Band alignment in Zn_(1−x)Mg_xO:Al/SiO_x/Si heterostructures for photovoltaic applications realized by atomic layer deposition: Effects of Al doping and Mg alloying

In the past years, promising efficiencies (⁠ $η$s) equal to $∼$13%–15% and $∼$16.5% were obtained in the case of n-ITO/n-Si isotype heterostructures^1,2 and n-ITO/p-Si anisotype heterostructures,³ respectively. These results triggered, also recently, searching for low-cost alternatives to ITO with ZnO/ZnO:Al being among the possible replacement candidates due to comparable electrical/optical properties combined with the possibility to realize Al metallizations with a specific contact resistance as low as $∼ 10 − 5Ω cm 2$⁠.^4,5 In addition, this device design, considering that Ag-based Ohmic contacts and ITO transparent layers for charge carrier extraction are among the current bottlenecks hindering the photovoltaic market penetration of silicon heterojunction solar cells (HJT), despite having reached a technological maturity and efficiencies up to $∼$26.8%, could be partially or in total be integrated in the current HJT solar cell structure.^6,7 This would make it possible to overcome their present reliance on Ag and In, thus resulting in an economically more affordable HJT technology also suitable for future multi-terawatt scale applications.⁸ Furthermore, using a ZnO/ZnO:Al-based emitter paves the way to the utilization of technologically less demanding deposition techniques like spin-coating or spray pyrolysis for its realization, a choice that would additionally lower the present manufacturing costs. However, both n-ZnO/n-Si and n-ZnO/p-Si devices showed generally lower performances in the $∼$1%–8.3% range, despite the wide variety of deposition techniques used, ranging from ultrasonic spray pyrolysis to DC magnetron sputtering on flat, texturized, and unpolished substrates.^9–13 Using nanostructured substrates,¹⁴ the introduction of an $Y 2 O 3$ or $La 2 O 3$ interlayer,¹⁵ varying the dopant, or alloying the ZnO layer^14,16,17 are routes that were also tested with $η$s achieved being anyhow well below the theoretical calculation limit that is $≳$20%.^18,19 In the previous listed works, little attention has been paid to the values of the ZnO/Si conduction band offset (⁠ $Δ E C$⁠), even though $Δ E C$ is anticipated to be one of the crucial parameters. In detail, the main impact of $Δ E C$ is on the open circuit voltage (⁠ $V O C)$ as well as the fill factor (⁠ $FF$⁠), thus determining the devices’ final $η$⁠. Indeed, theoretical calculations confirm that, even for interface recombination velocities as high as $∼ 10 6 cm s − 1$⁠, by turning the staggered (type II, $Δ E C>0$⁠) into a straddling (type I, $Δ E C<0$⁠) bandgap alignment, anisotype n-ZnO/p-Si based heterostructures could achieve $η≳$20%, as a consequence of a factor of $∼$2 and $∼$10% increase in $V O C$ and $FF$⁠, respectively (see also the inset in Fig. 4 for an energy band diagram sketch).¹⁸ Such $Δ E C$ tuning can be realized by alloying ZnO with Mg as shown in a previously published work.²⁰ However, the Mg content is practically limited to $∼$2 at. %, since increasing it further produces ZnMgO films with higher resistivity that are, therefore, not suitable to act as emitters. In the present study, the route of using $∼$2 at. % Al doping while increasing the Mg content up to $∼$12 at. % is investigated, and a detailed analysis of the n-ZnMgO/p-Si conduction band alignment that takes into account the presence of the $SiO x$ interlayer is provided. It is found that by Mg alloying, $Δ E C$ can be reduced from $∼$1 to $∼$0.6 eV in the investigated range. This turns out into a $Δ E C$ reduction factor vs Mg content equal to $∼$1.7 eV if the conduction band lowering related to the n-ZnMgO:Al high carrier concentration is taken into account. The extracted dependence is $∼$1 eV lower than the value used in previous simulations, set to 2.7 eV in that case.¹⁸ This, despite $Δ E C$ for the two extreme cases of ZnO and MgO vs Si being $∼$0.6–0.9 eV smaller (type II) and $∼$1.8 eV larger (type I) than the ones anticipated by theoretical calculations, respectively,^21–23 prevents turning the type II band alignment into a type I within the Mg alloying range in ZnO for which the wurtzite structural configuration is maintained. These findings explain the above-mentioned substantial difference between theoretically predicted and experimentally achieved $η$s and point to the residual $Δ E C$ as one of the main causes for the low photovoltaic performances not only obtained in the test structures realized in the present work but also reported in the literature.

As substrates were used quartz, soda-lime glass and single side polished (epi-ready), commercially available boron doped p-type silicon wafers with a (100) orientation, a nominal resistivity (⁠ $ρ$⁠) equal to 5–10  $Ω$ cm, and thickness of $250±10$  $μ$m bought from Siegert Wafer. An Al back contact of $∼$100 nm was sputtered using a Kurt J. Lesker PVD 75 thin film deposition system on the unpolished Si wafer side. Then, the wafers were annealed for 20 min in Ar at 500  $°$C using a AccuThermo AW610 rapid thermal process system. Glass and quartz substrates were cleaned in 2-propanol with an ultrasonic cleaner and dipped in de-ionized water (DI). Then, all the substrates were transferred into an atomic layer deposition (ALD) reactor Savannah 100 (Veeco). The $Zn ( 1 − x ) Mg x$O and $Zn ( 1 − x ) Mg x$O:Al layers were grown at 280  $°$C, if not differently stated, using diethylzinc (DEZ), de-ionized water (DI), bis(methylcyclopentadienyl)magnesium (MCp2Mg), and trimethylaluminum (TMA) as zinc, oxygen, magnesium, and aluminum precursors with pulsing times in the 0.02–0.7 s range, depending on the precursor. $N 2$ was used as the inert gas with 4 s long purging times after each precursor pulse. The number of cycles was adjusted to obtain films with a thickness of $∼$400 nm. The recipe represents a variation from the one used in previous studies (see, e.g., Ref. 20). The first TMA pulse was introduced after a total number of DEZ+DI and MCp2Mg+DI pulses varying in the 13–24 range, if not differently stated. Subsequently, on top of selected $Zn ( 1 − x ) Mg x$O and $Zn ( 1 − x ) Mg x$O:Al layers deposited on Si substrates, an aluminum-doped ZnO (AZO) film of $∼$200 nm thickness was grown with the same ALD deposition system using DEZ, DI, and TMA as zinc, oxygen, and aluminum precursors, respectively. The deposition parameters were chosen on the basis of an established ALD deposition routine with, in this case, purging times reduced to 4 s and 280  $°$C as growth temperature.^24,25 Similarly to what was previously reported in Refs. 24 and 25, the carrier concentration (⁠ $n$⁠), mobility (⁠ $μ$⁠), and $ρ$ of the AZO layers fluctuated in the $(3−6)× 10 20 cm − 3$⁠, $(9−11) cm 2 V − 1 s − 1$⁠, and $(1−2)× 10 − 3$  $Ω$ cm ranges, respectively.

Al/Au Ohmic contacts on the front AZO layer were deposited by sputtering/e-beam evaporation with the Kurt J. Lesker PVD 75 thin film deposition system. These top contacts were circles of $∼$1.5  $mm 2$ area with the Au top layer $∼$30 nm thick added only to prevent Al (film thickness $∼$100 nm) oxidation. The sample size was $∼$5 $×$ 5 mm [see the inset in Fig. 3(a) for a schematic view]. To realize the test solar cells, selected p-Si/ $Zn 1 − x Mg x$O:Al/AZO structures were cut into $∼1×1$ cm samples and a $∼$200 nm thick Al busbar with a finger contact pattern was sputtered by using the same PVD thin film deposition system. The active to dark/total area ratio of the realized test solar cells was $(87±3)$% [see the inset in Fig. 11(a) for a front picture]. Finally, Al/Au bilayers (⁠ $∼$100/ $∼$30 nm) on the corner of the ZnO-based films grown on glass and quartz substrates were also sputtered/evaporated using the same PVD setup to serve as Ohmic contacts for Hall measurements.

Energy-dispersive x-ray spectroscopy (EDS) performed with a scanning electron microscope (SEM) Hitachi SU-70 at 5 or 6 kV accelerating voltage was used to determine the Mg and Al atomic contents in the single layers. First, these measurements were used to verify that, as expected and shown in Fig. S1, the Mg content with respect to the sum of the Zn, Mg, and Al atomic contents increased linearly with a proportionality constant equal to 1 with respect to the ratio between the Mg number of ALD cycles and the sum of the Zn, Mg, and Al number of ALD cycles. Furthermore, the same measurements were also used to calculate the Mg content (⁠ $x$⁠), that is, the ratio between the Mg content and the sum of the Zn and Mg ones. SEM front and cross-sectional images after cleaving the layers were also acquired with the same setup operated in the secondary electrons mode (SE). Cross-sectional images were used to measure the film thickness that was confirmed by using a NanoCalc 2000s thin film reflectometer. The structural properties of the ZnO based layers were studied using a PANalytical X’Pert Pro Alpha1 MPD x-ray diffractometer (XRD) with locked-coupled $Θ−2Θ$ scans using the Cu K $α 1$ $(λ∼1.5406Å)$ monochromatic radiation. In order to extract the signal from the whole volume of the sample under study, the diffraction patterns were collected in the Bragg–Brentano geometry using a semiconductor strip detector in the scanning mode with a 2.122 $°$ $2Θ$ active range. The samples were spinning at an angular velocity of 12.6 rad  $s − 1$ during the measurements. The room temperature (RT) electrical parameters of the layers were extracted from Hall effect measurements performed in the dark using a RH2035 PhysTech system equipped with a permanent magnet producing a magnetic field of $∼$0.4 T. The measurements were executed in the van der Pauw configuration on the ZnO based films deposited on glass and quartz substrates. The experimental results were corrected for the contact size contribution as described in Ref. 26. Standard cross-sectional transmission electron microscopy (TEM) samples were prepared by grinding and ion-polishing with a PIPS ion polishing system from Gatan. The samples’ local structure and composition were investigated by high-resolution scanning transmission electron microscopy (STEM) using a monochromated FEI Titan G2 60-300 microscope operated at 200 kV and equipped with a Super-X EDS Bruker detector. Finally, the secondary mass spectrometry measurements shown in the supplementary material were conducted with a Cameca IMS 6f microanalyzer using a primary beam of 5.5 keV Cs ions. A constant erosion rate was assumed for depth calibration in this case.

The resulting devices were characterized by current density vs voltage (J–V) and capacitance vs voltage (C–V) measurements carried out at RT using a Keithley 2601A and an Agilent E4980A current and LCR meter, respectively. During the measurements, the temperature was monitored with a thermocouple and a HH11C thermocouple reader. In the latter case, $ac$-probing frequencies and an $ac$-probing voltage in the 1 kHz–1 MHz range and equal to 30 mV (root mean square value) were used, respectively. Successively, current vs voltage (I–V) under illumination, $η$ and external quantum efficiency (EQE) measurements were performed on the same devices and the test solar cells using a Bentham PVE300 photovoltaic QE system. The measurements were done under STC, i.e., under the standard terrestrial spectra of AM1.5G and temperature equal to 25  $°$C. The solar cells/test device parameters presented were calculated using the effective active area of the devices and extracted with the IV Curve Fitter v.1.8 ©software using the standard two-diode model.²⁷ 

The electrical properties of the $Zn ( 1 − x ) Mg x$O and $Zn ( 1 − x ) Mg x$O:Al films deposited on soda-lime glass vs $x$ are reported in Fig. 1(a) with the corresponding Al atomic content being shown in Fig. 1(b). By introducing Al, an order of magnitude increase in $n$ up to $∼$4 $× 10 20 cm − 3$ was achieved. To this increase corresponds a factor 2 decrease of $μ$⁠. Considering the low Mg alloying level in these layers, the observed drop is consistent with the increased ionized impurity scattering related to the Al donor activity. Furthermore, as discussed hereafter, charged compensating centers, neutral complexes are most probably present as well, thus providing additional scattering centers that contribute to further reduce $μ$⁠. No significant changes in the film’s electrical characteristics were found for samples in the $Zn 0.94 Mg 0.06$O:Al– $Zn 0.86 Mg 0.14$O:Al range, while a conductivity drop was observed to occur starting from the $Zn 0.84 Mg 0.16$O:Al sample related to both a reduced $μ$ and $n$⁠, with the latter indicating a less effective doping activity of Al for higher Mg contents. For the $Zn ( 1 − x ) Mg x$O:Al layers with $x$ varying in the $∼$0.20–0.24 range, $n$⁠, $μ$⁠, and $ρ$ resulted equal to $∼$1.7 $× 10 20 cm − 3$⁠, $∼$2.2  $cm 2 V − 1 s − 1$⁠, and $∼$1.9 $× 10 − 2$  $Ω$ cm, respectively. These values are similar to what was previously reported for films grown by ALD and CVD,^28,29 while a further annealing in In presence was found to be necessary to achieve comparable results in the case of sputtered $Zn ( 1 − x ) Mg x$O:Al layers.³⁰ The decrease in $n$⁠, $μ$⁠, and conductivity with increasing Mg content seen to occur from the $Zn 0.84 Mg 0.16$O:Al sample was already reported and attributed mainly to the increase in the reduced electron effective mass caused by Mg alloying.³⁰ However, as a look at Eq. (6) reveals, the effective mass increase is limited to a factor of $∼$1.2, which is not enough for explaining the observed factor $∼$3 decrease of $μ$⁠. In this respect, it should be noted that the Al content in the samples analyzed varied in the range of $∼$2.1–2.8 at. %, as shown in Fig. 1(b). This corresponds to an average Al concentration of $∼2× 10 21 cm − 3$⁠. Comparison with the $n$ values measured [see Fig. 1(a)] reveals that the Al doping efficiency is $∼$10%–20%, similarly to what was previously reported in the case of ZnO:Al (see, for example, Ref. 31 and references therein).³² Such a low activation ratio is partly specific to ALD grown films and has been attributed to the layered/non uniform Al distribution causing dopant clustering, Al incorporation in secondary insulating species, and/or Al segregation at the grain boundaries and interfaces.^33,34 Indeed, the STEM analysis revealed that the films present a columnar structure with the expected nanolaminate Mg and Al distribution within each column as shown in Figs. 2(a), 2(c), and 2(d) (see also Fig. S2 in the supplementary material). Here, it is worth pointing out that, however, the Al atoms’ distribution in the films is far from the idealized case, i.e., placed only one lattice plane. In fact, it has a full width half maximum (FWHM) of $∼$2 nm in the depth direction with a peak to peak distance of $∼$4 nm, as evidenced by the Al intensity profile extracted from the EDS signal displayed in Fig. 2(e) (for the Mg case, refer to Fig. S2 and the corresponding caption in the supplementary material). This widening of the Al distribution profile has been attributed to the TMA etching effect when reacting with the ZnO-based surface.³³ Furthermore, the DEZ and MCp2Mg steric hindrance could also significantly contribute to limit the layered growth, thus promoting the more random incorporation of Al.³⁵ In addition, in the present work, the depositions were performed outside the ALD window and for that matter, desorption of the main precursors also occurs.³⁶ Overall, the Al distribution FWHM was found to be independent from the Mg introduced as a comparison between Figs. 2(e) and S2(c) and S2(f) in the supplementary material reveals. Moreover, considering that the Al doping efficiency in the case of magnetron sputtered ZnO is $∼$30%–50%, it suggests that the localization related effects roughly account for a $∼$20%–30% deactivation by assuming a similar weight of the remaining mechanisms.³⁷ This, even though, in our case, STEM and XRD measurements did not provide any evidence of precipitates, additional phases or Al segregation, that are, therefore, if present, below the detection limit (see Figs. 7 and S2–S5 in the supplementary material). Another physical mechanism that is anticipated to limit the Al doping efficacy is related to the formation of Al- $V Z n$ complexes.^38,39 Concerning the latter, self-compensation effects involving $V Z n$ were shown to explain similar low Ga doping efficiencies in $Zn ( 1 − x ) Mg x$O layers.⁴⁰ Hence, even though a more detailed study would be required to address firmly the exact interplay between the physical mechanisms above mentioned, this is, most probably, a factor contributing significantly to the reduced Al donor activity also in the samples analyzed here. Then, the $n$ decrease for Mg contents $x>0.14$ suggests a reduction in the formation energy of the compensating/neutral center/s involved in lowering the Al doping efficiency with these defects also contributing to lower $μ$ in the same layers.

Here, it is worth noticing that in the approach used for extracting $Δ E C$⁠, $N a$ and $N d$ are considered uniform as evident from Eqs. (2) and (3) as well as differences between $n$ in the $Zn ( 1 − x ) Mg x O/ Zn ( 1 − x ) Mg x$O:Al films deposited on the soda-lime glass, which are the samples actually measured, and the emitter layers on the realized structures are neglected. Differences in $n$ between the $Zn ( 1 − x ) Mg x O/ Zn ( 1 − x ) Mg x$O:Al emitter layer and the films deposited on the soda-lime glass cannot be excluded, but they are anticipated to be comparable to the variation observed in the case of layers deposited on quartz and soda-lime glass during the same deposition process, which were found to be $≲$20% and thus similar to the measurement errors and therefore not relevant. On the other hand, non-uniformities in the interfacial region cannot be excluded. In detail, if the equivalent of Eq. (2) for heterostructures is used, depth intervals equal to $∼$0.4–1.5  $μ$m and < 0.4 nm in the p-Si and n-Zn $( 1 − x )$Mg $x$O/n-Zn $( 1 − x )$Mg $x$O:Al side of the heterostructure relate to the voltage range used.⁴¹ Non-uniformities in the effective acceptor and donor concentrations on the $p$-Si and $n$-Zn $( 1 − x ) Mg x O/ Zn ( 1 − x ) Mg x$O:Al sides of the heterojunction outside or with length scales larger than the probed depths by C–V measurements would result in an apparent $Δ E C$ since $ϵ F$⁠, as extracted from Hall measurements, would not correspond to the actual $ϵ F$ in the interfacial region with a similar argumentation holding for the Si part of the heterojunction. However, while such effect can be reasonably excluded to occur on the Si side, this might be the case for the $Zn ( 1 − x ) Mg x O/ n-Zn ( 1 − x ) Mg x$O:Al layers considering the limited depth investigated as well as the expected presence of defects in these layers close to the $SiO x$ interface. This represents a potential source of error in the determination of $Δ E C$ according to the procedure used here; this issue is discussed more in detail at the end of the following Subsection.

The results of the analysis outlined above are presented in Fig. 4. The most striking feature is the $∼$0.48 eV increase in $Δ E C$⁠, that is rising from $∼$0.5 eV up to $∼$1 eV following the Al introduction into the $Zn ( 1 − x ) Mg x$O layers, while increasing the Mg content from 0.04 to 0.06. As above mentioned, all the $Zn ( 1 − x ) Mg x O/ Zn ( 1 − x ) Mg x$O:Al layers are degenerate. Then, bandgap narrowing with the corresponding conduction band lowering is indeed expected in the samples. That is, the experimentally extracted $Δ E C$ is carrier concentration dependent. Hence, to derive more general conclusions, $Δ E C$ should be corrected for this contribution as described in the following Subsection.

As discussed in Sec. III A, only 10%–20% of the Al is actually acting as a donor with the remaining not providing free carriers because, most probably, it is involved in the formation of charged deep defects and/or neutral complexes at least partially. Considering that such defects are expected to present smaller capture cross sections with respect to shallow centers, then, their interaction with the conduction band electrons can be neglected, at least in the first approximation. Hence, in the calculations hereafter presented, $N D$ was considered equal to $n$⁠.

The values so obtained are displayed in Fig. 5, while the correction related to the conduction band lowering is shown in the inset.

It can be seen that the extracted $Δ E C 0$ after the abrupt increase of $∼$0.24 eV, as a consequence of Al introduction, was found to approximately linearly decrease with Mg content from $(0.59$ $±0.02)$ down to $(0.31±0.07)$ eV. From a least squares regression fit of the $Zn ( 1 − x ) Mg x$O:Al related data, a linear decreasing factor of $Δ E C 0$ vs Mg content and an intercept equal to $(−1.7±0.2)$ eV and $(0.69±0.03)$ eV were found, respectively.

Here, it is worth noticing that Eq. (10) should be modified in case the $Zn ( 1 − x ) Mg x$O:Al layer is strained and/or formation of an electrical double layer subsequent to the Al introduction, Al segregation at the interface, or polarity effects are present.^65,66 Each of these factors or a combinations of them could potentially explain the observed $Δ E C 0∼0.27$ eV difference between the $Zn ( 1 − x ) Mg x$O value and the expected one according to the $Zn ( 1 − x ) Mg x$O:Al layers’ linear fit. To estimate their contributions first, the morphological and structural properties of the $Zn ( 1 − x ) Mg x$O and $Zn ( 1 − x ) Mg x$O:Al layers were studied further by SEM and XRD with the collected XRD patterns, cross-sectional and front SEM views shown in Figs. 7 and S5(a)–S5(d), respectively [for a full overview of the dependence, the XRD patterns on the Mg atomic content, see Figs. S5(e) and S5(f) in the supplementary material]. Overall, the layers were polycrystalline and exhibited a wurtzite structure. In addition, it can be seen that no secondary phases related to, for example, clusters based on Al and/or Mg compounds, were observed indicating that, if present, their relative amount and/or size is, also in this case, below the detection limit or they are amorphous. In detail, the concomitant appearance of the 11.0 and 10.0 reflections was observed by introducing Al and keeping constant the Mg content with the presence of these peaks being consistent with the observed wedge-shaped crystallite morphology of the surface considering the wurtzite structure (see Figs. S4 and S5 and corresponding captions in the supplementary material for more details). On the other hand, the 10.1 reflection was found to be suppressed by Al introduction. Furthermore, the 00.2 reflection was still clearly detectable for $Zn ( 1 − x ) Mg x$O:Al layers with $≤0.1(6)$ Mg content. As shown in Fig. 7(e), increasing the Mg content further has been found to drastically suppress the growth in the $c$-direction, with the dominating peaks being the 10.0 and 11.0 reflections at the highest Mg content investigated in the present work. Thus, polarity corrections can be excluded in the layers with x > 0.16. The evolution of the lattice constants $a$ and $c$ vs Mg content is shown in Fig. 8. Overall, it was found that, in comparison to the relaxed material, the lattice constants of the not intentionally doped $Zn ( 1 − x ) Mg x$O layers are $∼−0.2%$ and $∼−0.3%$ strained in the $a$ and $c$ directions, respectively.^5,73 On the other hand, differently from what has been previously reported, the Al introduction appears to significantly affect only the lattice spacing in the $a$ direction with an additional $∼−0.1%$ contraction.⁷⁴ Moreover, further addition of Mg is found to shorten the $c$-axis length, while gradually increasing the $a$-axis length with, therefore, a minor impact on the unit cell volume, as expected, considering the similar ionic radius of $Zn 2 +$ and $Mg 2 +$⁠.⁷³ In Fig. 8, the measured values for the $c$-axis and $a$-axis are compared with the expected contraction and expansion reported in the case of $Zn ( 1 − x ) Mg x$O in Refs. 5 and 73. It can be seen that agreement was found between the experimental values and the expected trends. That is, it can be concluded that within the experimental errors, the Al introduction does not alter the expected variation of $a$ and $c$ with Mg incorporation. On the other hand, its impact due to the deformation can be roughly estimated to downshift the conduction band $≲$0.01 eV if the in-plane stress is considered isotropic and 6.05 eV, the conduction band deformation potential evaluated under hydrostatic pressure, is used.⁷⁵ That is, assuming that the observed strain is uniform up to the interface, the increase in the $a$-direction strain cannot account for the observed $∼$0.27 eV $Δ E C 0$ increase when Al is introduced. Second, over the full set of samples analyzed, a maximum downshift of $≲$0.05 eV corresponds to the maximum strain observed. Finally, as shown and discussed above and in the supplementary material, despite the morphological and structural changes related to the Al introduction, the 00.2 reflection is still the dominant one both in the $Zn 0.96 Mg 0.04$O and $Zn 0.94 Mg 0.06$O:Al case with nominally equal Mg content; hence, effects on the bands alignment related to the crystal orientation/polarity are expected to be similar if these two samples are compared. That is, in conclusion, the corrections related to the strain are, in the first approximation, negligible since of the order of the uncertainty on $Δ E C 0$ and polarity effects can be excluded as a main cause of its $∼$0.27 eV increase when Al is introduced.

As mentioned in Sec. II, the first TMA pulse was introduced after a total number of DEZ+DI and MCp2Mg+DI pulses varying in the 13–24 range. This corresponds to a distance from the interface of the first Al-doped layer in the $∼$2–4 nm range assuming a constant growth rate per cycle (GPC), value that is equal to or slightly larger than the Al distribution FWHM observed by STEM. Therefore, overall, an interplay between Al and the interface cannot be ruled out on the basis of the results presented so far even in the absence of interfacial segregation.^37,76 To exclude such an effect, two additional series of samples with the same nominal Mg content (fluctuating in the x $∼$0.18–0.20 range according to EDX measurements) and the first TMA pulse introduced after a total number of DEZ+DI and MCp2Mg+DI pulses equal to 96 and 384 were grown (hereafter labeled as II and III, respectively) and electrically characterized. In these films, considering a constant GPC, the first Al-doped layer is expected to be placed $∼$13 and $∼$57 nm far from the interface in the former and latter cases, respectively. SIMS measurements performed on this series of samples and shown in Fig. S6 in the supplementary material confirmed the presence of the $Zn ( 1 − x ) Mg x$O interlayers with its thickness clearly distinguishable in the case of the III series and equal to $≈$ 60 nm, in agreement with the predicted one and [Al] upper limit $≤ 10 18 cm − 3$⁠. Selected $1/ C 2$ curves used to extract $V d$ are shown in Fig. 9(a) with $n$ of the corresponding $Zn ( 1 − x ) Mg x$O:Al layers deposited on glass displayed in Fig. 9(b). As shown in Fig. 9(b), no dependence of the extracted $Δ E C 0$ on the Al position was observed.⁷⁷ 

In conclusion, no experimental evidence of a possible interplay of Al with the interface was found. In addition, corrections related to strain due to Al and Mg introduction are negligible/within the experimental errors, and polarity effects can be excluded for $x>0.14$ and, if present, their contribution is similar for $x$ $∼0.05$ in layers with and without Al. Overall, these findings rather support the $Δ E C 0$ values extracted in the case of the Al-doped layers, while suggesting that the $Δ E C$s for the undoped samples reported in Fig. 4 are apparent values. Indeed, reiterating the analysis based on Eqs. (3) and (9), it is found that the observed $∼$0.27 eV difference in $Δ E C 0$ between the $Zn 0.96 Mg 0.04$O and $Zn 0.94 Mg 0.06$O:Al layers can be explained if a $∼$0.46 eV downward band bending (⁠ $E b b$⁠) toward the interface with $SiO x$ is present. On the other hand, $ϵ F$ is found in the $∼$0.57–0.47 eV range in the Al-doped layers with $x≤0.14$⁠. That is, $ϵ F$ is large enough for compensating $E b b$ in these samples and, therefore, the extracted $Δ E C 0$s are reliable in this case. Here, it is also worth underlining that, beside variation in the effective $N d$⁠, other factors, like the presence of positive charges at the $Zn ( 1 − x ) Mg x O/ SiO x$ interface (externally introduced), in the SiO $x$⁷⁸ or polarity induced cannot be excluded on the basis of the data presented above and are most probably contributing to $E b b$ considering its relative large value. Furthermore, as shown in Fig. 5 the $Δ E C 0$ values are, within the experimental errors, in agreement with the expected trend also for $x>0.14$ where $ϵ F$ is in the $∼$0.38–0.28 eV range. This suggests a reduced $E b b$ in these samples pointing to an effective $N d$ reduction in the interfacial region and/or lower polarity contribution consistent, at least qualitatively, with the above presented effects of Mg introduction on the layers’ structural and electrical properties. However, further dedicated studies are required to firmly confirm this scenario.

Indeed, the extracted $Φ b p ZnO$ and $Φ b p MgO$ are in very good agreement with the values [ $(2.8±0.1)$ and $(3.9±0.3)$ eV, respectively] obtained by Mönch in Ref. 23 on the basis of experimental data from ZnO- and MgO-based heterostructures to a wide variety of semiconductor counterparts. Moreover, for the holes Schottky barrier height, a value of $∼$2.2 eV has been extracted from x-ray photoelectron spectroscopy (XPS) measurements in the case of metallic $RuO 2$ deposited in situ on ZnO.^79,80 The corresponding $Φ b p ZnO$ is found equal to $∼$2.8 eV applying the equivalent of Eq. (10) for Schottky contacts and using 5.4 Miedema-unit and 3.7 as Ru Miedema’s electronegativity and $ϵ ∞$ for ZnO, respectively.^65,81 Furthermore, the $Φ b p ZnO$ and $Φ b p MgO$ extracted here, considering that, as mentioned above, the intrinsic dipole contribution can be neglected in the case of ZnO/MgO heterojunctions, anticipate $Δ W v$ in the ZnO/MgO junction case equal to $∼$0.9 eV in agreement with the XPS measured value for such heterostructures.⁸² This further supports our $Φ b p$s estimates considering that XPS characterization is the reference technique for band-alignment studies. In addition, it justifies a posteriori the assumption of a linear $Φ b p ZnMgO$ dependence over the whole x range, i.e., Equation (11). That is, the above-mentioned theoretical correction due to the change in crystal structure is essentially within our experimental uncertainty even though it might explain the $∼$0.3 eV lower $Φ b p MgO$ estimate extracted here with respect to the value found by Mönch. On the other hand, on the basis of the analysis of Schottky contacts to ZnO using different metals, $Φ b p ZnO$ was placed $(2.45±0.05)$ eV above the valence band maximum.⁸³ This $∼$0.25 eV difference with respect to the value obtained here is larger than the generally reported $∼$0.1 eV $Φ b p$ fluctuation between values extracted from Schottky contacts and heterostructures²³ and requires further investigations. Overall, despite these fluctuations, it should be noted that the values of $Φ b p ZnO$ and $Φ b p MgO$ found here are $∼$0.6–0.9 and $∼$1.8 eV lower than the theoretical calculated ones for ZnO and MgO, respectively.^21–23 In this context, it is also worth stressing that in the case of ZnO, these calculated estimates have been backed up by XPS measurements that are often extended to other material systems with established branch point energies by using the transitivity rule. This approach fails when the electronegativity contribution in Eq. (10) is significant. Furthermore, possible polarity or reaction effects are implicitly assumed to be negligible if the transitivity rule is used, even though this might not be the case especially when lattice matching substrates/high growth temperatures are used. As an example, an $∼$0.67 eV difference in the ZnO/GaN valence band alignment is found if values originally extracted from ZnO/AlN and ZnO/ $Cu 2$O heterostructures are used.^84,85

As mentioned in the Introduction, the focuses of the majority of the previous attempts to achieve high yield n-ZnO/p-Si based solar cells were the technological procedures, the characteristics of the films deposited, and the photovoltaic response of the devices realized. On the other hand, little attention was paid to the correlation between these parameters and $Δ E C$⁠. To fill this gap and provide a touchstone for further improvements of the performances, the electrically characterized n-ZnO/p-Si based heterostructures were also measured under STC conditions with the extracted active area short circuit currents (⁠ $J S C$s), $V O C$s, and $η$s shown in Figs. 10(a) and 10(b), and in the inset of (b), respectively (for a full overview of the collected data from which they are extracted, see Fig. S7 in the supplementary material). The photovoltaic response of the devices based on $Zn ( 1 − x ) Mg x$O layers as well as $Δ E C$ was found to be close to what has been previously reported in the case of samples realized similarly but using different growth parameters for the emitter.²⁰ Furthermore, $J S C$ was observed to vary in the $∼$33–28 mA/ $cm 2$ range with no clear correlation with the Mg content [see Figs. 10(a), S7(f), and the corresponding caption for more details]. Similarly, no dependence on the Mg content was observed in the case of R $S$ and R $P$ that were found to oscillate in the $∼$1–6 and $≳$ 500  $Ω cm 2$ ranges, respectively (the values multiplied by the device active area were considered in the analysis of the photovoltaic response of the examined samples).⁸⁶ As evident from Fig. 10(b), the not intentionally doped samples presented the lowest $V O C$s that were found to be $≲$340 mV. Note that, as discussed in the previous Subsection, for these samples, the $Δ E C$ reported in Fig. 4 is the apparent one and $E b b$ should be added. This places $Δ E C$ at the same level as the $Zn 0.94 Mg 0.06$O:Al one with the corresponding $∼$40 mV larger $V O C$ observed in this case possibly indicating a slight underestimation of $E b b$⁠. A further comparison between Figs. 11(b) and 4 reveals that the observed $∼$0.1 eV reduction of $Δ E C$ following the increasing Mg content from $x∼$0.06 to $∼$0.09 and from $x∼$0.14 to $∼$0.16 correlates with the $∼$20 and $∼$40 mV larger $V O C$s measured, respectively. In addition, no significant changes in $V O C$ were observed for Mg contents $⪆$0.16 consistent with the extracted $Δ E C$ being constant in the $∼$0.2–0.25 Mg interval. On the other hand, to the $∼$20 mV $V O C$ increase occurring for $x$ in the $∼$0.09–0.14 range, a corresponding plateau in $Δ E C$ was found, even though the errors on $Δ E C$ can mask a possible reduction in this case. As mentioned in the Introduction, such dependence of $J S C$ and $V O C$ on $Δ E C$ is indeed expected. That is, overall the trends found are at least in qualitative agreement with what is expected with a detailed explanation of the possible minor discrepancies being behind the scope of the present work. However, the maximum achieved $V O C$s (⁠ $∼$450 mV) and $η$s (⁠ $∼$7.2%) are, as in the case of previous reports, far below the theoretically predicted ones.¹⁸ To exclude possible geometrical effects and consolidate the found results, full photovoltaic devices were realized [for a front view see Fig. 11(a)] and their photovoltaic response tested. This part of the study was limited to test solar cells with $Zn 0.8 Mg 0.2$O:Al layers since, on the basis of what is shown in Fig. 10 and discussed above, they are anticipated to be among the ones exhibiting the highest $η$⁠. Furthermore, the $Zn 0.8 Mg 0.2$O:Al layer was deposited at 280 and 300  $°$C since an improvement in the photovoltaic response by increasing the deposition temperature was reported in the previous published work.²⁰ 

Using for $E(λ)$ the AM1.5 global spectrum ASTM G-173-03 with an integrated power of 1000  $W / m 2$ corresponding to the illumination conditions, $J S C q e$ was found equal to $(29.0±0.1) mA / cm 2$ in both cases. The substantial agreement between $J S C$s and $J S C q e$s not only confirms the high $J S C$s found here and in similar structures²⁰ but also excludes a significant contribution from micro-shunts that act as a shunting load if the cell is irradiated only on a limited area as in the case of EQE measurements.⁶⁶ Within the analyzed samples, the $V O C$⁠, fill factor (⁠ $FF$⁠), and $η$s were observed to vary in the $(430±20)$ mV, $(61±2)$%, and $(7.2±0.3)$%, respectively. On the other hand, R $S$ and R $P$ were found equal to $(3±1)$ and $≳$1000  $Ω cm 2$⁠. That is, the realized solar cells exhibited a photovoltaic response similar to the test devices. In addition, no significant differences were found between the samples with the $Zn 0.8 Mg 0.2$O:Al layer deposited at 280 and at 300  $°$C. This finding is indeed consistent with the similar XRD patterns measured (as shown in Fig. S8 in the supplementary material) in the two cases indicating the absence of detectable structural changes differently to what is occurring when lower deposition temperatures are used.²⁰ Finally, overall, the photovoltaic response of the test solar cells were found to be stable after 7-month storage in air at RT. With respect to previously reported characteristics of structures realized with a similar procedure, the present ones exhibited similar $J S C$s but a $∼$100 mV larger $V O C$⁠,²⁰ thus representing a considerable improvement. Here, it is worth pointing out that, as mentioned in the Introduction, even though higher values of $V O C$ up to $∼$540 mV have been reported, overall, these values are anyhow well below the $∼$620–660 mV range indicated by theoretical simulations or the highest $V O C$s achieved in the case of HJTs solar cells that are in the $∼$0.71–0.75 V range.⁷ The same holds for the $FF$ that is reported to be $≳$10% lower than the expected one.^16,18,19 In our case, as shown in Fig. 4, the measured $Δ E C$ is $(0.61±0.02)$ eV for Mg contents varying in the (0.2–0.24) range with the conduction band lowering contribution due to the many body effects being $(0.31±0.02)$ eV, that is, accounting for $∼$50% of the total. Hence, lower than simulated, $V O C$ as well as $FF$ are indeed anticipated.¹⁸ In this respect, however, it is worth noticing that the relatively high $R S$s of the realized test solar cells contribute significantly in reducing the latter as well.⁸⁸ Furthermore, the fact that, as mentioned above, considerably lower $V O C$ and $FF$ are experimentally obtained even for devices that are optimized for photovoltaic purposes suggests that, also in these cases, a staggered type II band alignment was among the main factors limiting the obtained $η$s that have been reported to be $≥$10% lower than in the HJTs solar cells case.^7,16 Finally, here it is worth pointing out that the $Δ E C 0$ value for the n-ZnO/ $SiO x$/p-Si heterostructures of $(0.69±0.03)$ eV as well as the $Δ E C 0$ reduction vs increasing Mg content of $(1.7±0.2)$ eV extracted here are considerably different with respect to the ones used in the previously published simulations that were set to 0.3 and 2.7 eV, respectively.¹⁸ Thus, while the trends are preserved, the absolute values extracted from the simulations are overestimates for a specific Mg content and the theoretically calculated $V O C$ as well as $η$ represent indeed upper limits, unless further mechanisms for reducing the conduction band misalignment between the ZnO-based layer and the Si substrate are found.

In this work, a detailed study of the conduction band alignment between nominally undoped and Al-doped $Zn ( 1 − x ) Mg x$O layers and (100) p-Si in the presence of native $SiO x$ has been presented. Considering that even the nominally undoped $Zn ( 1 − x ) Mg x$O layers were degenerate and carrier concentration up to $4× 10 20 cm − 3$ could be achieved by introducing Al while keeping the Mg content $x<0.16$ conduction band lowering due to the many body effects was anticipated to have a significant impact on $Δ E C$⁠. By correcting the measured $Δ E C$ for this contribution taking into account the conduction band non-parabolicity of ZnO, the expected linear increase of $Δ E C$ vs Mg content was recovered with the extracted proportionality factor being equal to $(1.7±0.2)$ eV. From this dependence on the $Zn ( 1 − x ) Mg x$O:Al layers composition considering the presence of the $SiO x$ interlayer that accounts for 20% of the bands’ misalignments, and under the assumption of a linear variation between the respective values of the corresponding two binary compound semiconductors, ZnO and MgO, branch point energies for ZnO and MgO equal to $(2.7±0.2)$ and $(3.6±0.4)$ eV were extracted, respectively. Full 1  $×$ 1 cm test solar cells based on $Zn 0.8 Mg 0.2$O:Al layers exhibited short circuit currents, open circuit voltages, fill factors, and efficiencies fluctuating in the $(28±1) mA / cm 2$⁠, $(430±20)$ mV, $(61±2)$%, and $(7.2±0.3)$% ranges independent of the growth temperature that was varied in the 280–300  $°$C range. Besides further optimization of this kind of structures, the present study evidences that, even with $∼$12 at. % Mg content, the residual $Δ E C∼$0.61 eV can be reduced only down to $∼$0.30 eV by decreasing the carrier concentration in the $Zn ( 1 − x ) Mg x$O:Al layer, that is, by minimizing the conduction band lowering due to the many body effects and the electron–donor interaction. Furthermore, increasing the Mg content up to $x∼$0.35, the maximum Mg content reported for the $Zn ( 1 − x ) Mg x$O wurtzite phase would result still in a type II band alignment with $Δ E C∼0.11$ eV. Therefore, alternative routes should be investigated to further improve the photovoltaic performances of devices based on $n-Zn ( 1 − x ) Mg x$O:Al/ $SiO x$/p-Si.