In order to evaluate the passivation quality, values of surface recombination velocity,
SRV, have to be calculated. To this end, it is necessary to determine both phosphorus
profile and the surface doping concentration (Ns), so that it is possible to distinguish the recombination within the emitter from the recombination at the surface.
The process of determination of surface recombination velocity in a highly doped region (emitter) is summarized hereafter:
1. Determination of emitter profile, i.e. the doping concentration as a function of the depth. It is performed by SUPREM characterization tool, correlated with the measured sheet resistance, and compared with SIMS measurements (section 4.5.1).
2. Lifetime measurement of the symmetric structure and extraction of the saturation current density, J0e, as described in section 4.5.2.
3. PC1D simulation [26] of a p-n junction and generation of the corresponding dark
I-V characteristic. The structure for simulation is the following:
• Very thin p-type base (about 2 µm thick) with very high lifetime, about 10 ms (high resistivity and no SRH recombination). In this way, the recombination in the base can be neglected.
• The emitter profile determined from SUPREM, at the front side.
• A random value for the front surface recombination, Seff,front. We assumed a defect energy position located at the intrinsic level Ei and that the emitter is always working at Low Level Injection (LLI). This implies that surface recombination is dominated by the fundamental recombination velocity of the minority carriers (Sp0 in our case). Therefore we use
Seff,front = Sn0 = Sp0 = Seff at LLI as input parameters. • No rear surface recombination, Seff,back = 0.
Under these conditions, the recombination is only due to the front surface and the Auger plus radiative in the bulk emitter. The saturation current density for the I-V curve can therefore be fully attributed to the emitter, and the base component can be neglected: e b e J J J J0 = 0 + 0 ≈ 0 (4.2)
4. Simulation of the dark I-V and extraction of parameters. The model used for the fitting is described by:
( )
− + − = 0 T 1 0 rec T 1 qV n V rec qV V e I e I V I (4.3)Where V is the applied voltage, q is the electron charge, VT = k T /q is the thermal voltage, I is the total current, I0 and I0rec the saturation currents for the diffusion and recombination diodes, respectively, nrec is the ideality factor for the recombination diode (normally equal to 2). The corresponding saturation current
density, J0, is obtained by simply dividing I0 by the device area used in the PC1D simulation.
Steps 3 and 4 are repeated until the J0 matches with the measured J0e. Then, the surface recombination velocity in the emitter is the corresponding Seff,front obtained in step 3.
Concerning step 3, it is important to comment the effect of Band Gap Narrowing (BGN). As the doping concentration at the surface, Ns, increases, the band gap decreases, thus increasing the bulk recombination. Nevertheless, PC1D incorporates a lineal model for the shrinking of the Eg that is overestimated at high Ns values (see for example ref [138]). In PC1D it is not possible to modify this parameter. On the contrary, the excess of recombination provided by the wrong BGN model can be compensated by decreasing the Auger coefficient, Cn, which can be edited in PC1D. In order to obtain a suitable Cn term we simulated the p-n structure with the highest doped emitter (20 Ω/sq) and assuming no surface recombination (Seff,front = Seff,back = 0). Actually, in such very highly doped emitters the recombination is dominated by the bulk. Cn was lowered until the resulting J0e value was the same than that determined from lifetime measurements. Then, the same Cn value can be employed in the rest of pre-diffused emitters, as the surface doping concentration does not vary significantly with the sheet resistance in such structures. From this procedure we obtained Cn = 1.2 ×10-31 cm-6 s-1, while the default value in PC1D was
Cn = 2.2 ×10-31 cm-6 s-1.
Another argument to decrease the Cn term is provided by the experimental data of a passivated emitter with a drive-in process. The dependence of Ns with the sheet resistance in this case is significant. Because Ns has decreased, the error in the BGN committed by PC1D is smaller, so that the corresponding p-n junction could be simulated correctly. However, if we assume Cn = 2.2 × 10-31 cm-6 s-1 with Seff,front = Seff,back = 0, the J0e predicted is still higher than the experimentally measured, indicating that Cn value is anyway overestimated.
In addition to the experimental J0e(Rsh) curves, Fig 4.13 plots the resulting theoretical
J0e (Rsh) curves (solid lines) for given fixed values of SRV, ranging from 0 to 5 × 106 cm s-1, which is close to the thermal velocity limit for silicon [138]. It can be seen that for each stack configuration a single SRV value reasonably fits the whole J0e(Rsh) curve. This is again in agreement with the low variation of surface doping, Ns, due to the absence of a drive-in step, because normally SRV depends strongly on this value. For single ARC
layers we obtain a SRV of about 106 cm s-1, for the thinnest PAS layer (30 s deposited - 4 nm thick) it is about 105 cm s-1. Finally, the structures with the two thickest PAS layer (60 and 90 s deposited) keep SRV around 104 cm s-1 at low R
sh. At higher sheet resistance values, SRV tend to be 2 × 104 and 5 × 103 cm s-1 respectively.
At very low sheet resistance J0e converges to a single value, independently of the surface passivation treatment. Regarding the theoretical curves, the lowest Rsh achieved in our experiment (20 Ω/sq) could still show a certain degree of surface passivation. However, a complete de-passivated behaviour is observed for those samples, probably indicating a lattice damage within the emitter directly compared to high phosphorus concentration.