metallization. Three wafers were
93
Paul M. Jordan et al. / Energy Procedia 77 (2015) 91 – 98
prepared with defined inhomogeneity: For sample #A (p-type Si) a continuous gradient in Al2O3 thickness was realized by slowly dipping the sample into HF solution. After the etching step the layer thickness was mapped by spectroscopic ellipsometry. For sample #B (p-type Si) an additional 1 nm HfO2 interface layer was introduced between silicon and Al2O3. This thin HfO2 layer is known to reduce the fixed charge density of the passivation layer [13]. During the HfO2 deposition the substrate was masked by a quarter of a 150 mm wafer. The mask was applied at the wafer side, which was covered by the rear contact in a later process step. Before Al2O3 deposition, the mask was removed and the sample was completely covered by Al2O3. Sample #C (n-type Si) had some intentional and non-intentional surface damages. The 2D defect distribution was measured by a laser surface analyzer. After defect inspection, the wafer was passivated with an HfO2/Al2O3 stack with about 1 nm HfO2, according to Ref [13]. The effective carrier lifetime was measured by MDP using a commercial MDPmap setup from Freiberg Instruments GmbH. The photoconductivity of the samples was continuously detected within a special microwave cavity while the injection of photogenerated carriers was modulated by a pulsed infrared laser of 976 nm wavelength. Between the laser pulses, the lifetime was determined from the transients of the photoconductivity. In addition to the standard MDP setup, a bias voltage was applied between the Si substrate and the rear contact of the sample (Fig. 1a). In the BiasMDP measurement, the lifetime was determined as a function of the bias voltage. Fig. 1b shows the measurement result on a n-type Si sample with HfO2/Al2O3 passivation stack. At negative bias voltage the lifetime was above 4 ms. With increasing bias voltage this value reduced to a minimum lifetime (ʏeff,min) of 2.5 ms at 0.8 V. When the bias voltage was further increased, the lifetime recovered to values up to 3.5 ms. This voltage-dependent lifetime curve can be modelled by the extended Shockley Read Hall formalism developed by Girisch et al. [14]. With this model, Qf, Dit and capture cross sections can be extracted from the BiasMDP data [11]. However, the involved equations are not trivial to solve. For fast data processing of a high quantity of data points a simplified model was applied here. It was assumed that the lifetime drop is caused by the compensation of field effect passivation, i.e. the voltage at ʏeff,min is identical to Vfb. In general, this is only valid for mid-gap interface defects with equal capture cross sections for electrons and holes. However, previous investigations showed that the measured position of ʏeff,min correlated very well to Vfb determined in C(V)measurements [11]. To accurately determine ʏeff,min and Vfb, five measurement points around the minimum lifetime value were fitted with a quadratic function (Fig. 1b). The Qf value was calculated from ܳ ൌ
౮
ൈ ሺ߶୫ୱ െ ܸୠ ሻ
(1)
with ms the work function difference of Al and Si, q the elementary charge and Cox the oxide capacitance [5]. The ms values were determined in C(V) measurements on an Al2O3 thickness series (-0.6 V for p-type and +0.2 V for ntype Si).
Fig. 1. (a) Schematic of a BiasMDP measurement setup, which bases on carrier lifetime measurement with applied voltage bias. (b) BiasMDP measurement on n-type Si with HfO2/Al2O3 passivation layer. The curve was evaluated with a quadratic fit around the lifetime minimum.
94
Paul M. Jordan et al. / Energy Procedia 77 (2015) 91 – 98
The oxide capacitance was not measured for all samples but calculated from the layer thickness according to ܥ୭୶ ൌ
ఢబ ൈఢೝ ௧ೣ
(2)
with ݐ௫ the thickness of the oxide (Al2O3) and ߳ the permittivity of vacuum. The relative permittivity ߳ of Al2O3 (߳ = 8) was determined with C(V) on reference samples. The influence of the HfO2 interface layer was neglected due to its low thickness of 1 nm. The resulting capacitance error was < 3 % considering the higher permittivity of HfO2 (߳ = 17). The Dit values were calculated from ʏeff,min. At flat band condition, the bias voltage compensates the fixed charges and deactivates the field effect passivation. When neglecting bulk recombination, the carrier lifetime is only limited by recombination at interface states. It was assumed that Dit linearly correlated to 1/ʏeff,min ܦ௧ ൌ ܽ୬ǡ୮ ൈ ߬ୣǡ୫୧୬ ିଵ
(3)
with the proportionality factors an,p for n-type and p-type Si, respectively [14,15]. In general, these factors are defect type specific, i.e. the factors depend on energy distribution of the interface defects and their electron and hole capture cross sections. Additionally, ʏeff,min is a function of optical generation during the measurement. For this study, constant factors of an = 8 x 1010 ms eV-1 cm-2 and ap = 5 x 1010 ms eV-1 cm-2 were used for n-type and p-type Si, respectively. These factors refer to Dit at mid-gap, which were determined on reference wafers with HfO2/Al2O3 passivation stack using conductance measurements. This approach implies a constant interface defect type for all measured samples with only the density as variable. The validity of this assumption will be discussed in the result chapter. 3. Results For demonstration of the BiasMDP method, sample #A with wedge-shaped Al2O3 layer was investigated. This Al2O3 layer was mapped with ellipsometry and a continuous thickness gradient from 8 to 41 nm was found (Fig. 2a). Subsequently, a series of lifetime maps was recorded for different bias voltages on the complete wafer. The spatial resolution was 5 mm and steps of bias voltage were 125 mV. The map without applied bias voltage shows a high lifetime above 2 ms over the whole wafer (Fig. 2b). From the series of lifetime maps the minimum values and the Vfb were extracted for each data point. Fig. 2c shows all Vfb data as a function of the Al2O3 thickness. The Vfb values continuously increased from -0.4 V (8 nm) to values above 2.0 V (> 30 nm). This linear correlation is characteristic for dielectrics with fixed charges at the interface and uniform permittivity due to the increased voltage drop at the dielectric with increasing thickness. This conclusion is consistent with several other works on Al2O3 [16,17]. The data points were fitted with Eq. (1) and (2) and the parameters ms = -1.0 V and Qf = 4.7 x 1012 cm-2 with negative polarity were determined.
Fig. 2. (a) Ellipsometry map of Al2O3 thickness and (b) lifetime map at 0 V of sample #A with inhomogeneous Al2O3 thickness. The black rectangle marks the electrical contact to the Si substrate. (c) Extracted Vfb values (black dots) from BiasMDP as a function of the Al2O3 thickness and linear regression (red line) according to Eq. 1 and 2.
Paul M. Jordan et al. / Energy Procedia 77 (2015) 91 – 98
The extracted fixed charge density is typical for Al2O3 layers [1]. However, the work function difference was shifted by -0.4 V compared to the reference value, which was measured with C(V) on a sample with identical passivation stack. This small offset could be caused by process variations of the evaporated aluminium electrode. However, a more likely explanation is charging of the dielectric during application of the bias voltage. Positive bias voltages result in trapping of electrons into the dielectric [18-20]. As a consequence of additional negative charges, Vfb shifted to more positive values. This shift could be observed in BiasMDP hysteresis measurements after electron trapping during a sweep to positive voltages (data not shown). Since this effect increased with the electric field within the dielectric and the stress time, the influence of charge trapping increased with the Al2O3 thickness. This effect could also explain the high Vfb scattering at thicknesses above 30 nm. Furthermore, charge trapping could cause an additional contribution to the slope, which results in a virtual shift of the ms value. To avoid data distortion by charge trapping, different strategies are possible including reduced stress time by applying pulses instead of continuous bias voltages and discharging by reverse voltages pulses between the measurements. Sample #B was passivated partly with pure Al2O3 and partly with an HfO2/Al2O3 stack. Fig. 3 shows three lifetime maps with different bias voltages and with a spatial resolution of 3 mm. At 0 V (Fig. 3b), a rather homogeneous lifetime distribution with values above 3 ms was measured on the wafer. The different passivation stacks could hardly be detected in the lifetime map. Surface passivation with pure HfO2 usually does not reach the performance of Al2O3 [12,21]. However, an 1 nm thick HfO2 interface layer underneath Al2O3 is known to have little impact on the measured lifetime in high injection range [13]. At a negative bias voltage of -1 V the different passivation stacks became visible (Fig. 3a). The lifetime increased by about 1 ms in the center of the wafer. At the same time the outer area degraded below 2 ms. The lifetime map with +1 V bias voltage showed a similar separation in the two areas; however, the lowest lifetime of around 1 ms was detected in the center area (Fig. 3c). These three plots already show very clearly that the applied passivation stacks have different Qf values. For a quantitative study carrier lifetime maps for 40 different bias voltages were recorded and analyzed. The calculated Qf and Dit maps are shown in Fig. 4a and Fig. 4b. The signatures of both maps perfectly match to the mask position during the HfO2 deposition (Fig. 4c). The Qf map shows very low negative Qf below 0.5 x 1012 cm-2 for the HfO2/Al2O3 stack and negative Qf in the range from 3.5 to 4.0 x 1012 cm-2 for pure Al2O3. The high density of negative charges indicated a better field effect passivation of pure Al2O3 on p-type Si. A similar signature could be observed in the Dit map with values of around 4 x 1010 eV-1cm-2 for the HfO2/Al2O3 stack and about twice as much for pure Al2O3. However, this result differed from reference conductance measurements where Dit values of only 6 x 1010 eV-1cm-2 were measured for pure Al2O3. This deviation suggested that the proportionality factor ap overestimated the Dit values in Al2O3, which could be explained by a different energy distribution or capture cross sections of the interface defects in both stacks. The error in Dit values is a result of the simplified data evaluation applied in this study. To eliminate this error source, the lifetime would have to be modeled as a function of the voltage.
Fig. 3. Measured lifetime maps with applied bias voltages of (a) -1 V, (b) 0 V and (c) +1 V on sample #B (black rectangles mark the contacted areas).
95
96
Paul M. Jordan et al. / Energy Procedia 77 (2015) 91 – 98
Fig. 4. Maps of (a) Qf and (b) Dit determined with BiasMDP on sample #B (black rectangles mark the contacted areas). (c) Position of the mask during HfO2 interface layer deposition.
The effect of surface contaminations on the passivation performance was investigated on sample #C. A lifetime map without applied bias voltage is illustrated in Fig. 5a. The carrier lifetime distribution was inhomogeneous from below 1 ms up to 9 ms. A laser defect inspection revealed that the wafer surface was damaged (Fig. 5b). This damage could be partly linked to wafer handling. A chuck imprint was visible in the wafer center. Damage by handling with a vacuum tweezer was found at the wafer bottom side and some lines of defects were also found at various positions. Additionally, the name of the institute was scratched in with a diamond pen. The visual impression was that the surface damage map correlated to the lifetime map. To characterize the influence of mechanical damage on lifetime, the Dit and Qf values were determined with BiasMDP. Fig. 6a shows a correlation of lifetime and Dit. Each data point represents a measurement position on the wafer. The calculated Dit values varied by about one order of magnitude and the lifetime linearly decreased with increasing Dit in the double logarithmic plot. The Qf varied within a small range of ± 0.3 x 1012 cm-2, while ± indicates a change in the polarity. Fig. 6b shows that the correlation of lifetime and Qf was poor. Apparently, the influence of the small Qf variation on the passivation performance was low. The strong relation between lifetime and Dit suggested that mainly the chemical passivation was affected by the damage of the wafer surface.
Fig. 5. (a) Lifetime map without applied bias voltage and (b) map of surface damage on sample #C (black rectangle marks the contacted area).
Paul M. Jordan et al. / Energy Procedia 77 (2015) 91 – 98
Fig. 6. Correlation of lifetime measured with standard MDP to (a) Dit and (b) Qf determined with BiasMDP.
4. Summary 2D mapping of surface passivation layers was presented. The applied BiasMDP method made it possible to distinguish between chemical and field effect passivation with high spatial resolution. For this method, a metal electrode was deposited on the rear side of the wafer. A series of lifetime maps was recorded with different bias voltages applied between the rear contact and the Si substrate. The Qf and Dit values were extracted from minimum carrier lifetime, which occurred when the bias voltage compensated the electric field of the oxide charges. The 2D mapping was demonstrated on three different samples. First, a sample with oxide thickness gradient was mapped with BiasMDP and a linear correlation between the oxide thickness and Vfb was found. An Al2O3 passivated sample with spatially inhomogeneous HfO2 interface was mapped. The HfO2 distribution was not detectable in the standard lifetime map; however, it was clearly visible in the Qf and Dit maps. Thirdly, a wafer with surface damage was analyzed and it was shown that lifetime inhomogeneity mainly originated from Dit variation. These results demonstrated the capability of the BiasMDP for detecting inhomogeneities of passivation layers. Acknowledgements The authors would like to thank Nadine Schüler, Bastian Berger and Kay Dornich (Freiberg Instruments) for experimental support and valuable discussions. Jan Gärtner is acknowledged for performing the HF-dip. This work was funded by BMBF and BMU (MWT-plus, no. 03SF0420A). References [1] Dingemans G, Kessels WMM, Status and prospects of Al2O3-based surface passivation schemes for silicon solar cells, J Vac Sci Technol A 30(4); p. 040802, 2012 [2] Aberle A G, Glunz SW, Warta W, Impact of illumination level and oxide parameters on Shockley–Read–Hall recombination at the Si-SiO2 interface, J Appl Phys71(9); p. 4422, 1992. [3] Glunz SW, Biro D, Rein S, Warta W, Field-effect passivation of the SiO2-Si interface, J Appl Phys 86(1); p. 683, 1999. [4] Nicollian E H, Goetzberger A, MOS Conductance Technique for Measuring Surface State Parameters, Appl Phys Lett 7(8); p. 216, 1965. [5] Sze S M, Ng KK, Physics of semiconductor devices Hoboken, N.J.: Wiley-Interscience, 2007. [6] Hoex B, Schmidt J, Pohl P, van de Sanden MCM, Kessels WMM, Silicon surface passivation by atomic layer deposited Al2O3, J Appl Phys 104 (4); p. 044903, 2008. [7] Rommel M, Bauer AJ, Ryssel H, Quantitative oxide charge determination by photocurrent analysis, Microelectron Reliab 47(4–5); p.673, 2007 [8] Polignano ML, Surface Recombination Velocity from Photocurrent Measurements: Validation and Applications, J Electrochem Soc 146(12); p. 4640, 1999. [9] Haug H, Nordseth Ø, Monakhov ES, Marstein ES, Photoluminescence imaging under applied bias for characterization of Si surface passivation layers, Sol Energy Mater Sol Cells, 106; p. 60, 2012. [10] Haug H, Olibet S, Nordseth Ø, Marstein ES, Modulating the field-effect passivation at the SiO2/c-Si interface: Analysis and verification of
97
98
Paul M. Jordan et al. / Energy Procedia 77 (2015) 91 – 98
the photoluminescence imaging under applied bias method, J Appl Phys 114(17); p. 174502, 2013. [11] Jordan PM, Simon DK, Mikolajick T, Dirnstorfer I, BiasMDP: Carrier lifetime characterization technique with applied bias voltage, Appl Phys Lett 106(6); p. 061602, 2015. [12] Benner F, Jordan PM, Richter C, Simon DK, Dirnstorfer I, Knaut M, Bartha JW, Mikolajick T, Atomic layer deposited high-ț nanolaminates for silicon surface passivation, J Vac Sci Technol B 32(3); p. 03D110, 2014. [13] Simon DK, Jordan PM, Dirnstorfer I, Benner F, Richter C, Mikolajick T, Symmetrical Al2O3-based passivation layers for p- and n-type silicon, Sol Energy Mater Sol Cells 131; p. 72, 2014. [14] Girisch RBM, Mertens RP, De Keersmaecker RF, Determination of Si-SiO2 interface recombination parameters using a gate-controlled point-junction diode under illumination, IEEE Trans Electron Devices 35(2); p. 203, 1988. [15] Polignano ML, Caricato AP, Simultaneous characterization of bulk impurities and interface states by photocurrent measurements, Appl Surf Sci 235(3); p. 340, 2004. [16] Aboaf JA, Kerr D R, Bassous E, Charge in SiO2-Al2O3 Double Layers on Silicon, J Electrochem Soc 120(8); p. 1103, 1973. [17] Johnson RS, Lucovsky G, Baumvol I, Physical and electrical properties of noncrystalline Al2O3 prepared by remote plasma enhanced chemical vapor deposition, J Vac Sci Technol A 19(4); p. 1353, 2001. [18] Benner F, Jordan PM, Knaut M, Dirnstorfer I, Bartha JW, Mikolajick T, Investigation of the c-Si/Al2O3 Interface for Silicon Surface Passivation, Proc 27th PVSEC; p. 1793, 2012. [19] Werner F, Schmidt J, Manipulating the negative fixed charge density at the c-Si/Al2O3 interface, Appl Phys Lett 104(9); p. 091604, 2014. [20] Gielis JJH, Hoex B, van de Sanden MCM, Kessels WMM, Negative charge and charging dynamics in Al2O3 films on Si characterized by second-harmonic generation, J Appl Phys 104(7); p. 073701, 2008. [21] Dingemans G, Kessels WMM, Aluminum Oxide and Other ALD Materials for Si Surface Passivation, ECS Trans. 41(2); p. 293, 2011.
ScienceDirect Energy Procedia 77 (2015) 91 – 98
5th International Conference on Silicon Photovoltaics, SiliconPV 2015
2D mapping of chemical and field effect passivation of Al2O3 on silicon substrates Paul M. Jordana,*, Daniel K. Simona, Franz P.G. Fenglera, Thomas Mikolajicka,b, Ingo Dirnstorfera b
a NaMLab gGmbH, Nöthnitzerstr. 64, 01187 Dresden, Germany Technische Universität Dresden, Institut für Halbleiter- und Mikrosystemtechnik, 01062 Dresden, Germany
Abstract Fixed charge and interface defect densities are the critical material parameters for silicon surface passivation. These parameters are measured with high spatial resolution on 150 mm wafers using a novel method called ‘BiasMDP’. In this method the effective carrier lifetime is determined by means of microwave detected photoconductivity while a bias voltage is applied to an electrode on top of the passivation layer. The measured carrier lifetime strongly reduces when the external bias voltage compensates the electric field of the fixed charges. Based on this minimum lifetime fixed charge and interface defect densities are calculated. The sensitivity of BiasMDP is demonstrated on a silicon wafer with Al2O3 passivation, which is processed with thickness gradient. A continuous shift of flat band voltage is measured across the wafer. This shift linearly correlates with oxide thickness as predicted by theory. Furthermore, an Al2O3 passivation stack with local HfO2 interface layer is characterized. The 2D map of fixed charge density shows a variation between 0.5 x 1012 cm-2 and 4.0 x 1012 cm-2 of negative polarity. This inhomogeneity is attributed to the local presence or not presence of the HfO2 interface layer. Finally, a wafer with local surface damage and inhomogeneous carrier lifetime distribution is measured with BiasMDP. The measurement shows that the inhomogeneity is caused by degradation of chemical passivation. BiasMDP is shown to be a powerful method for detecting inhomogeneities of passivation layers. © 2015 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license © 2015 The Authors. Published by Elsevier Ltd. (http://creativecommons.org/licenses/by-nc-nd/4.0/). Peerreview reviewbybythethe scientific conference committee of SiliconPV 2015responsibility under responsibility of PSE AG. Peer scientific conference committee of SiliconPV 2015 under of PSE AG Keywords: passivation; Al2O3; HfO2; carrier lifetime; fixed charges; interface defects
* Corresponding author. Tel.: +49-351-2124990-38; fax: +49-351-47583-900. E-mail address: [email protected]
1876-6102 © 2015 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/). Peer review by the scientific conference committee of SiliconPV 2015 under responsibility of PSE AG doi:10.1016/j.egypro.2015.07.014
92
Paul M. Jordan et al. / Energy Procedia 77 (2015) 91 – 98
1. Introduction High efficiency solar cells require excellent surface passivation. The best passivation for silicon surfaces is currently achieved with dielectric passivation layers, such as Al2O3, SiO2 or SiNx:H [1]. Dielectric layers provide chemical and field effect passivation. The chemical passivation results from low density of open bonds and interface traps (Dit) at the silicon-dielectric interface and the diffusion of hydrogen towards these open bonds during the post deposition anneal. The low density of recombination centers results in a reduced surface recombination rate. The field effect passivation is caused by a high density of fixed charges (Qf), which are formed within the dielectric layer. These fixed charges change the carrier statistics at the silicon surface and suppress recombination [2,3]. For understanding and improving the passivation performance accurate characterization of Qf and Dit is crucial. The common characterization methods are capacitance-voltage [C(V)] [4,5] and corona charge technique [3,6]. In both methods, the silicon surface potential is modified by an applied electrical field. For C(V)-characterization, the passivation layer is sandwiched in a metal-insulator-semiconductor (MIS) structure with metal dot and silicon substrate. The parameters Qf and Dit are extracted from the voltage dependent capacitance curve as flat band voltage (Vfb) and frequency dependent stretch-out, respectively. In the corona charge measurement, a high electric field is applied to a tungsten tip, which ionizes air particles and molecules in a corona chamber. These ions are deposited on a passivated silicon wafer and change the silicon surface potential. Simultaneously, the effective carrier lifetime (ʏeff) is monitored. When the external charges compensate the field effect passivation, ʏeff drops to a minimum. This minimum is used to determine the Qf and Dit values. The increasing importance of dielectric passivation layers for solar cell manufacturing creates the demand for determining material homogeneity and process signatures with high resolution and high throughput. However, 2D mapping is difficult with state of the art characterization methods. Impedance measurements could be done on an array of local MIS structures, which are realized by metal evaporation with a shadow mask. The sequential measurement of local MIS structures is possible in automatic probe stations and is a standard routine in microelectronics. However, this method is extremely time consuming. The corona charge technique combined with highly spatially resolved lifetime measurements also allows fixed charge measurements; however, a reliable data mapping requires a homogeneous and time stable distribution of corona charges on the substrate. Though this technique is commercially available, reports about successful application on solar cells are difficult to find. Several alternative methods were suggested for fast 2D mapping of dielectrics. In the electrolytical metal oxide semiconductor method [7,8] the wafer is contacted with conductive liquids on both sides. A laser beam scans the front surface and generates charge carriers diffusing to the back side of the wafer. This diffusion current is measured as a function of a voltage applied to the dielectric layer on the front side. Modeling these data reveals the local surface recombination rate and flat band voltage. Haug et al. introduced another method based on photoluminescence (PL) imaging of a MIS structure with applied voltage [9]. This method is extremely fast since the camera system detects large areas. However, the surface recombination is deduced from the PL signal intensity and a careful calibration is required to achieve accurate results [10]. This work is based on the ‘BiasMDP’ method, which was recently published [11]. In this method, the lifetime is detected by spatially resolved microwave detected photoconductivity (MDP) while a bias voltage is applied at a back electrode. The parameters Qf and Dit are mapped with high spatial resolution on 150 mm wafers with different passivation stacks. The purpose of this work is to demonstrate the potential of BiasMDP for detecting inhomogeneities of passivation layers. 2. Experimental procedure All measurements were carried out on 150 mm float zone silicon wafers (2-3 cm) with a thickness of 250 μm. After a megasonic clean, 20 nm thick symmetrical Al2O3 passivation layers were deposited by thermal atomic layer deposition (ALD) at 150 °C. All samples were subsequently annealed in forming gas atmosphere (10 % H2, 90 % N2) at 350 °C [12]. Aluminium contacts were deposited in a thermal evaporator system to prepare MIS structures. The reference values for Qf and Dit were determined in C(V) measurements with an impedance analyzer. For BiasMDP, the external voltage was applied between a full area rear contact and the Si substrate, which was locally contacted at the wafer edge in order to minimize photocarrier diffusion towards the metallization. Three wafers were
93
Paul M. Jordan et al. / Energy Procedia 77 (2015) 91 – 98
prepared with defined inhomogeneity: For sample #A (p-type Si) a continuous gradient in Al2O3 thickness was realized by slowly dipping the sample into HF solution. After the etching step the layer thickness was mapped by spectroscopic ellipsometry. For sample #B (p-type Si) an additional 1 nm HfO2 interface layer was introduced between silicon and Al2O3. This thin HfO2 layer is known to reduce the fixed charge density of the passivation layer [13]. During the HfO2 deposition the substrate was masked by a quarter of a 150 mm wafer. The mask was applied at the wafer side, which was covered by the rear contact in a later process step. Before Al2O3 deposition, the mask was removed and the sample was completely covered by Al2O3. Sample #C (n-type Si) had some intentional and non-intentional surface damages. The 2D defect distribution was measured by a laser surface analyzer. After defect inspection, the wafer was passivated with an HfO2/Al2O3 stack with about 1 nm HfO2, according to Ref [13]. The effective carrier lifetime was measured by MDP using a commercial MDPmap setup from Freiberg Instruments GmbH. The photoconductivity of the samples was continuously detected within a special microwave cavity while the injection of photogenerated carriers was modulated by a pulsed infrared laser of 976 nm wavelength. Between the laser pulses, the lifetime was determined from the transients of the photoconductivity. In addition to the standard MDP setup, a bias voltage was applied between the Si substrate and the rear contact of the sample (Fig. 1a). In the BiasMDP measurement, the lifetime was determined as a function of the bias voltage. Fig. 1b shows the measurement result on a n-type Si sample with HfO2/Al2O3 passivation stack. At negative bias voltage the lifetime was above 4 ms. With increasing bias voltage this value reduced to a minimum lifetime (ʏeff,min) of 2.5 ms at 0.8 V. When the bias voltage was further increased, the lifetime recovered to values up to 3.5 ms. This voltage-dependent lifetime curve can be modelled by the extended Shockley Read Hall formalism developed by Girisch et al. [14]. With this model, Qf, Dit and capture cross sections can be extracted from the BiasMDP data [11]. However, the involved equations are not trivial to solve. For fast data processing of a high quantity of data points a simplified model was applied here. It was assumed that the lifetime drop is caused by the compensation of field effect passivation, i.e. the voltage at ʏeff,min is identical to Vfb. In general, this is only valid for mid-gap interface defects with equal capture cross sections for electrons and holes. However, previous investigations showed that the measured position of ʏeff,min correlated very well to Vfb determined in C(V)measurements [11]. To accurately determine ʏeff,min and Vfb, five measurement points around the minimum lifetime value were fitted with a quadratic function (Fig. 1b). The Qf value was calculated from ܳ ൌ
౮
ൈ ሺ߶୫ୱ െ ܸୠ ሻ
(1)
with ms the work function difference of Al and Si, q the elementary charge and Cox the oxide capacitance [5]. The ms values were determined in C(V) measurements on an Al2O3 thickness series (-0.6 V for p-type and +0.2 V for ntype Si).
Fig. 1. (a) Schematic of a BiasMDP measurement setup, which bases on carrier lifetime measurement with applied voltage bias. (b) BiasMDP measurement on n-type Si with HfO2/Al2O3 passivation layer. The curve was evaluated with a quadratic fit around the lifetime minimum.
94
Paul M. Jordan et al. / Energy Procedia 77 (2015) 91 – 98
The oxide capacitance was not measured for all samples but calculated from the layer thickness according to ܥ୭୶ ൌ
ఢబ ൈఢೝ ௧ೣ
(2)
with ݐ௫ the thickness of the oxide (Al2O3) and ߳ the permittivity of vacuum. The relative permittivity ߳ of Al2O3 (߳ = 8) was determined with C(V) on reference samples. The influence of the HfO2 interface layer was neglected due to its low thickness of 1 nm. The resulting capacitance error was < 3 % considering the higher permittivity of HfO2 (߳ = 17). The Dit values were calculated from ʏeff,min. At flat band condition, the bias voltage compensates the fixed charges and deactivates the field effect passivation. When neglecting bulk recombination, the carrier lifetime is only limited by recombination at interface states. It was assumed that Dit linearly correlated to 1/ʏeff,min ܦ௧ ൌ ܽ୬ǡ୮ ൈ ߬ୣǡ୫୧୬ ିଵ
(3)
with the proportionality factors an,p for n-type and p-type Si, respectively [14,15]. In general, these factors are defect type specific, i.e. the factors depend on energy distribution of the interface defects and their electron and hole capture cross sections. Additionally, ʏeff,min is a function of optical generation during the measurement. For this study, constant factors of an = 8 x 1010 ms eV-1 cm-2 and ap = 5 x 1010 ms eV-1 cm-2 were used for n-type and p-type Si, respectively. These factors refer to Dit at mid-gap, which were determined on reference wafers with HfO2/Al2O3 passivation stack using conductance measurements. This approach implies a constant interface defect type for all measured samples with only the density as variable. The validity of this assumption will be discussed in the result chapter. 3. Results For demonstration of the BiasMDP method, sample #A with wedge-shaped Al2O3 layer was investigated. This Al2O3 layer was mapped with ellipsometry and a continuous thickness gradient from 8 to 41 nm was found (Fig. 2a). Subsequently, a series of lifetime maps was recorded for different bias voltages on the complete wafer. The spatial resolution was 5 mm and steps of bias voltage were 125 mV. The map without applied bias voltage shows a high lifetime above 2 ms over the whole wafer (Fig. 2b). From the series of lifetime maps the minimum values and the Vfb were extracted for each data point. Fig. 2c shows all Vfb data as a function of the Al2O3 thickness. The Vfb values continuously increased from -0.4 V (8 nm) to values above 2.0 V (> 30 nm). This linear correlation is characteristic for dielectrics with fixed charges at the interface and uniform permittivity due to the increased voltage drop at the dielectric with increasing thickness. This conclusion is consistent with several other works on Al2O3 [16,17]. The data points were fitted with Eq. (1) and (2) and the parameters ms = -1.0 V and Qf = 4.7 x 1012 cm-2 with negative polarity were determined.
Fig. 2. (a) Ellipsometry map of Al2O3 thickness and (b) lifetime map at 0 V of sample #A with inhomogeneous Al2O3 thickness. The black rectangle marks the electrical contact to the Si substrate. (c) Extracted Vfb values (black dots) from BiasMDP as a function of the Al2O3 thickness and linear regression (red line) according to Eq. 1 and 2.
Paul M. Jordan et al. / Energy Procedia 77 (2015) 91 – 98
The extracted fixed charge density is typical for Al2O3 layers [1]. However, the work function difference was shifted by -0.4 V compared to the reference value, which was measured with C(V) on a sample with identical passivation stack. This small offset could be caused by process variations of the evaporated aluminium electrode. However, a more likely explanation is charging of the dielectric during application of the bias voltage. Positive bias voltages result in trapping of electrons into the dielectric [18-20]. As a consequence of additional negative charges, Vfb shifted to more positive values. This shift could be observed in BiasMDP hysteresis measurements after electron trapping during a sweep to positive voltages (data not shown). Since this effect increased with the electric field within the dielectric and the stress time, the influence of charge trapping increased with the Al2O3 thickness. This effect could also explain the high Vfb scattering at thicknesses above 30 nm. Furthermore, charge trapping could cause an additional contribution to the slope, which results in a virtual shift of the ms value. To avoid data distortion by charge trapping, different strategies are possible including reduced stress time by applying pulses instead of continuous bias voltages and discharging by reverse voltages pulses between the measurements. Sample #B was passivated partly with pure Al2O3 and partly with an HfO2/Al2O3 stack. Fig. 3 shows three lifetime maps with different bias voltages and with a spatial resolution of 3 mm. At 0 V (Fig. 3b), a rather homogeneous lifetime distribution with values above 3 ms was measured on the wafer. The different passivation stacks could hardly be detected in the lifetime map. Surface passivation with pure HfO2 usually does not reach the performance of Al2O3 [12,21]. However, an 1 nm thick HfO2 interface layer underneath Al2O3 is known to have little impact on the measured lifetime in high injection range [13]. At a negative bias voltage of -1 V the different passivation stacks became visible (Fig. 3a). The lifetime increased by about 1 ms in the center of the wafer. At the same time the outer area degraded below 2 ms. The lifetime map with +1 V bias voltage showed a similar separation in the two areas; however, the lowest lifetime of around 1 ms was detected in the center area (Fig. 3c). These three plots already show very clearly that the applied passivation stacks have different Qf values. For a quantitative study carrier lifetime maps for 40 different bias voltages were recorded and analyzed. The calculated Qf and Dit maps are shown in Fig. 4a and Fig. 4b. The signatures of both maps perfectly match to the mask position during the HfO2 deposition (Fig. 4c). The Qf map shows very low negative Qf below 0.5 x 1012 cm-2 for the HfO2/Al2O3 stack and negative Qf in the range from 3.5 to 4.0 x 1012 cm-2 for pure Al2O3. The high density of negative charges indicated a better field effect passivation of pure Al2O3 on p-type Si. A similar signature could be observed in the Dit map with values of around 4 x 1010 eV-1cm-2 for the HfO2/Al2O3 stack and about twice as much for pure Al2O3. However, this result differed from reference conductance measurements where Dit values of only 6 x 1010 eV-1cm-2 were measured for pure Al2O3. This deviation suggested that the proportionality factor ap overestimated the Dit values in Al2O3, which could be explained by a different energy distribution or capture cross sections of the interface defects in both stacks. The error in Dit values is a result of the simplified data evaluation applied in this study. To eliminate this error source, the lifetime would have to be modeled as a function of the voltage.
Fig. 3. Measured lifetime maps with applied bias voltages of (a) -1 V, (b) 0 V and (c) +1 V on sample #B (black rectangles mark the contacted areas).
95
96
Paul M. Jordan et al. / Energy Procedia 77 (2015) 91 – 98
Fig. 4. Maps of (a) Qf and (b) Dit determined with BiasMDP on sample #B (black rectangles mark the contacted areas). (c) Position of the mask during HfO2 interface layer deposition.
The effect of surface contaminations on the passivation performance was investigated on sample #C. A lifetime map without applied bias voltage is illustrated in Fig. 5a. The carrier lifetime distribution was inhomogeneous from below 1 ms up to 9 ms. A laser defect inspection revealed that the wafer surface was damaged (Fig. 5b). This damage could be partly linked to wafer handling. A chuck imprint was visible in the wafer center. Damage by handling with a vacuum tweezer was found at the wafer bottom side and some lines of defects were also found at various positions. Additionally, the name of the institute was scratched in with a diamond pen. The visual impression was that the surface damage map correlated to the lifetime map. To characterize the influence of mechanical damage on lifetime, the Dit and Qf values were determined with BiasMDP. Fig. 6a shows a correlation of lifetime and Dit. Each data point represents a measurement position on the wafer. The calculated Dit values varied by about one order of magnitude and the lifetime linearly decreased with increasing Dit in the double logarithmic plot. The Qf varied within a small range of ± 0.3 x 1012 cm-2, while ± indicates a change in the polarity. Fig. 6b shows that the correlation of lifetime and Qf was poor. Apparently, the influence of the small Qf variation on the passivation performance was low. The strong relation between lifetime and Dit suggested that mainly the chemical passivation was affected by the damage of the wafer surface.
Fig. 5. (a) Lifetime map without applied bias voltage and (b) map of surface damage on sample #C (black rectangle marks the contacted area).
Paul M. Jordan et al. / Energy Procedia 77 (2015) 91 – 98
Fig. 6. Correlation of lifetime measured with standard MDP to (a) Dit and (b) Qf determined with BiasMDP.
4. Summary 2D mapping of surface passivation layers was presented. The applied BiasMDP method made it possible to distinguish between chemical and field effect passivation with high spatial resolution. For this method, a metal electrode was deposited on the rear side of the wafer. A series of lifetime maps was recorded with different bias voltages applied between the rear contact and the Si substrate. The Qf and Dit values were extracted from minimum carrier lifetime, which occurred when the bias voltage compensated the electric field of the oxide charges. The 2D mapping was demonstrated on three different samples. First, a sample with oxide thickness gradient was mapped with BiasMDP and a linear correlation between the oxide thickness and Vfb was found. An Al2O3 passivated sample with spatially inhomogeneous HfO2 interface was mapped. The HfO2 distribution was not detectable in the standard lifetime map; however, it was clearly visible in the Qf and Dit maps. Thirdly, a wafer with surface damage was analyzed and it was shown that lifetime inhomogeneity mainly originated from Dit variation. These results demonstrated the capability of the BiasMDP for detecting inhomogeneities of passivation layers. Acknowledgements The authors would like to thank Nadine Schüler, Bastian Berger and Kay Dornich (Freiberg Instruments) for experimental support and valuable discussions. Jan Gärtner is acknowledged for performing the HF-dip. This work was funded by BMBF and BMU (MWT-plus, no. 03SF0420A). References [1] Dingemans G, Kessels WMM, Status and prospects of Al2O3-based surface passivation schemes for silicon solar cells, J Vac Sci Technol A 30(4); p. 040802, 2012 [2] Aberle A G, Glunz SW, Warta W, Impact of illumination level and oxide parameters on Shockley–Read–Hall recombination at the Si-SiO2 interface, J Appl Phys71(9); p. 4422, 1992. [3] Glunz SW, Biro D, Rein S, Warta W, Field-effect passivation of the SiO2-Si interface, J Appl Phys 86(1); p. 683, 1999. [4] Nicollian E H, Goetzberger A, MOS Conductance Technique for Measuring Surface State Parameters, Appl Phys Lett 7(8); p. 216, 1965. [5] Sze S M, Ng KK, Physics of semiconductor devices Hoboken, N.J.: Wiley-Interscience, 2007. [6] Hoex B, Schmidt J, Pohl P, van de Sanden MCM, Kessels WMM, Silicon surface passivation by atomic layer deposited Al2O3, J Appl Phys 104 (4); p. 044903, 2008. [7] Rommel M, Bauer AJ, Ryssel H, Quantitative oxide charge determination by photocurrent analysis, Microelectron Reliab 47(4–5); p.673, 2007 [8] Polignano ML, Surface Recombination Velocity from Photocurrent Measurements: Validation and Applications, J Electrochem Soc 146(12); p. 4640, 1999. [9] Haug H, Nordseth Ø, Monakhov ES, Marstein ES, Photoluminescence imaging under applied bias for characterization of Si surface passivation layers, Sol Energy Mater Sol Cells, 106; p. 60, 2012. [10] Haug H, Olibet S, Nordseth Ø, Marstein ES, Modulating the field-effect passivation at the SiO2/c-Si interface: Analysis and verification of
97
98
Paul M. Jordan et al. / Energy Procedia 77 (2015) 91 – 98
the photoluminescence imaging under applied bias method, J Appl Phys 114(17); p. 174502, 2013. [11] Jordan PM, Simon DK, Mikolajick T, Dirnstorfer I, BiasMDP: Carrier lifetime characterization technique with applied bias voltage, Appl Phys Lett 106(6); p. 061602, 2015. [12] Benner F, Jordan PM, Richter C, Simon DK, Dirnstorfer I, Knaut M, Bartha JW, Mikolajick T, Atomic layer deposited high-ț nanolaminates for silicon surface passivation, J Vac Sci Technol B 32(3); p. 03D110, 2014. [13] Simon DK, Jordan PM, Dirnstorfer I, Benner F, Richter C, Mikolajick T, Symmetrical Al2O3-based passivation layers for p- and n-type silicon, Sol Energy Mater Sol Cells 131; p. 72, 2014. [14] Girisch RBM, Mertens RP, De Keersmaecker RF, Determination of Si-SiO2 interface recombination parameters using a gate-controlled point-junction diode under illumination, IEEE Trans Electron Devices 35(2); p. 203, 1988. [15] Polignano ML, Caricato AP, Simultaneous characterization of bulk impurities and interface states by photocurrent measurements, Appl Surf Sci 235(3); p. 340, 2004. [16] Aboaf JA, Kerr D R, Bassous E, Charge in SiO2-Al2O3 Double Layers on Silicon, J Electrochem Soc 120(8); p. 1103, 1973. [17] Johnson RS, Lucovsky G, Baumvol I, Physical and electrical properties of noncrystalline Al2O3 prepared by remote plasma enhanced chemical vapor deposition, J Vac Sci Technol A 19(4); p. 1353, 2001. [18] Benner F, Jordan PM, Knaut M, Dirnstorfer I, Bartha JW, Mikolajick T, Investigation of the c-Si/Al2O3 Interface for Silicon Surface Passivation, Proc 27th PVSEC; p. 1793, 2012. [19] Werner F, Schmidt J, Manipulating the negative fixed charge density at the c-Si/Al2O3 interface, Appl Phys Lett 104(9); p. 091604, 2014. [20] Gielis JJH, Hoex B, van de Sanden MCM, Kessels WMM, Negative charge and charging dynamics in Al2O3 films on Si characterized by second-harmonic generation, J Appl Phys 104(7); p. 073701, 2008. [21] Dingemans G, Kessels WMM, Aluminum Oxide and Other ALD Materials for Si Surface Passivation, ECS Trans. 41(2); p. 293, 2011.