Doping strategies for small molecule organic hole-transport materials: impacts on perovskite solar cell performance and stability

Dopants for small molecule-based organic hole-transport layers impact both perovskite solar cells initial performance and long-term stability.


Perovskite solar cells
With global energy demand projected to grow nearly 50% by 2040, 1 it is imperative to divest from traditional greenhouse gasbased power production toward renewable energy sources such as solar. Perovskite solar cells (PSCs) represent the type of breakthrough solar technology to make solar cells and clean energy more ubiquitous. Hybrid organic/inorganic PSCs utilize an ABX 3 perovskite crystal structure active layer, typically prepared from methylammonium (MA), formamidinium (FA), and/or cesium cations, lead and/or tin, and halide(s). The result is a highly absorbing semiconductor material with excellent semiconducting properties, such as high absorption coefficient, high carrier mobility and lifetime, and long carrier diffusion length. [2][3][4][5] Common device architectures and device operation are summarized in Fig. 1. Briey, since halide perovskites are relatively high dielectric materials, free electrons and holes, rather than bound excitons, are generated in the active layer as light is irradiated on the solar cell. 6 The electrons ow through the electron transport layer (ETL) towards the anode, while holes migrate towards the cathode through the hole transport layer (HTL). The energetics and mobilities of each layer must be highly aligned and balanced to minimize charge accumulation and prevent recombination of positive charge carriers, or holes, and electrons. For instance, the HTL provides a pathway for both: hole conduction and electron blocking, which results in reduced recombination and increased ll factor (FF) and power conversion efficiency (PCE). 7,8 Through extensive work in the latter half of the past decade, [9][10][11][12][13] PSC efficiencies now rival commercial photovoltaic materials, [14][15][16] despite the fact that they are processed from solution.
Nevertheless, to realize commercialization, a number of barriers must be addressed that address the full lifecycle of this technology (e.g., improving module efficiency, stability and lifetime, and disposal of lead and other heavy metals, etc.). 17 For the purposes of this text, stability will be dened as a change in PCE over time with respect to specic set of conditions, while lifetime considers not only stability but also the total length of power generation. 17,18 While tremendous gains have been made in a number of areas, [19][20][21][22][23][24] since the HTL is generally on top of the perovskite layer it acts as a rst line of defense to protect perovskite structural integrity and therefore HTL engineering has a tremendous impact on PSC stability and lifetime.

Hole-transport materials
The rst breakthrough in PSC technology was the application of an amorphous solid state hole transport layer 25,26 to replace iodide liquid electrolyte used in the earliest reports. 27 With the application of the spiro-OMeTAD (2,2 0 ,7,7 0 -tetrakis[N,N-di(4methoxyphenyl)amino]-9,9 0 -spirobiuorene, Fig. 1) layer, the performance skyrocketed beyond 10% to the present day record of 23.3%. 14 The doped spiro-OMeTAD layer is not given enough credit in the development of this technology: without this easyto-process layer, perovskite solar cells may have never attracted so much attention.
A unique challenge in hybrid organic/inorganic PSCs is that the perovskite cannot withstand harsh processing upon deposition of a subsequent layer. For example, a number of inorganic materials 28 (e.g., copper iodide, nickel oxide, copper oxide) have been explored as replacement HTLs for PSC, and such materials could offer improved stability, but oen require incompatible or high processing temperature, making oxidebased HTLs desirable only for devices with p-i-n architecture (which implies substrate/HTL/perovskite/ETL) and thus they are typically not processed on top of the fragile perovskite lm. [29][30][31][32] Inorganic copper thiocyanate, CuSCN, 33 is one such HTL that can be used in the n-i-p (substrate/ETL/perovskite/HTL) architecture, but there are still fabrication challenges: performance is dependent upon a multi-step dynamic deposition on the perovskite lm for good surface coverage and crystallization with minimal perovskite dissolution. Aside from specic cases, the greater majority of perovskite solar cell research, and especially record efficiency, 14,34 has relied on an organic-based HTL PSC designs.
As mentioned above, spiro-OMeTAD ( Fig. 1) is the most commonly used OSM HTM for PSCs, and was originally developed for solid-state dye-sensitized solar cells. 56 While useful for fundamental research, 57 the synthetic costs, poor thermal stability, and low intrinsic conductivity and mobility make it unsuitable for commercialization. Researchers world-wide have been exploring lower cost and more efficient alternatives, although improving beyond the performance of spiro-OMeTAD has proven challenging. 7,8,44,47,48,50,52,53,[57][58][59][60][61][62][63][64][65][66][67][68] OSM HTMs (Fig. 2) can be synthesized inexpensively and designed to possess most of the desirable properties, but unlike polymeric HTMs, the majority of new small molecule HTMs still require dopants to increase the low intrinsic conductivity and mobility. These dopants oen contain hygroscopic and/or mobile ions, like Li + , and signicantly decreases PSC stability with and without encapsulation (vide infra). Therefore, there remains an excellent opportunity to develop dopants which accomplish the goal of improving the HTL conductivity for efficient hole extraction, but offer improved hydrophilic properties, thermal stability, synthetic ease, etc.
Among the groups utilizing OSM HTLs, there have been two clear research avenues: dopant-free (i.e., design a small molecule with sufficient conductivity/hole mobility without dopant 49, ) and specialized dopant design research for improving HTLs with intrinsically poor conductivity/hole mobility. Because device efficiencies are generally higher with a doped HTL, as record device efficiencies have been achieved with a doped HTL, 14,34 this review focuses upon the latter approachalternative dopants for small molecule organic HTLs used in standard (n-i-p) device architecture PSCs. Aer discussing general dopant principles, alternative chemical dopants (ionic liquids, metal-based salts, oxidized radical cation salts, TCNQ derivatives) and relationship to PSC performance/ stability will be discussed. A summary of reported PSC performance metrics are compiled chronologically in Table 1. We conclude with imminent research needs for HTL dopants for highly efficient and stable PSCs.
2. Organic small-molecule doping principles and common practice

Doping in small-molecules
A dopant is an impurity added to a bulk matrix (in this case a lm of "pristine" organic small molecules) to alter its semiconductor properties. 92 However, unlike in inorganic systems, the doping process in organic systems typically corresponds to chemical oxidation (specically in p-type doping), and charge transport involves redox reactions. 93 Fig. 3 highlights a proposed mechanism for charge hopping in p-doped organic thin lms. 94,95 The singly occupied molecular orbital (SOMO) of the dopant (oxidized spiro-OMeTAD) is slightly deeper (lower in energy) than that of an electron in the HOMO of the pristine material; thus, there is a thermodynamic driver for integer charge hopping from pristine to dopant. In essence, when the dopant is reduced, it becomes chemically identical to the bulk matrix, and the pristine molecule becomes chemically identical to the dopant. This process repeats across the thin lm as holes are accepted from the adjacent perovskite layer. For more detail on charge-transport mechanisms in organic systems, please refer to the discussions in Walzer, et al. 95 and Lüssem,et al. 92 Conductivity typically increases by multiple orders of magnitude as a function of dopant concentration as trap states are lled by the dopant, and then plateaus or decreases at higher dopant concentrations (Fig. 3). Dopant addition shis the HTL Fermi level (E F ) closer to the HOMO level of the pristine matrix, 97 and the observed HOMO level of the HTL lm to become slightly deeper (depending on doping concentration, $100-200 meV). 98 Note that observed conductivity, mobility, and energy level shis are dependent upon processing conditions (solution, vacuum), as morphology and impurity concentrations vary with each technique. 99 Overall, the combination of increased conductivity, mobility, and energetic alignment are key for current matching among the HTL and perovskite during device operation. In turn, this allows for improved hole injection into the perovskite and reduced recombination at the interface, and are key screening tools for dopant assessment. Moreover, HTL series resistance decreases concomitant with an increase in conductivity due to doping 100 which, in solar cells, leads to improved charge extraction and higher ll factor (FF).
For simplicity, the reactions involved in the doping process will be discussed with respect to spiro-OMeTAD as most studies investigate this system; however, it should be noted that many other HTL systems not based on spiro-OMeTAD utilize some or all of these standard dopants/additives. According to Abate et al. 96 the pristine spiro-OMeTAD reacts with O 2 to generate a weakly bound complex 99 (Scheme 1) in the presence of light or heat aer the thin lm is exposed to air. However, the TFSI À anion traps/stabilizes the radical cation, and the remaining metal cation (e.g., Li + , Co 3+ ) forms an oxide complex. Wang et al. 103 further suggest a spectral-dependence of radical cation generation; specically, in the presence of >450 nm light, radical cation generation can be initiated by the perovskite (Scheme 1). The lm is considered to be p-doped once a critical concentration of radical cations is trapped. A non-trivial consequence of this in situ protocol is that it is challenging, if not impossible, to control how much oxidized HTM is trapped. With regard to nomenclature, in the broader literature, LiTFSI and FK209 are regarded as dopants. We will continue to refer to the radical cation-generating additives as dopants throughout this text. However, it is the radical cation species, and not the "dopant" (i.e., LiTFSI, FK209), that is responsible for the improved HTL properties and ultimately dopes the HTL.
While doping of spiro-OMeTAD is largely dependent on ambient conditions, this conventional doping scheme is effective for decreasing HTL resistance. Compared to spiro-OMeTAD lms which are undoped, the mobility 104 and conductivity 102 of doped spiro-OMeTAD lms can be typically increased by at least an order of magnitude. Even though the charge-transport of spiro-OMeTAD is poor when compared to inorganic semiconductor materials, the thin nature of the HTL lms in general still allows for the attainment of PV devices with FF > 80%especially when doping with both LiTFSI and FK209. 102 Nevertheless, the low hole mobility of spiro-OMeTAD does appear to limit device FF to some degree. 29,105,106 This offers an opening for alternative HTL materials to prove advantageous as there is still signicant room for PSCs in FF improvement as the maximum FF under the Shockley-Queisser limit framework is $90% for band gaps between 1.5-1.6 eV. 106,107 In addition to LiTFSI and FK209, 4-tert-butylpyridine (tBP) is a near-ubiquitous additive with chemical oxidants, yet it is unclear as to whether tBP imparts purely electronic or morphological benets to the HTL. The main role of tBP is to prevent phase segregation of LiTFSI and spiro-OMeTAD, resulting in a homogeneous HTL. 108 Recently, cation-pi interactions between tBP and Li + ions were uncovered, 109 which corroborates what is observed in practice: LiTFSI solubility can be modulated in the presence of tBP. Pinholes in HTL lms have been shown to be related to the presence of small amounts of secondary solvents (e.g., water, 2-methyl-2-butene, or amylene). 110 Generally speaking, with tBP, HTL lms show reduced pinhole formation (Fig. 5). 111 With regard to electronic benets to the HTL, tBP has been shown to suppress charge-recombination, 112 and there is evidence that its addition makes the perovskite/HTL interface more hole-selective. 113 To understand this effect, Habisreutinger et al. propose tBP is protonated by methylammonium (from the active layer), resulting in a slight negative charge at the perovskite interface and increased hole attraction from resulting energetic band bending (Fig. 5). 113 In reality, tBP is likely involved in multiple processes.
Generally, dopants like LiTFSI and FK209 negatively impact PSC properties over time. While oxygen allows for radical cation generation upon LiTFSI addition to HTM, it is well known oxygen exposure is be detrimental to perovskite active layer (PAL) stability. 114,115 With LiTFSI, there are a number of reported phasesegregation challenges: 116,117 for instance, LiTFSI can migrate to the Au-HTL interface via lm pinholes (Fig. 6). Li + ions can    layer and device properties. [119][120][121][122] LiTFSI also contributes to HTL delamination from the PAL. 123 Little is known about direct effects of cobalt-based dopants on PSC stability, 124 as devices typically do not achieve as high of FF and PCE without LiTFSI co-doping and are thus less well-studied. 102 As researchers transition from cell-to module-level studies, scribe lines/module interconnects are susceptible to dopantrelated corrosion. 125 With respect to tBP, it has been shown to be volatile (Fig. 5) 111 and dissolve perovskite in high concentrations via formation of a complex with PbI 2 (Fig. 6). 126 However, it is unclear how much tBP in the HTL impacts device stability, 127 and changing pyridine functionalization (Fig. 6) or replacing with other salts 128 can reduce corrosion. In the context of this review, tBP will not be regarded as a "dopant"; however, we certainly believe that further investigation into tBP with respect to mechanistics and stability is warranted. To summarize, commonly-employed dopants allow for record-efficiency devices but they also directly preclude device stability.

Alternative chemical dopant schemes
While comprehensive design criteria for HTL dopants in PSCs remain elusive, an ideal chemical dopant should (1) Quantitatively and reproducibly generate highly stable radical cations at a reasonable rate for improved chargetransport (i.e., dopant oxidation potential that readily leads to radical triarylamine generation for high HTL conductivity and mobility), (2) generate inert byproducts or, if byproducts are inevitable, impart desirable properties (i.e., increase hydrophobicity or thermal stability, reduce phase segregation), (3) lead to highly efficient and stable PSC (i.e., adequate HTL energetic alignment with PAL for hole extraction) while (4) maintaining low cost (i.e., not limited to inert conditions, few synthetic steps, tunable redox properties, limited need for polar co-solvent upon HTL deposition) to promote widespread accessability to researchers.
The ubiquity of LiTFSI-doped HTLs in PSC research can likely be attributed to some combination of historical precedent, broad commercial availability, simplicity of methods, and demonstrated high PSC performance. Nevertheless, LiTFSI fails multiple metrics when considered against the framework enumerated above. While LiTFSI reproducibly generates spiro-OMeTAD radical cations and does yield high performance PCSs, the reaction yield is not consistent due to the dependence on ambient conditionsone explanation for the need to optimize dopant concentration from lab to lab, assuming other factors are held relatively constant. More problematic, dopant byproducts, lithium oxides, increase HTL hydrophilicity, and lithium ions can migrate through the device stack.
Therefore, in this section, we will expand upon efforts which aim to fulll each of these criterion towards universal dopant design criteria for HTMs in PSCs. Generally, they fall into the following categories: ionic liquids, Brønsted/Lewis acids, metalbased salts, oxidized radical cation salts, tetracyanoquinodimethane (TCNQ) derivatives, and other miscellaneous schemes.

Ionic liquids
Ionic liquids were one of the rst non-Li/Co-based dopants assessed for HTLs in PSCs to eliminate byproducts in the HTL. In 2013, Abate et al. doped spiro-OMeTAD with protic ionic liquids (PIL) of varying pH (all with TFSI À anions) (Fig. 7) leading to HTL conductivity increases by three orders of magnitude. 129 They proposed a slightly altered doping mechanism with PIL (as compared to Scheme 1): rst, the triarylamine nitrogen within spiro-OMeTAD is protonated by the PIL, and aer electron injection from an adjacent pristine/neutral spiro-OMeTAD molecule, hydrogen gas is released, which allows for the radical cation to be trapped by a TFSI À anion (Fig. 7). In contrast to LiTFSI/ FK209-based doping, the redox chemistry with PIL is thermally activated and does not require oxygen exposure. Of the three ionic liquids, bis(triuoromethanesulfonyl)imide (HTFSI)-doped spiro-OMeTAD HTLs in PSCs led to highest ll factor and PCE as compared to LiTFSI, 1-alkyl-3-methylimidazolium bis(triuoromethane)sulfonimide (Himi-TFSI), or tetraethylammonium bis(triuoromethane)sulfonimide (Et 4 N-TFSI)-based doping ( Table 1, entry 1). They attributed these performance differences to reduced HTL charge-transport resistance. Notably, tBP reduced PSC performance in the presence of ionic liquiddoped HTL, which may be due to side reactions that prevent spiro-OMeTAD oxidation and/or reduce radical cations. With any in situ doping method, benign products/ions that minimally impact bulk properties are desired. In this respect, hydrogen gas production from HTFSI is an ideal scenario because the reaction byproducts do not remain in the HTL. However, a drawback of many of these protic ionic liquids is inherent ionic liquid hydrophilicity, 130 and while a few follow-up studies have been reported, 131,132 the potential impact(s) of HTFSI on PSC stability is(are) unclear because co-dopants were used alongside HTFSI in these studies (i.e., Et 4 NTFSI, FK209 respectively). Nonetheless, Abate et al. 132 assert the importance of HTM selection with respect to HTL dopant and PSC stability ( Table 1, entry 8).
Zhang et al. utilized N-butyl-N 0 -(4-pyridylheptyl)imidazolium bis(triuoromethane)sulfonimide (BuPylm-TFSI) (Fig. 7) 133 as a lithium-free dopant, similar to a bromide analogue used for dye-sensitized solar cells (DSSC). 134 The molecular design was threefold to leverage radical generation properties and overall HTL stability: spiro-OMeTAD doping via PIL, increased anion thermal stability with increased anion molecular weight, and potential interface passivation via the pyridyl group affixed to the alkyl chain (akin to tBP, and later seen in HTM design 63 ). PSC with BuPylm-TFSI or LiTFSI-doped HTL generated comparable PCE (Table 1, entry 2). While the potential for increased thermal stability was purported, PSC performance at increased temperature was not reported. Additionally, BuPylm-TFSI synthesis requires pyrophoric diisopropylamide (LDA), so accessability may be limited to those with expertise. 134 To the best of the authors's knowledge, this is a stand-alone report.
Recently, a commercially available aprotic ionic liquid, 1butyl-3-methylpyridinium bis(triuoromethane)sulfonimide (BMPyTFSI, Fig. 7), was incorporated into a spiro-OMeTADbased HTL. 135 This ionic liquid is similar to quaternary ammonium cation-based Et 4 N-TFSI utilized by Abate et al. 129 except for the incorporation of the aromatic group, which increases the material's hydrophobicity. In contrast to PIL discussed previously, the HTL solutions with aprotic ionic liquids required oxygen exposure to generate spiro-OMeTAD radical cations before device fabrication and are not directly redox active. Interestingly, HTL conductivity increased from 4.9 Â 10 À8 S cm À1 (pristine spiro-OMeTAD) to 4.4 Â 10 À6 S cm À1 (7.8 mol% BMPyTFSI). This is in contrast to the reduced conductivity observed by Abate et al. 129 with Et 4 N-TFSI dopant at similar concentrations ($10 mol%). Nevertheless, devices with BMPyTFSI-doped spiro-OMeTAD performed similarly to devices with LiTFSI/FK209/tBP-doped HTL. FF and V oc were slightly lower than controls due to higher HTL series resistance (Table 1, entry 27). Device stability was assessed over 200 days. Devices were unencapsulated and stored in a humidity chamber ($50% relative humidity) in the dark and photovoltaic properties were periodically assessed. Devices with BMPyTFSIdoped HTL maintained up to 80% of initial PCE under these conditions, attributed to a J sc drop, while control devices lost over 50% initial PCE (Fig. 7). HTL hydrophobicity was monitored over time, and water contact angle measurements remained constant (>90 vs. >78 control). While this aprotic ionic liquid clearly has superior hydrophobicity for improved PSC stability, the degradation mechanism(s) for performance loss is unclear at this time, particularly for devices under operational conditions.

Brønsted/Lewis-acids
Acid doping is not limited to ionic liquids. In 2016, Li et al. show acid additives with a wide pK a range, such as phosphoric acid, sulfuric acid, acetic acid, or (4-(triuoromethyl)styryl) phosphonic acid, can improve the conductivity of spiro-OMeTAD lms (Fig. 8). In conjunction with LiTFSI/FK209/tBP, the acid-doped systems result in improved V oc , ll factor, and reduced J-V hysteresis (Table 1, entry 14). 136 Fig. 8 exemplies superior HTL properties with 10 mol% phosphoric acid: not only does HTL conductivity increase at a greater rate than with LiTFSI alone, but conductivity improvements are observed over many hours. Therefore, it can be inferred that radical spiro-OMeTAD cations are not appreciably reduced (quenched) in ambient conditions. Upon PSC integration, PCE increased from 15.2% to 17.6% when the HTL contained 10% phosphoric acid in addition to LiTFSI/FK209. In short, acid-doping eliminated Fig. 7 (a) Chemical structure of ionic liquids used in this study: H-TFSI and Himi-TFSI (n ¼ 5 for the data reported in this work) are PILs, with H-TFSI more acidic (free proton in red) than Himi-TFSI (the most acidic proton in orange), and Et 4 N-TFSI as an aprotic ionic liquid. The pH scale may be considered as an indication of how strongly a proton will be transferred from the PIL to a base, though it must be noted that the pH is usually considered for aqueous solutions and may not be appropriate for the nonaqueous PILs.  the need for traditional device aging with ambient oxygen. Intriguingly, varying the acid strength/proton lability facilitated similar performance improvements. It was proposed that weak hydrogen bonding between the acid and spiro-OMeTAD may increase positive charge carriers, or acid additives may improve interfacial contact with the PAL to reduce recombination losses. Shelf-life stability of PSCs with and without phosphoric acid doping were not signicantly different (Table 1, entry 14).
A Lewis acid, tris(pentauorophenyl)borane (BCF), in conjunction with LiTFSI also improved conductivity and device efficiencies, as reported by Ye et al. in 2017 (Fig. 8). 137 UV-Vis absorption spectroscopy and electron-pair spin resonance (EPR) spectroscopy affirm BCF's ability to generate radical spiro-OMeTAD + c species without a co-dopant. However, only in conjunction with LiTFSI or FK209 were improved HTL conductivies/PSC FF/PCEs observed ( Table 1, entry 15), which suggests BCF is not sufficient to stabilize radical species. Improved performance was attributed to increased dopant solubility and reduced HTL roughness.
In summary, acid co-doping can improve PSC performance with traditionally-doped spiro-OMeTAD-based HTL, yet there is currently little known on PSC stability impacts with these additional additives. Ultimately, acid-doping for improved PSC performance could have greater utility if the hygroscopic codopants (LiTFSI, FK209) were replaced with alternative, nonhygroscopic dopants.

Metal-based salts
A wide variety of metal-based salts as dopants have been popular in addition to Li-and Co-based salts because they are cost-effective, commercially available, and can promote redox reactions with spiro-OMeTAD. Moreover, many metal-based dopants do not require co-dopants, like LiTFSI or FK209, for radical cation synthesis.
Copper iodide (CuI) has been shown promote radical cation generation with LiTFSI. 151 Soon aer, both CuI and copper thiocyanate (CuSCN) were reported to act as p-type dopants for spiro-OMeTAD without LiTFSI. 148 In dark stability assessment at ambient temperatures and humidity, CuSCN-doped spiro-OMeTAD maintained $80% of its initial PCE, outperforming devices with CuI or LiTFSI-doped HTLs (Summarized in Fig. 10 and Table 1, entry 9). Improved performance with CuSCN compared to CuI was attributed to favorable HTL morphology: reduced aggregation, crystallization, and pinhole formation was observed in CuSCN-doped HTL lms. Even though spiro-based HTLs are amorphous, morphological changes induced by additives can have a tremendous impact on macroscopic properties. In this case, simply interchanging iodide for thiocyanate determined the nal HTL lm quality.
In 2017, bis[di(pyridin-2-yl)methane] copper(II) bis [bis(triuoromethyl-sulfonyl) imide] (Cu(bpm) 2 ) and bis [2,2 0 -(chloromethylene)-dipyridine] copper(II) bis[bis(tri-uoromethylsulfonyl)imide] (Cu(bpcm) 2 ) (Fig. 10), 149 were assessed by Chen et al. These dopants were synthesized in 3 and 4 steps respectively with good yields, but the amine ligand synthesis required pyrophoric n-butyl lithium. The main difference between the two dopants is the chlorine incorporation in Cu(bpcm) 2 . Chlorine is inductively electron withdrawing, and this ligand tuning deepens the highest occupied molecular orbital (HOMO) level as compared to Cu(bpm) 2 , which, in turn, increases the oxidation driving force in the presence of spiro-OMeTAD. Cu(bpcm) 2 readily oxidized spiro-OMeTAD at approximately 75% yield per UV-Vis absorption characterization without oxygen exposure, approaching quantitative conversion that traditional dopants do not consistently yield. Unfortunately, preliminary stability assessments in the   152 While the dopants presented in this review predominantly use TFSI À as an inert counteranion, this particular example showcases the counterion's less understood importance. In addition to dopant concentration optimization for best device-level performance, Zhang et al. 152 also considered (1) doping efficiency (ratio copper salt to spiro-OMeTAD + c cations), (2) dopant redox potential, and (3) "ideal" spiro-OMeTAD + c concentrations. They hypothesize that higher radical cation concentrations lead to increased charge-recombination at the PAL/HTL interface and lower V oc , and in the context of the copper-based dopants, they found 1 mol% spiro-OMeTAD + c correlated with best device-level performance regardless of copper ligand identity. While these ndings may not be directly applicable to all doping systems, it is one of the few studies with a focus on dopant optimization, subsequent radical cation concentration, and interface dynamics. While devices with JQ1doped HTLs displayed best initial performance, for the stability study, the HTL was doped with bis[2-methyl-6-(6-methylpyridin-2-yl)pyridine] copper(II) bis[hexauorophosphate] (JQ3) instead of JQ1. A 1 nm Al 2 O 3 interface was required to suppress recombination at the PAL/HTL interface, improve initial PSC PCE and stability of devices stored in ambient conditions in the dark at 25 C and 50% relative humidity (Table 1, entry 28). The required buffer layer for improved PSC stability suggests water ingress is an important degradation pathway to mitigate for devices with JQ3-doped HTLs.
Overall, copper-based metal organic complexes and salts are noteworthy candidates due to their low cost and wide availability, but again, more comprehensive stability and lifetime assessment are needed. 17 The rst zinc-based salt was reported as an effective dopant in HTL for PSC. Zinc bis(triuoromethanesulfonyl)imide (Zn(TFSI) 2 ) was used to dope spiro-OMeTAD, and increased HTL hole mobility by an order of magnitude as compared to LiTFSI (with TFSI À anion concentration constant). 150 Upon device integration, devices with Zn(TFSI) 2 -doped HTL generated increased PCE (21.52% vs. 19.48% with LiTFSI) resulting from V oc and FF gains (Table 1, entry 26). The best PSC performance required FK209 as co-dopant. In one stability assessment, PSC were subjected to one sun at 25 C at maximum power point under a N 2 atmosphere and HTLs were doped with Zn(TFSI) 2 or LiTFSI and FK209. Initial efficiencies are >20%, but over the course of 600 hours, the devices with Zn-doped HTL maintained 100% initial PCE, while devices with Li-doped HTL maintained 80%. Aer 100 hours at 50 C, PSCs with Zn-doped HTL maintained roughly 80% initial PCE, while PSC with Li-doped HTL maintained roughly 45% initial PCE (Fig. 10). In a shelf life assessment, devices were unencapsulated and stored in the dark at room temperature in between device parameter measurements, and FK209 was not added as a co-dopant. Over 800 hours in a dry atmosphere, PCEs observed in devices with Zn-doped and Li-doped HTLs remain relatively constant. When the relative humidity is increased to 40%, FF and PCE drop in both cases, but a slower rate of decay is observed in devices with Zn-doped HTL. Unencapsulated PSC stability with FK209 is not reported. Overall, a device performance and stability tradeoff is not observed with Zn-doped HTLs when devices are "perfectly encapsulated" (in a N 2 atmosphere).
Generally, metal-based dopants are attractive because they lead to high performing PSC with low-cost materials. With regard to other metals, iridium and iron salts have been reported in standalone reports. An IrCp*Cl(PyPyz)[TFSI]-based doping system 153 showed superior stability (96% PCE retention over 3 months in the dark in ambient conditions) over control devices (Table 1, entry 5). Benzoyl peroxide 154 and ironbased complexes 155,156 also successfully dope spiro-OMeTAD with LiTFSI co-dopant for efficiency improvements (Table 1, entries 19, 23, and 24 respectively). A lithium-ion endohedral fullerene (Li + @C 60 ) was also recently reported (Table 1, entry  25). 157 Unencapsulated devices under constant illumination produced power as much as 10 times longer than the control devices.
It is challenging to directly compare stability since conditions vary, but with regard to both high performance and low cost, Zn(TFSI) 2 -based doping is one of the most promising systems, as the dopant is low cost and PSC showed excellent stability in a variety of conditions. The most pressing concerns that remain include (1) inherent hydrophilicity (2) need for codoping (LiTFSI, FK209), (3) potential for ion migration as observed with a number of metal electrodes (e.g., Au, Ag, Li + , Na + , K + ), 22,158,159 and (4) the potential for redox reaction occurring between metal and the perovskite itself. 160 If these factors can be addressed, metal-based dopants could be a cost-effective solution for both highly efficient and stable PSC.

Oxidized radical cation salts
Reduced oxygen exposure, hygroscopic lithium elimination, and dopant concentration consistency are important for batch to batch consistency and PSC lifetime. Ionic liquids, such as HTFSI, fulll the rst two criteria. Oxidized radical cation salts fulll all of these requirements and can be engineered with greater hydrophobicity than ionic liquids.
Ultimately, for widespread use of metal-free, oxidized radical cation salts as dopants, HTL thermal robustness and overall efficiency must be addressed. Importantly, dopant structure, in addition to device stack optimization (i.e., PAL, ETL, contacts), are vital to mitigate materials-level degradation pathways observed in radical cation-doped HTL. 165,166 3.5 Tetracyanoquinodimethane (TCNQ) derivatives A drawback of the systems discussed thus far is that they are limited to solution-processing. While solution methods are certainly one attractive route for PSC scale-up, 167 tetracyanoquinodimethane (TCNQ) and TCNQ derivatives are dopants compatible with both vacuum and solution-processing. TCNQ is a known electron acceptor capable for forming charge-transfer complexes for improved optoelectronic properties, 168 and uorinated TCNQ derivatives have successfully doped spiro-OMeTADbased HTLs, and improved HTL hydrophobicity via hydrophobic uorine atom incorporation and metal cation elimination.
In addition to vacuum deposition techniques, TCNQ dopants are also soluble in organic solvents, like chlorobenzene, and can  (1) nitrogen atmosphere, never exposed to air; (2) exposed to air; (3) reintroduced to a nitrogen atmosphere after being exposed to air. Upon reintroduction to a nitrogen atmosphere, devices with spiro(TFSI) 2 maintained greater than 98% of initial efficiencies after 10 min of illumination compared to less than 90% for devices without. Reprinted with permission from ref. be integrated into fully solution-processed devices as reported by Huang et al. with F4-TCNQ. 172 The doping efficiency of F4-TCNQ in spiro-OMeTAD was measured by UV-Vis absorption spectroscopy (Fig. 12e). While doping was achieved, the initial PCE of devices containing the F4-TCNQ-doped spiro-OMeTAD were somewhat lower than control devices with LiTFSI-doped spiro-OMeTAD (10.59% vs. 12.66%) ( hours dark storage at room temperature without encapsulation, more than 60% of initial PCE was retained with F4-TCNQ interlayer and LiTFSI doped spiro-OMeTAD HTL.
Overall, the use of TCNQ-based derivatives is a promising doping technique as it allows for a non-hygroscopic additive using either solution or vacuum-processing, but the majority of examples still have the same limitation: lowered device efficiencies as a tradeoff for stability gains.

Additional chemical doping schemes
There are a few noteworthy doped OSM HTL systems that do not t into any of the previous categories. In 2015, Zhang et al. 174 sought to simplify the HTL solution by incorporating TFSI À anions into the chemical structure of the HTM (Fig. 13). It is of  note that quaternary ammonium ions with TFSI À counterions are present in 3,3 0 -(2,7-bis(bis(4-methoxyphenyl)amino)-9H-uorene-9,9-diyl)bis(N-ethyl-N,N-dimethylpropan-1-aminium) bis(triuoromethanesulfonyl)imide (X44), not radical triarylamine species, as presented previously. 129,135 Even without radical cations, upon PSC integration, X44 showed superior hole conductivity as compared to AS37 (ref. 164) and 3,3 0 -(2,7dibromo-9H-uorene-9,9-diyl)bis(N,N-dimethylpropan-1-amine) (X41). Interestingly, while the conductivities of AS37 and X41 were comparable, devices with X41 as HTL did not function. On the other hand, devices with X44 as HTL showed good device efficiency and stability at maximum power point in the dark over 15 days (PCE 16.2% aer aging from initial 15.2%) as compared to AS37 (PCE 7.8% aer aging from initial 7.4%) ( Table 1, entry 21). 174 In a separate stability test, aer 1 day of continuous light soaking, devices with X44-based HTL $60% initial efficiency was maintained, but stability suffered at elevated temperatures (>70 C). Remote doping is an alternative strategy for increasing hole carrier density. Xu et al. deposited MoO 3 on undoped, crosslinked N 4 ,N 40 -di(naphthalen-1-yl)-N 4 ,N 40 -bis(4-vinylphenyl) biphenyl-4,4 0 -diamine (VNPB). 175 The precise doping mechanism is unclear with certain metal oxides, like MoO 3 , as its energetic properties rapidly change in the presence of oxygen and metal electrode selection. 176 Xu et al. demonstrated chargetransfer complex generation at the interface, and that free hole density increases because metal oxide accepts electron(s) from the HTM at the HTM/MoO 3 interface. To summarize, the MoO 3 dopes VNPB. Critically, the cross-linked VNPB prevented pinhole formation that would lead to MoO 3 reacting with the active layer at the interface. 177 Encapsulated devices employing VNPB/MoO 3 HTL showed little change in FF as compared to spiro-OMeTAD/LiTFSI control aer 1 hour at 110 C in N 2 ( Fig. 13 and Table 1, entry 13). Aer dark storage in 70% relative humidity, decomposition is minimal as monitored by PbI 2 emergence via XRD (but device performance was not reported). To the best of the author's knowledge, this is the only explicit report of remote doping for an HTL system in the PSC literature.

Summary and outlook
Moving away from spiro-OMeTAD with LiTFSI/FK209/tBP is almost certainly required to meet hybrid organic/inorganic PSC stability goals. As discussed in this work, and compiled in Table  1, there are a number of promising doping strategies for small molecule-based HTLs for n-i-p architecture PSCs. Nevertheless, all of the methods discussed herein have shortcomings to overcome, some of which are perhaps insurmountable. First, many of the alternative dopant strategies highlighted in this review still utilize hygroscopic LiTFSI/FK209, making it likely that they will ultimately suffer similar stability issues as conventional spiro-OMeTAD-based devices even if some stability gains are achieved. Nevertheless, there are a few standout systems which we believe are interesting candidates for continued investigation. The oxidized radical cation salts as dopants, such as spiro(TFSI) 2 (ref. 98) and EH44-ox, 127,164 effectively eliminate metals and other byproducts of in situ doping strategies from the HTL. More specically, the EH44/EH44-ox transport layer, having low ionization-energy, to the interface with the high electron-affinity material, in this case transition metal oxide MoO 3 , thereby enhancing the hole carrier density throughout the thin HTL. Evolution of performance, morphology, and material under external stress. (d) (1) The performance of devices using spiro-MeOTAD as the hole-extraction contact tested at room temperature (light gray) and after a 110 C burn-in test (dark gray) [in the burn-in test, devices are annealed at 110 C for 1 h in an N 2 environment and tested after cooling down to room temperature]. (2) The performance of devices using VNPB-MoO 3 tested at room temperature (light gray) and after 110 C burn-in process (dark gray). Reprinted with permission from ref. 175  system shows very promising stability, 127 yet still suffers from degradation under high temperature and humidity, likely at least in part a result of its low glass transition temperature. Metal-based dopants are generally inexpensive and generate efficient PSCs, but stability oen suffers. Ultimately, widespread implementation of new dopants on the scale of LiTFSI/FK209 will depend not only on long-term performance but a number of factors, such as synthetic ease, cost, and fabrication compatibility (summarized in Fig. 14).
With respect to dopants for small-molecule HTLs, there are a number of imminent research needs that we believe will support the greater goals in the perovskite community regarding PSC stability and lifetime. We propose the following as clear research aims for concentrated efforts, analogous to efforts seen in stabilizing active layers, ETLs, etc. While many dopants can oxidize spiro-OMeTAD, improving charge-transport properties and PCE, there can be stark durability differences between doping schemes, even though the majority HTL constituent is composed on the same molecular matrix: spiro-OMeTAD. An increased understanding dopant-induced degradation mechanisms is required. As part of these experiments, operational cell-level PSC studies which probe operational stability will certainly be required over and above simple material stability or hydrophobicity questions. These studies will propel improved rational design and help to answer questions such as: are these chemical-induced degradation mechanisms, or morphological failures (e.g., dopant migration, pinhole formation, etc.)? Can we design or modify materials to impede these degradation pathways? Which dopant schemes are the most inert?
Moreover, much of what is known is related to spiro-OMeTAD doping, yet there are hundreds of alternative HTMs already reported in the literature and an effectively innite number yet to be explored or even synthesized. If organics are to be eventually used in commercial PSC modules these will likely be doped to improve transport properties, so we must therefore search for dopant and HTM design motifs. Synergistic efforts among HTM and dopant design will be necessary if they are to contribute to stable and efficient PSCs and ultimately to clean, inexpensive electricity generation.

Conflicts of interest
There are no conicts to declare.