Simulation of an optimised emitter

Simulation model – The PC1D model used for the simulations of epitaxial wafer-equivalents with POCl₃ diffused emitter (Section 4.1.1) was extended for simulations with epitaxial emitters. Figure 5-21 shows the doping profile used for the simulations. The arrows indicate the parameters which were varied. The model includes the two-layered emitter with high surface doping. The diffusion peak at the surface was taken from SIMS measurements of a sample that was cooled with 119 ppm PH3 (Figure 5-17). Additionally, the phosphorus and boron diffusions into the opposed doped region were included. For the short deposition times of maximum 10 minutes the diffusion constants do not differ significantly.

The pn-junction is shifted from the growth interface towards the base with increasing emitter thickness. The deviation of the sheet resistance as dependent on the base-emitter diffusion increases to 16% for thinner emitters [64]. The resulting deviation of the total series resistance was found to be less than 10%.

This value is still tolerable and the extensive calculation of the diffusion process was omitted. Furthermore, the dependence of the series resistance RS on the sheet resistance RSheet was incorporated in the simulations. The external quantum efficiency was adapted to a measured one sample (HC758, Figure 5-19) and a front surface recombination velocity of Sfront = 200 cm/s was assumed. In the first approach, the interface recombinations were not considered. The base doping, as well as the bulk emitter lifetime, thickness and doping were then varied.

0.5 1.0 15 20 25 10¹⁶

10¹⁷ 10¹⁸ 10¹⁹ 10²⁰

Carrier density [cm-3 ]

Depth [µm]

n-type p-type

Emitter Emitter-Base Diffusion

Base

BSF Substrate

Figure 5-21: Doping profile used for the PC1D model.

Simulation results – Figure 5-22 and Figure 5-23 show the simulated cell parameters in dependence of the emitter thickness for emitter doping varying from 1x10¹⁸ to 9x10¹⁸ cm^-3. For simplicity, only the base thickness of 20 µm and the base doping concentration of 2x10¹⁶ cm^-3 are shown. Similar results are found for the other doping levels. For emitter thicknesses below 1 µm, the short circuit current density J_SC and open circuit voltage V_OC are mainly determined by the base doping. A higher base doping results in lower short circuit current densities, due to the increased Auger recombination, whereas the open circuit voltages are increased.

J_SC – For thicker emitters, the emitter doping is more critical, as proportionally more light is absorbed in the emitter with increasing emitter thicknesses. Due to Auger recombination, the diffusion length is decreased for increasing doping levels and the electron-hole-pairs recombine before they can reach the pn-junction. Additionally, less light is absorbed within the base, in which the minority carrier lifetime is much higher. A dramatic decrease of the short circuit current density is therefore noticeable for emitter thicknesses above 4 µm.

V_OC – The explanation for the open circuit voltage is similar, as it is dominated by the short circuit current density (Appendix A.1). The open circuit voltage tends to decrease for thicker emitters, especially for higher emitter doping concentrations. For thin emitters, the influence of the emitter doping can be neglected.

FF – The fill factor increases for thicker emitters (see also experimental results in Table 5-4) and increasing emitter doping. The main reason is the reduction of the series resistances, due to lower emitter sheet resistances and better lateral conductivity in the emitter. The series resistance essentially dominates the fill factor in these simulations. Other factors, such as traps in the space charge region or shunts, which influence the fill factor, are not incorporated in the simulations. For this reason, no recombinations at grain boundaries have been investigated.

The efficiency is defined by the trends of the curves described above. An optimum efficiency is shown for all doping concentrations around 0.8 to 4 µm emitter thickness. Noticeably, it is mainly the emitter doping that influences the shape of the curves. For very thin emitters the efficiency is limited by the low fill factor, whereas the short circuit current density JSC and the open circuit voltage VOC are responsible for the decrease with thick emitters. A higher emitter doping concentration shifts the optimum emitter thickness toward a lower value.

0.2 0.4 0.6 0.8 1 2 4 6 8 10 20

15 20 25 30 630 640 650 660

Emitter doping 9x10¹⁸ 7x10¹⁸ 5x10¹⁸ 3x10¹⁸ 1x10¹⁸

Emitter thickness [µm]

J SC [mA/cm2 ]V OC [mV]

Figure 5-22: Simulated open circuit voltage VOC and short circuit current density JSC

depending on the emitter thickness and doping concentration for a base doping level of 2x10¹⁶ cm^-3.

0.2 0.4 0.6 0.8 1 2 4 6 8 10 20 8

10 12 14 16 76 78 80 82

0.6 0.8 1 2 4 6 8

15.0 15.2 15.4 15.6

η [%]

Emitter thickness [µm]

Emitter doping 9x10¹⁸ 7x10¹⁸ 5x10¹⁸ 3x10¹⁸ 1x10¹⁸

FF [%]

Figure 5-23: Simulated FF and η depending on the emitter thickness and doping concentration for a base doping level of 2x10¹⁶ cm^-3.

Process stability – As a large thickness deviation is present in the RTCVD reactors (Section 3.3.3), the simulations were also analysed for efficiency losses due to inhomogeneous emitters. In our cells, a thickness deviation of 40% is noticeable and the real emitter thickness across a wafer varies from 0.8 to 1.2 µm for a set emitter thickness of 1 µm. Fortunately, the simulations show that only slight variation (0.1% absolute) of the efficiency occurs for emitter thicknesses in this range. Apparently, such a large deviation in thickness homogeneity is totally acceptable.

Comparison to solar cells – The structural parameters of the simulation model were adapted to the properties of sample HC758. Table 5-7 compares the simulation to the measured solar cell parameters of this sample and of a cell with improved doping levels (HC776c). The simulated open circuit voltage VOC is about 3-4 mV higher than the measured value, which is within the measurement accuracy of the open circuit voltage. The actual short circuit current densities are about 0.4 mA/cm² lower than the simulated ones. The cause of this may be a lower lifetime of the epitaxial layer than the 25 µs that was assumed for the simulations. Recombination at stacking faults and other defects reduce the effective diffusion length and the carrier collection and are not taken into account in the simulations. The fill factor of HC776c is even higher than the

predicted one, which may arise from the different grid design of the smaller cell HC776c and from a lower deviation of the sheet resistance R_Sheet. The measured series resistance corresponds nicely with a deviation of 6% from the calculated value. In total, the measured values are in good agreement with the simulated ones, with deviation less than 2% relative. This model is suited for efficiency prediction and cell analysis of the epitaxial wafer-equivalent solar cells with epitaxial emitter.

Table 5-7: Comparison of simulated and measured solar cell parameters of epitaxial wafer-equivalent solar cells grown on Cz with epitaxial emitters. Both samples were calibrated at

CalLab ISE.

n° HC758 Simulation HC776c Simulation

Area [cm²] 92.1 4.2

[PH3]cooling [ppm] 119 119

demitter [µm] 1 1 0.6 0.6

RSheet [Ω/sq.] 85 85 110 110

NA [cm^-3] 5x10¹⁸ 5x10¹⁸ 6x10¹⁸ 5x10¹⁸

dbase [µm] 19.7 20 16.9 20

ND [cm^-3] 2x10¹⁶ 2x10¹⁶ 2x10¹⁶ 2x10¹⁶

VOC [mV] 655 658 651 655

JSC [mA/cm²] 28.4 28.9 28.9 29.3

FF [%] 79.9 79.7 80.6 79.2

η [%] 14.9 15.2 15.2 15.2

J01 [A/cm²] 2.1x10^-13 2.3x10^-13 J02 [A/cm²] 3.0x10^-7 9.6x10^-9

RS [Ωcm²] 0.5 0.3

RSh [Ωcm²] 3x10⁴ 5x10⁶ 5.4.4 Recombination in epitaxial emitters

Simulation – The two main recombination processes detrimental for epitaxial emitters are the recombination at the growth interface and the recombination in the emitter bulk. Simulations are an important tool to study the impact of these effects on the cell performance.

The growth interface is a critical surface since the generated carriers flow through this region. The recombination velocity at the interface was varied from 1 to 10⁶ cm/s, reflecting a wide range from barely any recombination to nearly the thermal limit of recombination. The second recombination path was studied by varying the bulk lifetime of the emitter. The Shockley-Read-Hall lifetime of the emitter was set to 1, 10 and 100 ns. Above 100 ns the lifetime is mainly

limited by Auger recombination. A front surface recombination velocity S_front of 1000 cm/s was assumed, the lowest for this surface doping concentration [118, 119]. Figure 5-24 shows the dependence of the open circuit voltage on the interface recombination velocity for different lifetimes. For velocities above 1000 cm/s the open circuit voltage is strongly reduced in all cases. When the emitter bulk lifetime is lowered from 10 to 1 ns a severe decrease is noticeable.

A similar behaviour is found for the short circuit current density while the fill factor remains almost unaffected. Therefore, the efficiency is highest for surface recombination velocities below 1000 cm/s and Shockley-Read-Hall emitter lifetimes above 10 ns.

10⁰ 10¹ 10² 10³ 10⁴ 10⁵ 10⁶ 540

560 580 600 620 640 660

V OC [mV]

Semitter-base interface [cm/s]

τ_emitter [ns]

1 10 100

Figure 5-24: Open circuit voltage VOC dependence on the recombination velocity at the emitter-base interface.

As already shown in Table 5-7, the comparison of the simulation with the experimental results on Cz shows a remarkable agreement of the parameters.

Therefore, the respective solar cells must have an interface recombination velocity below 100 cm/s and an emitter bulk lifetime larger than 0.1 µs, i.e.

emitter recombination does not play a detrimental role when monocrystalline substrates are used. In the case of epitaxy on mc-Si substrates (Table 5-6), the situation is more complicated because of the recombination at grain boundaries in the emitter bulk. However, by analysing the experimental results, an upper limit for the recombination can be deduced. The interface recombination must be lower than 1000 cm/s and the emitter bulk lifetime is at least 1-10 ns. The experimental fill factor is less than that predicted by simulations and other detrimental influences are present in the cell. For instance, no grain boundaries

or defects are included in the one-dimensional simulations. Only extensive simulations with e.g. SDEVICE include such shunts [120].

Dark saturation current density J02 in solar cells – On multicrystalline substrates, the epitaxial layer thickness in the region of grain boundaries is often reduced compared with the intra-grain thickness (see Figure 5-18). A direct contact between the n⁺ and p⁺ regions may be possible, which would result in leaky junctions [14, 121]. However, the opposite is observed as the shunt resistances are above 10⁴ Ωcm². In addition, minimum dark saturation current densities J02 around 3x10^-8 A/cm² (Table 5-6) are noticeable, indicating low recombination via traps in the space charge region. Similar low dark saturation current densities J02 below 1x10^-8 A/cm² are found for epitaxial emitters on Cz substrates (Table 5-7).

Figure 5-25-A shows the dark I-V curves of samples with epitaxial emitters on mc substrates (Table 5-6) and Figure 5-25-B on Cz substrates (Table 5-7), as well as the EpiWE with POCl₃ emitter (Table 5-6). It is noticeable that the dark saturation current densities J₀₂ and the shunt resistances R_Sh are improved for epitaxial emitters compared to POCl₃ diffused emitter. The following section discusses the possible reason for this apparent dissimilarity.

0.0 0.2 0.4 0.6

Figure 5-25: Dark I-V curves of epitaxial wafer-equivalents on Cz (A) and on mc (B) substrates. The open symbols are EpiWE with POCl3 emitters and the solid symbols EpiWE with epitaxial emitters.

It is known from literature [24, 122] that grain boundaries and defects provide fast diffusion paths. Deep peaks or emitter spikes are formed along these defects as shown schematically in Figure 5-26. Beaucarne et al. found that the longer the diffusion along the grain boundaries occurs, the better the carrier collection [115]. The emitter passivates the defects and the collecting surface is increased. However, the dead layer is thickened, which results in a shift of the internal quantum efficiency peak to red response [115]. Deep emitter spikes increase the open circuit voltage, as surfaces with high recombination velocities are passivated. However, simultaneously the space charge region is dramatically increased in a region where many defects act as traps. This results in an increased dark saturation current density J02.

The solar cells shown in Figure 5-25 with epitaxial emitters receive in total a much shorter heat treatment than POCl3

diffused wafers. The epitaxial emitters are grown in approximately 1 minute and cooled down within 4 minutes. The subsequent oxidation is about 1h. The 120 Ω/sq. POCl3 diffusion takes

approximately 1h with a subsequent oxidation of 35 minutes. It seems plausible that the phosphorus diffusion along grain boundaries and defects is deeper on wafers with POCl₃ emitters. As explained before, this results in an increased dark saturation current density J₀₂, which is observed for the POCl₃ samples.

Furthermore, this effect depends also on the total amount of defects, which is about 4x10⁴ EP/cm² on Cz substrates. Samples with lower defect densities should result in lower dark saturation current densities J₀₂. This is shown for epitaxial emitters on Cz compared to mc substrates. Following this argumentation, sample HC758 having a higher dark saturation current density J₀₂ should have an increased defect density compared to sample HC776c (Table 5-7, Figure 5-25-A).

The spikes along defects increase the open circuit voltage and short circuit current density and should be thus higher for POCl₃ diffused samples compared to these with epitaxial emitters. However, it is difficult to verify the improvement, as the open circuit voltage depends on many factors, such as the emitter profile (see 5.4.3). Furthermore, the thickness and the crystal quality

p⁺ n⁺

n⁺

Grain boundaries

p

Figure 5-26: Phosphorus diffusion along grain boundaries.

may change from sample to sample and makes an evaluation of the gain with the I-V parameters difficult.

The collection of current can be observed with an SEM in the electron beam induced current (EBIC) mode. Chu et al., for example, observed some improvement of the beam-induced current collection after a heat treatment of the epitaxial pn-junctions [24]. They attribute this once more to the increased collection along grain boundaries due to the emitter spikes. Preliminary Junction-EBIC measurements of EpiWE on highly-doped Cz and off-spec mc were performed. However, the sample preparation was problematic and no results are available yet. Further investigations should be performed to evaluate the collecting effect in the grain boundaries and the correlation to dark saturation current densities.

5.4.5 Implementation of texture

In order to enhance the short circuit current density, a front side texturing should be applied to increase the optical base thickness. The texturing of the EpiWEs with epitaxial emitter has to be planned carefully. There are several different ways to implement the texture into the EpiWE approach with emitter epitaxy.

They can mainly be grouped according to when the texturisation is performed:

before the epitaxial deposition, between the base and emitter deposition or after the epitaxial emitter deposition.

If the texturing occurs before the base epitaxy, the texture is flattened by facet formation during the deposition process and is therefore less effective.

Moreover, due to the resultant roughness of the surface, the epitaxial deposition in the RTCVDs shows an increased etch pit density of 10⁵ EP/cm² in contrast to the growth on a smooth surface with etch pit densities of approximately 10⁴ EP/cm². This reduces the open circuit voltages of the cells [14].

On the other hand, when the wafer is textured after the base and before the emitter deposition, the etching is more difficult. Most wet chemical etches for texture only start to work on damaged surfaces. On an epitaxial surface, no homogeneous texture is formed and deep holes are created at defects [123].

Additionally, the high etch removal would remove most of the deposited base.

Only a few microns should be removed, which can be accomplished by plasma texturisation [124-126]. This technique creates small structures with heights of approximately 500 nm and is well suited for EpiWE with diffused emitters.

However, a thick epitaxial emitter would overgrow the shallow plasma texture.

Furthermore, a texture process between the base and emitter deposition would eliminate the advantage of the situ emitter. An ideal texture would be done in-situ, such as HCl gas etching (see Section 6.2).

If the texture is performed after the epitaxial emitter deposition, the etching has to be homogeneous and shallow to prevent etching shunts through the emitter. A thick n⁺⁺-layer could be deposited, which is then etched back to a thickness of 20-100 nm. It was already proven, that epitaxial emitters etched by a smooth CP etch show high shunt resistances (Figure 5-15). However, it is very difficult to control the texturing with this accuracy and too much of the n⁺⁺-layer would remain within the structures of the texture, leading to high Auger recombination and losses in the short circuit current at short wavelengths [99].

Figure 5-27 shows illuminated lock-in thermography⁷ measurements of a cSiTF solar cell with epitaxial emitter and textured by plasma. It can be seen that big shunts are created at the edges from the isolation by laser. These shunts can be avoided using a better isolation technique. More importantly, many shunts in the middle of the cell are seen (A) which are probably created during the texturing.

The intensity of the shunts is not as strong (B), but the shunts still decrease dramatically the short circuit current density at short wavelengths.

All methods have disadvantages and so compromises have to be made. An alternative solution could be the combination of a moderately-doped epitaxial

7 Illuminated lock-in thermography (ILT) is a standard characterisation method for the laterally resolved characterization of leakage currents in solar cells at the operation conditions. The excitation is performed with a semiconductor laser as illumination source [127].

1.9

1

-0.03 0.6

0.03

-0.6

A B

1.9

1

-0.03 0.6

0.03

-0.6

A B

Figure 5-27: Illuminated lock-in thermography mapping of an EpiWE with epitaxial emitter and a subsequent plasma texturing. The cell was measured at open circuit voltage conditions.

The shunt location is shown in (A) and the intensity of the shunts in (B).

emitter etched by plasma and a quick POCl₃ diffusion or a highly-doped epitaxial top layer. As the lifetime in the top layer would be limited by Auger recombination and not by the crystal quality, the deposition could also occur at lower temperatures and therefore thin layers could be deposited in our reactors.

However, one has to analyse the effect on the dark saturation current density and intermediate layers should eventually be grown after the texturing in order to avoid high interface recombination [99].

5.5

n-type epitaxial emitters for cSiTF solar cells with screen-printed contacts

Solar cells fabricated by photolithography and evaporated contacts reach high efficiencies; however, the solar cell process is costly and time-consuming. In the PV industry, cell processes are made with screen-printed contacts. The applicability to these industrial processes is essential for the epitaxial wafer-equivalent to maintain the potential cost advantage. Therefore, the epitaxial emitter was adapted to the industrial cell process. The results reported in this section are the first with screen-printed contacts on epitaxial emitters.

5.5.1 Design of the doping profile

So far, the surface dopant concentration was optimised for a titanium-silicon contact. However, screen-printing applications need higher surface concentrations, as the contact formation is based on silver crystals, which grow faster on highly-doped emitters [128]. Therefore, a new emitter design with a higher surface concentration was required in order to achieve a low contact resistance. Two different emitter profiles were tested, as shown in Figure 5-28.

Both emitters were composed of two layers, each approximately 0.5 µm thick with phosphorus concentration of 4.5x10¹⁸ cm^-3 and 5x10¹⁹ cm^-3. One was cooled in a PH₃/H₂ atmosphere with 888 ppm PH₃, whereas the other only in H₂. The surface concentrations were 7x10¹⁸ cm^-3 and 4.5x10¹⁸ cm^-3, respectively, with corresponding sheet resistances of 29 and 40 Ω/sq. It is clear that the profiles are not optimised, as the higher doped layer is far too thick and Auger recombination will limit the cell performance at short wavelengths. However, as described previously, thin layers are difficult to deposit in our reactors and in a first approach it is necessary to test if a good contact can be established.

Therefore, solar cells on off-spec mc were fabricated as described in Section 4.1.3 with the emitter diffusion replaced by these epitaxial emitters.

0.0 0.5 1.0 1.5 10¹⁵

10¹⁶ 10¹⁷ 10¹⁸ 10¹⁹ 10²⁰

Epitaxial emitter without with

PH3 during cooling P concentration [atoms/cm3 ]

Depth [µm]

Figure 5-28: Doping concentration profiles measured by SIMS of two epitaxial emitters, both deposited with 5 ppm and 182 ppm PH3 flows, each layer approximately 0.5 µm thick. One

sample was cooled in 888 ppm PH3 in H2 (dots), the other only in H2 (squares).

sample was cooled in 888 ppm PH3 in H2 (dots), the other only in H2 (squares).

Im Dokument High-temperature CVD processes for crystalline silicon thin-film and wafer solar cells (Seite 84-0)