Roadmap on organic–inorganic hybrid perovskite semiconductors and devices

Lukas Schmidt-Mende, Vladimir Dyakonov, Selina Olthof

In the last decade, research on metal halide perovskites has made tremendous progress. The certified efficiency of perovskite solar cells (PSCs) has increased drastically to over 25% (see Fig. 1) and can now compete in this respect with Si-technology. The simple fabrication, including even processing from solution, opens the possibility to integrate perovskite processing into an industrial roll-to-roll fabrication, which allows mass production at low cost. Applications are not limited to single junction solar cells anymore. The chemical tuning of the bandgap of the material makes it especially interesting for application in tandem solar cells. Additionally, also other applications, such as light emitting devices, photodetectors, and memory devices, have already been successfully demonstrated.

Researchers have significantly increased their understanding of the material class, concerning, for example, the requirements to form stable 3D perovskites or employing larger cations for 2D structure formation. Intriguingly, the success of this novel high performance semiconductor revived the search by theoreticians and material scientists to look beyond and explore further promising complex perovskite compositions.

In this Roadmap, we cover important research aspects concerning metal halide perovskite semiconductors. We address in Sec. II some fundamental properties and the specific features of this material class. In Sec. III, we cover aspects of fabrication. The processing of efficient perovskite semiconductor films has been improved drastically over the last few years. Nevertheless, the solution chemistry is still an important research field, and processing of efficient perovskites films from solution, sublimation, and chemical vapor deposition (CVD) needs further investigation. Next to thin films, the growth of single crystals has been demonstrated.

The characterization and understanding of devices (see Sec. IV) has been drastically improved since the first report of a solar cell application. However, many details of the underlying physical mechanisms are still hardly understood. Next to the perovskite film itself, the interface formation and the adjacent transport layers (TLs) play a decisive role for the device efficiency and stability.

While perovskite solar cells are the main focus of this Roadmap, we also address tandem solar cell stacks in combination with non-perovskite absorbers as well as applications beyond solar cells, such as lasing and light emitting diodes (LEDs) in Sec. V. There are still major challenges, which we address in Sec. VI. In particular, if it comes to commercialization, many additional aspects need to be taken into account. Currently, the correct and precise measurement of perovskite-based devices is still an issue as well as the reproducibility. The role of ion migration and charge carrier mobility in the device and at the interface is still only partly documented. The attempts to replace the toxic lead in these thin films with less toxic materials have only been partly successful. One major challenge remains to be solved is the stability of the devices. Even though there has been tremendous progress, the device stability is still not comparable to silicon devices. This makes data-driven prediction of novel hybrid perovskites an exciting option to find materials overcoming the current limitations. When it comes to commercialization, one also needs to consider fabrication issues related to upscaling laboratory-scale devices to larger areas.

Despite these challenges concerning understanding device physics and functionality, there has been no material class in the recent past that attracted such high interest and introduced a paradigm change concerning the efficiency of solution processed semiconductors for solar cells. In this Roadmap, we summarize various important aspects to provide an overview of the current status of perovskite semiconductor research and indicate promising directions for future research efforts.

Feray Ünlü, Khan Moritz Trong Lê, Sanjay Mathur, Selina Olthof, Andrei D. Karabanov, DoruC. Lupascu, Laura Herz, Alexander Hinderhofer,Frank Schreiber, Alexey Chernikov,David A. Egger

Feray Ünlü, Khan Moritz Trong Lê, Sanjay Mathur

Single perovskites generally adopt the crystal structure similar to CaTiO₃, which is described with the ABX₃ formula, where the B-cation is coordinated octahedrally by X-anions and the A-cation fills the cuboctahedral voids and compensates the negative charge. Single metal halide perovskites (MHPs) adopt the ABX₃ structure consisting of a three-dimensional network of corner-sharing BX₆ octahedra, where the B atom is a group 14 metal cation (typically Sn²⁺ or Pb²⁺) and X is typically Cl, Br, or I. On the A-site, Cs⁺ or organic cations, such as CH₃NH₃⁺ (MA) and NH₂CH=NH₂⁺ (FA), can be used that are chosen to balance the total charge and to stabilize the lattice. For achieving structural stability in a 3D perovskite, the ionic radii of the cation and anions should follow the empirically derived Goldschmidt tolerance factor t given by the radii of A (r_A), B (r_B), and X (r_X),² 

Tolerance factors between 0.71 and 0.9 result in tetragonal or orthorhombic structures. For halide perovskites (X = F, Cl, Br, and I), the tolerance factor follows 0.81 < t < 1.11. If t lies in the range 0.89–1.0, a cubic structure is likely to be stabilized, whereas higher t values (>1.0) result in a hexagonal structure.^3,4 Recent analysis of structural data has shown that the tolerance factor t can differentiate between perovskite and non-perovskite materials in only 74% cases and a new tolerance factor (τ) was recently introduced that derives the structural predictions from the same parameters (composition, oxidation state, and ionic radii) and higher accuracy (>90%),⁵ 

Despite these parametric guidelines, structural transformation is common for any given perovskite composition, with higher symmetry structures being prevalent at high temperatures. Based on these considerations, the large ionic radii of Pb²⁺ (188 pm) and of halides (e.g., I⁻, 220 pm) limit the ionic radius of the A site cation. As a result, only organic cations with two or three C–C/C–N bonds or inorganic cations, such as Cs⁺ (188 pm), can form stable 3D perovskites.⁶ The crystal lattice of the perovskite is associated with a certain flexibility and defect-tolerance such that constituting ions can move and migrate within the lattice.⁷ The migration of ions and their mobility in the lattice causes the drift in ionic defects already at room temperature, which is responsible for the hysteresis of energy devices.^8,9 On the other hand, the geometrical stability and allowed range of tolerance offer a broad compositional space for chemical engineering of new perovskites by systematic substitution of cationic and anionic constituents and to obtain mixed-cation or mixed-anion perovskites through isovalent substitution in the parent (AMX₃) compound. In this context, the compositional engineering is shown to be a viable strategy for tuning the optoelectronic properties and to control the phase stability and check detrimental phase transformations.^10,11 Low formation energies allow facile exchange of ions within the A- and X-sites, whereas the formation energy of B-site doping is higher.^12,13

Given the fast anion exchange kinetics in hybrid perovskites, a broad range of compositions have been synthesized by in situ processing or post-synthesis chemical transformations. Intermixing of bromides or chlorides in the iodide perovskites has shown to be effective in changing the transport behavior of charge carriers and the bandgap energies.^14,15 For instance, the lattice parameters and bandgap energy of MAPb(I_1−xBr_x)₃ were observed to linearly scale-up in relation to the Br⁻ content.¹⁶ Furthermore, the inclusion of chloride into MAPbI₃ was found to significantly enhance electron and hole diffusion lengths over 1 μm in the mixed halide MAPb(I_1−xCl_x)₃, which is ten times higher compared to diffusion lengths in MAPbI₃.^17,18 The influence of halide anions on the perovskite’s bandgap is further described in Sec. II B.

A-site cation engineering includes the partial substitution of the parent organic or inorganic (e.g., MA or Cs⁺) cation by one or more co-cations, such as formamidinium (FA⁺) and/or other alkali metal cations (K⁺, Rb⁺), considering the appropriate sizes to maintain the tolerance factor. This approach was particularly successful in optimizing the stability of perovskites against undesired phase transitions, nonradiative defect sites, and was shown to be effective in enhancing carrier lifetimes and bestowing chemical stability under ambient conditions. The perspective of A-site cation engineering in tuning the functional properties and structural stability of halide perovskites were recently elaborated by Mathur and co-workers that can be found elsewhere.¹⁹ When considering the connectivity of the metal halide octahedra as the classification element, the perovskites also exist, besides the 3D configurations, in 2D layered structures, 1D chains, and 0D isolated structures.²⁰ Since there are no size rules for low dimensional networked perovskites, A-site cations of different lengths can be introduced, and therefore, low-dimensional perovskites offer tunable bandgaps and absorption/emission properties. Mitzi reported on 2D layered perovskites analogous to Ruddlesden–Popper phases with organic ammonium cations (R–NH₃⁺); here, the MX₄²⁻ anions are surrounded on both sides by organic cations stabilized by hydrogen-bonding.²¹ For 2D perovskites, the interlayer spacing between the inorganic layers is the determining factor for bandgap tuning, the longer the interlayer spacing, the wider is the bandgap. Hence, a combination of 2D and 3D perovskites is promising in terms of developing high efficiency solar cells with long-term stability. In 2D/3D systems, small organic cations are partially replaced by bulkier organic spacer cations; for instance, Grancini et al. reported one-year stable and hole transport layer (HTL)-free 2D/3D (AVA)_0.3(MA)_0.7PbI₃ [(AVA) aminovaleric acid, HOOC(CH₂)₄NH₃] PSC, without any obvious loss in power conversion efficiency (PCE 12.9%) in power conversion efficiency (PCE 12.9%) under ambient air.²² In addition, LEDs based on mixed layered perovskite materials were shown to display impressive performance and stability due to the mixed compounds acting as a photocarrier concentrator, funneling the charge carriers from high-bandgap materials to the lowest-bandgap ones to ultimately boost the external quantum efficiency (EQE).²³ More details on 2D perovskites and their properties will be discussed in Sec. II F.

B-site doping was shown to be very effective in tuning the bandgap and to influence the charge carrier dynamics of the perovskites. For example, substitution of the Pb²⁺ site in lead-based perovskite thin films or single crystals by isovalent Sn²⁺ was found to reduce the bandgap,^12,13 leading to broadening of the absorption range²⁴ and longer charge carrier lifetimes.²⁴ The Sn²⁺-doping also offers a plausible alternative to reduce the amount of Pb²⁺ content in perovskite compositions. In analogy, Ge²⁺ was recently incorporated into the double cation FA_0.83MA_0.17Ge_xPb_1−x(I_0.9Br_0.1)₃ where the solubility of GeI₂ was ameliorated by MACl additive achieving PCE over 22% and enhanced stability and photoluminescence (PL) lifetimes.²⁵ Moreover, Ge²⁺-doping in Sn based perovskites evolved to a strategic pathway to suppress the self-oxidation of Sn²⁺ to Sn⁴⁺, which in combination with A-site cation engineering and surface passivation has achieved a PCE of up to 13% so far.²⁶ Other bivalent dopants, such as earth alkali metals (Ca²⁺, Sr²⁺, or Ba²⁺), are reported to affect the bandgap and crystalline phase.^27,28 Looking at other main group elements, it was reported that Bi³⁺- or In³⁺-doping stabilizes the photoactive black perovskite phase at lower temperatures.^29,30 Transition metals, such as Mn²⁺, were investigated thoroughly in CsPbX₃ nanocrystals, leading to enhancement of long-lifetime luminescence by generating an additional exciton state associated with the Mn d–d emission.³¹ Rare-earth elements, such as lanthanoids, were used in perovskite nanocrystals achieving down- and up-conversion and quantum cutting.³¹ 

The emergence of perovskite solar cell (PSC) technology manifests the potential of chemical transformation in developing and designing new materials. In this context, the organic–inorganic hybrid perovskites have moved from phenomenological studies on compositional engineering and structure–property relations to establish themselves as drivers of an alternative photovoltaic (PV) technology as evident in their superior optoelectronic properties, summarized for a few representative compositions in Table I. While A-site cation engineering shows enhanced phase stability and durability under ambient conditions along with tunable bandgap properties through halide mixing, there are still stability issues under illumination, solar cell operation, or reverse bias conditions due to ionic migration and phase segregations in photoabsorber materials (Fig. 2). The ionic conductivity in perovskite halides was described as early as 1983³² where in MAPbI₃, the I⁻ anion was determined to be the major migrating species with four orders of magnitude higher mobility than the MA⁺ cation.³³ In particular, in mixed halide or multiple cation perovskites, the ion migration can lead to phase segregation. The phenomenon of J/V hysteresis (different voltage scanning direction with differing output current) is connected to the ion-migration (see Sec. VI C) induced poling of the perovskite,³⁴ causing reactions with adjacent (charge transport) layers and electrodes leading to overall device degradation.³⁵ Although belonging to the most efficient hybrid perovskite absorbers, mixed I/Br compositions tend to phase segregate under illumination leading to, e.g., I⁻-rich and Br⁻-rich phases.³⁶ Furthermore, organic–inorganic hybrid perovskites can suffer under thermal fluctuations, e.g., under reverse bias, which can lead to local heating and damage of the solar cell due to volatile organic components.³⁷ A possible solution against volatile components is to switch to the all-inorganic perovskites CsPbX₃; however, the wider bandgap due to mismatch of orbital energies and polymorphism are currently persisting challenges. One of the advantages of halide perovskite photovoltaics is their solution processability at low temperatures; however, the resulting polycrystalline thin films possess high defect density and pinholes that are detrimental for the functional performance. In particular, at the surface of perovskite films and grain boundaries, a large number of defects are present as the outermost ions are not saturated. The vacancies lead to trap states (energy levels in the perovskites bandgap) that reduce the overall charge transport and thereby the solar cell efficiencies, although these are shallow traps (near the band edges).³⁸ These limitations have been overcome by passivating the surface through chelation of molecular reagents.^39,40

The straightforward chemical synthesis was one of the triggers in witnessing the combinatorial explosion of compositionally engineered hybrid perovskites; however, the fundamental mechanism and insights into the formation of precursors, nature of chemical species in precursor solutions, existence of dynamic equilibria, and controlling effects of solvents (polar/non-polar) remain elusive. A better understanding of nucleation of perovskite crystals and their transformation on the substrate in a single phase perovskite composition is crucial for their application in devices and scale-up of the technology readiness level from prototypes to industrial production. Controlled synthesis of perovskite precursor ink, including precursor-intermediates that can be influenced both in situ (inks) and ex-situ (on-substrate) by coordinating ligands, is one of the deterministic factors for achieving a reproducible figure of merit for perovskite materials and devices produced in different laboratories. In particular, a better understanding of both homogeneous and heterogeneous nucleation of perovskite crystallites is inevitable to decipher the role of surface chemistry, purity of starting materials and employed solvents as well as additives in solution concentration, and precursor ratios and quality of chemicals to comprehend the various synthetic concepts that have been designated as novel and unique, albeit minimal differences in the underlying chemical interactions and nature of species. Since the interplay of above-mentioned factors is essential for homogeneous crystallization, the reproducibility of “ideal” inks continues to be difficult that aggravates with the increasing number of components, for instance, in multi-cation and anion systems. The high variability of compositions and properties makes it difficult to predict solar cell parameters, i.e., J_sc or V_oc, and associated energy losses when compared with the technological maturity of silicon photovoltaics. Current challenges of perovskite inks include a comprehensive understanding of nucleation and growth of perovskite crystals in solution and stabilization of perovskite inks to prevent uncontrolled growth and precipitation of thermodynamically preferred species from a mixture of components that is an intrinsic barrier in the scalability and shelf-life of perovskite inks that are crucial for deposition techniques, such as spin-coating, ink-jet printing, or slot die coating. In addition, relatively less attention has been devoted to fundamental studies on the chemical structure of perovskites based on mixed-cation compositions, for instance, there are no conclusive single crystal data for multi-component perovskites, which would elaborate the structural and positive influence of small amounts of dopants on the photoconversion efficiency. In addition, the homogeneity of distribution in such multiple-cation and mixed anion perovskites and its impact on the device are still an open question.

Long-term stability of perovskite materials and devices represents one of the crucial requirements besides the high efficiency to achieve large-scale production and commercialization of halide perovskite PV technology. To reach structural stability the A-site cation alloying has proved to be a viable concept to achieve a tolerance fit.⁵⁰ For example, Knight and Herz reported that the triple A-site cation (Cs, FA, and MA) and quadruple cation (Cs, FA, MA, and Rb) seem to be particularly stable. Similarly, the mixing of the 25% Sn/75% Pb perovskite can lead to increased photostability; however, more research is needed to evaluate the stoichiometric balance and its interrelation to structural and electronic stability [auto-oxidation of Sn(II)].⁵¹ Additionally, suppression of ion migration leading to phase segregation is mandatory to retain long-term stability under device operation and illumination. Among the common approaches to reduce ionic drift includes the implementation of polymers or organic molecules as barriers and surface passivating agents.^52,53 Implementation of 2D perovskites is another effective strategy to suppress ion migration, although this method is not applicable to all perovskite compositions as specific ligands are necessary. Furthermore, addition of chloride is shown to be effective in reducing the halide phase segregation, which is related to improvement of grain growth, crystallinity, grain boundary passivation, reduction in halide defect densities, and increased energetic barrier for halide migration.³⁶ Hence, compositional engineering is essential for designing stable and efficient perovskite absorbers; however, in order to progress faster and facilitate the screening of suitable chemical compositions, machine learning (ML) approaches are emerging as effective and effort-saving strategies.

With more powerful computing and advances in artificial intelligence, machine learning (ML) based approaches hold the promise for faster and more efficient investigation of new materials. Instead of empirical strategies, the optimal parameters are developed by training the machine with reliable data and a working algorithm. First-principle calculations (structure–property relation by quantum mechanical methods) are computationally demanding; however, an algorithm based ML can reduce the resources and make predictions by learning from existing data.⁵⁴ For example, 333 data points collected from 2000 publications were fed into a system to predict high performance perovskite compositions for solar cells based on the prediction of underlying physical phenomena to some extent.⁵⁵ Li et al. used similar approaches with data derived from high throughput density-functional theory (DFT) calculations to demonstrate alternative and economic pathways than iterative experimental attempts. They evaluated the compositional engineering with respect to perovskite stability to suggest more effective predictions that go beyond the conclusions drawn from the empirical tolerance factor.⁵⁶ Castelli et al. demonstrated the computational screening of oxide containing perovskites using the computational materials repository. Their work showed that out of a huge number of possible perovskites (around 19 000), only 47 were interesting for water splitting applications,^57,58 manifesting the power of computational methods in order to save experimental resources and time. However, also DFT calculations need a lot of computational energy and time; therefore, machine learning models can reduce these cost factors. For instance, Li and co-workers demonstrated the application of machine learning models to predict the thermodynamic stability of oxide perovskites using a DFT database of 1900 oxide perovskite energies, ending up with four stable compositions.⁵⁹ Similar studies for the discovery of new and stable perovskite structures were made by using machine learning approaches and the field is still under development.^60–62 Furthermore, Kirman et al. used machine learning to create a high-throughput system for perovskite single crystal synthesis.⁶³ Hartono et al. investigated organic halide capping layers for MAPbI₃ surface passivation by a machine learning framework, in which 21 different organo-halide capping layers were screened in order maintain perovskite stability under ambient conditions.⁶⁴ The further development of ML processing is beneficial for investigating mechanisms, predicting device performances, and facilitating the experimental challenges to synthesize perovskite materials with adequate compositions and even proposing non-hazardous alternatives.

The high photoconversion efficiencies and possible low production and material cost (attributed to solution processability, cheap perovskite materials, alternative low-cost electrodes, and charge selective layers) have established hybrid perovskites as a potential contender to challenge silicon or other semiconductor photovoltaics. The application of perovskite materials and fabrication of devices in industrial environment has been successfully demonstrated;⁶⁵ however, what is holding back the large-scale deployment of perovskite-based solar cells is the instability and shorter lifespan of perovskite absorbers. The challenge of environmental deterioration of perovskites has been tackled by encapsulation technologies,⁶⁶ although the intrinsic instabilities of the material under applied voltage, light-soaking, thermal stress reflected in hysteresis, and performance drift continue to compromise the long-term stability. Furthermore, the use of expensive charge selective materials, e.g., spiro-OMeTAD, and gold electrode in the highly efficient PSCs increases the levelized cost of electricity (LCOE) by up to 500%–800% compared to the FTO/SnO₂/perovskite/NiO/Cu module, which was calculated to be 4.43 US cents kW h⁻¹ (in comparison, the single junction Si-module has 5.50 US cent kW h⁻¹).⁶⁷ Therefore, a lot more research and concerted efforts are necessary to address the shortcomings of charge carrier transport, ionic mobilities, and main sources of losses in bulk and at the interfaces, which is a materials challenge.

Therefore, using the big dataset from the perovskite research community as a basis, compositional engineering coupled with modern machine learning systems can enable faster transition of perovskite science and technology toward commercialization. Beside this, investigations based on solution chemistry, crystal structure, nucleation, defect engineering, and device physics are essential to ensure a steady growth in understanding of the perovskite material as the class of perovskite material offers relatively unexplored versatility in applications.

Other potential applications of perovskites include near-infrared (NIR) solid-state and wavelength tunable 390–790 nm whispering-gallery nanolasers based on CH₃NH₃PbI_3−aX_a (X = I, Br, and Cl).^68,69 Compositional engineering of perovskites is also beneficial for increasing the stability of FET devices, as shown by Cs incorporation into the MAPbI₃ lattice and employing Rb as the passivating agent.⁷⁰ Hybrid perovskites were also found to be useful for photodetection⁷¹ and as sensors for a variety of gases, such as NH₃, O₂, and NO₂, and even some volatile organic compounds.^72–74 Within the last decade, interest in the lead halide perovskite based LED has rapidly increased, as the fabrication cost is comparable to organic LEDs (see Sec. V C). The operational stability, however, is far behind, and the half-lifetime of a few hours under continuous bias is significantly short when compared to the >10 000 h of lifetime needed to compete with organic light emitting diodes (OLEDs). Nevertheless, the beneficial effects of compositional engineering efforts as demonstrated in the improvement of the PV performance also underscore the potential for further development, especially in the exploration of new application fields (see Sec. V B) for this interesting class of functional materials.

Selina Olthof

Halide perovskites, with the structure ABX₃, can be composed of a variety of different cation and anion species, as introduced in Sec. II A. Depending on this composition, the bandgap E_g can vary from ∼1.2 to 3.6 eV, as summarized in Fig. 3(a), for all primary lead and tin compounds (discussion on Ge will be omitted in this section).⁷⁵ 

This variability in E_g is highly intriguing for the field of optoelectronics as the bandgap can be tailored to meet the individual need. For example, for single absorber solar cells, a bandgap around 1.3 eV is ideal, while as a wide gap cell in tandem applications, much larger bandgaps are desired (for example, ∼1.7 eV in combination with a silicon sub-cell; see Sec. V A for further examples). In the case of light emission (see Secs. V B and V C), appropriate bandgaps for display applications range from ∼2 eV for red emission to ∼2.8 eV for blue emission.

The changes in bandgap come about due to differences in the valence band (VB) and conduction band (CB) position that are influenced by the atomistic level positions, the lattice size, and lattice distortion. Band structure calculations show⁷⁶ that the top of the VB consists of hybridized metal s (M_s) and halide p (X_p) states, as sketched in Fig. 3(b). The VB is therefore directly influenced by the energetic position of the atomic M_s and X_p orbitals that are related to the element’s electronegativity χ [with χ(Pb) < χ(Sn) and χ(I) < χ(Br) < χ(Cl)]. Hence, a change in metal or halide component will lead to a shift in the VB, as depicted in Figs. 3(b-ii) and Figs. 3(b-iii). The bottom of the CB consists of metal p (M_p) and halide s (X_s) orbitals. The orbital overlap is smaller for the conduction band, and therefore, hybridization is not as pronounced, leading to a more ionic bond in which the CB is dominated by the metal state; it was found that the atomic p states of Sn (i.e., 4p) and Pb (5p) are rather similar in energy [Fig. 3(b-ii)].⁷⁵ 

The onsets of CB and VB are furthermore affected by the lattice size, which is determined by the ionic radii of the different anions and cations. As depicted in Fig. 3(b-iv), a smaller lattice will result in confinement effects, shifting the atomic levels slightly upward affecting both VB and CB. In addition, the hybridization will increase, which mainly affects the VB and shifts it upward.

Finally, also the bond angles will affect the degree of hybridization. The highest orbital overlap is achieved in a cubic perovskite structure, resulting in a small bandgap. In a phase transition, e.g., to a tetragonal or orthorhombic structure due to a lowering of the temperature or introduction of an ion with large mismatch in radius, the lattice becomes distorted and the degree of hybridization will reduce. This will lead to a lowering of the VB, as sketched in Fig. 3(b-v), thereby increasing the bandgap; again, the CB is likely not affected as much due to its stronger ionic character.

Obviously, not only pure ABX₃ perovskite structures can be prepared but also different ions can be mixed on the A, B, or X site. This way, the bandgap can be gradually tuned as was already shown in the 1970s by cooling melts containing appropriate proportions of precursors, e.g., by Barrett et al. for CsSnBr_xCl_1−x⁷⁷ or by Weber for MAPbBr_xI_1−x.⁷⁸ Reports regarding recent thin film measurements will be discussed below.

A critical issue when it comes to the comparability and validity of reported perovskite bandgaps is the inconsistency with which these values are determined. Often, the transmission T through a layer of material is analyzed, and from this, the absorbance A (with A = −log T) or decadic absorption coefficient α (with α = A/layer thickness) is reported. Less often, also the reflection is included in the analysis, which nevertheless is necessary to correctly determine A and α. Strictly speaking, from such a transmission and/or reflection measurement, the absorption/reflection onset can be extracted, which is not identical to the bandgap.

For a direct bandgap, where VB and CB exhibit a square root dependency on energy, the following relationship can be derived:

Therefore, plotting (αhν)² vs the photon energy in a so-called Tauc plot results in a straight line from which the optical gap can be extracted. In the case of an indirect semiconductor, the exponent of the left-hand side term changes to 1/2; for 3D halide perovskites, theoretical calculations consistently show direct transitions at the R point, so the equation shown above should be valid. Performing this Tauc plot analysis is commonly considered to be the more accurate approach for bandgap determination; however, there are limitations: (i) Issues can arise from the presence of trap or impurity states where the distributions of CB and VB electronic states do not terminate abruptly at the band edges. Using, e.g., photothermal deflection spectroscopy, which can measure down to very low values of α, transitions between localized and tail states can be seen as widening of the so-called Urbach tails or even as distinct sub-bandgap features; only above the Urbach tail region a Tauc plot analysis is meaningful. Even though gap state formation is not very pronounced in the defect tolerant halide perovskites, they have been observed, for example, in mixed halide systems.⁸⁰ Telling apart bandgap changes and defect state formation is not trivial. In the past, bandgap narrowing of a perovskite crystal by impurity Bi-doping has been reported.⁸¹ However, follow up experiments suggest that this change was caused by a change in defect concentration.⁸² (ii) Furthermore, if an exciton feature is present, then fitting the onset using a Tauc plot is inaccurate;⁸³ rather the theory described by the Elliott formula should be applied.⁸⁴ Here, the absorption of parabolic continuum states (band-to-band transition) is combined with an additional discrete absorption by an excitonic state. From such fits, values of the excitonic binding energy and bandgap can be deduced. However, due to the large number of fitting parameters, values extracted, e.g., for MAPbI₃ at room temperature using this approach vary between 5 and 29 meV. For a review over recent results, we refer to Ref. 85. In particular, lead-based halide perovskites show a rather distinct exciton feature at the band edge (see, e.g., the supplementary material of Ref. 75), which can lead to inconsistencies in reported values.

As a final remark, it should be checked whether the intended perovskite composition is indeed given in the prepared thin film. It is well-known that when processing a MAPb(I_xCl_y)₃ precursor solution, the incorporation of Cl ions into the perovskite crystal lattice is viable only in small quantities.⁸⁶ Furthermore, solubility issues can lead to deviations from the intended film composition, especially for Cs-containing precursors. Similarly, in the case of thermal evaporation, the severe issues related to monitoring the actual evaporate rates of the small organic cations⁸⁷ can lead to deviations in the final film composition.

Even though these issues regarding consistency in data evaluation prevail, we want to give an overview over the changes that can be introduced in the bandgap when the perovskite composition is tuned. For brevity, we focus on thin film measurements, therefore leaving out research on single crystals or nanoparticles.

The first work looking at mixed halide perovskite compositions was published by Noh et al. in 2013, investigating MAPb(I_1−xBr_x)₃.¹⁶ They found a gradual reduction in lattice size with the increasing Br content accompanied by an apparent change in color from black to red and yellow. The change in bandgap from 1.58 to 2.28 eV is not linear and can be described by a bowing equation,

Here, the first two terms describe a linear change in bandgap between E₁ and E₂, and the last term introduces non-linearity with a bowing parameter b. The reasons for the non-linear behavior is still a subject of debate and could be related to changes in bond angles and the mixing of atomic states; for further reading, we refer, e.g., to Refs. 88 and 89. A similar bowing behavior was observed for MAPb(Br_1−xCl_x)₃ (E_g varying between 2.42 and 3.16 eV)⁹⁰ and CsPb(I_1−xBr_x)₃ (E_g between 1.77 and 2.38 eV).⁹¹ For FAPbI_yBr_3−y, the gap can be tuned from 1.48 to 2.23 eV, but no 3D perovskite can be formed for the intermediate compositions of y = 0.5–0.7 due to lattice instabilities.^92,93 This can be circumvented by introducing a fraction of the smaller cations MA or Cs.^94–95 Less studies on halide mixing have been published for tin based perovskites so far; here, the changes seem to be rather linear, for example, in MASn(I_1−xBr_x)₃ (E_g 1.3–2.15 eV)^96,97 and CsSn(I_1−xBr_x)₃ (E_g 1.31–1.75 eV).⁹⁸ 

The bowing effect is more pronounced in the case of mixed-metal perovskites where it can even lead to a bandgap narrowing with respect to the values of the pure materials. This can extend the absorption into the infrared region, as first reported by Stoumpos et al. for MASn_xPb_1−xI₃.⁹⁹ Absolute values reported in the literature differ somewhat due to the issues discussed above, but it seems that a reduction by 50–130 meV can be achieved in this mixed system compared to pure MASnI₃;^89,99,100 in the case of FASn_xPb_1−xI₃, the reduction is 100–200 meV.^79,88,89 For Br compounds, the reported results are less consistent, but bandgap narrowing in thin films has also been observed.^89,101,102 In an extensive study, Rajagopal et al. showed⁸⁹ that the amount of bowing in the mixed-metal perovskite scales with the macrostrain present in the pure films, which affects octahedral tilting and lattice distortion.

Mixing A-site cations has only little effect on the bandgap, introduced by minor modifications in crystal size or variations in bond angles. Such a variation of the A site cations is rather of interest for changing the phase or temperature stability of a perovskite system (see Sec. II A). Changes in E_g are mostly less than 100 meV, as, e.g., reported for FA_xMA_1−xPbI₃^103,104 and FA_xMA_1−xPbBr₃.⁹⁴ However, in mixed perovskites with significant changes in lattice distortion, the effect can be larger than 200 meV.^79,105

Overall, there is a complex interplay between variations in atomistic energy levels, lattice size, and octahedral tilting/strain that leads, in most cases, to some degree of non-linearity in the bandgap change when the composition of a perovskite film is gradually tuned, as depicted in Fig. 4. With these different effects at play, it is not trivial to predict E_g values of mixed perovskite systems, and carefully evaluated and consistent measurements are needed to provide this information.

As a further remark, the bandgap can be influenced by other parameters, such as temperature, pressure, or confinement effects in nanoparticles. For brevity, these topics have not been discussed here but present further avenues that widen the utilization of this material class and can help to understand their fundamental properties.

Andrei D. Karabanov, Doru C. Lupascu

The high efficiency of perovskite solar cells relies on high carrier mobilities, effectiveness of exciton break-up, and a long lifetime of the generated charge carriers. In this section, we show that all three are directly related to dielectric effects.

In order to understand dielectric effects, one has to realize that these can be very scale dependent. Before entering into the details of halide perovskites, we will outline a few general aspects.¹⁰⁶ 

Dielectric effects reflect the response of matter to an electric field WITHOUT DC conductivity. So essentially, one has to consider the response of a capacitor to an electric field keeping in mind that this capacitive effect can occur on very different length scales. Under AC-electric fields, conductive components will arise also in the dielectric response, which is the reason why complex impedance is introduced.

The simplest dielectric effect is realized in the deformation of the electronic shell of an atom with respect to the position of the nucleus. Under an external field, the atom itself thus forms an electric dipole. After field removal, such an atomic dipole disappears. A similar effect can occur one scale up. In particular, in an ionic crystal, the positions of the ions are shifted under the action of an external field. Dipole formation now arises due to the shifting of the relative positions of the positive and negative ions of the lattice. After field removal, everything returns to its initial state: this is termed paraelectric behavior.

If a relative shift of ionic positions arises in a crystal lattice without the application of an external electric field, one is concerned with pyroelectric behavior because this shift is typically temperature dependent. Most systems display vanishing shifts at high temperatures and ordering when temperature is lowered. If such ordering generates permanent ordered dipoles, pyroelectric order arises. This is intrinsically related to a reduction in crystal symmetry. The most common mechanism driving the phase transition into an ordered electric state is the soft phonon effect, namely, a loss of stiffness of the crystal for one or more of its ions in the unit cell.¹⁰⁷ Crystals showing pyroelectric effects cannot have a center of symmetry in the point group of their crystal structure. This type of ordering is classified as long range because the ordered state extends over large parts of the prototype crystal, namely, a “domain” of electrical ordering.

If such an ordered state can be electrically reversed in between different stable orientations in the crystal, it is called the ferroelectric effect. Thus, a prerequisite for ferroelectricity is a non-centrosymmetric crystal symmetry and multiple stable orientations of mutually ordered dipoles within the same crystal.

Another contribution to dielectric effects arises if electric dipoles exist intrinsically, namely, if we are concerned with dipole molecules. The best known example is the water molecule. In halide perovskites, the methylammonium ion, among others, is an electric dipole molecule. Also many liquid crystals contain dipole molecules and show ordering. In certain cases, the ordering of the molecules can also be of long range character, and a ferroelectric effect is known in certain smectic liquid crystals.¹⁰⁸ 

The rotational motion of the MA ion was early on suggested to contribute to the dielectric response¹⁰⁹ as was later verified in experiment.^110,111 Chen et al.¹¹² showed in experiment that point defect interaction with MA ions highly influences photoconductivity and the Hall effect. Single crystals show much better photoconductivity than thin films. Solution grown films contain the highest number of defects.¹¹² Despite the encountered very high numbers of defects, the authors already evoked the dipolar screening by the MA ions to warrant good conduction nevertheless.¹¹² 

In a previous work, we showed that despite many theoretical assumptions in this sense, MAPbCl₃, MAPbBr₃, or MAPbI₃ are not ferroelectric. They exhibit an antiferroelectric phase transition in the range 100–150 K¹¹⁰ and show no long range ordering above this temperature. The ionic lattice shows paraelectric behavior with a very high dielectric constant of roughly ɛ_r = 30 up to very high frequencies in the gigahertz range (∼10¹⁰ Hz). In particular, we want to emphasize that these three solar cell materials are NOT ferroelectric at room temperature.

The fundamentally interesting effect in the methylammonium-containing materials is the dipole of the methylammonium ion itself. It is known that water molecules, due to their high dipole moment, strongly interact with ions and other matter. One of the effects is micellon formation, namely, the ordering of the molecules around a charged entity, typically in the radial direction and oriented to mitigate the field of the charge carrier at the center, a typical screening effect. This effect, e.g., dominates dissolution of ionic crystals. A micellon is a local state of ordering. It arises in suspensions, emulsions, ceramic slurries, biological systems, etc.

The interesting fact about the halide perovskites containing methylammonium or formamidinium is that these molecules can rotate in their lattice locations WITHOUT long range ordering. Thus, their response is very similar to a state in which they would be in a liquid state and can in a broader sense be compared to spin glass states in magnetic systems.

Now, the charged entities in a semiconductor are an electron or a hole. We proved experimentally¹¹⁰ that two mechanisms screen the electric charge of these electronic charge carriers. The first and classical mechanism is the formation of a Fröhlich polaron, namely, the relaxation of the crystal lattice of an ionic solid toward the charge. The second one arises due to the very local reorientation of the methylammonium dipole ions toward the respective charge.

We called the entity, charge carrier plus rotated dipole ion, a micellon. This micellon provides roughly 50% of the screening effect, namely, an increase in the relative dielectric constant. The remainder stems from ionic shifts in the unit cell, the classical Fröhlich polaron formation. The combination of both was termed hyperpolaron by us (Fig. 5)¹¹⁰ in order to illustrate the strong impact of this new type of charge carrier, which is solely due to dielectric screening on the properties of the methylammonium lead halides and potentially other crystals of this family that provide rotating molecules within the crystal structure.

The next question now concerns exciton breakup: How does a high local dielectric response alter the exciton binding energy? A useful analogous picture to understand the role of the dipole ions is to imagine this like a dissolution process of an ionic entity in water. The breakup of the ionic bond is much facilitated by the gain in energy due to the adherence of the surrounding water molecule dipoles. After screening, the two ions can separate. A similar process arises in the methylammonium lead halides: the two charge carriers formed during light absorption, forming the exciton, rapidly become screened. On the time scale of processes, likely the exciton forms first, then polaronic screening arises, and its time constant is slightly shorter than for molecular rotation. Then, the molecules rotate to add to the screening. Effectively, the Coulomb interaction of the electron and the hole is largely reduced, which is the same as saying that the exciton binding energy drops dramatically due to the dielectric effect.

One counter argument could be that effectively the binding energy of the charge to the lattice response is very large due to polaron formation. Thus, a small polaron should form that is localized (an electronic state deep in the bandgap). The ingenious detail is that there are two mechanisms providing the screening, each operating at other typical relaxation frequencies. Thus, the coupling arises to a much broader frequency spectrum than simple lattice relaxation. This avoids total localization of the deformation of the lattice because, effectively, a jump out of the energy well mechanism operating at another frequency helps to suppress full localization and small polaron formation. One could even imagine an energetic double well problem where the difference in intrinsic frequencies could drive a beat frequency, but theoreticians will have to deal with these details in future work.

Another important aspect of the dielectric screening is its effect on localized defects, such as dopants, ionic vacancies, interstitials, or anti-site defects. All these represent electrical charges in the lattice and thus form scattering sites for electron transport. On the time scale of lattice vibrations, these defects remain localized (even though defect jumps also arise due to collective excitation by phonons, but their constructive interference is a rare event). Hence, point defects are statically localized in the context of electron transport. Therefore, all dielectric contributions present in the crystal, even the slow components, will provide screening of the localized point defect. The charge effectively experienced by a bypassing electron or hole is smaller by a factor of 60 than in the unscreened scenario. Silicon only provides ɛ_r = 11. Dielectric defect screening in methylammonium lead halides is thus six times more effective than in silicon.

Another assumption was made that photoexcited electron and hole pairs could be separated in internal junctions between ferroelectric domains.¹¹³ After a long debate, it was concluded that tetragonal MAPbI₃ is not ferroelectric for the following reasons: most importantly, the centrosymmetry of the I4/mmm space group forbids ferroelectricity and no ferroelectric P(E) hysteresis is observed.¹¹⁰ Locally using scanning piezoforce microscopy, no dependence between the applied voltage and the resulting strain response and no polarity inversion with the change in the applied high voltage were seen, both being inherent to a ferroelectric.¹¹⁴ 

Dielectric properties of completely inorganic metal halide perovskites are also of great interest because such systems show better stability compared to hybrid halides.¹¹⁵ One of the prominent examples is CsPbX₃. This material also has a high dielectric permittivity (Fig. 6) leading to effective screening, but the contribution from dipole rotation (e.g., the MA-ion) is not available. Thus, at low frequencies, the dielectric constant amounts to merely half of the one for MAPbX₃. This reduced screening, in particular, of defects can explain the lower power conversion efficiency of CsPbX₃ perovskite solar cells, as more charge carriers are scattered at defects and the effective charge carrier mobility is reduced.¹¹⁶ 

Polarons form in any semiconductor due to the finite dielectric response, namely, its high frequency components. The higher the dielectric constant, the stronger is the polaron formation and its associated trapping energy. Due to their very high dielectric constant, the charge carrier mobilities in the lead halide perovskites are reduced in comparison to conventional semiconductors. Miyata et al.¹¹¹ used broadband dielectric spectroscopy in the frequency range 10⁶–10¹³ to see the response of polarons to the electric field in single crystals of all-inorganic CsPbBr₃ and hybrid MAPbBr₃.

As seen in Fig. 7, in CsPbBr₃, the dielectric constant is constant around ɛ_r = 20 from 1 MHz to 100 GHz and is around 60 in (CH₃NH₃)PbBr₃. The liquid-like rotational relaxation between 10¹⁰ and 10¹¹ Hz of the organic polar cations (dropping from 50 down to 20) is the same mechanism as called micellon before. The lifetimes of polarons in hybrid (CH₃NH₃)PbBr₃ are increased because their recombination is significantly hindered by this liquid-like polarization response.¹¹¹ The change in dielectric constant from 10⁶ to 10¹⁰ Hz by Δε_r = 10 can be related to a dynamic stiffening of the lattice to higher frequencies, being more pronounced in a hybrid lattice than in an all-inorganic system.

The dielectric response is also partly due to the lone-pair activity of Pb²⁺. These lone pairs can be responsible for the reduction of electron–hole recombination, leading to the increased carrier lifetimes. The polar distortions induced by lone pairs influence dielectric properties in these crystallographically centrosymmetric phases.¹¹⁷ 

The photoinduced change in the relative permittivity (Δε_r) gives information about the polarization in the perovskite systems. Conventional perovskites, such as MAPbI₃ and FAPbI₃ exhibit no measurable photoinduced changes in the real component of relative permittivity (Δε_r). Double and triple-cation perovskites show high values of Δε_r, meaning that more complex mechanisms lead to the polarization response of double and triple-cation perovskites.¹¹⁸ 

FA_0.83Cs_0.17Pb(I_0.9Br_0.1)₃ (FACs) and (FA_0.83MA_0.17)_0.95Cs_0.05Pb(I_0.9Br_0.1)₃ (FAMACs) enable high power conversion efficiencies. Hong et al.¹¹⁸ examined the frequency-dependent time-resolved microwave conductivity (TRMC) at frequencies of 10 GHz helping to reveal molecular motions of A-site cations. Δε_r of the above-mentioned compounds was measured as a function of time before, during, and after illumination by a nanosecond pulsed laser.

It was found that there is a proportionality of change in relative permittivity to the photogenerated carrier density (Fig. 8).¹¹⁸ ε_r decays with a shorter time constant than conductance, meaning that the photoinduced free charge carriers induce an immediate additional “dielectric” response that rapidly decreases while the lifetime of free carriers is significantly longer. The mechanisms explaining this effect locally are not yet determined. The photoinduced polarizability in FACs is larger in comparison with FAMACs.¹¹⁸ The local mechanisms behind this difference remain to be determined. The local polarizability could, e.g., be related to exciton lifetimes on the different organic cations.

Apart from conventional 3D perovskites, 2D and 1D crystal structures are studied nowadays. Claims on ferroelectricity in these systems could be justified due to the lower symmetries of crystals, but experimental verification does not typically meet the full requirements in the “ferroelectrics”-society. As we do not want to exclude the validity of these data, a brief summary of this work follows.

Several 2D perovskite ferroelectric structures were presented recently.^120–121 The change in the dimensionality is reached by the substitution of the A-cation by organic molecules, such as benzylamine(C₇H₉N),¹²⁰ hexahydroazepine (C₆H₁₃N),¹¹⁹ and others. Ferroelectricity can be obtained by fluorination of these structures. In this process, the substitution of one hydrogen atom (monofluorined substitution) or several hydrogen atoms (perfluorined substitution) with fluorine atoms is done. This leads to the structural change and enhancement of ferroelectric performance. One of the examples is (PFBA)₂PbBr₄ (PFBA = perfluorobenzylammonium)—multiaxial ferroelectric with the Aizu notation of 4/mmmFmm2(s), which shows a large piezoelectric response, a spontaneous polarization of 4.2 C cm⁻², a high Curie temperature of 440 K, and it is a semiconductor with a direct bandgap of 3.06 eV.¹²⁰ 

Another method is the substitution of A-cations by organic molecules containing organic functional groups R or S. This leads to the possibilities of changing the photofunctionality and conductivity properties of the perovskites. Moreover, the environmental stability of these materials could help in engineering the robust and highly efficient solar cell.

Yang et al.¹²² proposed [R-1-(4-chlorophenyl)ethylammonium]₂PbI₄ (R-LIPF) and [S-1-(4-chlorophenyl)ethylammonium]₂PbI₄ (S-LIPF) 2D lead iodide perovskites, R and S denoting the enantiomers (Fig. 9). Ferroelectricity of these materials was derived from piezoresponse force microscopy (PFM)-based hysteresis loop measurements. The spontaneous polarization was calculated using the point charge model. The bandgap from a Tauc plot is 2.34 eV, which is favorable for the application in photovoltaics.¹²² At present, it cannot be said whether we are concerned about real ferroelectricity or a PFM-tip induced effect.

Polarization can be increased when additional small A-site cations are introduced in the spaces of octahedral sheets, forming a quasi-2D metal halide perovskite (MHP; formula of R₂A_n−1B_nX_3n+1). The higher complexity of the temperature-dependent structural phase transition in these materials in comparison to the 2D MHPs is dictated by the distortion of the octahedra due to smaller A-cations, which lead to lattice displacement, providing an additional contribution to polarization.

The multi-layered (n = 3) quasi-2D MHP (BA)₂(MA)₂Pb₃Br₁₀ (BA = n-butylammonium, C₄H₉NH₃; MA = methylammonium, CH₃NH₃) was synthesized by Li et al.¹²³ by alloying the 3D MHP MAPbBr₃ perovskite with n-butylammonium. The organic components, namely, MA and BA, lead to the order-disorder transition, inducing electric polarization, which can be seen in Fig. 10, allowing for ferroelectricity. X-ray diffraction (XRD) analysis at room temperature showed that this material can be ferroelectric with a polar space group of Cmc2₁.¹²⁴ 

A decrease in dimensionality is also possible with systems containing other organic groups as A-cations, such as (R)-(–)-1-cyclohexylethylamine (R-CYHEA, C₈H₁₅NH₂) or (S)-(+)-1-cyclohexylethylamine (S-CYHEA, C₈H₁₅NH₂), which leads to the formation of 1D R-CYHEAPbI₃ and S-CYHEAPbI₃ perovskites. These organic groups remove the inversion symmetry of the material, making it a potential ferroelectric, while the PbI₆ octahedra are responsible for the semiconducting properties. The absence of the inversion symmetry was also proven by several methods, such as density functional theory and structural and spectroscopy analysis. Ferroelectricity was confirmed with dependent synchrotron XRD, differential scanning calorimetric, temperature-dependent dielectric constant, and temperature-dependent pyroelectric current. Such materials could open new possibilities, for example, in spin-orbitronics.¹²⁵ 

Dielectric studies play an important role in the understanding of materials for the use in the optoelectronics and photonics. The examination of different dielectric responses in a broad range of frequencies can reveal screening effects, the influence of the organic and inorganic cations on the polarizability, charge-carrier lifetime, effects stemming from the defect states, and charge carrier mobilities.

The examination of perovskite systems with different techniques, such as impedance spectroscopy, microwave or THz-time-domain spectroscopy and piezoforce microscopy, can help to reveal carrier mobilities, effectiveness of exciton breakup, and lifetime of charge carriers, as well as the presence of electric polarization and ferroelectricity or the absence of these.

The challenge of revealing ferroelectricity is due to several artifacts that can arise during PFM measurements, erroneous assignment of the space group, leakage current in the ferroelectric P(E) hysteresis, and many others. Irrespective of all these techniques, a thorough understanding of the symmetry group of the crystal is the basis of assigning pyroelectric, piezoelectric, or ferroelectric effects.

Laura Herz

The performance of solar cells critically depends on efficient photocurrent extraction, which, in turn, relies on sufficient mobilities of charge carriers. As a rough guide, charge-carrier diffusion lengths, which depend on the mobility-lifetime product of charge carriers, ought to exceed the thickness of the perovskite film required to absorb a substantial fraction of the incident sunlight. Much attention has therefore been devoted toward optimizing charge-carrier mobilities in metal halide perovskites and developing a fundamental understanding of the underpinning factors.^126,127 Here, a distinction has been made between intrinsic factors, such as charge–lattice interactions that cannot fundamentally be avoided, and extrinsic issues, such as grain boundaries, defects, energetic disorder, or background doping.¹²⁶ Specifically, for the class of metal halide perovskites, the intrinsic mechanisms limiting charge-carrier mobilities are now fairly well understood.^127–130 As a first approximation, intrinsic limits are well-described within the Fröhlich model of large polarons,¹³¹ which derives from the coupling of charge carriers to the electric polarization field generated by the collective longitudinal optical (LO) vibrational modes of the metal halide sub-lattice. Absolute charge-carrier mobility values derived from this model form reasonable upper limits when compared to the spread of available experimental data and seem to agree well with mobility trends with compositional variations.¹²⁶ For example, the Fröhlich model correctly predicts the experimentally observed¹³² decreasing trend in charge-carrier mobility along the iodide–bromide–chloride substitution line, which derives from the increasing ionicity of the lead-halide bond and the associated changes in the dielectric response of the metal halide sub-lattice.^129–130 Similarly, increases in charge-carrier mobilities with substitution of tin for lead^126,133,134 may result from the resulting increase in LO phonon frequencies,¹³⁵ given that the Fröhlich model depends on the dimensionless ratio of LO-phonon to thermal energies. A-cation substitution (e.g., MA, FA, or Cs) has, on the other hand, been found to have insufficient influence on charge-carrier mobilities because the electric field to which charges couple arises primarily from the LO phonon mode of the lead-halide sub-lattice.¹²⁷ Such models limit the maximum charge-carrier mobilities attainable for the prototypical MAPbI₃ at room temperature to 100–200 cm² V⁻¹ s⁻¹, which is substantially lower than the equivalent value for GaAs,¹²⁷ because of MAPbI₃’s lower LO phonon frequency (lead is a heavy atom), higher ionicity of the polar lattice, and, in the case of conduction-band electrons, significantly higher effective mass.

Aside from these intrinsic limits posed by electron–phonon coupling, charge-carrier mobilities in metal halide perovskites may also be subject to extrinsic effects arising from material imperfections, which have formed a persistent challenge for a range of specific materials.^{126,133–134} Charge-carrier mobilities in prototypical metal halide perovskites implemented in high-performance solar cells, e.g., APbI₃, where A is typically a combination of MA, FA, and Cs, are already likely to be close to what can be achieved in terms of optimized mobilities. However, lingering challenges persistently remain for specific cases, including mixed bromide–iodide lead perovskites with intermediate bromide fractions between 20% and 90%, for which poor crystallinity and charge-carrier mobility¹³² combined with light-induced halide segregation⁵¹ pose a barrier to the optimization of high-bandgap metal halide perovskites. Similarly, at the low-bandgap end, mixed lead–tin iodide perovskites have seen charge-carrier mobilities significantly suppressed by the formation of tin vacancies, which result in unintentional background hole doping that enhances the scattering of charge-carriers.^134–135 Such effects form a current hurdle to bandgap-tunability of metal halide perovskites that is particularly detrimental to the development of silicon-perovskite and all-perovskite tandem solar cells. Further challenges also remain to the development of fully accurate theoretical models of charge-carrier mobilities in metal halide perovskites that take account of the special characteristics of these materials. While the Frohlich model presents a good first approximation of electron–phonon coupling in metal halide perovskites, some difficulties arise from the inherent softness of these materials^136,137 and the presence of many atoms per unit cell, which leads to multiple LO phonon modes needing to be considered.¹³⁰ In this context, there has been much debate around the experimentally observed^138,139 T^−1.5 temperature-dependence of the charge-carrier mobility in MAPbI₃, which at first¹⁴⁰ appeared to be incompatible with a simple application of Frohlich theory. However, subsequent investigations have revealed that careful consideration of electronic couplings across the full range of low- and higher-energy LO phonon modes present in MAPbI₃ can replicate the experimentally observed temperature dependence suitably well.¹³⁰ In addition, anharmonicity of the lattice potential has been raised as a concern, given that metal halide perovskites tend to be fairly soft, meaning that the harmonic approximation employed by Frohlich theory could be particularly inaccurate for these materials.^51,137 Recent theoretical models¹³⁷ based on full nuclear dynamics combined with first-principles calculations avoided such implicit harmonic approximations, approximating the steep temperature-dependence of the carrier mobility in MAPbI₃ fairly well.¹³⁷ The extent to which such special characteristics of metal halide perovskites directly affect their optoelectronic properties will therefore form another challenge to be resolved over the coming years.

With regard to bandgap-tunable metal halide perovskites comprising mixed iodide–bromide or tin–lead perovskites, further advances in charge-carrier mobilities and stable performance will no doubt be made based on the ever-improving feedback loop between material synthesis and passivation protocols and advanced characterization techniques. A myriad of improvements in material morphology and reduction in defects are being made, while advanced non-contact probes of charge-carrier mobilities^126,141 have obviated the need for device fabrication to assess such parameters. However, while optimism in the field remains high with regard to raising these specific iodide–bromide and lead–tin compositions to the same optoelectronic and stability standards as their high-performing APbI₃ counterparts, an alternative strand has developed that focuses on the vast category of non-perovskite metal halide semiconductors.¹⁴² Such approaches acknowledge the substantial materials space available outside the much narrower ABX₃ space,¹⁴³ including Ruddlesden–Popper two-dimensional layered perovskites, new double perovskites with mixed metals (A₂BB’X₆), vacancy ordered materials, networks of corner- or edge-sharing [Metal-Halide₆]⁴⁻ octahedra, and general materials with metals in different oxidation states (e.g., 3+ and 4+). While opening a vast new materials space will offer many potential advantages not only restricted to improvements in charge-carrier mobilities but also, for example, to the pursuit of lead-free materials (see Sec. VI F), it also offers fresh challenges. Here, computational screening based on high-throughput first-principles calculations^141,143 will help to narrow down the huge parameter space, identifying promising candidates to explore for synthesis (see Sec. VI H). Continuous development of first-principles computational methods and increasing power of supercomputing infrastructures now allow for rapid materials screening, transforming the discovery of new semiconductors from the current slow, trial-and-error, “needle-in-a-haystack” search into a rapid, targeted, and systematic exploration of a vast group of potential candidate materials.¹⁴¹ Further improvements in this area may also ultimately derive from the increased use of machine learning or “artificial intelligence” in such screening attempts. While such fast-screening approaches already succeed in evaluating relatively straight-forward parameters, such as chemical stability, band structure, and the resulting charge-carrier masses, more complex calculations still pose difficulties. In particular, full evaluation of charge-carrier mobilities relies on the accurate calculation of electron–phonon couplings, which still needs to be brought within reach of genuine fast-screening approaches. Such calculations may be further complicated if an enhanced presence of surfaces and increased structural flexibility potentially enhance polaronic effects in certain new metal halide semiconductors,¹⁴² requiring a level of computational resource even higher than those currently employed for metal halide perovskites.

While much progress has been made over the past five years with regard to an understanding of charge-carrier mobilities in metal halide perovskites, significant work remains to be done within the wider class of metal halide semiconductors. While prototypical metal halide perovskites are now widely accepted to fall within the Frohlich picture of large polarons, some theoretical work is still required to clarify the specific couplings to existing phonon modes and the relative importance of anharmonic contributions of the lattice. Certain extrinsic material challenges also remain, in particular, for large-bandgap bromide–iodide lead perovskites and narrow-gap lead–tin perovskites. In contrast, the factors governing charge-carrier mobilities for the much wider space of non-perovskite metal halide semiconductors are still relatively unexplored. Such materials bear some resemblance to ABX₃ perovskites in that they often comprise similar octahedral building blocks. However, they may differ substantially in terms of lattice flexibility and polarizability, surfaces, and interfaces, thus opening up a vast new area of research.

Alexander Hinderhofer, Frank Schreiber

Scattering methods (e.g., x-ray scattering and neutron scattering) are versatile tools for investigating the structure and morphology of hybrid perovskite materials. With the various scattering methods, many different characteristics of a given sample can be retrieved, including crystal structure, domain sizes, defect densities, roughness, electron densities, unit cell orientation, and phase composition.

X-ray and neutron scattering on single crystals and powder samples or thin-film samples with various degrees of texturing were used to determine the crystal structure and phase behavior of different hybrid perovskite formulations. The continuously tunable element composition of hybrid perovskites leads often to correspondingly shifting unit cell sizes and corresponding characteristics.¹⁴⁴ 

In particular, the application of neutron scattering also allows for the determination of the position of hydrogen atoms within the unit cell and has shown that the central organic cation, often methylammonium or formamidinium, can be present in several different orientations.¹⁴⁵ This is related to the extremely important question of structural dynamics in these systems.¹⁴⁶ 

More challenging is the structure determination for perovskite thin films, which are relevant for device applications. Such thin films are typically polycrystalline with a varying degree of texture (preferred unit cell orientation).

As a standard method for thin film characterization, often x-ray diffraction (XRD) in the classical 2θ Bragg Brentano geometry is applied or, alternatively, x-ray reflectivity (XRR) with parallel beam optics to determine the position of out-of-plane Bragg reflections. In particular, XRR from surfaces and interfaces has developed into a versatile tool, now being frequently used for the characterization of the surface/interface morphology. In the simplest version, the method measures the specular reflection, i.e., the flux of the specularly reflected radiation as a function of the angle of incidence. Specular XRR contains important information, in particular, the thicknesses and electron densities of individual layers in a layered sample as well as root-mean square roughnesses of the interfaces. For more complex structures, the interference between the layers can provide information about the overall coherence of the structure, such as the formation of superlattices.

Two additional experimental configurations are used for complete reciprocal space mapping. Using grazing-incidence small-angle x-ray scattering (GISAXS), island size distributions (small Q) can be probed and analyzed. For large values of the in-plane scattering vector Q_||, lateral structures with very small sizes (down to few nm, including in-plane lattice spacings) can be probed using grazing-incidence wide angle x-ray diffraction (GIWAXS). Since, in a standard experimental configuration, the coherence width of the partially coherent primary x-ray beam is much smaller than the irradiated footprint on the sample surface, the measured signal is averaged over a statistical ensemble of all microscopic configurations of the interface.¹⁴⁷ 

The demanding properties for stable highly efficient solar cell materials present several structural challenges. Even slight changes in the preparation conditions of thin perovskite thin films can lead to significant changes in the structure, texture, and morphology, which in return impact the device characteristics. This makes it essentially compulsory to include structural characterization as a matter of common practice.

Another challenge is the phase behavior of highly efficient formamidinium based perovskites. FAPbI₃ is unstable at room temperature and decays into a hexagonal structure. To avoid the decay, several stabilization methods can be used, for example, changing the perovskite composition or and/or solvent and temperature treatment or capping layers.¹⁴⁸ However, the specific phase composition of a given perovskite structure poses still a problem.

There are several challenges in the area of perovskite structure formation or conversion: for example, in high performance multiple cation perovskite materials, the cation distribution is often unknown. If a respective cation is included in the lattice, how the cations distribute (e.g., statistically vs phase-separation tendency) is sometimes unclear. This poses an interesting challenge to the structural characterization.

Another big challenge for hybrid perovskites is the fast degradation in the presence of water or oxygen. To meet this challenge, 2D-perovskites or a combination of 2D on 3D perovskite structures are applied.¹⁵⁰ Such structures have additional peculiarities from a structural viewpoint. The pure 2D structures include additional spacer molecules (see Sec. II F on 2D perovskites). The formation of an alternating 2D/3D structure is not possible for all kind of spacer molecules, and the domain orientation is of great importance for charge transport properties. Finally, in 2D or 3D perovskites, the 2D layer functions as a passivation against degradation. For the structural characterization, the challenge is here to obtain information on the interface between two layers. This can be achieved by applying GIWAXS and by the variation of the angle of incidence. With such a setup, depth dependent information on the crystal quality, lattice parameters, domain orientation, and the coherence length of a 2D superstructure can be obtained.¹⁴⁹ Some of the above issues can be addressed using scattering methods with time resolution.

Real-time in situ scattering measurements during sample preparation or modification can be used to follow the structural evolution with a temporal resolution of down to 10 ms. The solvent layer during sample preparation can be penetrated by applying high photon energies. This allows for a detailed tracking of intermediate phases and conversion time scales, which improves the understanding of perovskite formation. Figure 11 shows, for example, the perovskite conversion of MAPbI₃ into MAPbCl₃ by the application of MACl solution on MAPbI₃. Figures 11(b)–11(d) shows the corresponding GIWAXS data obtained in situ.¹⁵¹ 

One method to determine the element and ion distribution in mixed perovskite lattices could be anomalous x-ray scattering. Anomalous x-ray scattering (i.e., scattering employing energies near core level of specific elements) can be used to probe the distribution of the lattice site occupation of these specific elements in alloys and doped crystals. In mixed perovskites, e.g., in a mixed iodine–bromine perovskite, the correlation between iodine-occupied lattice sites and bromine-occupied lattice sites can be determined. So far, this method was mainly applied to study well-ordered inorganic perovskite structures.¹⁵² We expect hybrid perovskite layers to be highly ordered and that anomalous dispersion x-ray scattering can be successfully applied.

We believe that in view of the sensitivity of the performance of perovskites to structural details, employing scattering methods is and will remain indispensable for routine characterization. Furthermore, some of the challenges to the understanding of these materials will require the application of advanced tools, such as anomalous scattering, time-resolved methods, or depth-dependent detection schemes. These require typically support from neutron or synchrotron facilities.

Alexey Chernikov, David A. Egger

In addition to three-dimensional systems, the family of metal halide perovskites hosts a rich variety of nanostructures, such as two-dimensional (2D) Ruddlesden–Popper compounds. These materials have been studied and explored for many decades¹⁵³ but have recently re-emerged with renewed interest in their potential for high-efficiency photovoltaic and light-emitting applications¹⁵⁴ as well as in view of their intriguing fundamental properties. This is strongly motivated by the design flexibility associated with their structural characteristics that presents opportunities from the perspectives of technology and fundamental solid state science. As schematically illustrated in Fig. 12, 2D perovskites exhibit a layered structure common for the broader family of van der Waals materials. They adopt the generic structural formula (RNH₃)₂(A)_n−1BX_3n+1, where RNH₃ is usually an aliphatic or aromatic cation that constitutes the spacer between perovskite layers formed by the A, B, and X ions with the number of sublayers denoted by the integer n. The release of the structural constraints that are otherwise implied for the formation of the 3D lattice offers excellent tunability of 2D metal halide perovskites, which further involve multiple types of subclasses, depending on the stacking directions in the crystal. Conceptually, these systems can be understood as natural quantum wells where the electronic excitations are confined in the inorganic layers surrounded by the organic barriers.

This structure leads to a central distinction in the optoelectronic properties of the 2D metal halide perovskites from their 3D counterparts that originates in the strongly enhanced Coulomb interaction due to both quantum confinement and, most importantly, weak dielectric screening from the organic spacer layers. It leads to the formation of highly robust exciton states with binding energies on the order of several hundreds of meV¹⁵⁵ that dominate the optoelectronic response of these materials. Consequently, a rather small spatial proximity of the electron and hole constituents in an exciton also results in a very strong light–matter coupling that is beneficial for efficient absorption and emission. In contrast to more traditional quantum wells or inorganic 2D semiconductors, however, the lattice of the 2D perovskites is unusually soft with characteristic features of anharmonic effects around room temperature,¹⁵⁶ thus promoting the coupling between excitons and lattice vibrations. Altogether, this particular combination of strong electron–electron and electron–phonon interactions renders 2D perovskites a very interesting platform for studies of fundamental many-particle phenomena in addition to their rich potential for applications.

In view of more recent research activities on 2D perovskites, there are several directions that are currently being explored. From the perspective of applications, these materials were found to show an improved structural stability, in particular, with respect to their exposure to moisture, including reports of photovoltaic devices with robust long-term operation.¹⁵⁷ 2D perovskites can also be prepared as nanoplatelets and ultra-thin layers, in close analogy to inorganic van der Waals crystals, such as graphene. Interestingly, a number of general properties of much thicker crystals, including the optical response, can be kept largely intact in these nanometer-thin structures. Among notable differences, however, is an enhancement of the exciton binding energy up to 0.5 eV and the occasional suppression of the structural phase transition characteristic for certain types of 2D metal halide perovskites.¹⁵⁸ More recently, fundamental properties of the excitons in these materials, such as effective masses, were also shown to strongly depend on the crystal structure and composition with distinct experimentally and theoretically demonstrated changes.¹⁵⁹ Furthermore, it is also interesting that 2D perovskites were found to show increased carrier lifetimes compared to their 3D analogs.¹⁶⁰ The key mechanisms determining their radiative and non-radiative lifetimes remain among open questions in the field. Such differences in the lifetimes, for example, could be well impacted by the changes in structure, composition, and phase-transition behavior that the pure 2D as well as the 2D–3D interfaces exhibit under various thermodynamic conditions appearing in working devices.¹⁶¹ 

Of particular importance remains the question regarding the spatial propagation of optical excitation in 2D metal halide perovskites. Exciton mobility is closely tied to the general nature of both exciton band structure and exciton–phonon coupling and is highly relevant for the understanding of light-harvesting and light-emitting devices based on excitonic materials. Interestingly, the diffusion of excitons in these systems has been addressed only very recently in contrast to the vast literature on this subject for their 3D counterparts. Selected results of these studies are summarized in Fig. 13. In the work by Deng et al.,¹⁶² it was shown that both the exciton–exciton interactions and the effective dimensionality of the system play a crucial role for the exciton diffusivity at ambient conditions. The chemical composition of the materials and, in particular, the choice of the organic spacer turned out to be equally important, as demonstrated by Seitz et al.¹⁶³ and Xiao et al.¹⁶⁵ Both phenomena are found to stem from the strong variations in the strengths of the exciton–phonon coupling that depends on the stiffness of the crystal lattice. Temperature-dependent measurements of the exciton diffusivity by Ziegler et al.¹⁶⁴ then revealed a wide range of conditions from room temperature down to about 50 K, where the exciton propagation exhibits characteristic behavior of free particles in the regime of conventional diffusion. At lower temperatures, however, exciton diffusion was found to be highly anomalous, including observations of effectively negative diffusivity. The latter appeared to stem from an interplay of a rapidly expanding exciton cloud, followed by a long-lived, less mobile exciton fraction with indications of localization.

Despite much progress in the field of 2D metal halide perovskites, a number of conceptional and technological challenges remain. For example, while there are improvements in terms of stability of 2D metal halide perovskites, it still seems necessary to continue to rely on various methods employed for material protection, especially when the samples are subjected to optical excitation. Individual approaches, such as different ways of encapsulation, turned out to be rather successful, yet it also remains unclear whether they may also affect some of the studied properties, especially for very thin layers. Further associated with that are the requirements of reproducibility and material system control that are still being explored (also see Sec. VI B). Moreover, a number of conceptual challenges to describe opto-electronic properties of 2D perovskites are intimately tied to the combination of a very strong Coulomb coupling and a highly efficient exciton–phonon interaction. Closely related is the question of how the transport of tightly bound exciton states should be fundamentally described. While there is evidence of free diffusion that resembles semi-classical quasiparticle propagation in traditional quantum wells, there are also features associated with exciton localization and thermally activated diffusion. It seems to be strongly connected to the questions of the exciton–phonon coupling, application of the polaron picture, or even effects of quantum interferences that could appear between the regimes of semi-classical diffusion and hopping for strongly localized states.¹⁶⁶ 

These questions motivate a number of interesting challenges for future theoretical studies. It should be emphasized that 2D metal halide perovskites are challenging to be treated from first-principles considering that their crystalline unit cells involve many atoms. Furthermore, it is not clear whether their complex structures can be much simplified considering that, for example, the organic spacer cation cannot be neglected in the theory since it plays important roles in various phenomena as discussed above. Taken together, this renders theory work that would be relatively straightforward for 3D crystals, such as vibrational spectra, and quite challenging for the case of 2D perovskites. To address some of the fundamentally challenging topical questions regarding electron– and exciton–phonon effects that we emphasized above may thus very well require the development and exploration of alternative theories that are simpler numerically, such as parameterized models to complement first-principles studies.

In summary, 2D metal halide perovskites represent a highly interesting class of low-dimensional hybrid materials with an array of fascinating properties associated with their structural flexibility and very strong interactions of their electronic and vibronic systems. Representing a conceptual link between inorganic quantum wells and organic molecular crystals, these natural 2D structures hold much promise to explore the realm of mobile, robust quasiparticles coupled to a comparatively soft crystal lattice. Altogether, these systems offer a highly interesting material platform to study fundamental many-particle phenomena and interactions in condensed matter, motivated by both basic science and technology.

Oleksandra Shargaieva, Caterina Cocchi, Eva Unger

Oleksandra Shargaieva, Caterina Cocchi, Eva Unger., Michael Saliba, Mahdi Malekshahi Byranvand, Martin Kroll, Frederik Nehm, Karl Leo, Alex Redinger, Julian Höcker, Vladmir Dyakonov

A fascinating aspect of metal halide perovskites is that they are the best solution-processed materials for photovoltaic application, as illustrated in Fig. 1.¹ The success of these materials is attributed to their high defect tolerance in solution processed thin films. This opens up a myriad of processing strategies.¹⁶⁷ Processing procedures for metal halide perovskite thin-films as well as precursor solution compositions, here referred to as “inks,” have been developed through an empirical approach based on solar cell device performance results. Most developed processing strategies are based on spin-coating as the deposition technique due to its wide availability. One of the most successful strategies involves the utilization of a mixture of solvents, and the crystallization is induced by quenching with an anti-solvent or gas-flow during the process.^169–170 Without the use of such quenching methods, the crystallization mechanism and kinetics are determined by the solvent evaporation rate.¹⁷¹ For high-quality large area thin films, specially designed inks are developed in which solvents and additives fulfill the specific purpose of controlling the chemical and physical formation mechanism to result in thin-films with high optoelectronic quality and coverage.

The solvent plays a quite complex role in perovskite processing as it affects the properties of (i) the precursor inks (e.g., solubility and complexation), (ii) their processing (e.g., viscosity, evaporation rate), and (iii) the crystallization mechanism (e.g., by becoming incorporated into intermediate phases) (see Fig. 14). When preparing hybrid perovskite precursor inks, the solvent molecules interact with the precursor inducing dissociation, which is often accompanied by the coordination of solvents with the precursors and the formation of lead halide complexes of a general formula (PbI_2+n)ⁿ⁻· (Solvent)_m.¹⁷² The level of halide coordination often depends on the choice of solvent and is described in terms of coordination strength defined by solvent polarity and various solubility parameters.^173,174 Since the interaction with the solvent can be weak or strong, the choice of solvent defines the properties of the solution. The strongly binding solvents, such as dimethyl sulfoxide (DMSO), lead to the formation of complexes with a larger number of solvent molecules in the coordination sphere of Pb. The weakly interacting solvent tends to induce a stronger interaction among the precursors themselves and lead to the formation of higher coordinated polyiodide plumbates. These complexes are the key building blocks for the formation of the material from the solution. The formation of complexes and the size of colloids in the solution can also be manipulated by the addition of small molecule and polymer additives.^53,176,177

During the deposition, the precursor concentration, solvents, and additives affect the rheological properties of precursor inks. In particular, usage of low viscosity solvents and polymeric additives is reported to improve flow properties and wetting of the solutions on the substrate aiding the nucleation and the formation of homogeneous films.^180–180 The volatility of the solvents or solvent mixtures then critically affects the crystallization kinetics.^181,182 A strong interaction between solvents and precursors can lead to the formation of crystalline intermediate phases, which then transform into perovskite phase upon full solvent removal. Weakly coordinating solvents much more readily allow for the formation of the perovskite phase upon solvent evaporation,¹⁸³ while stronger coordinating solvents often require annealing.¹⁸¹ The formation of a limited amount of intermediate phases, such as (CH₃NH₃)₂Pb₃I₈·2DMSO, has been found to be beneficial for the formation of high-quality perovskite layers.¹⁸⁴ Therefore, fine-tuning the ratio between strongly coordinating and weakly coordinating solvents is an important approach to controlling the kinetics and formation mechanism of metal halide perovskites during crystallization.

The quality of the thin-film depends on the chemical transformation and association of chemical building-blocks in solution into solid-state semiconductor thin-films.^174,185 Thus, the key to high-quality reproducible thin-films is understanding and, as a result, controlling the crystallization. In part, this can be achieved by designing new inks with each component of the ink fulfilling a defined function to control the chemical equilibrium between solution-complexes and thin-film formation mechanism and kinetics. It has been shown that the optimization of each parameter of the precursor solutions can lead to beneficial effects. The deliberate engineering of precursor inks requires a more in-depth understanding of the complex role of solvents and additives in precursor inks.

Establishing the key parameters for the rational engineering of precursor solutions could be achieved with the understanding of specific interactions that occur between precursors and solvents or additives. For this purpose, quantum-mechanical methods and specifically ab initio calculations play an essential role. Not only calculations based, for example, on (time-dependent) density-functional theory allow one to model complex systems with effective treatment of the solution environment.¹⁸⁶ Most importantly, they offer the flexibility to turn on or off specific coupling mechanisms. A comparison with experiments can, thus, reveal the presence and role of the underlying interaction in the examined systems and processes. The efficiency achieved by ab initio methods and their coupling with novel automatization tools, such as AiiDA, enable processing a large configurational space at once.¹⁸⁷ In this way, a rational screening for new solvents is possible. Furthermore, machine learning techniques can be applied to correlate calculated and measured data, thereby accelerating the discovery of new solvents and compounds that can be used in the role of solvent.

An additional benefit from the solvent screening is a substitution of currently used solvents for less harmful, greener, and sustainable ones. Currently, none of the commonly used solvents for the preparation of perovskite solutions (e.g., DMSO, DMF, NMP, and GBL) is satisfying both conditions of health and environmental safety.^188,189 This is especially important when moving to industrial-scale production of devices.

In recent years, a number of reports have illuminated detailed mechanisms of thin-film formation during various stages of processing. In particular, the use of optical and x-ray-based techniques for monitoring structural changes during the formation of perovskite thin-films has highlighted the occurrence of crystalline intermediate phases and their phase transformation kinetics to solid-state semiconductors.^190,191 Theoretically, various models, such as the LaMer mechanism, von Weimarn crystallization rules, and Ostwald ripening, have been utilized to describe the nucleation and crystal growth processes in hybrid perovskites.¹⁹² Combined approaches where in situ monitoring by means of x rays or laser reflectometry is used together with modeling of the evaporation rate demonstrate a promising direction toward understanding perovskite formation.^175,193 However, the experimental methodology to investigate the early stages of perovskite crystallization still needs to be explored. This information can potentially be obtained by employing multi-parameter monitoring where various properties (structural, optical, and electronic) are detected simultaneously. One of the promising steps toward implementation of such an approach is shown in the work of Hu et al., where authors employed in situ grazing incidence wide-angle x-ray scattering, Fourier-transformed IR spectroscopy, and optical microscopy to monitor perovskite formation.²⁹ Additionally, there have been several attempts to rationalize chemical interactions in solutions and the formation of perovskite by implementing different computational methods. One of the approaches is to compare experimental data on chemical interactions in solutions and properties of precursor solutions to fundamental theoretical calculations. Recent reports indicate the presence of agglomerated iodoplumbates that are formed in high-concentration solutions.^175,194 The predicted structure of these larger complexes is similar to crystalline intermediate clusters and might act as an intermediate structure between solution complexes and crystalline intermediate phases. This is particularly important as precursor solutions likely to undergo supersaturation prior to nucleation and the formation of intermediate phases.

To identify alternative “greener” solvents and solvent systems, computational methods for solvent screening have delivered promising results. A recent study by Gu et al. demonstrates how a combination of high-throughput screening (HTS) and a combinatorial synthesis executed with a robot can accelerate the search for suitable anti-solvents and rationalize the chemical interaction between solution complexes and anti-solvents.¹⁹⁵ Another interesting example is the use of machine learning combined with robot-assisted synthesis to predict the optimal conditions for the synthesis of new perovskite single crystals shown in the work of Kirman et al.⁶³ 

We predict that furthering the in-depth understanding of the solvent–solvate interactions in precursor solutions and during crystallization will, in combination with high-throughput experimental methods to identify interesting solvent-mixtures and additive, catalyze the development of greener precursor solutions to process metal halide perovskites.

Being able to produce high-quality perovskite thin-films from solutions is a unique selling point of these materials. While a lot of developments in processing strategies and precursor solution compositions have been seemingly empirical, an increased understanding of the role of solvents and additives in the formation mechanism of metal halide perovskite thin-films from solution-complexes enables a more and more rational ink design. High-throughput synthetic methods to screen for novel precursors, solvent, and additive combinations will likely enable a swift identification of greener precursor solutions to process metal halide perovskites. We believe that a coordinated effort to illuminate the fundamental solution chemistry transformation processes through the combination of experimental and computation methods will prove highly valuable.

Michael Saliba, Mahdi Malekshahi Byranvand

Organic–inorganic perovskites with unique optoelectronic properties emerged as an efficient light absorber for achieving high power conversion efficiency (PCE) of perovskite solar cells (PSCs). However, the quality of the perovskite films depends on many processing parameters that require precise control during the fabrication to achieve pinhole-free (dense) films with large grains.¹⁹⁶ 

So far, solution-processed one-step antisolvent and two-step methods utilizing spin coating techniques have been investigated for perovskite film deposition frequently (Fig. 15).¹⁹⁷ In the one-step, perovskite ink deposits in one step onto a substrate; however, the solvent of perovskite ink should be extracted by one of the solvent quenching approaches, such as antisolvent, gas, or vacuum quenching, to promote perovskite crystallization more efficiently.²⁰⁴ In the two-step method, PbI₂ ink deposits onto a substrate at the first step, followed by the deposition of the organic cations at the second step by dipping the PbI₂ film into organic cations precursor solution¹⁹⁸ or by spin-coating this solution onto the PbI₂ film.¹⁹⁹ To date, the best PCE record of 25.2% has been achieved for small areas n-i-p mesoporous²⁰⁰ and planar²⁰¹ PSCs by the one-step antisolvent method. However, considerable PCEs >24% have been reported for fabricated PSCs with a two-step method.^202,203 It should be mentioned that very promising results have been achieved by the thermal evaporation method for perovskite film deposition, presented in Sec. III C.

Figure 16 shows the typical procedure for the perovskite film crystallization. In step 1, perovskite ink must be carefully prepared based on the optimized formulation to achieve a uniform and stable perovskite film. Afterward, the perovskite ink will be deposited on the substrates (TCO-coated glass, for example) utilizing the different deposition techniques, including laboratory-scale methods (i.e., spin-coating) or large-scale methods (i.e., blade-coating, slot-die coating, inkjet printing, and spray coating) (step 2). The perovskite ink should be forced into a supersaturated state, i.e., solvent quenching, by applying antisolvent, gas quenching, vacuum solvent extraction, and air blading, on wet perovskite films (step 3), which promotes the nucleation process (step 4). Finally, crystal growth will be facilitated by annealing (step 5) with different methods, resulting in completed perovskite films. It is well known that the quality of perovskite film depends strongly on controlling the nucleation and growth rates, i.e., steps 4 and 5, which are related to the concentrations of the precursor and inducing supersaturation in the wet film. To achieve a highly uniform and dense perovskite film, fast nucleation and slow crystal growth have been beneficial. Altogether, given the crucial role of high-quality perovskite films for achieving high-performance PSCs, developing a reliable and repeatable crystallization process is of the essence. Here, we provide the challenges, progress, and open questions in different steps of perovskite processing.

A double cation perovskite, i.e., (FAPbI₃) × (MAPbBr₃)_1−x, has been introduced as an efficient formulation for perovskite ink preparation because it helps facilitate the crystallization process and assists in stabilizing the “black phase” of FAPbI₃.²⁰⁴ Later, Saliba et al.^46,47 demonstrated that adding a small amount of alkali metal cations, such as Cs⁺ and Rb⁺, to perovskite ink can improve the perovskite film quality further, resulting in high stabilized PCEs of 21.1% and 21.6%, respectively. Importantly, a more repeatable stabilization of the FAPbI₃ was achieved, which was demonstrated in various processing environments. Moreover, it has been demonstrated that adding the long-chain organic cations, such as butylammonium (BA⁺) or phenylethylammonium (PEA⁺), to perovskite ink converts some parts of the three-dimensional (3D) perovskite to two-dimensional (2D) phase, i.e., 2D/3D structure, resulting in a stability improvement of perovskite films.²⁰⁵ Moreover, all inorganic perovskite inks also have been suggested to achieve highly stable perovskite films.²⁰⁶ Solvent engineering is also critical to achieve high-quality perovskite ink. DMSO as the co-solvent of DMF in a 1:4 volume ratio is often used to obtain a stable intermediate phase of the PbI₂–DMSO–MAI adduct during the perovskite deposition, resulting in pin-hole-free and dense perovskite films.²⁰⁷ However, Lee et al. demonstrated for the FAPbI₃ perovskite that the NMP co-solvent yields a more stable FAI–PbI₂–NMP intermediate phase compared to DMSO due to stronger interaction with FAI.²⁰⁸ Therefore, each new perovskite component needs to consider the available co-solvents and their subsequent effect on the final film quality.

After the solution preparation step, perovskite ink should be deposited on the surface of TCO substrates properly. The spin-coating technique has been thoroughly studied for the deposition of perovskite ink with controllable and optimized film morphology (Fig. 17). This technique has been very successful for fabricating highly efficient PSCs in the laboratory-scale, limiting to a scale of <10 cm and wasting a significant portion (>90%) of perovskite ink during the process, which is not applicable for industrializing perovskite photovoltaic (PV) technology.²⁰⁹ However, the achieved science from spin-coating could be transferred to large-scale processing techniques for the deposition of perovskite ink on large substrates, enabling the industrialization of PSCs. As shown in Fig. 17, well-developed large-scale deposition techniques from other solution proceed optoelectronic technologies, i.e., organic photovoltaics (OPVs) or copper indium gallium selenides (CIGS) solar cells, such as blade coating, slot-die coating, inkjet printing, and spray coating, have been used for perovskite film deposition. These are very promising approaches as they allow roll-to-roll manufacturing processes and upscaling the perovskite photovoltaic fabrication. In the blade-coating technique, a certain amount of ink solution is poured in front of the blade and moves over the substrate at a specific speed.^210,211 This technique attracts significant attention for scale-up of this kind of solar cells, owing to its easy compatibility with a wide variety of perovskite compositions, substrates, and device structures. Slot-die coating is another promising technique for large-scale perovskite deposition, similar to the blade-coating technique (see Fig. 17).²¹² Unlike blade-coating, in this technique, the perovskite ink comes out from a thin slit between two pressed metal blocks of a coating head and consumes a bit more perovskite ink; however, the resulting perovskite film showed higher quality and reproducibility.²¹³ Inkjet printing emerged as a fast, facile, and economical technique to form desired patterns of large-scale perovskite film deposition.^214,215 Originally, this coating technique is a digital management technology inspired by typical office printers. Spray coating has also been used to produce thin-film photovoltaics, an appropriate method for mass production of perovskite films.^218–218 Similar to other large-scale techniques, slow solvent evaporation rate and dewetting of the substrates are the most critical challenges, leading to a low surface coverage and poor crystallinity.

Despite the impressive progress of perovskite film deposition by these techniques, the PCEs of large-scale deposited PSCs are lower than small area devices. Providing a fine and homogeneous perovskite film formation is critical in large-scale deposition techniques to reduce series resistance and increase shunt resistance, improving the PSC performance. Therefore, as detailed below, understanding the perovskite crystallization process is very important for the further development of PSCs.

Typically, perovskite inks have a moderate rate of solvent evaporation during the deposition due to the low vapor pressure of its solvents, i.e., DMF and DMSO, leading to a low density of heterogeneous nuclei and inhomogeneous perovskite films after the annealing step (step 5). Therefore, various solvent quenching techniques have been developed to increase the nucleation rate, leading to compact perovskite film with full surface coverage.

Antisolvents have been frequently used for quenching or solvent extraction of perovskite ink from the wet films, leading to rapid supersaturation of the perovskite precursor, i.e., fast nucleation, and achieving smooth and compact perovskite films [see Fig. 18(a)]. The antisolvents should be miscible with the perovskite solvents, i.e., DMF and DMSO, and should not dissolve the components of perovskite composition, i.e., PbI₂, PbBr₂, MAI or FAI, and so on. So far, various antisolvents, such as toluene,²⁰⁴ chlorobenzene,²¹⁹ ethyl ether,²²⁰ ethyl acetate,²²¹ and anisole,²²² have been established. As mentioned, the PCE record was achieved for antisolvent assisted spin-coating deposition by dripping the antisolvent onto the wet perovskite film during the spin-coating process. However, antisolvents need the spinning effect to perform maximum solvent extraction from wet films; therefore, it is challenging to use this technique for scalable deposition.

Heat quenching has been determined as an efficient technique for solvent extraction from the perovskite wet films [see Fig. 18(b)].^223,224 Typically, this technique speeds up the film drying rate, accelerating the supersaturation condition and promoting the nucleation step in perovskite films. In this approach, different temperatures of 30–90 °C have been applied to wet films for the fabrication of high-quality of perovskite films.^225,226 However, heat quenching can be affected by other parameters, such as the solvent and solutes of the perovskite ink. At low temperatures (<30 °C), a discontinuous perovskite film with low surface coverage will be formed due to a slow solvent extraction rate. At high temperatures (>90 °C), substrate dewetting occurs very fast due to fast solvent extraction, leading to non-homogeneous perovskite films. Therefore, optimizing the applied temperature is critical for efficient solvent extraction, achieving high-quality perovskite films.

Nitrogen (N₂) or dry air gas quenching technique has been introduced for solvent extraction from the as-coated wet film within a few seconds [see Fig. 18(c)], mainly used for large-scale deposition, such as blade-coating or slot-die coating of perovskite films.^229–229 In this technique, a gas knife, i.e., N₂ or air, moves with the blade simultaneously and blows the gas on the wet perovskite film to quench or extract the solvent from the ink, consequently inducing fast nucleation and achieving a full surface coverage by the perovskite film.²³⁰ After gas blowing, a post-annealing process is needed to ensure the complete liquid-to-solid conversion. In this technique, the gas flow rate and gas temperature are the two critical parameters that need to be controlled carefully for high-quality thin films. Moreover, the solvent composition in perovskite ink should be considered as an essential factor to form a stable intermediate phase after gas quenching treatment. For example, Conings et al.²³¹ demonstrated that adding a small amount of DMSO to a DMF-based perovskite ink plays a crucial role in forming a stable intermediate precursor–DMSO complex, achieving dense and smooth perovskite films.

Vacuum quenching has been introduced as another innovative technique to reach the supersaturation level of perovskite wet film quickly by extracting the solvent from the perovskite ink effectively [see Fig. 18(d)].^232,233 This technique introduced by Li and co-workers^234,235 for solvent extraction from the spin-coated wet film is producing uniform and highly crystalline perovskite films. The vacuum degree and vacuum duration are the two critical parameters in this quenching technique that should be considered to find the optimum deposition condition. Afterward, vacuum quenching has been frequently used for large-scale deposition methods. Similar to other quenching techniques, the solvent composition of perovskite ink is also an important parameter to achieve a stable intermediate phase, leading to fully covered and compact perovskite film. This technique was used for both the one-step and two-step perovskite deposition methods. For example, Guo et al.²³⁶ used vacuum quenching as an efficient technique to extract the superfluous solvent of the freshly deposited inorganic component PbI₂ in a two-step sequential deposition approach, achieving high-quality perovskite films.

After the nucleation step, i.e., supersaturation point, by extracting the solvent from the perovskite ink with antisolvent, heating, gas quenching, and/or vacuum quenching processes, crystal growth should be controlled carefully by annealing the films. Typically, perovskite films are annealed (at 60–150 °C on a hot-plate for 30–60 min) to expel residual solvents from the intermediate adduct, i.e., DMSO, thereby aiding the crystallization process. However, it has been confirmed that the quality of perovskite films can also be improved by performing an annealing step in the presence of vapor of perovskite solvents, such as DMSO, DMF, or GBL.²⁰⁹ New, fast, and contact-less annealing approaches, including rapid thermal annealing (RTA),^237,238 near-infrared (NIR),^239,240 flash infrared annealing (FIRA),^242–243 photonic curing or intense pulse light (IPL),^244,245 radiative thermal annealing,^246,247 laser annealing,^248,249 microwave annealing,²⁵⁰ and ultraviolet annealing,²⁵¹ have been introduced to improve or simplify the perovskite crystallization. The perovskite grain sizes and film quality can be significantly affected by annealing temperature and duration.^252,602 On the one hand, increasing the annealing temperature or the annealing time can increase the grain sizes.^253,254

On the other hand, it can increase the risk of perovskite decomposition by evaporating volatile organic components.^252,603 Therefore, finding an optimum annealing temperature and annealing time is very important to achieve efficient perovskite crystallization. It has been found that applying high temperatures in a few seconds on freshly spincoated perovskite films utilizing mentioned methods could be considered as an efficient annealing approach, improving perovskite crystallinity and cell efficiencies.

In conclusion, we describe the typical steps in perovskite processing applicable for different types of PSCs. We point out the critical steps during the perovskite process to achieve high-quality perovskite films and highly performing PSCs. Various laboratory- and large-scale deposition techniques are discussed. Moreover, we explore different solvent quenching and annealing methods for attaining high-quality perovskite films.

Martin Kroll, Frederik Nehm, Karl Leo

After solution processing, sublimation in vacuum is the most used alternative for perovskite thin film preparation. Most commonly, multi-source thermal evaporation is used. Here, precursor materials are loaded into separate crucibles and heated in ultra-high vacuum (10⁻⁶ mbar), with the deposition rates monitored by separate quartz microbalances (QMBs). The substrate (ideally rotating for better uniformity) is mounted above the crucibles where the evaporation cones overlap. Depending on the process, the crystalline perovskite can form directly on the substrate without annealing. However, most studies include an annealing step afterward to improve film properties.

Similar to solution processing, methylammonium (MA) lead triiodide has dominated the field for a long time and is still used in the most efficient evaporated perovskite solar cells (PSCs). Since the first report of an evaporated perovskite solar cell by Liu et al.²⁵⁷ in 2013 with an efficiency of 15.4%, several studies have improved the efficiency to over 20%.²⁵⁸ 

In the large-area mass production of similar technologies, such as OLED displays, thermal evaporation has been realized with very good upscaling properties, such as precise thickness control and homogeneity. Furthermore, thermal evaporation allows for highly customized multilayer stacks without the risk of damaging already deposited layers (i.e., with solvents) as well as homogeneous coating of textured substrates, such as silicon solar cells for tandem use.

In 2020, Li et al.²⁵⁹ showed that evaporation processes show promising results for larger area solar cells. Here, an efficiency of 18.2% for 21 cm large minimodules was reported. The modules are based on an n-i-p structure with MAPbI₃ as the absorber. To achieve these high efficiencies, the perovskite was treated with a solution containing potassium acetate and MAI. By the post processing treatment, a defect passivation and reduction of non-radiative recombination is observed.

In recent years, thermal evaporation was successfully realized on most perovskite compositions, such as mixed halide triple cation [CsFAMAPb(IBr)₃],²⁶⁰ MA free,²⁶¹ tin, and tin–lead based perovskites²⁶² and all inorganic CsPbI₃⁶⁹⁰. The cell efficiencies in those studies do not exceed 19%.

The challenges of thermally evaporated perovskites differ strongly from their solution-processed counterparts. In order to fabricate high quality films, the control of stoichiometry and microstructure of the deposited films is a key factor. For an accurate stoichiometry, rate control for the organic cation is crucial. The most commonly used methylammonium iodide is a small molecule with high vapor pressure, known to resublimate from the vacuum chamber walls. As a result, a gas-like behavior is observed upon evaporation, leading to a difference between QMB readings and the actual amount condensing on the substrate. Moreover, adhesion on the QMB surface and the substrate is dependent on the temperature and material purity, further increasing uncertainties in rate control and ultimately stoichiometry. These problems have led several groups to control the MAI evaporation by either vapor pressure or crucible temperature rather than the rate.

Beyond stoichiometry, film quality and device performance are strongly governed by the perovskite microstructure. For MAPbI₃, the crystal grain sizes are found to be significantly smaller in thermally evaporated perovskite than in solution processed films. Although the exact role of crystal grain boundaries is still under investigation, it is assumed that larger crystal grains are beneficial for the photovoltaic device performance. This also limits the development of evaporated PSCs.

Another limiting factor is the more complex process control during the deposition of the now commonly used the triple cation mixed halide perovskite. For these perovskites, the different evaporation behaviors and deposition characteristics of the organic and inorganic precursors have a strong influence on the stoichiometry and the crystallization of the perovskite film. Although triple cation perovskites were successfully deposited by,²⁶⁰ no increase in performance was reached.

For solution based processing, additives such as acids as well as solvent engineering methods have led to an improvement in device stability, performance, and reproducibility. Besides these key properties for good perovskite solar cells, the room temperature stabilization of the low bandgap cubic phase of FAPbI₃ was achieved. For vacuum processes, similar techniques are more complicated and different approaches need to be developed. Post-processing surface treatments have shown improvements in the film, but this negates the advantages of the solvent-free nature of vacuum deposition.

Due to the high initial cost of vacuum systems and comparably slow deposition process in the laboratory, progress on vacuum-evaporated perovskites is significantly slower than for their solution-processed counterparts. This has also resulted in much fewer publications on evaporated perovskites over the past few years. However, similar trends were initially seen during the early stages of OLED development, which are almost exclusively produced in vacuum nowadays.

As widely known but rarely mentioned in the community, results vary when using precursor materials from different commercial suppliers. In 2019, Borchert et al.⁸⁷ investigated methylammonium iodide from several different companies. They observed different evaporation properties in dependence of compound purity using MAI synthesized in-house as well as commercial products. By using mass spectroscopy during the evaporation process, it was possible to correlate the accuracy of QMB signals to material impurities. Interestingly, a higher amount of impurities led to a more reliable rate control, and after optimization, similar photovoltaic performances were reached. Consequently, knowledge about the precursor material is essential for an accurate deposition control.

To further control the deposition, Ross et al. investigated the sticking behavior of the organic compound to the substrate in dependence of the substrate temperature.²⁵⁸ The tooling of MAI changes drastically when varying the substrate temperature and the underlying transport layers. Using this enhanced deposition control, an efficiency of 20.6% was achieved for p-i-n structures with self-assembling monolayers as the hole transporter. This better understanding of sticking properties should help to further stabilize the deposition process. Furthermore, an optimal process window around room temperature was determined in this work.

This is in good agreement with the work of Lohmann et al.²⁶³ and Kottokkaran et al.²⁶⁴ Both investigated the influence of the substrate temperature on the crystal grain size over different temperature ranges. Lohmann showed that the crystallite size increases when cooling the substrate to 0 C, compared to room temperature. However, the device performance decreased with the increasing crystal grain size. Kottokkaran et al. observed larger crystal grains for elevated temperatures but still have the highest performing devices when depositing at room temperature. Another difference in film crystallization was found by Palazon et al. in 2019.²⁶⁵ They reported that MAPbI₃ can crystallize in the thermodynamically unexpected cubic phase at room temperature although the phase transition from tetragonal to cubic phase is expected to happen at 50–60 C. By changing the deposition rate of lead iodide, they observed a change of the as deposited film to the tetragonal phase. The cubic phase has a slightly lower bandgap, which leads to an improved photovoltaic performance of 19.7%.

Although lower efficiencies are reported for small-area vacuum deposited perovskite solar cells compared to their solution processed counterparts, vacuum sublimation yields excellent performance on the module scale and shows unique properties. The increased insight into the crystallization process leads to a better understanding of the process parameters.

The control over the evaporation process is developing and with it the potential for more basic research, improvements, and finally industrialization. Many challenges remain and the development has been slower over the last years, but with increasing interest and research, the gap in PCE could be reduced and other important properties might prevail.

Alex Redinger

Chemical vapor deposition (CVD) comprises a class of highly versatile deposition methods to produce epitaxial or polycrystalline films with a high degree of reproducibility, combined with excellent opto-electronic properties. An impressive recent example is the six junction solar cell grown epitaxially on GaAs via metal organic vapor phase epitaxy.²⁶⁶ Unlike physical vapor deposition (PVD) or molecular beam epitaxy (MBE), which need to be carried out under high vacuum conditions, CVD is a technique that can be operated up to atmospheric pressure. This additional degree of freedom allows for a better control of the reaction and re-evaporation rates of volatile species.

A large variety of different CVD techniques have been developed over the last few years, which allow to deposit high performance solar cell absorbers, extraction layers, and electrodes (for a recent review, see Ref. 267). For the synthesis of the halide perovskite absorbers, the field is largely dominated by solution based methods, which allow for a fast combinatorial screening and rapid improvement of the opto-electronic properties and device efficiencies in the laboratory.^268,269 At first sight, it seems that most of the advantages of CVD, namely, the ability to produce semiconductors with low impurity concentrations, excellent compositional control including compositional gradients via multiple stages, and the ability to upscale combined with a highly conformal deposition, are not necessary for halide perovskites. In fact, the current record efficiencies for CVD based perovskite devices are substantially lower than their chemically produced counterparts. Nevertheless, proof of concept studies have been reported for halide perovskites. Leyden et al. showed that efficient methylammonium (MA) and formamidinium (FA) based perovskites^275–272 can be produced via hybrid CVD, where in a first step PbCl₂ is deposited via PVD, followed by an exposure to a MA/FA-halide vapor. Reports on one-step CVD methods are also available but have never been pushed to a competitive device efficiency level.^273,274

The current halide perovskite record devices are mostly based on multi-cation mixed halide absorbers.^269,275 These solar cells exhibit an improved stability and performance compared to prototypical methylammonium lead iodide⁴⁷ but impose challenges for CVD, since many precursors and deposition steps need to be developed and controlled. Consequently, reports on multi-cation/halide perovskite absorbers produced via CVD methods are scarce,²⁷⁶ and a fair comparison between CVD and solution based methods in terms of device performance is not yet feasible.

The application of CVD and atomic layer deposition (ALD) techniques is more common in the perovskite community to produce high performance carrier selective layers, electrodes, and buffer- and encapsulation-layers (see Sec. IV D for a more detailed discussion on extraction layers). Most of the work on electron extraction layers has been devoted to TiO₂, ZnO, and SnO₂, whereas the most common HTL produced via CVD is NiO. Passivation layers, such as Al₂O₃, have also been developed for perovskites in order to reduce the interface recombination velocity and prevent ionic mobility induced degradation.²⁶⁷ 

It is generally accepted that Si-based single junction solar cells will continue to dominate the PV market. The true benefit of halide perovskites are closely linked to the ability to make high performance tandem devices with improved performance compared to single junction devices. PVD and CVD methods are ideal tools to make efficient tandem devices since large area conformal coating on any type of planar and textured substrates can be carried out. Consequently, a further thorough development of these deposition techniques is highly desirable.

The current challenges for CVD based high performance devices are first of all the speed and dynamics at which the field is developing. New processes, doping and passivation strategies, are being developed at high speed via solution based methods, which makes it difficult for researchers to design custom built evaporation/reaction chambers and processes that allow for highest reproducibility and device performance.

Two different strategies can be envisioned. Researchers wait until the race for the presumably best perovskite composition is over and then PVD and CVD based large area deposition processes for industrial applications can be developed. This strategy, however, assumes that the most stable high-performance halide perovskite is similar for the case of chemical deposition methods and CVD based methods. This may not be the case since impurity concentrations, interface defects, and their passivation will certainly depend on the deposition process and thereby affect the device efficiency and stability.

Therefore, the best approach is to be inspired by the current race for ultimate efficiency via solution based methods but nevertheless study as many material combinations as possible to identify the most stable and best performing CVD based halide perovskite composition. At present, not enough data are available for CVD based halide perovskites and more work is necessary. The focus should lie on developing perovskite absorbers that can be produced with a minimum number of precursors and reaction steps in order to be industrially relevant.

It needs to be emphasized that CVD is a technique that can be used very effectively to produce layers at higher processing temperatures as in PVD (60C²⁵⁸ vs 140C²⁷⁰). This unique advantage, which is a direct consequence of the ability to use high partial pressures of the volatile organic component, may allow forming perovskite absorbers that are more temperature resistant since the semiconductor can be grown closer to equilibrium. CVD and PVD techniques have the advantage that very clean semiconductors can be grown with a low number of impurities that might deteriorate the PV performance. However, both techniques require very clean raw materials, which is not always the case as discussed in more detail in Sec. III C. A disadvantage of CVD compared to PVD is certainly the flexibility to grow intentional gradients in composition over the thickness of the absorber. Dedicated metal organic vapor phase epitaxy systems tuned to the properties and precursors of for halide perovskites would need to be developed, as they are currently not available.

Over the last few years, very powerful diagnostic tools have been developed that allow for early material screening, such as time resolved photoluminescence and photoluminescence quantum yield (PLQY) measurements.²⁷⁷ Unfortunately, such measurements are scarce for CVD based perovskites. If perovskite layers with PLQY values and minority carrier lifetimes similar to solution based perovskites can be achieved on glass substrates, then there is no fundamental limit that impedes this material to be a suitable candidate for high efficiency devices. In large samples with intentional gradients in element concentration and temperature, combinatorial methodologies should be used to screen the whole parameter space in an efficient manner (as done, for example, for coevaporated CsPbI₃²⁷⁸). This is particularly simple for CVD based methods where the individual zone temperatures and flow rates can be adjusted precisely to produce compositional gradients along the sample.

In particular, the impact of the elemental composition on stability and device performance needs to be studied carefully. This, however, requires developing extraction layers that are tailored for CVD based absorbers. A simple copy of the extraction layers from solution based approaches will not lead to ultimate efficiencies. Similar combinatorial strategies combined with luminescence techniques²⁷⁹ can be applied to find the best performing extraction layers that reduce interface recombination losses while providing excellent selectivity. The layers should be chosen such that they can be transferred to tandem configurations. A strong focus on low temperature CVD and ALD methods is essential, especially if these layers will be grown on top of the perovskite surface. In tandem configurations, the impact of ALD and CVD layers on the bottom cell also need to be taken into account.

Finally, CVD based perovskite absorbers should be used to design the reference material to study fundamental physical and chemical properties of the semiconductors. The growth and characterization of the epitaxial material on lattice matched substrates or via van der Waals epitaxy has only been studied in some rare cases.²⁸⁰ These films would be ideal to study the impact of grain boundaries on moisture induced degradation or could be used to characterize the band structure via angle resolved photoemission measurements (see Sec. IV F for details). In addition, CVD based films can be grown under very clean conditions, which would allow us to understand the impact of impurities on degradation, thereby excluding external factors, such as solvents. It is currently not clear to which extent the stability of different perovskite layers is altered by the deposition process itself. Here, PVD and CVD based methods offer an ideal playground to study the fundamental limits of the perovskite absorbers.

In order to meet these challenges, the following criteria need to be considered and carefully evaluated. Studies on solution based devices have shown that the MA based perovskites are too unstable to meet the necessary stability criteria.²⁸¹ Dedicated experiments for CVD based absorbers with optimized extraction layers and diffusion barriers need to be carried out to corroborate that MA based CVD perovskites exhibit similar disadvantages as their chemically produced counterpart. Some results on evaporated²⁸² and CVD based MA based absorbers²⁷⁰ show promising stability and this needs to be further analyzed.

In the next step, formamidinium based absorbers need to be developed to a competitive device efficiency level. Here, a combinatorial approach with FA, Cs, and Sn should be used followed by dedicated surface and interface passivation strategies and tailored extraction layers. A critical mass of studies needs to be carried out and the bottlenecks need to be evaluated via PLQY measurements after the deposition of each layer. This will finally allow to corroborate if CVD based absorbers can meet the stability and efficiency requirements for large scale manufacturing.

Strain engineering could be a valuable tool to study how the metastable photovoltaic active phase can be induced and retained in FA-based devices produced by CVD. Preliminary results of FA-based absorbers show that depending on the processing conditions, the photovoltaic active phase can be grown quite effectively.²⁷¹ Bandgap engineering via intentional compositional gradients in the perovskite absorber has already been reported to be beneficial for device performance.²⁷⁶ This technique, which has been proven to be very important for other thin film solar cell technologies,^283,284 needs to be investigated and developed further.

Novel architectures via a combination of PVD/CVD based perovskites with graphene or other transition metal dichalcogenides, which may lead to van der Waals epitaxy, should be studied in order to reduce interface recombination combined with impermeable charge extraction layers. In that respect, solution based large area graphene/perovskite solar cells show very promising results, and further improvements via CVD and vapor deposition methods can be expected.²⁸⁵ 

Furthermore, the field of low dimensional perovskites should also be investigated further via CVD based techniques (for details, see Sec. II F). Here, the number of publications is still very low, and the true potential of CVD based low dimensional perovskites is still to be exploited. Again, a combination with other 2d materials seems to be a very attractive pathway to limit recombination and produce interfaces with low defect concentrations.

Finally, a strong focus of CVD based methods to produce high efficiency stable devices with a scalable technology is mandatory. Although the current record tandem devices are currently produced with a Si-bottom layer,²⁶⁹ other high efficiency technologies, such as Cu(In,Ga)Se₂, offer unique properties, such as the ability to produce flexible devices^286,287 combined with a low energy payback time. CVD based extraction layers at low processing temperatures need to be developed with special emphasis on low permeation rates for water in order to improve stability. These extraction layers need to be tested on CVD based perovskite absorbers, and detailed band alignment studies via photo-electron spectroscopy methods need to be carried out.

CVD based methods have shown very promising results over the last few years despite the fact that the method is only used in some research groups and efficiency is driven mostly by solution based methods. However, the true potential of this technique becomes only visible when reproducibility on large scales for large volumes becomes important. It is very probable that a compromise between efficiency and processibility needs to be found. In that respect, the research should focus on developing high performance devices with the lowest possible number of elements in the solar cell in order to be as competitive as possible. The number of precursors and processing steps in CVD based perovskites needs to be reduced to a minimum. Dedicated inline process control techniques need to be developed combined with an optimization of all the layers in the device. The true potential of CVD to produce competitive high performance devices is still to be explored. Given the potential and the unique advantages of this technique, more scientific effort in that direction is likely to be very fruitful.

Julian Höcker, Vladmir Dyakonov

The studies of perovskite single crystals with high crystallographic quality is an important technological area of the perovskite research, which enables to estimate their full optoelectronic potential and thus to foster their future applications. It is therefore essential to grow high-quality single crystals with lowest structural as well as chemical defect densities and with a stoichiometry relevant for their thin-film counterparts. Optoelectronic devices, e.g., solar cells, are highly complex systems in which the properties of the active layer (absorber) are strongly influenced by the adjacent layers, so it is not always easy to define the targeted properties and elaborate the design rules for the active layer. Currently, metal halide perovskite single crystals with the structure ABX₃ are the most studied crystalline systems. These hybrid crystals are solids composed of an organic cation, such as methylammonium (A = MA⁺) or formamidinium (A = FA⁺), to form a three-dimensional periodic lattice together with lead (B = Pb²⁺) and a halogen, such as chlorine, bromine, or iodine (X = Cl⁻, Br⁻, or I⁻). Among them are methylammonium lead tribromide (MAPbBr₃) and methylammonium lead triiodide (MAPbI₃) as well as methylammonium lead trichloride (MAPbCl₃).^288,289 Important representatives with the larger cation FA⁺ are formamidinium lead tribromide (FAPbBr₃) and formamidinium lead triiodide (FAPbI₃).²⁸⁹ Besides the exchange of cations as well as anions, it is possible to grow crystals containing two halogens to obtain mixed crystals with different proportions of chlorine to bromine and bromine to iodine, as shown in Fig. 19. By varying the mixing ratio of the halogens, it is therefore possible to vary the color and thus the absorption properties of the crystals,²⁹⁰ as it can be done with thin polycrystalline perovskite films. In addition, since a few years, it is also possible to grow complex crystals that contain several cations as well as anions. These include the double cation–double halide perovskite, formamidinium lead triiodide methylammonium lead tribromide [(FAPbI₃)_0.9(MAPbBr₃)_0.1 (FAMA)]²⁹¹ and formamidinium lead triiodide methylammonium lead tribromide cesium lead tribromide [(FAPbI₃)_0.9(MAPbBr₃)_0.05(CsPbBr₃)_0.05 (CsFAMA)],²⁹² which have made a significant contribution to increase the efficiency in thin-film photovoltaics. The growth of crystals to this day is performed exclusively from solution. Important preparation methods are the cooling acid-based precursor solution crystallization,²⁹³ the inverse temperature crystallization (ITC),²⁹⁴ and the antisolvent vapor-assistant crystallization (AVC).²⁹⁵ In the cooling crystallization, the precursor salts AX and PbX₂ are dissolved in an aqueous halogen-containing acid at high temperatures.²⁹⁶ Controlled and slow cooling finally results in a supersaturated precursor solution, which leads to spontaneous nucleation of crystal nuclei followed by subsequent crystal growth. The ITC method is based on the retrograde solubility of a dissociated perovskite in an organic solvent.²⁹⁴ With the increasing temperature, the solubility of the perovskite decreases and mm-sized crystals can be grown within a few hours.²⁹⁴ In the AVC method, the precursors are also dissolved in an organic solvent as well.²⁹⁵ By slow evaporation of a so-called antisolvent,²⁹⁶ the solubility of the perovskite in the now present solvent mixture decreases and it finally precipitates. In addition, there are many other methods with the goal of growing high quality and large crystals in a short period of time.²⁹⁷ 

The investigation of high-quality crystals not only offers the advantage of studying the pure perovskite material system but also that the crystals have fewer lattice defects in contrast to their thin film counterparts. Their excellent optoelectronic properties range from high carrier mobilities,²⁹⁸ long charge carrier lifetimes,²⁹⁹ and diffusion lengths³⁰⁰ to high photoconductivity³⁰¹ and high sensitivity to high energy radiation.³⁰² For these reasons, they are already used as semiconducting components in solar cells,²⁹⁶ x-ray-detectors,³⁰² gamma-ray-detectors,³⁰³ and photodetectors.³⁰⁴ However, there are some challenges on the way from crystal growth to the device prototypes: Many crystallization methods, such as cooling crystallization, but also the AVC method, need a much longer period of time for single crystal growth. Moreover, it is difficult to adjust the nucleation rate to grow few but large-sized crystals. The ITC method can overcome these challenges. However, the inverse solubility²⁹⁴ effect can be a problem in growing defect-free crystals because the probability of lattice defects is rising with the increasing growth temperature. The use of seed crystals to obtain large crystalline solids can also be a problem, as this leads to irregularities in the crystal lattice. For future crystal growth, the requirements of time, temperature, controllable nucleation rate, and the associated crystal size must play an important role. In addition, another overriding goal must be to pull large-sized single crystals from the melt rather than from solution, as is already possible with the Bridgeman method for the growth of CsPbBr₃,³⁰⁵ in order to find large-scale applications, such as with gallium arsenide (GaAs) and silicon (Si). Besides preparative challenges, there are challenges in interpreting the experimental data obtained on perovskite crystals. In the literature, there are often strongly deviating values for the device-relevant optoelectronic parameters, for example, charge carrier mobility, which can be determined by time-of-flight (TOF), space charge limited current (SCLC), or Hall effect methods.²⁹⁶ Similarly, the experimentally determined values of carrier lifetime and diffusion length can vary by up to two orders of magnitude according to the literature.²⁹⁶ Finally, many optoelectronic parameters, especially for more complex crystals, such as FAMA or CsFAMA, are not yet known.

The number of techniques for growing metal halide perovskite single crystals is constantly increasing. The above-mentioned crystallization techniques can be combined, and different chemical reactions are applied to grow high quality single crystals. A mechanical approach to grow complex and large-sized crystals without the use of seed crystals is realized by the re-fill crystallization method (RFCM),³⁰⁶ which is based on the inverse solubility of ITC.²⁹⁴ By introducing new precursor solution and simultaneously removing the depleted solution, it is possible to grow in situ large-sized crystals with excellent structural quality, as confirmed by the out-of-plane X-ray diffraction (XRD) measurements and the rocking curve of a FAMA single crystal in Fig. 19.³⁰⁶ These quality characteristics of the crystals are not only visible from structural measurements but also show up in the physical parameter of the trap density. In contrast to perovskite thin films, the trap densities in lead-halide perovskite single crystals are usually 4–5 orders of magnitude smaller²⁹⁶ due to the high quality of the crystals. With such a variety of crystallization techniques, it is now possible to grow all semiconducting perovskites known from thin-film device applications as mm-sized crystals. As mentioned above, by selectively changing the halogen ratio, it is possible to obtain crystals with a wide variety of colors and optical bandgaps, which may serve as photodetectors in the UV/VIS range.^288,296 However, the variation of the halogens not only results in a change of the optical properties but also results in an increase in the lattice constant with increasing bromine to chlorine and iodine to bromine ratios, respectively (Fig. 19). In addition, the lattice structure changes from a simple cubic lattice^288,293,294 for MAPbCl₃ and MAPbBr₃ to a body-centered tetragonal lattice for MAPbI₃.^293,294 An important outcome of these efforts is the successful fabrication of single crystal solar cells, which achieves over 21% power conversion efficiency, representing a new benchmark in the field of metal halide perovskite single crystal solar cells.^312–309 While mm-thick crystals were found less suitable for this purpose, μm-thick MAPbI₃ wafers were shown to be more promising for the production of photovoltaic devices.³¹⁰ As shown in Fig. 19, a crystal wafer can be easily grown on a poly[bis(4-phenyl) (2, 4,6-trimethylphenyl) amine] (PTAA)-coated silicon wafer, while the growth height of the crystal is limited by spacers.

The field of metal halide perovskite single crystals is both inspired and driven by the fact that the most efficient silicon solar cells are made with the highest quality large area crystalline absorbers. Encouragingly, the power conversion efficiency over 21% for solar cells fabricated with single-crystalline MAPbI₃ was reported.³⁰⁸ Interestingly, other major activities in the field are focused on single-crystal-based x-ray-detectors,³¹¹ gamma-ray-detectors,³⁰³ and photodetectors,³⁰⁴ which show promising performance in terms of sensitivity, response speed, and lowest detectable dose rate. In the future, large-area single crystalline wafers with controlled thickness and area are logically expected to come into focus. One of the possible barriers to studying optoelectronic properties even in high-quality crystals remains ion migration with the resulting complex electric field distribution, which is a similar problem to what is known for perovskite thin films and devices.³¹² We would like to add that a general goal of growing large-sized crystals with a complex composition should be done together with a critical evaluation of the term “single crystal” in relation to these crystals. On the other hand, the crystallographic and stoichiometric qualities alone do not allow yet any conclusions about the “electronic quality,” which should also be verified. Nevertheless, the development of large-area, highly crystalline, and low-defect density perovskites remains an important platform to design the easy-to-grow highly functional materials, on the one hand, and to understand their optoelectronic properties, on the other hand, to enable new applications, beyond what we are mainly dealing with to date. High optical bulk quality and compositional variability can make them interesting, for example, for quantum applications, spin-based ones or as coherent photon emitters, which have not yet been intensively worked on.

Thomas Kirchartz, Jonathan Warby, Emilio Gutierrez-Partida, Dieter Neher, Martin Stolterfoht, Uli Würfel, Moritz Unmüssig, Jan Herterich, Clemens Baretzky, John Mohanraj, Mukundan Thelakkat, Clément Maheu, Wolfram Jaegermann, Thomas Mayer, Janek Rieger, Thomas Fauster, Daniel Niesner

Thomas Kirchartz

High photovoltaic power conversion efficiencies require a combination of a high photocurrent and a high free energy per extracted electron, i.e., a high voltage at a given current. The typically used figure of merit to compare the ability of the solar cell to extract a large free energy per carrier is the voltage V_oc in the limit of infinitely slow charge extraction, i.e., at the open circuit. This voltage can be understood as the electrostatic potential difference between the two metal electrodes of the solar cell, but qV_oc is furthermore also typically identical to the quasi-Fermi level splitting side of the photovoltaic absorber layer. In efficient lead-halide perovskites, this open-circuit voltage V_oc is remarkably high relative to its thermodynamic limit (V_oc,SQ) given by the Shockley–Queisser model.³¹³ The value of V_oc,SQ is ∼280 mV less than E_g/q (the bandgap divided by the elementary charge), which gives around 1.32 V for typical perovskite bandgaps of 1.6 eV. As shown in Fig. 20(a), the highest open-circuit voltages reached so far in the literature approach the thermodynamic limit by about 60 mV,³¹⁴ which is an excellent value for any solar cell technology, let alone a technology mainly based on solution-processed polycrystalline thin films. Figure 20(b) shows that the evolution of non-radiative voltage losses over the last few years has seen substantial improvements from >300 mV losses to ∼60 mV.

The key to high open-circuit voltages relative to V_oc,SQ is to allow a high concentration of charge carriers to build up, which leads to a high quasi-Fermi level splitting and subsequently a high voltage at the open circuit. In the absence of charge extraction, generation and recombination of electron hole pairs are the only mechanisms that control the concentrations of electrons and holes. The key is therefore to slow down recombination. Given that radiative recombination as the inverse process of photoabsorption must happen due to detailed balance,³¹⁵ the only recombination types that can be slowed down by proper device design are non-radiative recombination processes. Non-radiative recombination typically occurs via multiphonon transitions from the conduction band to a localized state in the bandgap (defects) and from there to the valence band. In the harmonic oscillator approximation, multiphonon transitions should become very unlikely if large energy steps in units of the phonon energy have to be overcome in a single transition.³¹⁶ Therefore, shallow defects are generally considered less detrimental in terms of accelerating recombination because the transition of a shallow defect to the other band edge has to bridge an energy of nearly the bandgap in a single transition. Lead-halide perovskites have antibonding valence band edges leading to a higher likelihood of shallow intrinsic defects and a lower likelihood of deep intrinsic defects. This fact is typically called “defect tolerance”³¹⁷ and could, in combination with the fairly low phonon energies of lead-halides, lead to very slow recombination as compared to other polycrystalline semiconductors. The figure of merit for slow recombination in a semiconductor is typically the charge-carrier lifetime—a quantity that will depend on charge-carrier concentration in an intrinsic or highly excited semiconductor. Therefore, the determination of charge-carrier lifetimes may not always be done at exactly the same conditions. Nevertheless, it is clear that bulk lifetimes in lead-halide perovskites can exceed several microseconds,^277,653 i.e., values that are much longer than those of other polycrystalline direct semiconductors, such as Cu(In,Ga)Se₂. A key requirement for such long lifetimes is a well-passivated surface of the perovskite film, which is often achieved by using passivating molecules, such as n-trioctylphosphine oxide (TOPO).³¹⁸ 

By measuring the external quantum efficiency of photoluminescence (PL) of a semiconductor film, one can calculate back the quasi-Fermi level splitting ΔE_F in the film that must have led to the measured PL intensity. The same method would also work for layer stacks including interfaces to charge transport layers (CTLs). Therefore, steady-state PL is a widely used method to assess the quality and the ability of semiconductor layers or layer stacks to enable a high open-circuit voltage if used as part of a full device stack. A key finding in the literature is that TOPO passivated methylammonium lead iodide (CH₃NH₃PbI₃, MAPI) films achieve ΔE_F of up to 1.28 eV, i.e., 40 meV less than what would be expected from a sample without any non-radiative recombination.³¹⁸ Additional losses to the V_oc of the finished device can be extremely small in some rare cases, but for the bulk of the published data in the literature, the V_oc is substantially lower than what might be possible based on the quality of the bulk material. The key culprit causing these additional losses is interfaces to charge transport layers. At interfaces, the conduction and valence bands are often not perfectly aligned, thus leading to an increased product of electron and hole concentrations at a given quasi-Fermi level splitting. This is equivalent with saying that at the interface, the energy gap separating electrons and holes across the interface is lower than in the bulk. Thereby, even for reasonably low interface recombination velocities, the recombination rate at the interfaces might be substantial enough to limit the achievable quasi-Fermi level splitting and eventually V_oc in a finished device.

The importance of interfacial recombination for limiting V_oc is a result of detailed sample characterization efforts,²⁷⁹ but it is also an implicit consequence of the success of surface passivation efforts to improve efficiencies and open-circuit voltages. Examples that resulted in a successful reduction in interfacial recombination are the surface treatment with salts, such as phenethylammonium iodide (PEAI),²⁰² n-butylammonium bromide (BABr),³¹⁹ or guanidinium bromide,³²⁰ to name just a few. Other successful efforts to improve V_oc were achieved by exchanging the hole or electron transport layer (ETL) and thereby tuning of energy level alignment or introduction of interfacial dipoles.³²¹ 

Thanks to the current level of success in implementing surface and bulk passivation strategies, the open-circuit voltages of perovskite solar cells have been approaching levels of ∼10% external luminescence quantum yield and non-radiative V_oc-losses of around 60 mV over a range of bandgaps from around 1.5 eV to around 1.6 eV, which is better than any photovoltaic technology except for GaAs.³²² Toward higher bandgaps, however, the voltage losses usually increase, as shown in Fig. 20(a), with only rare exceptions showing lower losses.⁶⁵³ Thus, one further strategy in the field should be to reduce recombination losses over a wider range of bandgap values.³²³ Further reducing the voltage losses below 60 mV might require tackling not only recombination losses but also parasitic absorption losses that reduce photon recycling and thereby the charge carrier concentration and voltage at constant charge generation.³²⁴ Reduction of parasitic absorption by, e.g., improving the optical quality of the back mirror is an important topic for achieving higher efficiencies in GaAs solar cells and might become more relevant for perovskites as well. Finally, the role of ion movement, the electric field, and the built-in voltage on charge recombination and open-circuit voltage has to be better understood. While there are some reports exploring ion movement in experiment and simulation on JV curves and photovoltage transients, the employed cells were not efficient, given today’s standards.³⁶⁵ In contrast, there are also publications that provide evidence that ions are hardly relevant for understanding the open-circuit voltage of efficient solar cells.⁶⁵² Furthermore, the origin of changes in V_oc over seconds and minutes of illumination is not fully understood. Thus, a combination of systematic experimental and modeling studies³³⁷ on high performance cells will hopefully provide evidence on how to optimize the electrostatic potential distribution for maximum V_oc.

The open-circuit voltages of lead-halide perovskite solar cells relative to V_oc,SQ have always been high in comparison with other photovoltaic thin-film technologies and in consideration of the maturity of the technology. Over the last few years, the gap to the thermodynamic limit has been continuously shrinking with a wide variety of passivation methods and improvements in the alignment between the absorber and charge transport layers being mainly responsible for the improvement. The underlying reasons for the remaining losses are non-radiative recombination via defects and parasitic absorption of luminescence. However, the understanding of the defect physics of perovskite solar cells is currently still at an early stage. One challenge is that the measurement of low defect densities and energetic positions of defects is difficult with most methods being prone to misinterpretation.⁶⁵⁴ In addition, the calculation of multiphonon recombination processes between extended and localized states in highly anharmonic semiconductors, such as lead-halide perovskites, is difficult and has so far been only performed in few cases for specific defects in specific perovskite compositions.^343,655 An additional challenge for optimizing V_oc is the huge variety of compositions with different bandgaps and energy levels that each require slightly different contact layers as well as processing and passivation strategies.

Jonathan Warby, Emilio Gutierrez-Partida, Dieter Neher, Martin Stolterfoht

The study of non-radiative recombination has long been at the forefront of academic research on various photovoltaic (PV) technologies. This is because non-radiative recombination is a loss mechanism in solar cells, lowering the external radiative efficiency [in the following, called the photoluminescence quantum yield (PLQY)], which, in turn, lowers the FF and V_OC of the cell. Non-radiative recombination in semiconductors is typically caused by defects and impurities, which leads to trap-assisted recombination through midgap states. In silicon PV, extremely high-quality wafers with low defect densities in the bulk (≪1 ppb) were achieved through extended growth of single crystals at high temperatures (∼1100 °C) using high purity precursors. Contrastingly, perovskites with low deep trap state densities can be fabricated from relatively impure materials and low temperature processes, which has allowed the field to rapidly progress as researchers can make efficient perovskite solar cells using simple laboratory equipment. However, in contrast to silicon where the PLQY of the “neat material” or as-cut wafers (1 × 10⁻⁸–1 × 10⁻⁷)³²⁶ is generally enhanced by orders of magnitude (up to >1%) upon addition of tailored passivation and charge transport layers (TLs) [Fig. 21(a)], for perovskites, this has not been achieved yet.²⁷⁹ In fact, for a large number of perovskite systems, photoluminescence measurements on partial cell stacks and other transient techniques have demonstrated that the recombination losses in the complete cell remain stubbornly dominated by interfacial recombination at one or both perovskite/transport layer heterojunctions [Fig. 21(b)].^{279,332–330} Despite this, the study of non-radiative interfacial processes has been largely left out of the limelight in a field that has strongly focused on improving the bulk quality and the passivation of electronic defects on the surface. Clearly, to make progress further toward the thermodynamic limits, the field needs to disentangle the underlying physics that causes interfacial recombination.

Today, state-of-the-art perovskite solar cells have open-circuit voltages that are approaching the radiative limit.^202,331 Jiang et al.²⁰² showed an exceptionally high V_OC of 1.18 V at a bandgap of 1.54 eV, and they measured an electroluminescence quantum yield (EQE_EL) of ∼7% at an injection current of J_SC, which means a non-radiative V_OC-loss[$kT/q⁡lnEQEEL$] of only 69 meV. Similarly, in the paper with the highest certified efficiency of 24.4%, we can predict their EQE_EL of 6.8% from their V_OC of 1.162 V and an estimated bandgap of 1.5 eV.³³¹ As shown in Fig. 21(c), these values for EQE_EL are outperforming the best EQE_EL values reported for monocrystalline Si (1.6%) values and approach the highest EQE_EL of III-V semiconductor solar cells, e.g., GaAs (35.7%).³²² Very encouragingly, the perovskite as prepared following the work of Jiang et al. exhibits a PLQY of 25% with a Shockley–Read–Hall lifetime of over 18 s even without additional passivation. In such a material, the losses with respect to the radiative limits [see Fig. 21(d)] are likely related to parasitic absorption and imperfect light outcoupling management. This underlines further that perovskite solar cells can closely approach their thermodynamic limits and surpass silicon in power conversion efficiency if the losses associated with the interface recombination can be controlled.

The problem of interface recombination is particularly evident in (but not limited to) pin-type solar cells that have lower efficiencies (25.4% vs 22.3%) than nip-type cells. Yet, pin-type cells are more desirable for application in perovskite tandem solar cells, partially due to lower temperature processing (100 C). An interesting observation is that in both nip- and pin-type architectures, the transport layer deposited on top of the perovskite is often limiting the total recombination loss of the device, while near lossless transport layers, such as SnO₂,²⁷⁹ different self-assembled monolayers (SAMs)³²¹ or PTAA/PFN-Br³³² can be employed below the perovskite [Fig. 21(b)]. This leads to the question of whether (i) there is a difference in the surface states between the front and back side of the device, which causes the penalty, or (ii) if it is due to engineering considerations of what can be deposited on top without damaging the perovskite.

In pin-type cells, C₆₀ is commonly used as the charge transport layer, which is deposited on the top surface, and it lowers the PLQY of the neat material by 1–2 orders of magnitude to 1 × 10⁻⁴–1 × 10⁻³, as shown for many different perovskites.^{279,330,337–334} It is therefore of expressly high importance to address the losses at the perovskite/C₆₀ interface, which currently remain poorly understood. It is likely that the loss is related to midgap states or a broad density of states in C₆₀ acting as efficient non-radiative recombination centers. However, at present, an energetic misalignment at the perovskite/C₆₀ interface or C₆₀-induced midgap states on the perovskite surface cannot be excluded. The potential loss mechanism is shown in [Fig. 21(e)]. In contrast, in nip-type cells, heavily doped spiro-OMeTAD is typically used as a hole transport layer. The strong doping and the expected increase in the built-in field³³⁵ over the perovskite or increased back-surface field could explain why interfacial recombination can be suppressed to a minimum in record nip-type cells (cf. pin-cells). However, n-doping the C₆₀ or PCBM has not yet been very effective at suppressing interfacial recombination, which might be due to the instability of n-dopants and their potential reactivity.³³⁶ In order to clear up this confusion, the nature of the built-in potential distribution needs to be better understood, which is significantly complicated by the presence of mobile ions in the perovskite absorber layer.^337,338 Many groups have reported mobile ion densities significantly exceeding the electronic or the capacitive carrier density.^344–341 In this case, the built-in potential would drop across a relatively small portion of the absorber layer by ion accumulation at the interfaces, which impedes efficient charge extraction.³⁴² Therefore, the seemingly conflicting information between high performance and attenuation of the built-in field by mobile ions and their impact on the interfacial recombination remain a pressing and open topic. All this currently complicates the design rules for tailored charge transport layers that reduce surface recombination while rapidly extracting the majority carriers—again highlighting the need for deeper understanding of interfacial recombination.

In order to reduce the recombination rate in the bulk or at the surface of a semiconductor, there are several distinct possibilities that the experimentalist has: (i) reducing the number of defect states via chemical passivation, (ii) reducing the interfacial transfer rate through an energetic barrier for minority carriers, e.g., with wide-gap interlayers, and (iii) reducing the concentration of the minority carrier at that surface by creating a back surface field [Fig. 21(f)], e.g., through surface or TL doping or molecular dipoles. Until today, the perovskite solar cell community has predominantly focused its efforts on chemical passivation, which may be partially due to the distinct ionic nature of the perovskite that creates electronically charged defects opening a myriad of possibilities for chemical passivation (e.g., coordinate bonding, addition of Lewis bases or acids, and many more).^343,344 Moreover, the simple solution processability and defect tolerant nature of perovskites allow the addition of chemical additives to the precursor solution, which may ultimately lead to a reduced defect density in the crystallized films. However, although these methods have resulted in impressive PCE improvements, it remains unclear if additive-based passivation can push the technology to its practical PCE limit of over 30% in the next years. An alternative strategy is the use of thin insulating layers as tunnel barriers to reduce the minority carrier concentration from the critical interface (ii).^327,345,346 While early works on such wide-gap insulators have demonstrated promising performance improvements, their applicability appears to be limited to certain perovskite compositions for which the benefit of the reduced transfer of the minority carriers outweighs the negative effects associated with the increased transport resistance. Similarly, a wide-gap interlayer can also be formed through compositional engineering of the perovskite, for example, by increasing the Br content at the surface.³⁴⁷ More recently, the application of wide-gap two-dimensional perovskite layers has become popular to reduced hole density at the critical perovskite/spiro-OMeTAD interface by creating 3D/2D “homojunctions” with negligible interfacial recombination and reduced surface recombination.³⁴⁸ Ultimately, this approach has allowed us to improve the open-circuit voltage to over 1.3 V for 1.72 eV nip-type cells.³¹⁹ In contrast, the third strategy of repelling minority carriers from the interface through graded doping and the creation of a back-surface field or increased built-in voltage has proved to be challenging, although a more n-type surface has been found to correlate with a higher V_OC in pin-type cells.^320,349,350

As perovskites have benefitted from a prolonged “honeymoon period” where improved processing of the perovskite and transport layers has allowed efficiencies to continually increase, interfacial recombination at perovskite/transport layer heterojunctions has largely been left out of the limelight. Today, as the opto-electronic quality of perovskite semiconductors is essentially at the radiative limits, the efficiency gains made by the field are beginning to plateau, necessitating the need of a clear understanding of the processes causing interfacial recombination. In particular, for pin-type cells, the losses at the top interface to C₆₀ appear to be omnipresent and universal to various perovskite compositions. However, similar losses are also observed for various other transport layers underlining the difficulties in creating ideal perovskite/semiconductor heterojunctions. In contrast, perovskite/transport layer homojunctions (e.g., 3D/2D perovskites) may prove to be effective at suppressing the minority carrier recombination at the critical interface while allowing rapid charge extraction. Overall, understanding and solving these problems will ensure further progress toward realizing near-ideal perovskite cells at the radiative limits, possibly for different bandgaps required for multi-junction perovskite solar cells.

Uli Würfel, Moritz Unmüssig, Jan Herterich, Clemens Baretzky

Interfaces (between different materials) play a decisive role for the performance of perovskite solar cells. In contrast to conventional crystalline silicon solar cells, devices based on perovskite absorbers do not make use of a pn-junction within the perovskite material. In contrast, perovskite solar cells are devices comprising a p–i–n or n–i–p heterojunction. The “i” means an intrinsic absorber sandwiched between an n-doped and a p-doped material, respectively. Note that several studies indicate that perovskite absorbers are, in fact not, very intrinsic and doping levels up to 7.6 × 10²⁰ cm⁻³ have, e.g., been derived from Hall effect measurements.^358–358 However, note that it is not intentionally doped and that there is no pn-junction within the absorber layer itself.

The perovskite absorber is sandwiched between the electron and hole transport layers, ETL and HTL. They are used to achieve the following four conditions:

1. passivation of surface states, i.e., ensuring a low interfacial density of states with energies within the bandgap of the perovskite absorber;

2. lossless extraction of majority carriers, achieved by providing a high corresponding electric conductivity;

3. an energetic barrier for the minority carriers in order not to extract both type of charge carriers from the absorber, achieved through a large bandgap and favorable band alignment; and

4. very low absorption in the transport layers as to maximize the absorption in the perovskite absorber layer, usually achieved by application of a thin layer with a rather large bandgap.

ETL and HTL comprise in most cases a rather large bandgap and are chosen such that the conduction (valence) band of the n-type (p-type) material matches well with the conduction (valence) band of the perovskite material. Then, due to their bandgaps being larger than that of the absorber, an energetic offset results for the holes (electrons) at the absorber/n-type (absorber/p-type) interface. These energetic offsets impede the extraction of the “wrong” type of charge carrier. To ensure a good conductivity for the electrons (holes), the ETL (HTL) is usually n-doped (p-doped). Moreover, doping has also another beneficial effect, which will be discussed below.

Although an energetic offset of several hundreds of meV would suffice, it has to be considered that such materials and, especially such interfaces, are not ideal, i.e., condition (1) is not fully met. In fact, every interface constitutes an abrupt interruption of the periodicity in which atoms are arranged within the material. Consequently, there are states not only for electrons in the respective bands but also within the bandgap, the so-called interface (trap) states. For this reason, to impede significant interface recombination, the minority carriers have to be prevented from reaching that interface. In a fundamental study, it was shown that the decisive ingredient to achieve this is a very low conductivity of the holes (electrons) in the absorber close to the absorber/ETL (absorber/HTL) interface.³⁵⁹ This is the second effect of the doping of the transport layers: it keeps the concentration of the minority carriers in the absorber close to the interface at a low value.

Processing techniques to apply ETL and HTL materials are either solution- or vacuum-based. Most organic materials, such as MeO-2PACz, spiro-OMeTAD, PTAA, PEDOT:PSS, and PCBM, and some inorganic materials, such as ZnO, TiO_x, ZrO₂, and graphite, as well as organic–inorganic 2D-perovskites are deposited from solution via spin-coating, blade-coating, slot-die coating, spray coating, or inkjet-printing, each requesting separate optimization.^171,321,360 Fullerenes and many inorganic materials are deposited by evaporation in ultrahigh vacuum (UHV).^361,362 Moreover, atomic layer deposition is also frequently used in the context of perovskite solar cells, e.g., for SnO_x.^363,364 Much of the impressive progress in the efficiency of perovskite solar cells in recent years is, in fact, due to improvements in the quality of the interfaces between the absorber and transport layers.

As mentioned in Sec. IV B, the presence of mobile ionic species (ions and ion vacancies) in the absorber makes the situation at those interfaces in perovskite solar cells even more complex.^{9,339,370–368} These ionic species move within the perovskite absorber to reach electrochemical equilibrium, which includes their reaction on the electric field. Without ions, the latter originates mostly from the different work functions of the electrodes (contacts and transport layers), the so-called built-in field. There are different ionic species with (largely) differing mobilities.^{33,339,373–370} Therefore, the time it takes for the ionic species to equilibrate, e.g., in the dark under thermal equilibrium, can range from ms to days.^370,371 Furthermore, if now such a device is illuminated (e.g., under open-circuit conditions), the build-up of the photovoltage changes the electric field in the absorber and so the ionic species again move to reach their new equilibrium positions. In different studies, it has been identified that the recombination occurring at the interfaces between the absorber and ETL or HTL depends sensitively on the distribution of the ionic species. This is because the recombination rate is governed by the electron and hole densities near such an interface and these densities, in turn, depend on the electric field that changes when the ion distribution changes. There are two effects that must be distinguished here: (i) the actual loss of photogenerated charge carriers due to interfacial recombination and (ii) potential voltage losses due to gradients of the quasi-Fermi levels of the majority carriers close to their respective electrode. Whereas the former loss mechanism reduces the carrier densities and hence both voltage and photoluminescence (PL), the latter almost exclusively reduces the voltage but has hardly any impact on the intensity of the PL. Hence, by measuring both V_OC and PL intensity, these two effects can be separated from each other. To illustrate this, Fig. 22 shows an example of both, an experiment and a simulation, where the redistribution of mobile ions in the absorber during illumination leads to a strong increase in V_OC over time, while the PL stays rather constant. The experiment was done with a p-i-n perovskite solar cell with a Cs_0.17FA_0.87Pb(I_0.73Br_0.27)₃ absorber between PTAA and PCBM. The simulations are based on a model outlined in detail in our recent work.³⁷² 

Analogously, a measurement of the current–voltage characteristics of a perovskite solar cell constitutes a dynamic change in the electric field in the absorber and causes ionic species to move. Whether they will react within the time frame of the measurement depends on their mobility. This can clearly be seen by a variation in temperature where lower temperatures lead to slower responses of the device as the ionic mobility depends strongly on temperature.^339,373,374

Further improving our understanding of the interplay between mobile ions and interfacial recombination is crucial to correctly characterize the effects of interface engineering. One approach to eliminate hysteretic effects is to decrease the density of mobile ionic species, potentially also decreasing stability issues due to phase segregation.⁵¹ On the other hand, if the selectivity of the electrodes (including, transport layers) can be increased to a high degree, the actual distribution of ionic species within the perovskite absorber will hardly have any impact and recombination occurring at those interfaces will be minimized under all operational conditions. Different materials are used to improve surface passivation, such as PEAI or polymeric interlayers, such as PMMA.³⁷⁵ The latter has also shown increased resistance to moisture-based degradation.^52,376,377 While the increase in moisture resistance can be attributed to the hydrophobic nature of PMMA, there is strong evidence that it also passivates surface defects that would otherwise lead to increased recombination. However, the interlayer thickness must be carefully calibrated as not to hamper charge transfer. In this way, the long-term stability can be improved by interface engineering, either due to blocking the influx of contaminants or by passivating defects that nucleate the degradation process.

The application of 2D perovskites allows for the incorporation of larger, more hydrophobic, and stable organic molecules between multiple planes of inorganic perovskite octahedrons. As the charge transfer is usually anisotropic and specifically reduced perpendicular to the layers, the resulting lower performance must be balanced against the increased stability and selectivity. Decreased performance can be partially compensated by aligning the 2D layers more perpendicular to the perovskite surface to allow increased charge transfer along the normal direction.^378,379 This is discussed in greater detail in Sec. II F.

For the passivation of interfaces and grain boundaries, additives can be used in the perovskite precursor solution. Recent examples are ionic liquids, such as 1-butyl-3-methylimidazoliumtetrafluorborat, and surface-anchoring alkylamine ligands, phenylethylamine and butylamine.^350,380 Another approach to control surface recombination that has been successfully demonstrated is to dope the perovskite absorber in the interfacial region.^657,658

Interfaces in perovskite solar cells, in particular, those between the absorber and the transport layers, are of utmost importance in order to maximize the fill factor and the open-circuit voltage of these devices. This can also be seen in the research efforts to properly characterize and control interface properties. This is very similar to all other types of heterojunction solar cells. What makes perovskite solar cells special is the fact that the absorber comprises mobile ionic species, which redistribute according to the electric field. As the latter not only depend on the device stack but also on light intensity and applied voltage, these ionic species move under solar cell operation and thereby influence the distribution of electrons and holes, especially near the interfaces to the transport layers. This can have strong effects on two types of losses correlated with interface recombination: the actual reduction of charge carrier densities due to interface recombination as well as voltage losses due to gradients of the quasi-Fermi levels of the majority charge carriers close to their respective electrode. This must be accounted for during characterization in general (see Sec. VI A) as well when engineering interfacial properties.

John Mohanraj, Mukundan Thelakkat

In 2012, the incorporation of solid p- and n-layers in a p-i-n solar cell involving perovskite as the intrinsic semiconductor has revolutionized the field of solution processed solar cells and triggered the evolution of perovskite photovoltaics.^381,382 Charge transport layers (CTLs), in addition to extracting charges from the active layer, are crucial for perovskite solar cells (PSCs) as they define the built-in field in the device along with the contacts and control the charge carrier recombination events at the active layer/CTL interfaces. Thus, a myriad of efforts has been devoted over time to develop and optimize electron transport layers (n-layer; ETL) and hole transport layers (p-type; HTL) for PSCs, which led to define the following criteria for suitable CTLs: (i) matching energy levels with the conduction and valence bands of perovskite; (ii) reasonable charge carrier mobility (>10⁻⁴ cm² V⁻¹ s⁻¹) and conductivity properties;^383,384 (iii) high transmittance in the spectral region in which perovskite absorbs, and (iv) ability to form compact and stable films. By adopting a variety of CTLs satisfying these conditions from established technologies, such as organic heterojunction photovoltaics, dye sensitized solar cells, and organic light emitting devices, perovskite photovoltaics has achieved unparalleled progress in a short time.

Historically, PSCs have derived their device geometry from dye sensitized and organic heterojunction solar cells, in which the photoactive material is deposited either on a compact TiO₂³⁸¹ or ZnO or mesoporous TiO₂ layer to facilitate electron extraction. Later, it was demonstrated that the perovskites themselves possess high electron transport properties; hence, insulating Al₂O₃ was also used as a scaffold.³⁸² A regular PSC contains a thin compact TiO₂ as ETL (<40 nm) with or without a mesoporous TiO₂ layer, and until now, this structure remains as a popular choice for highly efficient devices as there is a favorable conduction band energy level alignment with that of the perovskite, facilitating efficient electron transfer from the latter. Recently, SnO₂ has emerged as a promising ETL with better energy level alignment at the ETL/perovskite interface, a deeper valence band that blocks holes effectively, and offers high transparency and photostability in combination with the low temperature preparation method, which enables the device fabrication even on flexible substrates.³⁸⁵ While the above device structure is considered as normal n-i-p type,²⁵⁷ PSCs are further classified into inverted p-i-n³⁸⁶ type based on the direction at which electrons and holes are extracted. In this device geometry, fullerene derivatives, in particular, phenyl-C₆₁-butyric acid methyl ester (PCBM) known from organic photovoltaics, are employed on top of the perovskite layer as standard ETLs.

As HTL for normal n-i-p devices, many inorganic p-type materials, such as CuI, NiO_x, CuSCN, and CuO_x,³⁸⁷ have shown to be effective due to their high hole mobility values (0.01–0.1 cm² V⁻¹ s⁻¹). However, the conventional triarylamine based organic small molecules and polymers, in particular, air-doped 2,2′,7,7′-tetrakis(N,N-di-p-methoxyphenylamino)-9,9′-spirobifluorene (spiro-OMeTAD), poly triarylamines (PTAA), and poly-3-hexyl-thiophene (P3HT), are still dominating as hole transport materials, resulting in PSCs with >20% efficiency.^320,392 As p-doping of these materials improves their bulk hole mobility as well the built-in potential of the devices and reduce the minority charge carrier concentration at the HTL/perovskite interface as discussed in Secs. IV A and IV B, many exotic p-dopants are also simultaneously developed.^388,389 In the case of p-i-n type devices, PEDOT:PSS was employed as HTL for a long time, and recently, it has been replaced by nano-crystalline NiO_x to obtain stable and efficient PSCs.³⁹⁰ With a same perovskite active layer, mesoporous n-i-p type PSCs excelled with high efficiency and some p-i-n type devices displayed parasitic I–V hysteresis-free performance. Such complementary outcomes combined with the demand for stable and efficient PSCs formed the basis for further research in the field of CTLs.

The CTL/perovskite interfaces play a crucial role in dictating the device performance and stability. Until now, the non-radiative charge recombination events occurring at these interfaces, which are majorly controlled by the intrinsic properties of the CTLs, such as energy level alignment, carrier mobility, and trap density, are attributed to the loss in maximum achievable V_oc in PSCs. Optimizing such interfaces with appropriate transport materials and through passivation techniques remains as a big challenge to achieve highly efficient PSCs.

On the other hand, the bulk properties of CTLs are considered as one of the major reasons for the long standing functional photo and thermal stability issues of devices. For instance, TiO₂ and NiO_x are established photocatalysts, which could impact the stability of perovskites under UV-light irradiation.^390,391 Similar decomposition events are also observed with other metal oxides, in which the superoxide radicals formed at their intrinsic defect sites initiate the redox reactions.³⁹⁰ Chemical reactions of ZnO³⁹³ and CuSCN³⁹⁴ with perovskites are also observed at elevated temperatures, causing thermal stability issues in PSCs.

With fullerene based ETLs, their affinity toward halide ions is claimed to suppress the I–V hysteresis; however, the same facilitate the halide ion migration and, consequently, perovskite degradation over time.³⁹⁵ Similarly, organic HTLs permit ion/atom diffusion from perovskites³⁹⁶ and electrodes³⁹⁷ due to their poor mechanical properties at high temperatures (>80 C), causing stability issues. Additionally, the hygroscopic and corrosive nature of PEDOT:PSS directly influence the stability of the active layer and transparent electrodes,³⁹⁸ whereas spiro-OMeTAD and PTAA play their part indirectly. These HTLs need to be p-doped due to their poor conductivity and is achieved through air-doping by adding cumbersome amounts of hygroscopic Li-TFSI and tert-butyl pyridine additives. The consequence of this process is many folds: (i) exposing uncompleted devices for air-doping facilitates moisture and oxygen ingression into the perovskite, initiating the device degradation process even in the dark; (ii) both additives can chemically interact with the perovskite and permanently damage the perovskite/HTL interface;^399,400 (iii) inhomogeneous films with pin-holes generation over time,⁴⁰¹ and (iv) as the air-doping efficiency varies with ambient conditions, the performance reproducibility of PSCs is jeopardized.⁴⁰² 

Considering the commercialization prospects, the scaled up PSCs (>10 cm²) have reached only 16% efficiency up until now, which is significantly lower than the state-of-the-art small area devices. Such a huge disparity in performance between small and large area devices is majorly attributed to the heterogeneity in perovskite and CTL films deposited on large areas.²⁰⁹ At present, the scalability of PSCs without compromising their device performance remains as a pivotal challenge in the path of commercialization, which is yet to be addressed.

Concerning CTLs, there is a steady progress in materials development and layer engineering with a spotlight on improving their photostability, charge carrier extraction, and transport properties and suppressing intrinsic defects density, which all have direct consequences on the performance and durability of PSCs. In this line, the concept of using ternary metal oxides, such as ZnTiO₃, Zn₂SnO₄, SrTiO₃, and BaSnO₃, and multilayer binary metal oxides, such as ZnO/TiO₂ and WO_x/SnO₂, as ETL is gaining interest as the resulting PSCs display over 20% efficiency and long-term stability.³⁸⁴ Elemental doping of existing binary and ternary metal oxides based CTLs is also investigated to improve the performance of PSCs as this process largely suppresses the intrinsic defect density in transport layers.^404,405 Alternatively, surface passivation has also been practiced on CTLs through depositing a thin layer of organic Lewis acids or bases, which limits the defect density at the CTLs/perovskite interfaces (see Sec. IV C) This is reported to improve FF and V_oc of PSCs due to reduced parasitic charge recombination processes at the interfaces, and interestingly, the same has enhanced the device stability in ambient conditions as these layers form a physical barrier against moisture and oxygen ingression.^389,405 In this context, depositing a thin layer of PCBM or other fullerene derivatives on TiO₂, ZnO, and SnO₂ has become a common practice in device preparation.

In PSCs with organic CTLs, cross-linking the organic semiconducting material in both ETL and HTL has shown to improve their mechanical strength even at higher temperatures.^406,407 On the other hand, with regular organic HTLs, the key issues have been addressed in three ways: (i) small quantities of a variety of direct non-conventional dopants are tested in spiro-OMeTAD and PTAA HTLs to avoid device exposure to moisture and oxygen;^{388,389,408,409} (ii) new, hydrophobic, dopant-free hole transport materials with improved intrinsic electronic properties are being developed;⁴¹⁰ and (iii) hybrid HTLs, such as carbon nanotubes mixed with spiro-OMeTAD or P3HT^411,412 or gradient mixture of spiro-OMeTAD and P3HT,⁴¹³ are introduced to improve stability against moisture and thermal and mechanical stresses.

Toward upscaling of PSCs, a few reported examples demonstrate high performance large area devices with CTLs, such as SnO₂, NiO_x, CuSCN, PTAA, and P3HT,²⁰⁹ and indeed, further developments are foreseen in device and process engineering methods. In this context, ETL- or HTL-free PSCs technology is emerging as a viable option for mass production as they demonstrate the combined benefits of reduced CTL related issues and cost-effective production possibilities.^414,415

Undoubtedly, CTL development has significantly contributed to the growth of perovskite photovoltaics in the last few years. It is also evident that some of them caused performance and stability issues in PSCs. However, recent progress in materials, device engineering, and scalable techniques are encouraging as the durability of PSCs continues to improve over thousands of hours, maintaining reasonably high efficiencies even in large area devices. On the other hand, a fundamental understanding on CTLs/perovskite interfaces to improve the performance of PSCs close to their thermodynamic limit needs further studies. Alongside, a few paradoxical observations, such as high efficiency of PSCs containing TiO₂ and air-doped spiro-OMeTAD CTLs with poorly defined electronic properties and ETL- or HTL-free PSCs superior performance, question the role played by CTLs in these devices. Gaining such basic knowledge on CTLs will certainly promote the large scale production of PSCs with improved durability and efficiency and, consequently, facilitate the commercialization of this technology.

Clément Maheu, Wolfram Jaegermann, Thomas Mayer

Photoelectron Spectroscopy (PES) is a rare characterization technique that can give access to both spectroscopic (i.e., chemical) information and the electronic structure at the same time. PES has been widely used to characterize perovskite solar cells (PSCs). Béchu et al. recently reviewed most of the work that has been conducted in the past few years.⁴¹⁶ PES has been used to confirm the synthesized perovskite phases and to determine their stoichiometry.^416,417 PES was also used to determine their electronic structure and to understand the mechanism of some degradation process.

However, the strength of photoemission, namely, its surface sensitivity, prevents the investigation of buried interfaces in material stacks. The latter plays a crucial role in a device by governing its overall efficiency and its resulting I–V characteristics. The open-circuit voltage (V_OC) and thus the efficiency depend on the built-in potential created at these interfaces in thermodynamic equilibrium. The short-circuit current (J_SC) and thus the efficiency depend on the optimized charged carrier transfer from the perovskite absorber to the charge extraction layer toward the contacts. It means that recombination at the interface states should be minimized.

Therefore, to investigate buried interfaces with PES, appropriate strategies are required. They can be divided into two main families: the bottom-up approaches and the top-down approaches.

The bottom-up approach starts from a layer to progressively grown full device stacks. The so-called step-by-step interface experiments alternate step up contact layer deposition with PES analyses. The shifts of the PES spectra indicate band bending at the interfaces. This was used for several years on PSCs and decades on organic or thin-film solar cells; many of these experiments are described in Ref. 416. Recently, Hellmann et al. investigated the electronic structure of classical and inverted PSCs. They evidenced that the photovoltage mainly occurs at the typically n-type perovskite|p-type charge extraction layer interface. To reach such a conclusion, they compared MAPbI₃|spiro-OMeTAD and NiO_x|MAPbI₃ interface experiments in the dark and under illumination.⁴¹⁸ 

The top-down approach starts directly with a full device stack. X-ray photoelectron spectroscopy (XPS) depth profiling alternates ion gun etching cycles with XPS measurements.⁴¹⁹ For instance, Cacovich et al. addressed the effect of solar illumination on the optoelectronic properties of a state-of-the-art PSC. Hard X-ray PES (HAXPES) can also probe different depths (from 10 to 20 nm) of a given layer via photon energy tuning.^418,420 Finally, high local resolution methods probe the cross section of a full device. This is the case for Kelvin Probe Force Microscopy (KPFM) as well as Electron-Beam Induced Current (EBIC). They can inform on the electronic structure of a full device, but these methods lack the chemical information.^338,421

Top-down and bottom-up approaches suffer from the challenges that are intrinsic to PES. The long time exposure of the samples to vacuum, light, and x rays may induce significant chemical modifications of the hybrid organic inorganic metal halide perovskites.⁴²² It can be avoided by minimizing the time of analysis and by performing careful PES measurements. For instance, Cacovich et al. recorded two survey spectra, one before and one after the full PES acquisitions.⁴¹⁹ It would reveal potential x-ray induced degradations. Another issue is the possible, unintentional, photovoltage. It can be induced by residual light in the ultrahigh vacuum (UHV) system and/or the PES source itself, leading to misinterpretation of the electronic structure. Hellmann et al. reported several precautions for limiting the amount of residual light and performed real “dark” measurements.⁴¹⁸ In addition, each of the two approaches also has limitations. The step-by-step interface experiments are limited to the study of layers that can be prepared by vacuum deposition. The PSCs with a perovskite layer prepared by vacuum deposition are often lagging behind their solution-processed counterparts. Getting information on the latter is also crucial to investigate the state-of-the-art devices. This is also the case for the extraction layers. Interface experiments with spiro-OMeTAD are numerous,^416,418 but promising alternative charge extraction materials can be found in the literature (see Sec. IV D). Some of these candidates cannot be prepared inside the UHV setup. Therefore, the bottom-up approach cannot be applied to all the buried interfaces composing PSCs.

In the top-down approach, the sample needs to be modified for revealing the buried interfaces. For KPFM or EBIC analyses, a normal cross section needs to be prepared by focused ion beam (FIB). To address the potential modification that FIB preparation may induce, Weber et al. evidenced that, for their system, the efficiency and the hysteretic behavior are similar before and after FIB polishing.³³⁸ While KPFM and EBIC inform on the built-in potential along the device, PES can provide electronic information as well as chemical composition. It, however, lacks in local resolution, and it is in the range of 10 μm, while the thicknesses of the perovskite and space charge layers are in the nm range. XPS depth-profile is one alternative, but conventional argon sputtering may induce chemical and concomitant electronic variations. Damage can be minimized by sputtering with large organic molecules, such as C₆₀ or coronene.

The Tapered Cross Section PES (TCS-PES) methodology is one answer for analyzing a cross section by PES without using ion etching in vacuum. This top-down approach was recently proposed to study PSCs.^423,424 Polishing a sample with a shallow angle extends the normal cross section (nm scale) to a tilted cross section (mm scale). Then, PES line scans can be taken on the resulting device. In Fig. 23, each line scan corresponds to a position of this TCS of a mixed ion lea halide PSC. Several core levels are analyzed; the detection of the elements gives an idea of the solar cell architecture (see the scheme on the left). Au mainly contributes between positions 0 and 6, the clear onset of strong Pb and I emissions at position 10 indicates the spiro-OMeTAD | perovskite interface, the Ti2p_3/2 line intensity starting at position 36 indicates the mesoporous TiO₂ and then the block the TiO₂ layer, and finally, the Sn3d_5/2 emission at position 49 corresponds to the FTO front contact.

The XPS line scans also inform on the variation of the composition and of the stoichiometry among the device. For instance, a bromine enrichment was evidenced toward the perovskite|m-TiO₂ interface.⁴²³ 

In addition to the chemical information, the steady variation of the binding energy of a certain core orbital indicates band bending and/or possible chemical variations. Under bias light, chemical and compositional modification will persist while band bending is flattened due to induced photovoltage.

For instance, the binding energy shift of the C1s orbital in spiro-OMeTAD (Fig. 23, positions 6–10) is due to band bending. On the contrary, the Pb4f_7/2 shift at the perovskite|m-TiO₂ interface (Fig. 23, positions 36–46) corresponds to variations, such as the Br/I ratio in the perovskite and to the TiO₂/perovskite ratio within the m-TiO₂. Thus, the band diagram across a wet chemically prepared perovskite PSC has been deduced.

In contrast to the mechanical polishing, Yang et al. used a chemical strategy for exposing the buried interfaces.⁴²⁵ They reported a lift-off process to bring the perovskite part of the buried PTAA|perovskite interface to the top of the sample. They performed numerous characterizations, such as XPS, KPFM, or advanced microscopy, on the fresh layer and on the lifted-layer to confirm that this process has a small influence on the morphology of the layer and its chemical composition. Such work helps to conclude on the morphology of the perovskite layer at the interface. It is inhomogeneous and has sub-microscale imperfections that result in more rapid recombination.

Unfortunately, similar questions than as for the XPS depth profile or the investigation of the normal and the tapered cross section with KPFM, EBIC, or PES must be addressed. Lifting-off the perovskite layer, FIB, or the preparation of a TCS by polishing may modify the chemical composition and the electronic structure of the buried interfaces.

Buried interfaces played a crucial role in PSCs and in solar cells, in general. Techniques such as KPFM, EBIC, or PES can investigate such interfaces. The latter has the asset to provide both chemical and electronic information.

To access the buried interfaces of a full-stack, few methods were listed: chemical lift-off, FIB, ion gun, or mechanical polishing. The main challenge consists in accessing these buried interfaces of a device with minimized modification of the chemical composition and electronic structure.

A combination of KPFM or TCS-PES with the HAXPES for which the interfaces can remain buried would be one solution.

Janek Rieger, Thomas Fauster, Daniel Niesner

Angle-resolved photoelectron spectroscopy (ARPES) gives direct access to the surface electronic band structure of a semiconductor. Hence, it is an important tool to determine essential electronic parameters of metal halide perovskites. So far, studies focused, in particular, on the occupied part of the band structure. MAPbBr₃, MAPbI₃, MAPbCl₃, and CsPbBr₃ were investigated, and they all exhibit a characteristic spectral shape of the valence bands.^433–429 The width of the dispersing valence bands, surface doping, and hole effective masses around 0.25 m_e were determined. Results are compared in a recently published review.⁴³⁰ In most cases, the periodicity of the surface electronic structure reflects the pseudocubic building block of the metal halide perovskite, even if the bulk-terminated unit cell is larger due to superstructures related to the tetragonal and orthorhombic distortions. There are also some discrepancies: The bandwidths of the valence bands and the surface doping vary significantly.⁴³⁰ Two studies found a Rashba effect on MAPbBr₃, but others did not.^{427,437–432} The surface periodicity observed with low-energy electron diffraction (LEED) shows the large unit cells expected for the bulk-terminated tetragonal and orthorhombic crystal structures in contrast to the surface Brillouin zones from ARPES.

As a complementary technique, time-resolved two-photon photoemission spectroscopy (2PPE) can be used to study the unoccupied part of the band structure and relaxation dynamics. Time-resolved studies were carried out to track the charge carrier dynamics at the conduction band minima of MAPbBr₃, MAPbI₃, CsPbBr₃, and a quadrupole cation mixed metal halide perovskite.^{49,426,433,434} Very slow cooling of hot electrons was observed and explained by the formation of large polarons.^426,434

ARPES and momentum-resolved 2PPE require atomically well-ordered periodic surfaces with a defined surface orientation. Experiments must be conducted in ultra-high vacuum (UHV) because of their high surface sensitivity. Preparation of suitable surfaces under UHV conditions is the main challenge of ARPES and 2PPE. Most of the ARPES studies were done on cleaved single crystals.⁴³⁰ Cleaved single crystals can deliver well-ordered surfaces but exhibit several experimental issues: Cleaving can generate varying surface terminations, which changes the surface properties.^431,435 The solution processing and subsequent cleaving of single crystals induce defects at the surface.⁴³⁶ Strong doping is an indicator of high defect concentrations. All investigated MAPbI₃ samples were heavily n-doped, while the doping level of the measured MAPbBr₃ surfaces varied. Initial defect concentrations cause further degradation and change surface band bending during measurement.⁴³⁷ Single crystals also suffer from charging, especially at low temperatures. This can pose limitations for temperature-dependent studies and methods using higher light intensities, such as synchrotron- and laser-based ARPES measurements as well as 2PPE.

An alternative surface preparation technique is needed, which reliably produces defect-free surfaces of a defined and ideally controllable surface orientation and termination. This would permit systematic investigations with multiple techniques, such as ARPES, 2PPE, scanning tunneling microscopy (STM), and LEED, at different measurement temperatures. Reproducibly prepared surfaces could also serve as a platform for investigation of adsorbates, e.g., their adsorption geometries and interface level alignment, or segregates, such as dopants.

A common preparation approach in surface science is the deposition of layers onto well-defined single-crystalline and atomically clean substrate surfaces under UHV conditions. Layers may be deposited using thermal sublimation (see also Sec. III C) or chemical vapor deposition (see also Sec. III D). The approach has several advantages: UHV conditions avoid external contamination. Accumulation of charge during experiments can be circumvented by a conductive substrate and a finite controllable film thickness. The stoichiometry of the surface can be controlled by the calibration of evaporation rates, allowing, for example, for bandgap engineering (see also Sec. II B). Ordering of the surface can be optimized by the substrate temperature during deposition. This approach was already applied to prepare locally ordered lead-halide perovskite surfaces for STM studies.^15,438,439 Perovskite films with a single (surface) orientation can be obtained by epitaxially matching lattice parameters of lead-halide perovskite and substrate.

CsPbBr₃(001) and Au(100) have surface lattice parameters (a_S) of 5.75 Å and 2.87 Å, respectively. One unit cell of CsPbBr₃ fits on four unit cells of the substrate with a lattice mismatch of <1% [see Fig. 24(a)]. This epitaxial relationship supports reproducible growth of CsPbBr₃ with long-range order in one (surface) orientation.

Figure 24(b) shows the LEED pattern of 9 ML (50 Å) CsPbBr₃ on Au(100) at 90 K. The spots corresponding to the square bulk-terminated unit cell of the Au(100) substrate are highlighted by golden circles. Additional spots can be assigned to CsPbBr₃. Their sharpness and the relatively low background reflect the high surface quality. The most intense diffraction spots (black circles) form a 2 × 2 superstructure with respect to the substrate. These spots reflect the basic pseudocubic building block of the perovskite, containing one CsPbBr₃ unit.

The surface shows an additional superstructure, which cannot be explained exclusively by superstructures of a bulk-terminated orthorhombic crystal but stems presumably from additional surface reconstructions.

Figure 24(c) shows an angle-integrated photoemission spectrum of 9 ML CsPbBr₃ on Au(100) at 90 K. The valence bands show a spectral shape similar to the one of other lead-halide perovskites.⁴³⁰ Three intense peaks at binding energies between 6 and 3 eV are followed by a less intense feature close to the valence band maximum (VBM). Their sharpness indicates a high crystalline quality of our epitaxial films. The binding energy of the VBM (E_B = 1.1 ± 0.1 eV) was determined with a parabolic fit to the angle-integrated spectrum spectrum.⁴²⁷ No energetic shift or change of the spectral shape could be observed after several hours under illumination. This indicates a constant Fermi level close to the middle of the bandgap and no significant doping. Surface-sensitive x-ray photoelectron spectroscopy (XPS) does not show any detectable defect density in the epitaxially grown samples. One representative spectrum of the Pb 4f levels is shown in the inset of Fig. 24(c). The peaks consist of only one component. No metallic lead can be detected.

ARPES was carried out for a further comparison of the surface electronic structure to the one of single crystals. We applied a 3D curvature method to our three-dimensional I(E_kin, k_x, k_y) dataset to visualize the dispersion of the topmost valence band. The resulting dispersion along a high-symmetry cut in k-space [indicated in Fig. 24(b) as a guide to the eye] is shown in Fig. 24(d). The periodicity of the topmost valence band reflects a pseudocubic structure. A hole effective mass of m* = 0.25 ± 0.1 m_e can be estimated at the VBM at $M̄$⁠, in line with m* = 0.26 ± 0.2 m_e determined for the single crystal.⁴²⁸ 

The preparation of ultrathin epitaxial films of metal halide perovskites provides a platform to apply momentum-resolved techniques, such as LEED and ARPES. Initial experiments imply that it has the potential to be extended to a variety of metal halide perovskites, in particular, lead-free modifications. Detailed momentum-resolved measurements of the valence-band structure and conduction-band dynamics will aid in the understanding of the exceptional properties of metal halide perovskites. Using low and high photon energies, the properties of the films at their surface, bulk, and substrate interface may be investigated. Interface level alignment and dynamics could be explored by controlled exposure to adsorbates or by varying substrates.

Fengjiu Yang, Steve Albrecht, Thomas Riedl, Azhar Fakharuddin, Maria Vasilopoulou

Fengjiu Yang, Steve Albrecht

Although metal halide perovskite single-junctions have achieved enormous efficiency improvements within the last few years, further improvement will become more and more challenging when approaching fundamental limits. The main losses in any single junction solar cell at given bandgap (E_g) are non-absorbed photons with energy below the absorber E_g and thermalization of high energy photons toward the band edges (see Fig. 25). A tandem architecture with complementary absorber materials is a promising solar cell concept to reduce thermalization losses while covering a wide range of sun-light absorption (see Fig. 25). In a tandem cell, perovskite based absorbers can used as the top cell, facing the sunlight. The perovskite composition can be tuned to vary the E_g around 1.6–1.8 eV (see Fig. 4). The low E_g tandem partner, i.e., the bottom cell, can be composed of crystalline silicon (1.12 eV), copper-indium-gallium-selenide (CIGS) (∼1.10 eV), organic bulk heterojunctions (∼1.41 eV), and tin-based perovskite itself (∼1.25 eV).^{321,447–442} This tandem architectures can overcome the typical Shockley–Queisser limit of single-junction solar cells being around 33% and have radiative limits of 45% for perfect E_g combination (see Fig. 25). In the case of multi-junctions with 3, 4, or an infinite number of absorber materials with perfect E_g combination, even higher radiative limits of 50.1%, 54.0%, and 65.4% efficiency could be reached at normal sun-light (1-sun) illumination.⁴⁴³ Combining the high ease of production and low fabrication cost together with the high efficiency and E_g tunability, metal halide perovskites are the best partner for tandem and multi-junction applications. This fascinating material class might achieve high efficiencies at reasonable costs in future solar cell module applications.

The different tandem architectures can be divided into the simplest, mechanical stacked, individually operating four-terminal (4T) structure that needs four contacts.⁴⁴⁴ In addition, the series-interconnected subcells in the three-terminal⁴⁴⁵ or two-terminal (2T) and also called monolithic architecture can be utilized.⁴⁴⁶ For common 2T structures, both subcells are connected via a recombination (tunnel) contact and connected in series. This recombination contact is needed, as electron–hole pairs from different absorption events are generated and half of these electrons and holes recombine at the recombination junction, while the other half is extracted at the outer contacts (see Fig. 25). When this selectivity in the recombination contact due to appropriate choice of energy levels is ensured, the voltage of both subcells is summed so that the tandem V_OC is that of the sum of the subcells (V_top + V_bottom). In addition, the overall photocurrent is limited by that of the limiting subcell, and thus, the so-called current matching between the subcells is needed for the highest J_SC. The tandem FF is determined by the photocurrent limiting subcell and by the amount of current mismatch.⁴⁴⁷