Review Article: Atomic layer deposition of doped ZnO films

Atomic layer deposition (ALD) exerts the ultimate control over film thickness and composition at the nanoscale. The sequential pulsing of precursor molecules leads to self-limited chemisorption from the gas phase on to surfaces and results in monolayer growth of films. Mixing and matching pulse chemistries results in new tailor-made compositions, harder to achieve by other deposition techniques. These process advantages in ALD have been realized through more than two decades of fundamental and applied research and highlighted in numerous review articles.^1–15 Today, ALD is a manufacturing-at-scale process, adopted by industry and projected to reach a market value of $3.01 billion by 2025.¹⁶ 

ZnO is a wide bandgap (3.30 eV) semiconductor that is a versatile materials platform used in thin film electronics, gas sensors, light-emitting devices, photodiodes, solar cells, and catalysts.^17–22 These applications are made possible by a rich compositional diversity achievable by doping ZnO with various elements. For example, Mg-doped ZnO and Cd-doped ZnO are two compositions typifying the use of doped ZnO for bandgap engineering in optoelectronic devices.^23,24

Thus, given (1) the process advantages of ALD and (2) the material importance of ZnO, ALD of ZnO presents immense opportunities for future devices and technologies. ALD chemistry for ZnO was first demonstrated by Tammenmaa et al. using zinc acetate and water as precursors.²⁵ Since then, a large body of work has been published, detailing fundamental mechanisms of ALD of ZnO using diethyl Zn (DEZ) as a precursor and the ability to dope ZnO using a combination of various precursor molecules.^26–32 There are 22 elements that have been used to dope ZnO via ALD. This is shown in Fig. 1(a). Applications of ALD doped ZnO have included thin film transistors, solar cells, light-emitting diodes, and gas sensors.^33–36 The main motivation for developing ALD doped ZnO as a transparent conducting oxide (TCO) is to replace currently used TCOs such as tin-doped indium oxide (ITO).

Given that ZnO is a wide bandgap semiconductor metal oxide, it is easy to appreciate how doping can create excess charge carriers in ZnO while maintaining optical transparency.^37,38 Group III oxides and, in particular, Al₂O₃ is used to synthesize Al-doped ZnO TCOs [Fig. 1(b)]. The Kröger–Vink reaction can be represented as

The two electrons generated in the above reaction are donated to the conduction band. This results in improved electronic conductivity. One can extend the rationale to group IV (e.g., TiO₂) and group V oxides (e.g., Ta₂O₅), assuming that a higher number of electrons may be available for conduction. However, the actual scenario is more complicated than what the above reaction implies. The presence of intrinsic defects such as O vacancies, Zn interstitials, and hydrogen and charge compensation mechanisms create a complex electronic environment that has been extensively studied in the literature.^39–44 

From a process perspective, the key takeaway is that ALD represents a powerful platform to realize a range of doping compositions by (1) elemental type and (2) atomic percent (at. %), which allows for a systematic investigation of process-structure-properties relationships of doped films. By varying the order and ratio of pulsing sequence, the entire compositional phase space can be mapped, and fundamental processes related to doping mechanisms can be unraveled. If the deposition rate (in nanometers/cycle) of the parent film and dopant is known, the estimated atomic percent dopant can be written as

Here, $λMOx$ and λ_ZnO are deposition rate per ALD cycle for the dopant oxide MO_x and ZnO, respectively, and Γ is the cycle ratio of ZnO to MO_x. The above formula assumes no chemical “cross talk” between the parent and dopant films and is an oversimplification. Yet, a majority of papers use similar formulae for estimating atomic percent in ALD doped ZnO films. Because of the ease with which ALD doped ZnO can be fabricated and their previously stated role in key applications, there now exists a wealth of literature that needs to be collectively assessed.

Therefore, the aim of this paper is to review current data and extract underlying trends in ALD doped ZnO films that may aid in our understanding to design and synthesize the next generation of doped ZnO films and other multicomponent films via ALD. Table I provides a comprehensive list of ALD doped ZnO films from the perspective of TCOs.^45–207 This table lists the doped element, the precursor molecules used, the composition (or cycle ratio) at which the TCO properties are optimized, the thickness deposited, and the optical and electrical properties.

With a view to aid practitioners of ALD design better processes, this review focusses on the processing-structure-properties relationships in ALD doped ZnO films. In Sec. II, we describe the process parameters that are important for producing efficiently doped ALD ZnO films. Next, the characterization of these films is discussed. In particular, x-ray diffraction (XRD), x-ray photoelectron spectroscopy (XPS), and photoluminescence (PL) data are discussed for various ALD doped ZnO films. Finally, optical and electrical properties are highlighted. Where there is availability of extensive data, generalized trends in ALD doped ZnO films are presented. At the end, limitations in the processing schemes, knowledge gaps in the compositional maps, and perspectives on the future of ALD doped ZnO are described.

ALD process parameters strongly dictate properties of doped ZnO. In this section, the effect of pulsing sequence, deposition temperature, and choice of dopant precursor molecules is considered.

Dopant distribution can be precisely tuned along the thickness direction of an ALD film. Injecting a dopant pulse after a few, fixed cycles of Zn and oxidant precursor molecules is the most common method for synthesizing doped ZnO.^27,89,207 The entire sequence, or “super-cycle,” is then repeated to achieve the desired thickness. This is schematically shown in Fig. 2(a) for the cation Al³⁺ doped in ZnO, although the efficiency of doping of anions such as N is known to be sequence dependent as well.¹³⁴ By considering the “pre” dopant pulse (Zn or oxidant, “O”), the dopant pulse, and the “post” dopant pulse (Zn or O), one can see that there are at least four (i.e., 2 × 2) combinatorial ways of achieving the pulse sequences. These sequences are represented from n₁ to n₅ in Fig. 2(b), based on the recent work by Le Tulzo et al.¹²⁷ Sequence n₁ has two consecutive oxidant pulses that differs from sequence n₅ which has only one postdopant H₂O pulse. Additionally, we have included the pulse sequence “co,” i.e., a co-injection pulse where both DEZ and dopant are pulsed into the reactor simultaneously.⁹⁰ 

There are marked differences in electrical properties observed between these pulse sequences. We specifically discuss the widely published case of Al-doped ZnO.¹²⁷ According to Le Tulzo et al., the sequences n₁ and n₅ described above, where the DEZ pulse immediately precedes a trimethyl aluminum (TMA) pulse, create lower resistivity films for films deposited at 200 °C and a TMA:DEZ cycle ratio of 1:10. This has been suggested to be due to the etching of TMA on ZnO surfaces, which leads to superior Al incorporation inside the ZnO film. On the other hand, Pollock and Lad⁹⁹ have shown that sequence of type, n₂ lowers film resistivity for films deposited at 160 °C with a TMA:DEZ cycle ratio of 1:11. Additionally, the Al-doped ZnO film crystallinity and texture is enhanced along the [002] direction. It is also suggested that the DEZ step after TMA helps to localize and bind the defect site produced as a result of prior TMA etching. Variation of pulse sequence has not been tried for dopants other than Al.

In contrast to periodically layered dopants, homogeneous dopant distribution can also be achieved using ALD.^162,202 The co-injection pulse sequence is labeled as “co” in Fig. 2(b). Yuan et al.⁹⁰ have achieved co-injection by tuning the partial pressure of DEZ and TMA to maintain an appropriate mixture ratio during the precursor pulse. For similarly doped Al atomic percent, the homogenously doped films show better crystallinity and mobility as compared to the periodically doped films. Improved electrical performance of homogeneous Hf-doped ZnO has been reported as well.¹⁸² 

Understanding interaction effects of the precursors with the film come from both in situ and ex situ analyses of processes and films. In situ quartz crystal microbalance,^27,87 quadrupole mass spectrometry,²⁰⁷ and in situ conductance measurements⁸⁷ provide insights into how the dopant pulse interacts with the film. Ex situ characterization of films using TEM has shown the quasi-superlattice-like, periodically layered microstructure as shown in Fig. 2(c).⁹² Auger electron spectroscopy can determine the periodicity of the dopant layers [Fig. 2(d)].⁹⁰ Recently, the Kessels group¹²⁵ has shown the periodicity of Al-doped films using atom probe tomography [Fig. 2(e)]. Surprisingly, the Al concentration profile along the depth of the film is not as discrete as previously thought and shows considerable mixing with the ZnO matrix.

Ideally, the dopant pulse should not interact with the underlying film. However, in practice there is significant chemical “cross talk” between the dopant pulse and the underlying film. Elam et al. were the first to show the impact of TMA etching on ZnO films. In Fig. 2(f),²⁷ the thickness difference of an AZO film between measured and estimated values based on the “rule of mixtures” is shown. The measured thickness of Al-doped ZnO is lower than the calculated thickness. As the Al percentage increases, the difference between measured and calculated thicknesses increases until the Al cycle ratio reaches 20% (i.e., 80% ZnO cycle). This thickness difference is a result of the etching effect of TMA on ZnO surfaces. Incidentally, trimethyl indium shows a similar effect as well.¹⁶⁶ Phenomenologically, it has been shown that it takes ∼6 cycles of DEZ + H₂O to recover 63% of the original ZnO deposition rate.⁸⁹ The etching mechanism is responsible for the Al incorporation into the ZnO matrix. Recent data show that a cation exchange mechanism, where Al substitutes Zn in the film, is possible.²⁰⁸ 

In contrast to TMA, Ti-doped ZnO using tetrakis-isopropoxide (TTIP) results in enhanced growth rates.¹⁸⁵ This is suggested to be due to the enhanced adsorption of TTIP molecules on ZnO surfaces. However, the end result is similar to the TMA case, where excess Ti is detected beyond the rule of mixtures.

ALD of ZnO can be performed across a wide temperature window. For example, the lowest temperature for ZnO deposition is reported to be 60 °C using DEZ and H₂O as co-reactants.^71,209 The highest temperature of ZnO deposition is reported to be 300 °C using DEZ and H₂O as co-reactants.¹⁶⁸ The dopant incorporation must be activated between these two extrema in temperature. However, this is not the only consideration when choosing a suitable deposition temperature as highlighted in Fig. 2(g).⁸² Here, the change in resistivity is shown as a function of deposition temperature. It is noted that the nominal temperature for ALD ZnO and Al₂O₃ using H₂O as an oxidant is around 200 °C.^27,89,209 However, in ALD Al-doped ZnO, the lowest resistivity sample is synthesized at 250 °C. The higher deposition temperature enhances the removal of precursor ligands and carbonaceous species from the film. Postdeposition annealing to temperatures as high as 550 °C could further improve film resistivity by redistributing dopant atoms uniformly in ZnO and simultaneously increasing the grain size of the film.^90,105,158

Dopant incorporation and hence film properties are affected by preconditioning the surface prior to the dopant pulse or carefully choosing different precursor molecules. Essentially, the isotropic and homogeneous distribution of dopants is a key factor for improved transparency and conductivity of ALD doped ZnO films. The dopant “effective radius model” can help explain how surface poisoning and homogeneous doping distribution affect the electrical property and the doping efficiency.⁹² Here, we focus on ALD process strategies for achieving isotropic and homogeneous doping.

“Surface poisoning” using alkyl alcohols or passivation using titanium tetra-isopropoxide²¹⁰ after each TMA pulse reduces density of surface reaction sites.⁹³ This implies that less Al is incorporated in each layer but counterintuitively helps to increase the carrier density of an Al-doped ZnO film [Fig. 2(h)].⁹³ Alternately, using a larger Al precursor molecule, such as dimethyl-aluminum isopropoxide (DMAI), can also affect Al-doped ZnO film resistivity.⁹⁵ DMAI molecules are much larger than TMA and have a larger steric hindrance when attached to the ZnO surface. This leads to less Al concentration in a single layer. The doping efficiency using DMAI is much higher than TMA [Fig. 2(i)].⁹⁵ 

In this section, the impact of extrinsic dopants on the structural characteristics of ALD doped ZnO films will be presented. Dopant effects on ZnO crystallinity, chemical bonding, and defects will be discussed while presenting x-ray diffraction, x-ray photoelectron spectroscopy, and photoluminescence data, respectively.

There are two primary growth directions for ALD ZnO. ZnO, which has a wurtzite crystal structure, has a (002) plane which is a charged, polar surface due to alternate layers of Zn²⁺ and O²⁻ along the c-axis. The a-axis (100) plane is a charge neutral surface that consists of alternate rows of Zn²⁺ and O²⁻ [Fig. 3(a)].⁹² Atomic layer epitaxy of ZnO on (0002) sapphire produces a c-axis oriented film. Alternately, it has been shown that at low temperatures, ALD of intrinsic ZnO leads to a preferential growth of the (100) plane, while at high temperatures the (002) plane is favored.²¹¹ This is attributed to the distribution of surface charge which in turn determines surface energies of the crystal planes during the sequential surface reactions in ALD. Thus, the prevalence of either the charged (002) or neutral (100) plane and their subsequent presence as a function of dopant concentration provide insights into the effectiveness of dopant incorporation in ALD doped ZnO.

To understand dopant behavior on crystallinity, we look at ALD doped ZnO films deposited on glass or thermal SiO₂ on Si, except for the case of N-doped ZnO which is conducted on sapphire. Substrate effects are not considered in our analysis, although it has been shown that carefully chosen substrates can enhance conductivity.⁸⁸ 

We have analyzed the effective ionic radius²¹² r under tetrahedral coordination and the charged state (+Z) of the dopants used in ALD ZnO. The linear charge density which can be a measure of the strength of interaction of the cation dopant with the ZnO matrix can be approximated by the parameter “Z/r,” that is, the charge density per unit length of the ionic radius. Higher this parameter, stronger the interaction. The linear charge density parameter is plotted as a bubble graph in Fig. 3(b) for Zn²⁺, Mg²⁺, In³⁺, Ga³⁺, Zr⁴⁺, Hf⁴⁺, Al³⁺, Ta⁵⁺, Ge⁴⁺, Nb⁵⁺, Si⁴⁺, and B³⁺, where the size of the bubble equals 1000 × Z/r. Independent of the effective ionic radius (i.e., y-axis), there are two regions that can be demarcated by the bubble size. The bubbles of Type I indicate linear charge densities of cations with 1000 × Z/r > 90, significantly higher than Zn²⁺(1000 × Z/r = 33). The bubbles of Type II indicate linear charge densities of cations with 1000 × Z/r ≤ 77 and similar to Zn²⁺(1000 × Z/r = 33). This demarcation using linear charge density is a useful way to analyze the XRD data as will be shown below.

In Figs. 3(c)–3(g), XRD of doped ZnO films as a function of atomic percent of type I cations is shown. In all the cases, the crystallinity of the ZnO matrix monotonically degrades as atomic percent doping increases for Nb⁵⁺ (Ref. 205), B³⁺ (Ref. 102), Si⁴⁺ (Ref. 184), Ta⁵⁺ (Ref. 207), and Ge⁴⁺ (Ref. 178). The XRD peaks shift to higher 2θ, implying that lattice constants decrease with atomic percent. The peak intensities decrease and the full-width-at-half-maximum (FWHM) widens. Irrespective of the effective ionic radius, the strong electrostatic interaction of these cations disturbs the charge state of the ZnO matrix and hinders crystalline growth of doped ZnO films.

In contrast, in Figs. 3(h)–3(m), XRD of doped ZnO films as a function of atomic percent of type II cations is shown. Here, the crystallinity of the ZnO matrix initially improves. Peak intensities increase and FWHM narrows, before film crystallinity starts to degrade. For example, for Al³⁺, the primary peak intensity maximizes at 3 at. %,⁸⁹ Hf⁴⁺ at 6.7 at. %,¹⁸⁰ Ga³⁺ at 1.1 at. %,¹⁵⁰ Mg²⁺ at 10 at. %,²⁰⁴ Zr⁴⁺ at 6.7 at. %,¹⁷³ and In³⁺ at 4.7 at. %.¹⁶⁶ It is noted that the linear charge density (Z/r) of the type II cations is similar to Zn²⁺. Thus, the type II cations initially improve crystallinity of the films. At high dopant percent, past solubility limit, interconnected islands of amorphous or secondary phase oxides in the ZnO matrix, degrade ZnO crystallinity.¹⁸⁰ 

For anionic dopants such as F⁻ (Ref. 194), Cl⁻ (Ref. 202), N³⁻ (Ref. 142), and S²⁻ (Ref. 162), the trends in crystallinity differ from cationic dopants.¹⁴² The anionic dopants are incorporated in the oxygen vacancies (⁠$VO∙∙$⁠) or substitute oxygen sites. In general, peak intensities shift to lower 2θ degrees, indicating the presence of tensile stress and an increase in lattice constants. For F⁻ doping [Fig. 3(n)],¹⁹⁴ a change in crystallinity is not observed though film texture appears to change with increasing atomic percent doping. Since the ionic radii of F⁻ (1.31 Å) and O²⁻ (1.38 Å) are almost the same, no change in the lattice constants is observed. For Cl⁻ doped ZnO [Fig. 3(o)],²⁰² the crystallinity of the film remains intact as Cl⁻ concentration increases to a maximum of 0.65 at. %. The crystalline orientation of the film changes from a-axis to c-axis. Further, the c-axis and a-axis lattice parameters increase. This is expected as the ionic radius of Cl⁻ is 1.81 Å, while O²⁻ is 1.38 Å. For N³⁻ doping [Fig. 3(p)],¹⁴² the peak intensity appears to diminish rapidly for post rapid thermally annealed films and at 12.9 at. %, formation of Zn₃N₂ is observed. A shift of the peak to lower 2θ is observed. Finally for S²⁻ doping [Fig. 3(q)],¹⁶² the entire phase space from ZnO to ZnS has been explored. Pure ZnO starts as wurtzite with diminishing peak intensities and shifts to lower 2θ as S atomic percent increase. Fully amorphous films are observed for >31 at. %. However, at 77 at. %, ZnS (111) with zinc blende structure is observed.

XPS is a powerful surface analysis technique to study the oxidation state of the Zn and dopant cation in doped ZnO. By studying the perturbation of the charged state (i.e., “2+” for the Zn cation), one can understand the degree to which the dopant is incorporated into the ZnO matrix. Alternately, by studying the XPS fine spectra of oxygen, the impact of dopants on the anionic sublattice can be understood. In particular, the variation of $VO∙∙$ can be monitored as a function of dopant concentration. This is important because $VO∙∙$ determines, to varying degrees, electrical properties of doped ZnO films. Herein, we will present XPS data as a function of dopant atomic percent for the Zn 2p and O 1s fine spectra.

In pure ZnO, Zn 2p_3/2 is between 1021.8 and 1022.5 eV and Zn 2p_1/2 is 22.97 eV higher than 2p_3/2.²¹³ We find that there is a consistent trend in the Zn peak when a dopant is added via ALD. This trend depends on the oxidation state of the dopant cation. In Fig. 4(a), the shift in the binding energy (BE) is plotted (z-axis) as a function of dopant atomic percent (x-axis) and the oxidation state (Z) of the dopant cation (y-axis). The dopant cations for which data are available as a function of atomic percent are Z = +2 for Mg;²⁰⁴ Z = +3 for Al and In;^31,119 Z = +4 for Zr, and Z = +5 for Ta and Nb,^205,207 whereas, for Z = +2 and Z = +3, the Zn 2p peaks do not shift in their B.E. (i.e., shift = 0 eV), for Z = +4 and Z = +5, the shifts to higher BE are measurable. This implies that Zn²⁺ loses part of its valence electrons in the presence of these highly charge cations, namely, Zr⁴⁺, Ta⁵⁺, and Nb⁵⁺. An example of Zn 2p_3/2 peak unperturbed with increasing dopant concentration is presented in Fig. 4(b) for Al-doped ZnO.¹¹⁹ Alternately, in Fig. 4(c), Ta-doped ZnO shows a consistent trend of peak shifts to higher BE as the atomic percent Ta increases.²⁰⁷ Thus, based on the results above, we find that the Zn 2p spectra responds to the dopant cation for Z ≥ +4.

O_1s spectrum in ZnO XPS can be deconvoluted^{119,180,195,202,205,207} into three peaks, namely, O associated with ZnO (i.e., O²⁻) at 529.6 eV, O in the vicinity of $VO∙∙$ at 531.8 eV, and adsorbed oxygen at 532.8 eV. For Al-doped ZnO, systematically increasing Al concentration leads to an initial increase of $VO∙∙$ till 3.41 at. % [Fig. 4(d)].¹¹⁹ A subsequent increase in Al concentration leads to a decrease in $VO∙∙$⁠.¹¹⁹ Al³⁺ creates more $VO∙∙$ at low doping and this helps in improving film conductivity. However, other cation dopants have a different influence on $VO∙∙$ in ZnO. For example, in Ta-doped ZnO the strong $VO∙∙$ initially observed in intrinsic ZnO [Fig. 4(e)] is clearly reduced at 5.2 at. %. As the Ta percentage further increases, the shoulder defect peak reappears.²⁰⁷ In Nb-doped ZnO shown in Fig. 4(f), the $VO∙∙$ decreases from bulk ZnO to 3.8 at. % and then increases for 12.6 at. % Nb doping.²⁰⁵ For Hf-doped ZnO shown in Fig. 4(g), the concentration of $VO∙∙$ reaches its lowest value at 6.7 at. % and increases again as Hf concentration increases.¹⁸⁰ One possible explanation about differences in vacancy behavior between Al³⁺ compared to Ta⁵⁺, Nb⁵⁺, and Hf⁴⁺ dopants could be that the corresponding oxides of these cations, Ta₂O₅, Nb₂O₅, and HfO₂, provide a more oxygen rich environment than Al₂O₃. Thus, the intrinsic $VO∙∙$ associated with ZnO is annihilated by use of the higher oxidation state (Z ≥ 4+) dopants. However, at higher concentrations of these dopants, density functional theory calculations predict acceptor like metal vacancies to form.²⁰⁷ The acceptor like nature of these defects implies their energy levels to be closer to the valence band of ZnO, and thus the defects perturb the O orbitals.

For anion dopants, the changes to $VO∙∙$ are straightforward. In both Cl-doped ZnO and F-doped ZnO, the concentration of $VO∙∙$ monotonically decreases with increasing Cl and F concentration. The fine spectra and the trends for Cl-doped ZnO are shown in Figs. 4(h) and 4(i), respectively.²⁰² The fine spectra and the trends for F-doped ZnO are shown in Figs. 4(j) and 4(k), respectively.¹⁹⁵ The likeliest explanation for this behavior is that the direct substitution of $VO∙∙$ by Cl or F in ZnO leads to the annihilation of $VO∙∙$⁠.

Photoluminescence spectroscopy is another way to investigate ALD doped ZnO films. The near band edge (NBE) emission of ZnO at 385 nm (=3.22 eV) corresponds to its direct bandgap. The sharpness of this peak reflects the crystallinity of the doped ZnO film. Shift in the peak position represents changes to the bandgap. At the same time, the broad emission on the low energy side correlates with the defects in the ZnO due to the doping process. Thus PL, in conjunction with XRD and XPS, can be used to study the structure of doped ZnO films and its impact on optical and electronic properties.²¹⁴ Figure 5 shows PL spectra of ALD doped ZnO with different dopants. For cationic dopants and specifically in Fig. 5(a) for Al-doped ZnO,⁹¹ Fig. 5(b) for Hf-doped ZnO,¹⁸⁰ Fig. 5(c) for Zr-doped ZnO,¹⁷³ and Fig. 5(d) for Ta-doped ZnO;²⁰⁷ as the doping percentage increases, the intensity of NBE of ZnO is quenched and the peaks become broader. The broader shape of NBE indicates a degradation in film crystallinity. This crystallinity change due to the high percentage of cation doping is in line with the XRD data, though a quantitative analysis is harder to interpret. Furthermore, as the cation doping percentage increases, the NBE position shifts to higher energy.

Besides NBE emission, green band emission (502 nm, 2.47 eV) is also observed. The green band emission is suggested to be a result of $VO∙∙$⁠.^41,42 In Hf-doped ZnO [Fig. 5(b)] and Ta-doped ZnO [Fig. 5(d)], the green band emission decreases first at low doping percentage. As the doping percentage is increased, the green band emission reappears again. This phenomenon indicates that $VO∙∙$ is annihilated first, when the cation dopants are introduced into ZnO. Continued increase in the cation concentration leads to regeneration of defects.

For anion dopants, Cl-doped ZnO [Fig. 5(e)]²⁰² and F-doped ZnO [Fig. 5(f)],¹⁹⁴ the change of NBE is not as strong as for the cationic dopants. Similarly, there is little change in NBE observed in ALD F and Ga co-doped ZnO in Fig. 5(g) with varying F⁻ atomic percent.²¹⁵ However, the quenching of NBE in N-doped ZnO is stronger than Cl-doped ZnO and F-doped ZnO as shown in Fig. 5(h).²¹³ One hypothesis for stronger NBE quenching of N-doped ZnO would be the high negatively charged state of the N³⁻ compared to Cl⁻ and F⁻ which could strongly perturb the anionic lattice of ZnO.²¹³ 

The substitution of oxygen by chlorine and fluorine causes the intensity of $VO∙∙$ to reduce to the minimum as shown in Figs. 5(e)–5(g). The higher the doping percentage, the lower the intensity of $VO∙∙$ peak. This reduction of $VO∙∙$ in anion doped ZnO is similar to XPS data presented earlier. The effect of N-doping on the defect band has not been reported.

Thus, the cation dopants and anion dopants behave differently in the ZnO matrix and affect crystalline quality and defect. Whereas the cations degrade the crystalline quality, anions such as the halides do not affect the crystalline quality. On the other hand, cations reduce $VO∙∙$ concentration at low atomic percent doping, before increasing defect concentration at high atomic percent doping. Anionic dopants universally reduce $VO∙∙$⁠.

Optical properties of ALD doped ZnO films will be discussed using UV-vis spectroscopy data. The electrical properties consisting of resistivity, carrier concentration, type, and mobility will be discussed presenting data from Hall measurements.

The bandgap is an important TCO property to ensure that ALD doped ZnO is transparent to incoming visible light (400–800 nm) while maintaining sufficient electrical conductivity. Thus, optical bandgap and transmittance of the film are the two main parameters of interest. These data are obtained via UV-vis spectroscopy of the ALD films, which measures the spectral response of optical transmittance. The resulting Tauć plot for direct allowed transitions (as is the case for ZnO) yields the bandgap.²¹⁶ The room temperature direct bandgap of ALD ZnO is reported to lie between 3.19 and 3.30 eV.^{89,102,146,157,162,172,178,180,184,186,194,202,204,207} This corresponds to a photon wavelength of 390–377 nm. As a result, ZnO is transparent across the entire visible spectrum. The addition of dopant atoms leads to a change in the bandgap. This change is a function of dopant type and the atomic percent doping.

Because ALD can systematically dope ZnO, trends of bandgap change with atomic percent doping can be obtained. Here, it is noted that the Burstein–Moss effect, which causes apparent bandgap increase due to increased carrier concentration, will be discussed under the section on electrical property. In the absence of any phase segregation, the change in bandgap is a continuous function of composition and can be empirically modeled as²¹⁷ 

where E_g(c) is the bandgap of the alloy semiconductor, c is the atomic fraction of the alloy AO_x, and b is the bowing parameter that depends on the electronegativities of ZnO and AO_x.

In Fig. 6(a), conduction and valence band edges and bandgap of ZnO and the cationic oxides and anionic Zn compounds are compared. The bandgap (E_g ∼ 1.23 eV) of Zn₃N₂ is noted to be particularly different. The bandgap of ZnS ∼3.56 eV, ZnF₂ between 7 and 8 eV, ZnCl₂ ∼3.74 eV, MgO ∼7.8 eV, Al₂O₃ ∼8.7 eV, Ga₂O₃ ∼4.85 eV, SiO₂ ∼8.9 eV, GeO₂ between 5.35 and 6.00 eV, ZrO₂ ∼5.1 eV, HfO₂ ∼5.9 eV, and Ta₂O₅ ∼3.9 eV, which are larger than ZnO.^218–227 Because the conduction band edge position of ZnF₂ and B₂O₃ is unclear, the band structure of these is not shown.

In order to observe the effect a dopant has on the ZnO bandgap, we plot the bandgap change as a function of atomic percent doping. Figure 6(b) shows the atomic percent doping from 0% to 10% for Cl⁻ (Ref. 202), F⁻ (Ref. 194), B³⁺ (Ref. 102), Si⁴⁺ (Ref. 184), and Zr⁴⁺ (Ref. 172). All dopants, except for Si⁴⁺, show a linear increase in bandgap with atomic percent. For Si⁴⁺, the peak in bandgap change (from 3.25 to 3.55 eV at 2.1 at. %) is attributed to a maximum in the carrier concentration. Subsequently, as the atomic percent Si⁴⁺ increases, the carrier concentration drops and a drop in bandgap is also observed. This is attributed to the Burstein–Moss effect.

In Fig. 6(c), the effect of S²⁻ (Ref. 162), N³⁻ (Ref. 146), Mg²⁺ (Ref. 204), Al³⁺ (Ref. 89), Ga³⁺ (Ref. 157), Ge⁴⁺ (Ref. 178), Ti⁴⁺ (Ref. 186), Hf⁴⁺ (Ref. 180), and Ta⁵⁺ (Ref. 207) on ZnO bandgap is shown and the x-axis scale varies from 0 to 100 at. %. Because of the completeness in data, we take the case of S²⁻, first.¹⁶² The two compositional edges of ZnO and ZnS were found to have bandgaps of 3.2 and 3.4 eV, respectively. From the XRD shown previously, the films undergo a transformation from wurtzite (ZnO) to amorphous (ZnO + ZnS) to zinc blende (ZnS) structure. Frijters et al.¹⁶² have used appropriate adjustments in their Tauć plots to account for these structural changes. Using the formula presented above, a good fit to the variation of bandgap as a function of S doping (c) can be obtained. This is given as E_g(c) = 3.20(1 − c) + 3.4c − 3.02(1 − c)c. N³⁻ addition to ZnO is the only other dopant that produces a decrease in bandgap. This is to be expected as the bandgap of zinc nitride (Zn₃N₂) is 1.23 eV.²²⁷ 

The case for Hf⁴⁺ deserved attention as well. Here, it is reported that the bandgap initially increases and reaches a maximum at 6.7 at. %.¹⁸⁰ As the Hf⁴⁺ atomic percent increases further, the bandgap starts to decrease. When the Hf atomic percent is about 39.3 at. %, two clear bandgaps are observed in transmittance spectrum. It is proposed that these two bandgaps are due to the phase separation as a result of heavy Hf⁴⁺ doping. It is reasonable to hypothesize that this effect would manifest in other ALD doped ZnO systems, though the separated bandgaps have not been observed in other dopants. In all other instances of doping ZnO, a monotonous increase in bandgap is obtained as a function of dopant atomic percent.

In Fig. 6(d), we present the maximum change in bandgap obtained for various dopants and their atomic percent at which these changes are observed. It can be seen that S²⁻ and Ti⁴⁺ are the two most effective dopants to induce decrease and increase in the bandgap of ZnO, respectively.

Lastly, the transmittance of ALD doped ZnO needs to be mentioned. It is reported that the average transmittance in the visible wavelength range is 80%–85%. Thicknesses are in the range between 30 and 400 nm. High doping does not appear to have a strong effect on the transmittance of the film.

In this section, we will discuss the electrical properties of ALD doped ZnO with respect to the following parameters: (1) resistivity (ρ), (2) electron concentration (n), and (3) mobility (μ); noting that electrical resistivity is given as ρ = (neμ)⁻¹. We only discuss single dopants even though several papers show improved electrical properties of ALD doped ZnO with multidopants (i.e., more than one dopant). For example, the resistivity of Al and Ga co-doped ZnO is 7.6 × 10⁻⁵Ω cm for a 30–40 nm film.²²⁸ We further limit our data to published reports that have comprehensive electrical data via Hall measurements without post-treatment of the sample.

Figure 7(a) illustrates the resistivity (ρ) of ALD doped ZnO with various dopants as a function of atomic percent. With the exception of N-doped ZnO, the trend in ρ shares similar behavior.^{89,94,144,158,166,174,178,180,184,186,194,202,207} The ρ decreases first as the dopants are introduced and reaches a minimum. However, the specific atomic percent at which ρ is minimized varies among dopants [Al³⁺ = 3 at. %,⁸⁹ Ga³⁺ = 3 at. %,¹⁵⁸ Si⁴⁺ = 2.1 at. %,¹⁸⁴ Ge⁴⁺ = 1.3 at. %,¹⁷⁸ Ti⁴⁺ = 1.2 at. %,¹⁸⁶ Zr⁴⁺ = 4.8 at. %,¹⁷⁴ Hf⁴⁺ = 6.7 at. %,¹⁸⁰ and Ta⁵⁺ = 5.2 at. % (Ref. 207)]. Among all dopants, Ga-doped ZnO has the lowest ρ of 3.30 × 10⁻⁴Ω cm at 53 nm thickness.¹⁵⁸ We also note that Lee et al. report In-doped ZnO with a ρ of 3.9 × 10⁻⁴Ω cm.¹⁶⁶ However, this doping is for 80 at. % indium doping; which should be considered as Zn-doped In₂O₃ and not In-doped ZnO.¹⁶⁶ Finally, for N-doped ZnO the n-type ZnO switches to p-type semiconductor as the N percentage increases, which leads to a monotonous increase in ρ.¹⁴⁴ 

The carrier concentration (n) of ALD doped ZnO is shown in Fig. 7(b). The n increases first as the dopant percentage increases and reaches a maximum value (except N-doped ZnO). The highest n is 1.37 × 10²¹cm⁻³, which is obtained in Ga-doped ZnO.¹⁵⁸ It is also observed that ZnO with cation dopants have generally higher carrier concentration as compared to anion doped ZnO.

Figure 7(c) illustrates the change in the mobility (μ) of ALD doped ZnO as a function of atomic percent doping. In cation doped ZnO, μ decreases severely from ∼25 cm²/V s for undoped ALD ZnO to <5 cm²/V s, as the atomic percent doping increases. This is true for Al³⁺ (Ref. 89), Ga³⁺ (Ref. 158), In³⁺ (Ref. 166), Si⁴⁺ (Ref. 184), Ge⁴⁺ (Ref. 178), Ti⁴⁺ (Ref. 186), Zr⁴⁺ (Ref. 174), Hf⁴⁺ (Ref. 180), and Ta⁵⁺ (Ref. 207). The μ loss is particularly strong for Ge³⁺ and Ti⁴⁺.^178,186 It is known that μ is affected by (1) electron-electron scattering, (2) ionized impurity scattering, and (3) grain boundary scattering in the film. At lower atomic percent, the electron-electron scattering will be enhanced due to an increase in n. Dopants due to their highly charged state are considered ionized impurities in the film, which lead to strong scattering as the atomic percent increases. Grain boundary scattering is a little difficult to assess due to the variability in grain sizes observed in the XRD data in Fig. 3. On the other hand, anionic dopants individually behave quite differently. We highlight the case of Cl⁻ doped ZnO,²⁰² where the μ increases as the atomic percent increases. This effect is attributed to the fact that Cl⁻ substitutes the O²⁻ sites and oxygen vacancies, passivating dangling bonds. Further, the Cl⁻ substitution does not perturb the conduction band, which is primarily made of Zn 2p orbitals.

Finally, we discuss the apparent increase in bandgap measured with UV-vis spectroscopy with improved electrical conductivity. This effect is called the Burstein–Moss effect.^229,231 The effect manifests itself in degenerate semiconductors where the Fermi level is pushed beyond the conduction band edge. Thus, electronic band-to-band excitations must energetically surpass the intrinsic bandgap of the materials and the excess filled states in the conduction band as a result of the degeneracy. While details of this phenomenon are well covered in the literature, in Fig. 7(d) the relation between carrier concentration change (n^2/3) and the optical bandgap change (ΔE_g) is plotted for ALD doped ZnO. For identifying the Burstein–Moss effect, this trend must be linear. This is because the linearized form between n^2/3 and ΔE_g is given as

Here, h is Planck’s constant and m* is the reduced effective mass. It can be seen that the slope of the linear relationship contains information on m* which can be given as

Here, $meV∗$ and $meC∗$ are conduction and valence band density-of-states effective masses. Thus, used effectively, the Burstein–Moss plot can provide estimates on how strongly a dopant is able to electronically perturb the band structure in a host lattice. To illustrate this, in Fig. 7(d) the highest slope (for Ti⁴⁺) at m* = 0.72 and the lowest slope (for Si⁴⁺) at m* = 0.45 are illustrated. This can be compared to the undoped ZnO reduced effective mass of only m* = 0.17 obtained using $meV∗$= 0.59 and $meC∗$= 0.24.²³¹ 

We note that care must be taken when interpreting optical properties of ALD doped ZnO and discerning between chemical effects that lead to a change in bandgap and the Burstein–Moss effect.

In this review article, the current state-of-the-art in ALD doped ZnO films has been surveyed. From a materials perspective, doped ZnO films are potentially strong candidates as TCOs. While ITO is the most widely used TCO due to its high conductivity [1–2 × 10⁻⁴Ω cm for polycrystalline ITO (Refs. 232 and 233) and 0.77 × 10⁻⁴Ω cm for epitaxial ITO (Ref. 234)]; polycrystalline Al-doped ZnO and Ga-doped ZnO synthesized via RF-magnetron sputtering can achieve comparable resistivities of 2.0 × 10⁻⁴Ω cm and 2.7 × 10⁻⁴Ω cm, respectively.^235,236 Similarly, epitaxially grown [via pulsed laser deposition (PLD)], Al-doped ZnO and Ga-doped ZnO can reach resistivities of 0.85 × 10⁻⁴Ω cm and 0.51 × 10⁻⁴Ω cm, respectively.^237,238 Thus, doped ZnO films have comparable resistivities with ITO films.

From a process perspective, the lowest reported ALD film resistivity is 3.30 × 10⁻⁴Ω cm for epitaxially grown Ga-doped ZnO on c-axis sapphire.¹⁵⁸ The next lowest resistivity reported is for an amorphous In-doped ZnO film (or rather Zn-doped In₂O₃) with a value of 3.90 × 10⁻⁴Ω cm.¹⁶⁶ Thus, PLD doped ZnO films are ∼6× lower in resistivity than ALD doped ZnO films. Precursor molecules with cleaner chemistries and low residual C and N impurities, and postdeposition treatment of films can help narrow this performance gap. On the other hand, the process advantages that ALD has over other deposition techniques may offset the higher resistivity of ALD doped ZnO films. These process advantages include (1) pin hole-free films at ultrathin (a few nanometers) thicknesses, (2) superb conformality in high aspect ratio nanostructures, (3) low deposition temperatures, (4) monolayer fidelity over thickness control, (5) flexibility and ease of tuning film composition, and (6) manufacturing scalability.

Based on the current review of the ALD doped ZnO literature, we believe that there remain knowledge gaps in the synthesis and choice of dopants. Some of these are listed below:

1. As was shown in Fig. 1(a), a limited set of 22 elements have been used to dope ALD ZnO films. However, ALD chemistries for metal oxides are widely available.² For example, dopants such as Cd²⁺ have shown to reduce ZnO bandgap and work as an n-type with increased conductivity.²³⁹ Thus, there are opportunities to rationally screen and explore more dopants and integrate them with ALD ZnO films. The descriptor, linear charge density (Z/r), can be a valuable metric for initially screening dopants that lead to improved crystallinity of ALD doped ZnO films.

2. It is becoming increasingly important to synthesize new molecules, specifically tailored for dopant incorporation in parent films. This is an often-overlooked requirement. However, current precursors are limited in their ability to effectively dope ZnO. For example, the use of dimethyl-aluminum isopropoxide—a considerably larger molecule than trimethyl aluminum—results in 54% lower film resistivity.⁹⁵ The idea of adding steric hindrance, thereby spatially separating the dopant centers can be extended to Ga- and In-based chemistries which may result in even lower film resistivities than currently reported.

3. We have intentionally limited our reporting on multidopants in ALD ZnO due to the relative infancy of this subfield. However, initial reports are promising. For example, when 2 at. % Al-doped ZnO is deposited on epitaxially grown GaN, the Ga diffusion into the Al-doped ZnO results in a film resistivity of 7.6 × 10⁻⁵Ω cm.²²⁸ On the other hand, mixed InGaZnO have been deposited as amorphous semiconductors for thin film transistor applications.⁸⁴ These reports suggest a highly complex, mixed metal oxide environment where the cationic sublattice plays a crucial role in determining structure-property relationships in the films.

4. Doped ALD films usually form nanolaminate structures. On the other hand, there is a growing body of work that shows dopant activation is most effective if the dopants are homogeneously distributed through the film. As a result, a high temperature postdeposition anneal is commonly employed to redistribute dopants. This step is not compatible for applications involving transparent conducting ZnO films on flexible, polymeric substrates. Thus, co-injection of dopants should be widely adopted as means to dope ZnO.⁹⁰ This is not a trivial process to control. It implies that precursor delivery characteristics (volatility, vapor pressure, and chemical compatibility) of various molecules need to be closely matched, monitored in situ and, reliably dosed to achieve precise but homogeneous doping in the film.

5. Finally, we observe that there is a lack of modeling work for ALD doped ZnO films. ALD reaction modeling from ab initio principles is important, as experimental data suggest that there is significant chemical cross talk between the chemisorbed dopant molecule and the ZnO surface (and vice versa). As more complex precursor molecules are formulated, reaction modeling will be increasingly critical to understand surface reactions and substrate interactions a priori. At the same time, dopant incorporation into the ZnO matrix results in significant changes to the electronic band structure. An important issue not extensively discussed in the surveyed papers is the impact of the dopant atoms on point defects, especially oxygen vacancies.

   Here, it is noted that apart from ab initio modeling, there currently exists two models to explain the impact of dopants in ALD ZnO films. The first model is the “effective radius model” by Lee et al.⁹² and adopted by various groups.^81,93,95,102 The second model is by Saha et al.,¹¹⁶ who propose a dimensionless parameter, K_Fl_e (K_F is the Fermi wave number and l_e is the elastic mean free path of electrons) to evaluate the doping performance of ALD doped ZnO. Work should continue on testing the validity and refinements to the models.

In conclusion, therefore, ALD doped ZnO films show promising transparent conducting properties when compared to doped ZnO films via other deposition methods. The materials “palette” for designing new ALD doped ZnO films remains extensive and continues to grow. In addition to the traditional process advantages of ALD over other deposition techniques, new innovations in precursor molecule design and delivery and improved understanding via modeling may lead to ALD as a process and, doped ZnO as a material of choice for future TCO applications.

Z.G. acknowledges support from the Preeminent Postdoctoral Program (P3) fellowship at the University of Central Florida. P.B. acknowledges support from the National Science Foundation (NSF) under Award No. 1808625.