Boosting Charge Carrier Mobilities in Upgraded Metallurgical Grade Silicon by Phosphorous Diffusion Gettering

Abstract

Herein, it is demonstrated how the carrier mobility and carrier lifetime of upgraded‐metallurgical grade silicon (UMG‐Si), a feedstock alternative to electronic‐grade, high‐purity polysilicon dominating photovoltaic technology, can be largely improved upon the gettering action of industrially compatible phosphorous diffusion gettering (PDG) process at the wafer level. The results, based on ultrafast THz spectroscopy and inductively coupled photoconductive decay measurements, show outstanding increments in the carrier lifetime of PDG‐treated multi‐crystalline UMG‐Si wafers from 50 ps to 25 μs and a boost in intra‐grain charge carrier mobility from 283 ± 37 up to 726 ± 23 cm² Vs⁻¹. Most remarkably, the latter figure parallels the carrier mobility observed in monocrystalline wafers manufactured from polysilicon, thereby demonstrating the effectiveness of a simple pre‐conditioning step in upgrading the electronic properties of UMG‐Si up to the device level.

Introduction

Photovoltaics (PV) represents nowadays the largest share of newly installed power capacity for electricity production, including both renewable and conventional sources.^[1] However, current capacity figures still lag well behind those needed to meet the targets set on net-zero emissions, as stated in the Paris agreement and recently updated in COP26.^[2] Within this scenario, any advancement towards an improvement in silicon-based PV efficiency/cost ratios will be valuable for furthering the PV deployment in the market over the next decades. While high-purity (in the range of 9–11 N) crystalline polysilicon (poly-Si) is a very robust and efficient material that dominates the PV market, still a significant ≈12% fraction of the silicon-based PV-module cost and about ≈20% of its energy investment is linked alone with silicon purification.^[3] As a matter of fact, such a high degree of purity is mostly required for fulfilling the demands of the electronic sector, where a single impurity present in a nano-device contained within a computer chip can compromise its whole functionality. In this respect, finding environmentally friendly, cheaper, and less energy-demanding alternatives to high-purity Si for PV, while still meeting electronic quality indicators compatible with high-efficiency device manufacturing (what has been termed as solar-grade silicon), represents a neat avenue for furthering PV deployment.

Upgraded metallurgical grade silicon (UMG-Si) is obtained from the metallurgical route and results in lower economic costs and environmental impact as compared to conventional poly-Si.^[4–6] However, this lower cost is typically achieved at the expense of incorporating a higher concentration of impurities during crystallisation.^[4] This aspect can be detrimental to the performance of related PV devices, particularly affecting charge carrier lifetimes and charge carrier mobilities. To overcome these drawbacks, different low-energy input treatments have been studied with the aim of improving the electronic quality of low-cost UMG-Si-wafers with high impurity contents. Among them, phosphorous diffusion gettering (PDG), an industrial compatible process commonly used during emitter formation in standard solar cell technology, has proved effective in improving the performance of UMG-Si.^[7–11] The PDG process is used for driving diluted impurities, particularly transition metals, towards the wafer surface, where they can be effectively removed following an etching process. As a result, PDG-treated UMG-Si wafers have shown improved charge carrier lifetimes when compared with reference samples;^[10,12,13] however, little is known yet about the impact of specific PDG treatments on UMG-Si-related charge carrier mobilities. Disentangling mobility-lifetime products is often not straightforward and requires either the application of suitable models through iterative calculations to the experimental results,^[14] or alternatively the utilization of specialized characterization tools capable of discerning between both parameters.

In this work, we have employed time-resolved Terahertz spectroscopy (TRTS), in combination with inductively coupled quasi-steady-state photoconductance decay (QSS-PCD), to study the impact of PDG processes in both critical parameters: the charge carrier lifetime and the charge carrier mobility of UMG-Si wafers, resolving both parameters independently. As a result of the PDG treatment, we find that effective carrier lifetimes in UMG-Si are boosted from ≈50 ps up to ≈25 μs. By means of TRTS, we could also retrieve the complex frequency-resolved conductivity in the THz region, from which we can infer a notable improvement in the charge carrier mobility figures of UMG-Si treated samples, changing from 283 ± 37 up to 726 ± 23 cm² Vs⁻¹ upon the PDG treatment. Most remarkably, the recorded mobility values of PDG-treated UMG-Si wafers parallel those observed in commercially available high-purity poly-Si float zone reference wafers. These results validate PDG processing as a meaningful treatment for upgrading the electronic quality of UMG-Si wafers to levels commensurate with poly-Si wafers produced by the Siemens process.

Results and Discussion

Multi-crystalline wafers consisting of p-type UMG-Si analyzed herein were manufactured via the metallurgical route by FerroGlobe and crystallized as multi-crystalline ingots by FerroSolar.^[5,6] UMG-ingots were sliced in 15.6 × 15.6 cm², 200 μm thick wafers, with resistivity values in the range of 1–1.7 Ω cm. The original, as-received wafers were subsequently cut into nine 5 × 5 cm² wafers for these studies (see Figure 1a). The material is moderately compensated due to the presence of P and B dopants introduced during the crystallization stage, with a compensation level defined as (N_D + N_A)/(N_A-N_D) ranging between 2 and 4 (N_A and N_D referring to the concentration of acceptors and donors respectively). After chemical cleaning of the UMG-Si as-received samples (see methods), PDG treatments were conducted on UMG-Si wafers in a tubular furnace, using liquid POCl₃ as P-source, at 800 °C for 1 h, letting the samples cool down passively outside the furnace. The near-surface region, including the resulting P-doped emitter, was subsequently chemically etched (see methods), recovering thereby a p-type wafer with upgraded electronic quality, as a result of the gettering action associated with P-diffusion. To interrogate carrier lifetimes and to prevent surface recombination effects, the PDG-treated samples were provided with a thin SiN_x passivation coating (see Figure 1b), deposited by plasma-enhanced chemical vapor deposition (PECVD). Although not optimized here for PV applications, this industrially compatible SiN_x treatment is typically employed also as an anti-reflection coating, due to a proper refractive index matching between air and the silicon substrate underneath (see Figure 1c). From now on, we will refer to wafers provided with both, the PDG treatment and the SiN_x coating as “UMG-Si treated” samples. Analogously, “UMG-Si untreated” samples will refer to the bare wafers.

[IMAGE OMITTED. SEE PDF]

To study charge carrier lifetimes of UMG-Si samples, we have combined two powerful contactless techniques, namely inductively coupled photoconductance decay (PCD)^[15–18] and time-resolved THz spectroscopy (TRTS),^[19–21] which together span the accessible range of carrier lifetime values from pico- up to milli-seconds. Figure 2a represents TRTS data showing the normalized time-resolved response as a function of pump-probe delay of the real conductivity for treated and untreated UMG-Si samples (green and black dotted traces, respectively). Signal normalization in Figure 2a is justified here as the employed fluences provide first-order dynamics (see Figure S1 and S2, Supporting Information). The samples were excited with a 387.5 nm laser pump beam (2.6 μJ cm⁻² with 150 fs pulse width) and probed with a ≈3 mm diameter and ≈1 THz bandwidth probe impinging on single grains of the multi-crystalline UMG-Si wafers (see Figure 1a,b). Within the same wafer, grain-to-grain TRTS responses were practically identical, with just minor changes in amplitude that we attribute to reflectivity differences among randomly oriented grains (see Figure S4, Supporting Information).

[IMAGE OMITTED. SEE PDF]

As it is evident from Figure 2a, the PDG treatment on UMG-Si wafers has a significant impact on charge carrier lifetimes as monitored by TRTS over the probed 1 ns time window. Right after above-bandgap pump excitation (at 0 ps pump-probe delay), a rise in the real conductivity is resolved by TRTS, which represents the instantaneous generation of free carriers in the sample ($\text{Re}[\sigma (t)]=eN(t)\mu$ ; where e is the electron charge, N(t) the photoinduced carrier density as a function of time and μ the charge carrier mobility). For the untreated UMG samples, the initial TRTS rise is followed by a decay over the probed window. The decay can be well fitted by a phenomenological bi-phasic exponential function $N(t)=N_{\text{hν}}(e^{-k_{1}t}+e^{-k_{2}t}$), providing rate constants k₁ and k₂ of 1.9 × 10⁹ ± 1.92 × 10⁸ and 1.75 × 10¹⁰  ± 2.6 × 10⁹ s⁻¹, respectively. From the rate constants we can estimate charge carrier decay times of τ₁ = 524 ± 53 and τ₂ = 57 ± 8 ps (where τ_i = 1/k_i; i = 1,2). For the case of the treated UMG-Si wafer (green trace in Figure 2a), a clear improvement in charge carrier lifetime is resolved; however, it cannot be quantified by TRTS within the recorded time window (τ >> 1 ns).

To quantify the carrier lifetime improvement of the UMG-Si-treated wafer, the same samples were characterized by PCD in quasi-steady-state (QSS) mode, thereby spanning the observed lifetime window up to the ms scale (Figure 2b). In QSS-PCD measurements, the photoconductance of the sample is recorded as a photovoltage developed in an inductively coupled coil while a slowly varying illumination impinges upon the sample. If the characteristic time of the flashlight source (typically 10 ms) is much longer than the carrier lifetime in the sample, QSS conditions can be invoked to solve the continuity equation1$\frac{\mathrm{d}\Delta n}{\mathrm{d}t}=0=G-R=G-\frac{\Delta n}{\tau }$where Δn is the excess photogenerated carriers, G and R are the generation and recombination rates, and τ is the carrier lifetime. The generation rate is monitored along the flash duration in a reference photodiode of the PCD setup, which allows tracing the carrier lifetime dependency on the injection level. The characteristic apparent lifetime increase observed at low injection levels in both samples is a well-known effect associated with carrier trapping.^[22,23] Carrier trapping effectively delays the consumption of excess carriers upon recombination, which is reflected in the aforementioned equation as an apparent increase in τ. Bulk carrier lifetimes are therefore better estimated at sufficient injection levels so as to avoid entering the exponential-like, trap-related curvature. In the flat region, the effective lifetime curve of the treated sample (green dotted line in Figure 2a) shows a neat improvement of nearly two orders of magnitude with respect to the untreated sample (which lies at the detection limit of the technique), with a value of 25 μs at an injection level of 1 × 10¹⁵ cm⁻³, consistent with the results observed by TRTS results. The resolved improvement in charge carrier lifetimes observed in PDG-treated samples is largely related to the action of the PDG process; a conclusion that can be drawn from the comparison of carrier dynamics between PDG samples with and without the SiNx treatment (see Figure S5, Supporting Information).

Having demonstrated a substantial boost in charge carrier lifetimes upon PDG treatment, we focus our attention now on its impact on the charge carrier mobility. For this purpose, we have analyzed the frequency-resolved complex conductivity σ(ω) that can be retrieved from TRTS data at any pump-probe delay. Figure 3 shows the real (black solid dots) and imaginary (open blue circles) components of σ(ω) obtained for the untreated and treated samples, panels a and b, respectively. The data have been obtained at a pump-probe delay of 3 ps, thereby ensuring the complete cooling of hot charge carriers generated after ≈400 nm excitation.^[24,25] The first panel (Figure 3a), representing σ(ω) for untreated UMG-Si, shows a characteristic negative imaginary conductivity component that increases with frequency and changes sign to positive values within the probed THz window, while the real part of the conductivity remains always positive. This line shape can be well described by the Drude–Smith (DS) model, a phenomenological extension of the free carrier Drude model, where long-range transport in the DC limit is limited by charge carrier localization.^[26] The DS model can be defined as $\Delta {\sigma }_{\text{DS}}=\Delta {\sigma }_{\text{Drude}}\times [1+c/(1-i\omega {\tau }_{\mathrm{s}})]$where $\Delta {\sigma }_{\text{Drude}}$ represents the Drude conductivity2$\Delta {\sigma }_{\text{Drude}}={\epsilon }_0{\omega }_{\mathrm{p}}^2{\tau }_{\mathrm{s}}/(1-i\omega {\tau }_{\mathrm{s}})$where ${\tau }_{\mathrm{s}}$ is the averaged scattering time and ${\omega }_{\mathrm{p}}^{}$ is the free-carrier plasma frequency ${\omega }_{\mathrm{p}}^2{}_{}=N{e}^2/{\epsilon }_0{m}^{*}$, with N the charge carrier density, $e^{}$ the electron charge, and $m^{*}$ the effective mass. In addition, in the DS model, the c parameter represents the coefficient of backscattering, ranging from 0 (pure Drude behavior) to −1 (total charge carrier localization and null conductivity in the DC limit). Furthermore, according to the phenomenological DS model, the charge carrier mobility is expressed as $\mu =(e{\tau }_{\mathrm{s}}/m^{*})(1+c$). From the best fit to the DS formula (solid lines in Figure 3a), we obtain an average scattering time of 99 ± 5 fs, a c parameter of −0.58 ± 0.04, and a value for the plasma frequency of 104 ± 2 THz, from which, taking into account an electron effective mass for the conductivity of $0.26m_{\mathrm{e}}$,^[27] we obtain a value for the mobility of 283 ± 37 cm² V⁻¹ s⁻¹ and a photodoping of 8.9·10¹⁷ ± 3.6·10¹⁶ cm⁻³. The appearance of a DS response in the case of the untreated UMG-Si sample within a single crystal domain (i.e., probing a local area of 3 mm diameter within a way larger grain) reveals that the restoring force limiting the long-range motion of the photogenerated charge carriers cannot be linked with physical backscattering at grain boundaries, but rather, it should be a consequence of a restoring force induced by a “corrugated” energetic landscape seen by the probed free carriers. We speculate that this could be related to the presence of doping/impurity aggregates; further work is needed to ascertain its origin.

[IMAGE OMITTED. SEE PDF]

Figure 3b shows the frequency-resolved conductivity of the treated UMG-Si sample. The plot reveals a decreasing real and increasing imaginary conductivity components (both positive) in the analyzed THz range. In this case, we can reproduce the data very well invoking the simple Drude model for free carriers (see Equation (2)). From the best fit (solid lines in Figure 3b), we obtain a value for the scattering time and plasma frequency of 107 ± 5 fs and 54 ± 1 THz, respectively. Considering once more the value $0.26m_{\mathrm{e}}$ for the electron effective mass for conductivity, the corresponding mobility is 726 ± 23 cm² V⁻¹ s⁻¹, for a pump-induced carrier density of 2.34 × 10¹⁷ ± 4 × 10¹⁶ cm⁻³. This figure shows an almost threefold improvement in charge carrier mobility for the PDG-treated UMG-Si samples. To place this figure in context, we measured by TRTS under the same experimental conditions a highly pure float zone (FZ) monocrystalline Si wafer (see Figure S7, Supporting Information). As summarized in Table 1, and remarkably, the charge carrier mobility obtained for PDG-treated UMG-Si samples parallels, within the experimental error, the value obtained from the FZ-Si reference. By analyzing a sample without the SiN treatment under the same excitation conditions, we conclude that the PDG process is the main contributor to the observed increase in charge carrier mobility (see Figure S6, Supporting Information). This result demonstrates that the charge carrier mobility of UMG-Si after the employed PDG treatment is at the same level as state-of-the-art conventional poly-Si wafers.

Table 1 Summary of results obtained from OPTP from UMG-Si wafers and FZ-Si crystalline silicon reference wafer

Sample	Charge carrier density [cm−3]	Mobility [cm2 V−1 s−1]
UMG-Si untreated	8.9 × 1017 ± 3.6·1016	283 ± 37
UMG-Si treated	2.34 × 1017 ± 4·1016	726 ± 23
FZ-Si	5.6 × 1017 ± 1.4·1016	718 ± 19

Conclusion

In summary, we have demonstrated an important improvement of the carrier mobility in UMG-Si wafers as a result of a gettering process. Phosphorous diffusion can effectively upgrade the carrier lifetime and mobility, leading to improvements as large as about four orders of magnitude in the case of the lifetime, and pushing the charge carrier mobility up to 726 cm² V⁻¹ s⁻¹, at the same level of semi-insulating, single crystalline, commercially available polysilicon. The improvement observed in the carrier lifetimes after PDG treatment is associated with the removal of bulk impurities; correlated to this effect, the improvement in charge carrier mobility could be attributed to a reduction of the impurity scattering in the samples.^[28,29] Sample to sample variability in both lifetimes and mobilities could be a priori expected on wafers coming from different UMG ingots and different heights within an ingot, if so, the effectiveness in absolute terms of the proposed PDG process will depend on the specific nature and concentration of the limiting impurity. In any case, our results demonstrate the effectiveness of simple pre-conditioning steps adapted for UMG-Si, which are compatible with current solar cell technology manufacturing. The lower economic and environmental impact of UMG-Si in comparison to conventional polysilicon can thus be fully exploited from a material no longer penalized by its starting electronic quality. Further work is nevertheless needed regarding the optimization of the gettering and passivating processes, accounting for, e.g., sample and grain-to-grain heterogeneity and the impact of dislocations and grain boundaries on the expected performance of UMG-Si-based devices.

Experimental Section

Materials

Solar-grade silicon wafers have been provided by Aurinka PV. Samples were manufactured via the metallurgical route by FerroGlobe^[5] and then crystallized as multi-crystalline ingots by FerroSolar. UMG-ingots were sliced in 15.6 × 15.6 cm², 200 μm thick wafers, with resistivity values in the range 1–1.7 Ω cm, as measured by the four-point probe method. The original wafers were subsequently cut into nine 5 × 5 cm² wafers for these studies. The material is moderately compensated due to the presence of P and B dopants introduced during the crystallization stage, with a compensation level defined as (N_D + N_A)/(N_A− N_D) ranging between 2 and 4. All wafers were chemically etched with CP4 (H₂O/HNO₃/HF: 450/150/30) and RCA1 (H₂O/H₂O₂/NH₃: 500/100/100) before the PDG.

Semi-insulating silicon crystalline wafer (float-zone) used as a reference was obtained from Sigma-Aldrich with ID: 646687.

PDG-Treatment UMG-Si Wafers

PDG processes were performed in a tubular furnace under O₂ and N₂ atmosphere. Liquid POCl₃ was the P source in a process carried out at 800 °C. After PDG, the near-surface region, including the n-type emitter formed after P-diffusion, was removed by chemical etching with CP4 (HNO₃/HF) and RCA1 (H₂O₂/HF) surface cleaning. SiN_x passivation coating was deposited via a low-temperature Plasma enhanced chemical vapor deposition (PECVD) process in a Meyer Burger AK-400 reactor.

TRTS

The ultrafast photo-conductivity response was determined using TRTS. The setup used is similar to the one described in ref. [19]. A Ti:sapphire laser amplifier system generation provided <150 fs laser pulses centered at 775 nm at a repetition rate of 1 kHz. The laser output was used for the optical pump by doubling its frequency with a beta barium borate (BBO) crystal, obtaining a pump wavelength of 387.5 nm. For the optical generation, THz light was generated via optical rectification on a 1 mm slab of ZnTe crystal cut along the (110) plane. The detection was performed via electro-optical sampling on an identical ZnTe crystal. THz signal gives access to monitor the change in the photoconductivity according to: $-\Delta T(\omega )/T(\omega )\alpha \Delta \sigma =Ne\mu$, where N is the charge carrier density and $\mu =e\tau /m^{*}$ is the mobility, dependent on the carrier scattering time, τ, and on the effective mass, $m^{*}$. The optical penetration depth is estimated in 80 nm,^[30] which sets the volume of charge carrier photogeneration. Since the penetration depth, $d_0$, is small compared to the THz wavelength and the sample thickness, the system can be considered as a photoexcited thin film. The frequency-dependent complex conductivity σ(ω) can be obtained from the expression below,^[19] measuring in the time domain the electric field of the THz pulses transmitted through the non-excited sample $T(\omega )$ and the pump-induced change $\Delta T(\omega )$^[19]3$\Delta \sigma (\omega )=-\frac{\Delta T(\omega )}{T(\omega )}\frac{(n_1+n_2)}{Z_0{d}_0}$where $Z_0$ is the intrinsic impendance of free space, $n_1$ is the refractive index of the medium before the sample, and $n_2$ is the refractive index of the medium with photoexcitation. This allows inferring the complex photoconductivity of the system assuming a thin region near the surface which is photoexcited.

Minority Carrier Lifetime

QSS-PCD measurements were obtained using the Sinton Instruments WCT-120 tool, using a slow flash decay mode with a characteristic constant decay of 1.8 ms, as measured at the reference cell. For the acquisition of the lifetime curves, an optical constant of 0.7 accounted for reflectivity losses and the Dannhäuser carrier mobility model was used to translate PC versus time curves into τ vs Δn traces.

Reflectance

Global reflectance was measured with the Shimadzu spectrophotometer UV-2401 PC using the integrating sphere ISR-240 A and with a white plate of BaSO₄ as reference.

Acknowledgements

The authors acknowledge financial support from Comunidad de Madrid (2017-T1/AMB-5207, 2021-5A/AMB-20942, and Y2018/NMT-5028), from the Spanish Government (PID2019-107808RA-I00), and from the Max Planck society. MCIN/AEI/ 10.13039/501100011033 is also acknowledged for financial support through the GREASE project (PID2020-113533RB-C31) and the project MADRID-PV2 (S2018/EMT-4308) funded by the Regional Government of Madrid with the support from FEDER Funds. The authors would like to also acknowledge Javier Álvarez Conde and Dr. Eva María Garcia-Frutos (ICMM-CSIC) for letting us use the UV-2401 equipment.

Conflict of Interest

The authors declare no conflict of interest.

Data Availability Statement

The data that support the findings of this study are available from the corresponding author upon reasonable request.

References

International Renewable Energy Agency, in Renewable Capacity Statistics 2021, International Renewable Energy Agency, Masdar City, Abu Dhabi 2021.

International Energy Agency, in Renewables 2021 Analysis and Forecast to 2026, International Energy Agency, Paris, France 2021.

D. Feldman, V. Ramasamy, R. Fu, A. Ramdas, J. Desai, R. Margolis, in U.S. Solar Photovoltaic System and Energy Storage Cost Benchmark: Q1 2020, National Renewable Energy Laboratory, Golden, CO, USA 2021, pp. 1–120.

L. Méndez, E. Forniés, D. Garrain, A. Pérez Vázquez, A. Souto, T. Vlasenko, Sci. Total Environ. 2021, 789, 147969.

E. Forniés, B. Ceccaroli, L. Méndez, A. Souto, A. P. Vázquez, T. Vlasenko, J. Dieguez, Energies 2019, 12, 1495.

E. Forniés, C. del Cañizo, L. Méndez, A. Souto, A. Pérez Vázquez, D. Garrain, Sol. Energy 2021, 220, 354.

N. Dasilva-Villanueva, S. Catalán-Gómez, D. Fuertes Marrón, J. J. Torres, M. García-Corpas, C. del Cañizo, Sol. Energy Mater. Sol. Cells 2022, 234, 111410.

J. S. Kang, D. K. Schroder, J. Appl. Phys. 1989, 65, 2974.

W. Schröeter, R. Kuehnapfel, in Point and Extended Defects in Semiconductors (Eds: G. Benedek, A. Cavallini, W. Schröter), Springer US, Boston, MA 1989, pp. 95–104.

A. Bentzen, A. Holt, Materials Science and Engineering B 2009, 159, 228.

A. Y. Liu, S. P. Phang, D. Macdonald, Sol. Energy Mater. Sol. Cells 2022, 234, 111447.

H. B. Xu, R. J. Hong, B. Ai, L. Zhuang, H. Shen, Appl. Energy 2010, 87, 3425.

A. Jiménez, C. del Cañizo, C. Cid, A. Peral, J. Electron. Mater. 2018, 47, 5068.

A. Cuevas, D. Macdonald, Sol. Energy 2004, 76, 255.

R. A. Sinton, S. Consulting, S. Jose, A. Cuevas, M. Stuckings, in Conf. Record of the Twenty Fifth IEEE Photovoltaic Specialists Conf., IEEE, Piscataway, NJ 1996, pp. 457–460.

R. A. Sinton, A. Cuevas, Appl. Phys. Lett. 1996, 69, 2510.

M. Goodarzi, R. Sinton, D. Chung, B. Mitchell, T. Trupke, D. Macdonald, in IEEE 44th Photovoltaic Specialist Conf. (PVSC), IEEE, Piscataway, NJ 2017, pp. 2707–2710.

H. Nagel, C. Berge, A. G. Aberle, J. Appl. Phys. 1999, 86, 6218.

R. Ulbricht, E. Hendry, J. Shan, T. F. Heinz, M. Bonn, Rev. Mod. Phys. 2011, 83, 543.

P. Kužel, H. Němec, Advanced Optical Materials 2020, 8, 1900623.

C. A. Schmuttenmaer, Chem. Rev. 2004, 104, 1759.

D. Macdonald, A. Cuevas, Appl. Phys. Lett. 1999, 74, 1710.

S. Catalán-Gómez, N. Dasilva-Villanueva, D. Fuertes Marrón, C. del Cañizo, Phys. Status Solidi RRL. 2021, 15, 2100054.

J. O. White, S. Cuzeau, D. Hulin, R. Vanderhaghen, J. Appl. Phys. 1998, 84, 4984.

F. E. Doany, D. Grischkowsky, Appl. Phys. Lett. 1988, 52, 36.

T. L. Cocker, D. Baillie, M. Buruma, L. V. Titova, R. D. Sydora, F. Marsiglio, F. A. Hegmann, Phys. Rev. B 2017, 96, 205439.

B. Van Zeghbroeck, in Principles of Semiconductor Devices, University of Colorado, Boulder, CO 2011.

D. Chattopadhyay, H. J. Queisser, Rev. Mod. Phys. 1981, 53, 745.

H. Brooks, C. Herrings, Phys. Rev. 1951, 83, 879.

G. De Graaf, R. F. Wolffenbuttel, IEEE Trans. Instrum. Meas. 2004, 53, 238.