ARTICLE

Unravelling the role of vacancies in lead halideperovskite through electrical switching ofphotoluminescenceCheng Li1, Antonio Guerrero 2, Sven Huettner1 & Juan Bisquert 2

We address the behavior in which a bias voltage can be used to switch on and off the

photoluminescence of a planar film of methylammonium lead triiodide perovskite (MAPbI3)

semiconductor with lateral symmetric electrodes. It is observed that a dark region advances

from the positive electrode at a slow velocity of order of 10 μm s–1. Here we explain the

existence of the sharp front by a drift of ionic vacancies limited by local saturation, that

induce defects and drastically reduce the radiative recombination rate in the film. The model

accounts for the time dependence of electrical current due to the ion-induced doping

modification, that changes local electron and hole concentration with the drift of vacancies.

The analysis of current dependence on time leads to a direct determination of the diffusion

coefficient of iodine vacancies and provides detailed information of ionic effects over the

electrooptical properties of hybrid perovskite materials.

DOI: 10.1038/s41467-018-07571-6 OPEN

1 Department of Chemistry, University of Bayreuth, Universitätstr. 30, 95447 Bayreuth, Germany. 2 Institute of Advanced Materials (INAM), UniversitatJaume I, 12006 Castello, Spain. Correspondence and requests for materials should be addressed to S.H. (email: sven.huettner@uni-bayreuth.de)or to J.B. (email: bisquert@uji.es)

NATURE COMMUNICATIONS | (2018) 9:5113 | DOI: 10.1038/s41467-018-07571-6 |www.nature.com/naturecommunications 1

1234

5678

90():,;

The advent of hybrid perovskite solar cells has given rise toextraordinary photovoltaic performances, causing the riseof a solar energy conversion technology. However, the

original physical characteristics of these materials are not yetcompletely understood, and many significant experimentalobservations have not been satisfactorily explained so far. Slowtime-scale variations of photoluminescence (PL) phenomenahave been widely reported since an early stage in the research ofperovskite photovoltaics and optoelectronics1,2. Sanchez et al.3

observed either increasing or decreasing PL intensity accordingto the preparation method of the methylammonium lead triio-dide (MAPbI3) layer. Hoke et al.4 interpreted the observation oflight-induced transformations of PL in mixed MAPb(BrxI1–x)3 interms of photoinduced phase segregation. Subsequently, Leitjenset al.5 observed that an applied electrical field across the per-ovskite layer can either enhance or suppress the luminescence inlateral interdigitated electrode devices. These findings weredescribed in terms of a simple mechanism common to both fullyinorganic and organic semiconductors6,7, in which photo-generated electrons and holes drift to opposite sides of thedevice, reducing the bulk recombination rate and hence PLintensity. However, some aspects of the observations cannot beexplained by this simple model. Furthermore, it is expected thatunder an applied field, a massive ion drift will strongly affect thePL characteristics. In fact, the ion displacement has beenreported to modify PL properties under different conditions8,9,and a variety of reversible and irreversible PL transient responseshave been obtained under modification of the applied biasmagnitude and electrode polarity of laterally contacted MAPbI3layers10–14 that still require a general explanation. In this paper,we will present a dynamic transport model, based on the mod-ification of electronic concentrations by the displacement ofhalogenide vacancies in the perovskites. We show that the modelmatches well with experimental data of the advancement of adark edge in a wide variety of material samples, and moreimportantly, it directly relates the PL effects to the ionic driftcontrolled by a local saturation effect of defects, allowing the

direct determination of defect densities and ionic diffusionconstants.

ResultsObservation of advancing front characteristics. Recently, aproperty was uncovered, revealing striking features of the tran-sient PL in perovskite layers15. Using a wide-field PL imagingmicroscope, it was observed that the PL is progressively sup-pressed from the positively biased electrode. Here, additional datahave been recorded for interdigitated electrodes with a widerchannel length of ~150 μm, and they are shown in Fig. 1a. Thedarkened area in the left of the image forms a sharp front thatadvances at a slow velocity of ca. 10 μm s–1 toward a negativeelectrode, shown in Supplementary Movie 1. In some cases, thefront reaches the negative electrode and PL is completelyremoved. It can be later restored by biasing the perovskite film inthe contrary direction, as shown in Supplementary Movie 2 orletting the system to relax in the absence of bias and dark con-ditions, although recovery occurs at a much slower rate. Theseobservations were interpreted in terms of ionic redistributioncaused by the applied field11,16, but a quantitative understandingin terms of specific ionic species and a mechanistic suppression ofPL could not be achieved. To further understand the mechanism,we carry out the following experiments and establish the electricalmodel.

In the experiments presented in this work, the PL was opticallyrecorded while the electrical current was measured at a constantbias voltage from less than 1 V for impedance spectroscopy (IS)measurements, ~5 V for MAPbIxCl3–x and up to 100 V for multi-quantum wall (MQW) perovskites. Note that the electrical field inthe 150 μm channel is significantly lower than the fieldexperienced in a perovskite layer of 350 nm in photovoltaicdevices, as illustrated in Supplementary Table 1. In Fig. 1, weobserve that the overall initial PL intensity within the channeldecreases during the initial 12 s. In parallel, the measured currentshows a rapid decrease from 14 μA to 3.0 μA (Fig. 1g) after a

0 sec

32 sec 44 sec

8 sec 16 sec 24 sec

–

Z

+

14

a b c d

e f g h

Cur

rent

(μA

)

1/l 2 (

1010

A–2

)

12

10

8

6

4

20 10 20

Time (s)30 40 0 10 20

Time (s)30 40

10

8

6

4

2

0

Fig. 1 Observation of a dark advancing front in a light-soaked perovskite. a–f Time-dependent PL images of a perovskite film CH3NH3PbI3–xClx under anexternal electric field (~2 × 104 Vm–1). The “+ ” and ‘–‘ signs indicate the polarity of the electrodes. The excitation intensity is ~35mW cm–1 with awavelength of 440 nm, and the exposure time per image is 200ms. The channel length is ~150 µm. z(t) represents the PL quenched areas. The time-dependent z(t) is displayed in Supplementary Figure 1. The scale bar represents 100 μm. The color bar is a gray value with arbitrary units, indicating the PLintensity. g, h Electrical current I and 1/I2, respectively, monitored as a function of time during the measurement of experiment a

ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/s41467-018-07571-6

2 NATURE COMMUNICATIONS | (2018) 9:5113 | DOI: 10.1038/s41467-018-07571-6 | www.nature.com/naturecommunications

small initial transient due to ion drift and interfacial capacitivecontribution17,18, details are shown in Supplementary Figure 2. Adark front becomes evident in the left side of the channel, whichadvances toward the right of the channel during the measurementtime. Slowly with time, the current decreases to stabilize at about~3 μA after a measurement of 40 s with the front covering ¾ ofthe channel width. After this time, the advance of the dark frontslows down and after 50 s, the PL is almost completely quenchedwith a few inhomogeneities (Supplementary Movie 1). We shouldremark that the observation of a sharp moving front has beenconfirmed in independent devices that all show this generalbehavior, especially at applied electrical fields similar to thosepresent in working photovoltaic devices. However, the point atwhich the dark front slows its advance and current saturatesvaried for different perovskite material systems, which shows theoverall sensitivity to defect densities (Supplementary Figure 3).There is a clear relationship between the applied electrical field,the velocity in the advance of the dark front, and the measuredcurrent. Different perovskite samples, including mixed halide andMQW perovskites have been analyzed using different conditionsto prove its validity and are shown in Supplementary Table 2 andSupplementary Figure 3 and 4.

A variety of effects can be invoked to explain a decrease orincrease of PL1 and include mobile charges19, mobile ions11,20,defects, and passivation14,21, solid-state reactions such as theinteraction with iodide-related defects and iodine vapor22, etc.However, the challenge is to establish the mechanism that causesthe existence of a sharp front that moves at velocity v. This effectcannot be provoked by removal of electronic carriers swept by theelectrical field that would produce a homogeneous decrease of PLacross the film. The low velocity of the sharp front indicates thatwe must consider the displacement of ions. Diffusion of ionicspecies ascribed to the ion concentration gradient, makes oneassume a gradually decaying spatial distribution of PL23 ratherthan an advancing edge toward an opposite electrode.

In the following, we develop a quantitative model indicated inFig. 2 based on the idea that the current across the film (except fora short initial transient effect) is predominantly electronic. Thereason for this is that the electronic mobility in hybrid perovskitesis about ten orders of magnitude larger than the ionic mobility.However, the applied electric field has the effect that ionic defectswill move and change the local charge distribution in the bulk ofthe sample, decreasing and sometimes increasing the overallcurrent density. In addition, the moving defects promote localsolid-state reactions that enhance non-radiative recombinationproducing a dark area.

Physical model for the advancing front velocity. In general,clarifying the dominant moving species in a mixedionic–electronic conductor is quite challenging. Starting from theseminal work of Azpiroz et al.24, it has been recognized thatiodine vacancies readily diffuse in the perovskite, much fasterthan MA and Pb vacancies. Several previous works have analyzedthe perovskite films with lateral symmetric contacts and studiedthe quenching of PL when a large bias is applied. These studiesconcluded that iodide vacancies are the dominant ionic speciesthat moves under the influence of an electrical field, in combi-nation with an electronic current that flows through the sample.Huang et al.16 observed a moving thread in MAPbI3. Ho-Bailieet al. reported the growing regions of quenched PL dependenceon the size of electrical field and humidity10,11. In all these cases,the sample becomes darkened at the anode as in our experiments.A recent work by Senocrate et al.25 provided an interpretation ofmoving ionic species based on a number of techniques. It wasconcluded in all cases that iodide vacancies are the main moving

species. However, the main ionic conductivity component occursby hopping conductivity of iodine vacancies and not by iodineinterstitials. Transient luminescent phenomena due to ionicrearrangement were reported by Bolink et al.26 and Gingeret al.27, where the effect of electronic carriers is studied bymodification of the injection properties of the contacts. It wasshown that in order to have PL quenching in the lateral config-uration as a function of the applied bias, not only ion migration isneeded but also the presence of electronic current injected fromthe contacts.

In order to provide a very direct evidence of the p-typecharacter of the herein-used perovskite films, we determined theSeebeck coefficient (Supplementary Figure 5). The sign of theSeebeck coefficient directly points out the p-type majority chargecarriers prevalent in a semiconductor and was measured to be 3.4± 0.5 mVK–1. This value is consistent with previous work by Yeet al.28 In addition, after biasing 50 V for 180 s, the Seebeckcoefficient increases with a factor of 2, implying a change ofdoping by an external field. Therefore, electronic conductivity ismainly attributed to electronic holes.

Based on the ionic–electronic properties of MAPbI3, weformulate a dynamic model that is outlined in Fig. 2. Here, wepropose that a high concentration of Vþ

I vacancies is present atthe interface with the contacts as a result of lead beingundercoordinated by iodine atoms. These types of

Region 1

Vacancy Vı+ InterstitialIodide ion lı

–

–

–

–

+

+

+

e– en

ergy

Region 2

Ev

qV2

qV1p1 = p0-Cm

p0

dz z

Ec

v→

Fig. 2 Scheme of a physical model. Optoelectronic effects of iodinevacancies drifting in a perovskite layer with symmetric contacts, under theapplied field caused by positive bias at the left side. In this model of a p-doped material, the dominant electrical current flowing across thesymmetrical electrodes, is an electronic current. The unshaded region2 shows the original situation of majority carrier holes of density p0 (due tothe intrinsic doping by iodine interstitials I�i ) that cause PL underphotogeneration of electrons by light soaking. The applied voltage inducesan ion movement: iodine vacancies Vþ

I drift in the electrical field toward theright. In the shaded region 1, they fill the space up to a density Cmcompensating the p-doping and reducing majority carrier density to p1=p0–Cm. With applied voltage V, the conduction band Ec and valence band Evbend by an amount qV= qV1+ qV2, where q is elementary charge. Theionic current is negligible compared with electronic current (due to theenormous difference of ionic and electronic carrier mobilities) but iondisplacement modifies the local extent of doping and influences themajority carrier density and thus the total current changes (decreases) withtime. Due to the drift of ions from the contact, the shaded region growswith time, so that point z advances with time at a velocity v

NATURE COMMUNICATIONS | DOI: 10.1038/s41467-018-07571-6 ARTICLE

NATURE COMMUNICATIONS | (2018) 9:5113 | DOI: 10.1038/s41467-018-07571-6 |www.nature.com/naturecommunications 3

undercoordinated species have previously been detected andidentified by absorbance measurements29. As mentioned pre-viously under a constant applied bias, iodine vacancies Vþ

I drift inthe electrical field toward the opposite electrode. The additionalvacancies correspond to iodine ions that pile up at the contactsurface, in a characteristic accumulation region that has beenobserved in many cases by IS30 and photovoltage decays31. Whenthe Vþ

I vacancies move to the right, they compensate the negativecharge from iodine interstitials reducing the hole density.Vacancies can fill the space up to a maximum concentrationvalue Cm, reducing hole density to p1=p0–Cm (assuming p0 > Cm).The reduction of doping and the creation of non-radiativerecombination sites induces a dark region 1 of size z(t) thatincreases with time. Meanwhile, we assume that region 2 (z � x � d) remains largely undisturbed with approximately theinitial hole density p0. In the case that photogenerated carriersexceed the majority carrier density, p0 indicates the density in thebright region and p1 refers to the lower carrier density in the darkregion due to enhanced recombination.

The electrical current in each region ji (i= 1, 2) is driven by therespective electrical field Ei and since the mobility μp>>μC theionic current component can be neglected. We have

j1 ¼ qp1μpE1 ¼ j2 ¼ qp0μpE2 ð1Þ

Hence, the division of voltage between the two regions inV=V1+V2 is governed by the equation

V2 ¼p1p0

d � zz

V1 ð2Þ

and we obtain

V1 ¼ 1þ p1p0

dz� 1

� �� ��1

V ð3Þ

The ionic drift current in region 1 is

jC ¼ qCmμCV1

zð4Þ

The ionic flux causes the increase of region 1 as follows:

jC ¼ qCmdzdt

ð5Þ

Combining the previous expressions, we obtain the equationfor the variation of z

v ¼ dzdt

¼ z�1 1þ p1p0

dz� 1

� �� ��1

μCV ð6Þ

By integration, we get the dependence on time of the velocity ofadvance of the front

v tð Þ ¼ 1þ 2γdv0t

� ��1=2

v0 ð7Þ

where the initial velocity is

v0 ¼p0p1

μCVd

ð8Þ

and we have introduced the quantity

γ ¼ p0p1

� 1 ð9Þ

The electrical current depends on time as

j ¼ 1þ 2γdv0t

� ��1=2

j0 ð10Þ

where the initial current is

j0 ¼ qp0μpVd

ð11Þ

Accordingly, the following representation of jðtÞ1j2¼ 1

j20þ at ð12Þ

forms a straight line up to the point where current saturates. Theslope in Equation (12) is given by

a ¼ 2γv0dj20

ð13Þ

The final current at z=d has the value

j ¼ 11þ γ

j0 ¼ qp1μpVd

ð14Þ

It is smaller than the initial current, since the initial amount ofp-doping has decreased by the spread of vacancies. That thedoping has changed is also reflected in the increased Seebeckcoefficient after biasing, which increases from 3.4 ± 0.5 to 6.2 ±0.8 mVK–1 (Supplementary Figure 5).

Let us consider alternative scenarios for the modification ofdefect distribution in the biased film. Instead of a p-doped systemassumed above, we take a system that is n-doped with intrinsicdensity n0 and undergoes the same experiment. The model is verysimilar to Fig. 3, but in this case, the moving iodine vacancies willreduce the doping of region 1 and this zone will get the largerelectrical field. The results are given by the same previousequations but now

γ ¼ n0n1

� 1 ð15Þ

Thus, the current increases with time, in contrast to the caseabove, as now the overall doping is increased. Finally, we considerthe case in which the drifting vacancies change the type of dopingso that n1=Cm–p0. Then region 1 and 2 conduct by electrons andholes, respectively, and the layer operates in a diode mode bydouble electronic carrier injection from electrodes. The currentsmeet at z establishing a strong recombination zone at the edge, asin a light-emitting electrochemical cell (LEC), where theredistribution of ionic charges forms the recombination diodestructure32. Indeed, some perovskite systems show an increasedphotoluminescence right at the boundary between regions 1 and 2(Supplementary Figure 6). This observation could indicate thepresence of electroluminescence, i.e., the direct recombination ofholes and electrons occurs at that place and electron conductionoccurs as well, however, it can also be explained by a decrease ofnon-radiative recombination as commented later. In the herein-

ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/s41467-018-07571-6

4 NATURE COMMUNICATIONS | (2018) 9:5113 | DOI: 10.1038/s41467-018-07571-6 | www.nature.com/naturecommunications

presented model, it was completely sufficient to choose a singlecarrier model of Fig. 2 to describe the experiments.

As it is observed in Fig. 1g that the electrical current decreaseswith time, we are in the situation of majority hole carriers, wherebias-induced excess Vþ

I in region 1 reduces the carrier density.Plotting the current as j–2(t) in Fig. 1f, an excellent agreement isobtained with the model, as the figure displays a straight line inthe first half of the length d= 150 μm. We obtain the resultsj0=12 μA, a= 4.6×10–3 μA–2 s–1, v0= 10.5 μm s–1, and γ= 3.3.By Equation (9), it follows that

p1 ¼1

1þ γp0 ¼ 0:23p0 ð16Þ

This result indicates that defects traveling to the bulk layercreate a 30% decrease of doping density, as it is also evidentcomparing the initial and final current density. Using Equation(8), we obtain the mobility of iodine vacancies µC= 9.1 × 10–7

cm² V–1 s–1 that corresponds to a diffusion coefficient

Dv1=2.4×10–8 cm2 s–1. The current analysis has been applied to

different samples, as shown in Supplementary Table 2 andSupplementary Figure 8, and calculated DVI

ranges between 5 ×10–8 and 6 × 10–9 cm2 s–1 values are in good agreement with thevalue 2 × 10–9 cm2 s–1 of Senocrate et al.25. It is important to notethat in some samples, there is a deviation from the modeled curveafter half of the electrode is dark and the measured current eitherslows down the reduction or even increases. The reason is thatsecond region 2 cannot be assumed to remain undisturbed withp0 at this point.

Alternative interpretations. In previous work, the interpretationof the front advance in MAPbI3 was attributed to direct migrationof ionic defects11. Using the velocity of the front, the calculationprovides a value DVI

of the order 10–11 cm2 s–1. This result dis-regards the local charge compensation that influences electroniccurrent and consequently it is rather low. In contrast to this,applying our model based on local saturation effect, we obtainresults in agreement with other techniques where DVI

rangesbetween 5 × 10–8 and 6 × 10–9 cm2 s–1, as pointed out before.

Another approach to explain a decaying current across a mixedelectronic–ionic conducting film is the method of stoichiometric(galvanostatic) polarization, which assumes that the initialcurrent is mixed electronic and ionic, while the final current isonly electronic. However, these experiments are usually per-formed25,33,34 at lower current density. It has been shown that theexistence of a moving front that creates a dark region requires alarge applied electrical field, higher than a threshold value, that isa function of ambient humidity11. It is likely that galvanostaticpolarization conditions25,33,34 should occur below the thresholdfor the formation of a dark region advancing front. In the otherextreme, a larger applied electrical field may damage the MAPbI3sample11,35. It is worthwhile to mention that CsPbI2Br showsenhanced stability to an applied electrical field and largeractivation energies for ion hopping34.

Evidence on the dominance of electronic current. There are twoexperimental evidences that suggest that the measured currentin our experiment is due to the change in the electronic con-duction properties of the bulk perovskite layer under illumi-nation, rather than to the disappearance of an ioniccontribution. On the one hand, when the applied voltage isreversed, the current is maintained just with a change of sign,Fig. 3a. These results clearly indicate that the dominant currentis electronic current passing through the film, as opposed to asignificant ionic contribution, as reversing the applied voltagewould instantaneously increase the initial transient current dueto accumulated ions. Different examples of voltage reversal areshown in Supplementary Figures 8, 9, and 10. It should bepointed out also that electronic current in laterally contactedelectrodes has been observed to increase with time27, which isimpossible in stoichiometric polarization conditions but quitelikely in our model of ion-induced doping modification in thebulk. On the other hand, IS measurements have been carriedout with the double function of providing a source of electricalfield, by the application of a DC bias, and to probe the electricaleffects that take place at different time scales. In this mea-surement, the high-frequency signal is due to the properties ofthe perovskite layer2. Interestingly, the arc at high frequencyincreases the resistance as the PL intensity decreases by theapplication of increasing potential bias, Fig. 3b, c. Overall, theseevidences point to a modification of the bulk properties of theperovskite layer as responsible to the modification of the elec-tronic properties.

4a

b

c

2

0

Cur

rent

(μA

)P

L (a

.u.)

–2

–4

0.25

0.20

0.15

0.10

0.05

0.00

250

200

150

100

50

0600 700 800

Wavelength (�)900 1000

0.0 0.1 0.2 0.3 0.4 0.5

0 10

+7.5 V

–7.5 V –7.5 V

0 V0.2 V0.4 V1.0 V

0 VInitial

0.2 V0.4 V1.0 V

Resistance

20 30Time (s)

Z ′ (MΩ)

–Z ″

(MΩ

)

40 50 60

Fig. 3 Physical measurements of the biased perovskite films. a Currentdecay curves as a function of the polarity of the applied voltage on aCH3NH3PbI3–xClx perovskite film with ± 7.5 V across an ~150 μm channel. Achange in polarity maintains the current with a different sign, indicating thationic transport is modifying the bulk properties of the perovskite. bComplex impedance plot of an interdigitated electrode measured in thedark as a function of the applied DC bias. c PL decay in intensity of thedevice measured in b

NATURE COMMUNICATIONS | DOI: 10.1038/s41467-018-07571-6 ARTICLE

NATURE COMMUNICATIONS | (2018) 9:5113 | DOI: 10.1038/s41467-018-07571-6 |www.nature.com/naturecommunications 5

DiscussionIn this section, we analyze in detail the mechanism whereby themoving ionic defects provoke a total disappearance of PL in a partof the sample. Since the first observations of PL quenching inbiased and light-soaked perovskite films, mobile ion-inducedenhanced non-radiative recombination10 has been suggested asthe main cause of film darkening in regions that grow with time.So far, it has not been explained mechanistically how theadvancing front of the iodide vacancies causes the suppresion ofPL. It is natural to assume that the ionic flux that advances fromthe positive electrode consisting of anion vacancies is starting alocal solid-state ionic reaction that creates abundant non-radiative recombination centers. While the opposite ionic reac-tion occurs often at the edges, removing the non-radiative localspecies, and enhancing, rather than decreasing, PL.

First of all, we must analyze the type of defects that can bepresent in our samples. In this respect, Seebeck effect clearlypoints to p-type doping in the initial state. This type of dopingcan be due to either an excess of interstitials negatively chargedions (I−), vacants of positively charged native ions (MA+, Pb2+),or ions placed in the lattice arising from redox reactions thatsomehow are able to stabilize the perovskite structure (i.e., Pb(0)). From X-ray photoelectron spectroscopy (XPS) measure-ments, metallic lead Pb(0) is frequently observed, as a degradationproduct. Therefore, Pb(0) should be stable in the film36. Thesespecies would include Ii, VMA, VPb, and lattice Pb(0).

The origin of these defects is not central in our work but onecan attribute them to the generation of Schottky or Frenkeldefects. Our observations are compatible with the presence ofFrenkel defects involving displacement of an atom from the lat-tice position to an interstitial position and generation of avacancy. Other works have endorsed the conclusions of theseobservations37,38 and the presence of iodine interstitials haverecently been measured by X-ray and neutron scattering39.Schottky defects indeed have been shown to have reasonablysmall formation energy40. However, very recently, Kelvin probemicroscopy experiments using a similar lateral device geometry asin this work, show the major prevalence of Frenkel defects andtheir long-range motion in an electric field27,41.

Next, we must consider the species that can be present in ourfilms and could lead to PL quenching. There are different species

that have been reported to quench the PL in the perovskite filmforming effective recombination centers38. Iodide interstitialshave been recently highlighted as one predominant non-radiativerecombination center. A second possibility is the presence of Pbvacancies. However, due to the large size of the ion, thisquenching ability of Pb would be expected to be observed asbackground for all our experiments, as it is not expected that theelectrical field would lead to migration of the ion. Alternatively,formation of interstitial Pb(0) has also been suggested by Birk-hold et al. as the dominant recombination center27. Furthermore,the oxidation state of the halogen vacancy has a dominant effecton the extent of recombination at a defect site42.

Finally, we can provide an interpretation of migrating speciesand its relationship with PL quenching, outlined in Fig. 4. Wehave an initial background doping due to any of the speciesdescribed above (Ii, VMA, VPb, and lattice Pb(0)). We can assumethat all of them will be present in our films. From these species,only Ii will be responsible for PL quenching, as suggested bydensity functional theory (DFT) calculations. The initial PLintensities of the samples are high, but in many cases, the PLincreases with the time under illumination (and no appliedelectrical field). Then, we cannot rule out the presence of chargedIi as one of the initial dopants that could be neutralized by thereverse of Frenkel-type defect chemistry, this is the combinationof vacancy and iodine interstice to eliminate the quenching site.As vacancies advance toward the negative electrode, they com-pensate the charge of the background doping induced by Ii, VMA,VPb, or Pb(0) reducing the conductivity. In addition, the excessiodine vacancies (compensated by electrons) can lead to redoxreactions as those described by Birkhold et al., in which interstitialPb2+ is reduced to Pb(0)27. Both Ii and interstitial Pb(0) will bethe recombination centers that lead to the dark area in the PL.Overall, the positive electrode will be enriched of iodine atoms, assupported by XPS measurements carried out previously by ourgroup43.

It can be observed in different images that there is a bright edgeclose to the contacts, indicating that the actual species present atthe contacts are different to those in the bulk, with less con-centration of PL quenching centers close to the contacts. Here, wepropose that at the bright edge, there is a large concentration ofVI susceptible to be transported under an applied electrical field

Cm

h+

h+

h+

IiPb(0)

VPb2+

VMA+

h+ h+ h+ h+ h+ h+ h+h+h+

p0 =

Cm =

p0

h+

h+

a b c

d e f

+ – –+

Vı+

X–

X– X– X– X–

X–

X–

X–

X– X– X– X– X– X– X– X– X– X–

h+ h+ h+ h+ h+

X–Vı

+

Au

Au

Au

Au

Excess

Pb(0)

p1 = p0-Cm

Au

Au

Vı+

Vı+

Vı+

Vı+ Vı

+ Vı+ Vı

+ Vı+Vı

+ Vı+

e–

e– e–

e–e– e– e– e– e–

Vı+

Vı+ Vı

+Vı

+

Ii–

Ii–

Vı+

Vı+

Vı+

Pb2+

Fig. 4 Interpretation of the dominant ionic effects. Diagrams with a proposal of the processes taking place during the application of an electrical field in thelateral electrode configuration. a and d represent the initial state with a background of p-dopants (p0) and iodine vacancies present at the interface with theelectrode. b, e When the electrical field is applied, vacancies migrate toward the negative electrode, in this transit, they partially compensate the charge ofthe background doping, including some non-radiative recombination defects (Ii). c, f Excess vacancies carry excess of electrons that can lead to generationof other PL quenching defects such as Pb(0)

ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/s41467-018-07571-6

6 NATURE COMMUNICATIONS | (2018) 9:5113 | DOI: 10.1038/s41467-018-07571-6 | www.nature.com/naturecommunications

as dominant moving species, as proposed by Senocrate et al.25

and low concentration of Ii as quenching species. As can beobserved in Supplementary Movie 1 (from 8 s onward), at theedge of the advancing front, there is an increase in PL whichcould be the result of neutralization of Ii recombination centersby VI, just the opposite to Frenkel defect generation.

It is important to highlight that the direction of the advancingfront will be highly dependent on the actual doping speciespresent in the films. Our results clearly show an advance of thedark front from the positive to the negative electrodes in agree-ment with different authors11,44. Alternatively, there are otherauthors who report that the dark front advances from negative topositive12,27,45. If the VI are initially abundant, the negativeelectrode will be enriched with vacancies and when these areconfined at the electrode, they can lead to formation of PbI2, asdescribed by Huang et al.16 in agreement with a Pb/I ratio lowerthan three, also supported by XPS43. Indeed, it has been reportedthat the doping nature of the MAPbI3 will strongly depend on theprocessing conditions to prepare the films and imbalance of thereagents with excess of MA+ or PbI2 can lead to n-doping or p-doping, respectively46.

In conclusion, an electrical bias in a perovskite layer can switchon and off the photoluminescence in a timescale of seconds. Weexplain the advance of the dark area with a sharp front in terms ofthe drift of iodine vacancies that fill the space up to a criticaldensity turning the material more intrinsic-like and enhancingnon-radiative recombination. An interplay between interstitialand vacancy defects determines the dominant electronic densityand subsequently allows a control of electro-optical properties.

The model fits perfectly with the experimental data describingthe initial dynamics of the drift of vacancies in an electric fieldand directly allows us to determine the mobility of iodinevacancies. This method provides a direct and visual way to trackthe migration of vacancies and obtain a value for the defectdensity and is applicable to a broad range of organometal halideperovskite systems as well as Ruddlesden–Popper halide layeredperovskites. The implications are significant as it shows thatvacancies (i.e., ions) redistribute in the presence of an electric fieldas it is prevalent in a solar cell device and change the respectiveelectron/hole densities affecting the overall electronic current. Itwill greatly help to understand how to compensate defect andmigration processes in order to eliminate hysteresis effects anddevice degradation.

MethodsDevice fabrication. CH3NH3PbI3–xClx precursor solution was prepared by dis-solving 0.88M lead chloride (PbCl2) (Sigma-Aldrich) and 2.64M CH3NH3I (MAI)(Tokyo Chemical Industry company) in anhydrous N,N-dimethylformamide(DMF) (99.8%, Sigma-Aldrich). The precursor solutions were filtered through a0.2 μm polytetrafluoroethylene (PTFE) filter (Carl Roth GmbH+ Co. KG). Glasssubstrates were washed with acetone and isopropanol for 10 min each. Then theseglass substrates were treated with ozone for around 10 min. In a nitrogen gas-filledglovebox, this precursor solution was spin-coated on the glass substrates at 3000rpm for 60 s. Following that, the as-spun films were annealed at 100 °C for 80 minin the glovebox. The characterization of the perovskite film is shown in Supple-mentary Figures 11 and 12. The performance of the photovoltaic device using thisperovskite film is shown in Supplementary Figure 13. The perovskite film onglasses was transferred into an evaporation chamber with pressure of ~3 × 10–6

mbar, and ~70 nm thickness of Au was deposited by thermal evaporation througha shadow mask. This interdigitating shadow mask defined the electrode geometry:the electrode distance was 200 μm and a ratio between channel width W and lengthL was of 500. External conducting wires were connected to the device using anUltrasonic Soldering System (USS-9200, MBR electronics GmbH). In the end, toprotect the film from oxygen and water, 40 mgmL–1 poly(methyl methacrylate)(PMMA) solution dissolved in butyl acetate (anhydrous, 99%, Sigma-Aldrich) wasspin-coated on the film at a speed of 2000 rpm for 60 s in the glovebox.

Photoluminescence imaging microscopy. This measurement was conducted on ahome-build PL microscope, as shown in Supplementary Figure 7. Based on a com-mercial microscopy (Microscope Axio Imager.A2m, Zeiss), the sample was allocated

in the focal plane of an objective lens (10 × /0.25 HD, Zeiss), and the sample positionwas adjusted by a motorized scanning stage (EK 75*50, Märzhäuser Wetzlar GmbH &Co. KG). The sample was illuminated by an internal LED illuminator using a dichroicmirror and filter combination (HC 440 SP, AHF analysentechnik AG) with theexcited wavelength of around 440 nm. The excitation light intensity can be controlledand was set to ~35mWcm–1. The PL signal was filtered (HC-BS 484, AHF analy-sentechnik AG) to suppress residual excited light and directed onto a CCD camera(Pco. Pixelfly, PCO AG) with the exposure time of 200ms. A constant voltage of4–10 V was applied between the Au electrodes (236 Source Measure Unit, KeithleyCompany), and the current was monitored and recorded by a LabVIEW program.During the measurement, the excitation illumination was kept on continuously.Under the above illumination and bias condition, the ratio between light current anddark current was measured as Ilight/Idark= ca. 102. Additional measurements coupledPL imaging with impedance spectroscopy measurements using an Autolab PGSTAT204 with an impedance analyzer module. Measurements were carried out using DCbias, as described in the text with a 20 mV (RMS) AC voltage perturbation withfrequencies ranging from 20 mHz−1 to 1 MHz.

Seebeck coefficient measurements. The Seebeck coefficient was measured onperovskite films, which were processed in exactly the same way as above. As shownin Supplementary Figure 5(b), lateral single-channel gold electrodes with a spacingof 1 mm were evaporated maintaining the same sample geometry. The sample wasplaced on a differential temperature stage with a hot and a cold side enclosed in asmall vacuum chamber at 0.1mbar. The temperature on one side was changedbetween 50 and 75 °C while keeping 24 °C on the other side and an equilibrationtime of 20 min was allowed. The thermal voltage was measured with an AutolabPGSTAT 204 averaging over 300 single measurements.

Data availabilityThe experimental data that support the findings of this study are available from thecorresponding author upon reasonable request.

Received: 13 March 2018 Accepted: 8 November 2018

References1. Panzer, F., Li, C., Meier, T., Köhler, A. & Huettner, S. Impact of structural

dynamics on the optical properties of methylammonium lead iodideperovskites. Adv. Energy Mater. 7, 1700286 (2017).

2. Lopez-Varo, P. et al. Device physics of hybrid perovskite solar cells: theory andexperiment. Adv. Energy Mater. 8, 1702772 (2018).

3. Sanchez, R. S. et al. Slow dynamic processes in lead halide perovskite solarcells. Charact. Hysteresis. J. Phys. Chem. Lett. 5, 2357–2363 (2014).

4. Hoke, E. T. et al. Reversible photo-induced trap formation in mixed-halidehybrid perovskites for photovoltaics. Chem. Sci. 6, 613–617 (2015).

5. Leijtens, T. et al. Modulating the electron–hole interaction in a hybrid leadhalide perovskite with an electric field. J. Am. Chem. Soc. 137, 15451–15459(2015).

6. Greenham, N. C., Peng, X. & Alivisatos, A. P. Charge separation and transportin conjugated-polymer/semiconductor-nanocrystal composites studied byphotoluminescence quenching and photoconductivity. Phys. Rev. B 54,17628–17637 (1996).

7. Morana, M. et al. Bipolar charge transport in PCPDTBT-PCBM Bulk-Heterojunctions for photovoltaic applications. Adv. Func. Mater. 18,1757–1766 (2008).

8. Galisteo-López, J. F., Li, Y. & Míguez, H. Three-dimensional opticaltomography and correlated elemental analysis of hybrid perovskitemicrostructures: an insight into defect-related lattice distortion andphotoinduced ion migration. J. Phys. Chem. Lett. 7, 5227–5234 (2016).

9. Yoon, S. J., Kuno, M. & Kamat, P. V. Shift happens. How halide ion defectsinfluence photoinduced segregation in mixed halide perovskites. ACS EnergyLett. 2, 1507–1514 (2017).

10. Chen, S. et al. Mobile ion induced slow carrier dynamics in organic–inorganicperovskite CH3NH3PbBr3. ACS Appl. Mater. Interfaces 8, 5351–5357 (2016).

11. Deng, X. et al. Electric field induced reversible and irreversiblephotoluminescence responses in methylammonium lead iodide perovskite. J.Mat. Chem. C. 4, 9060–9068 (2016).

12. Jacobs, D. L., Scarpulla, M. A., Wang, C., Bunes, B. R. & Zang, L. Voltage-induced transients in methylammonium lead triiodide probed by dynamicphotoluminescence spectroscopy. J. Phys. Chem. C 120, 7893–7902 (2016).

13. Xu, Z., De Rosia, T. & Weeks, K. Photoluminescence–voltage (PL–V)hysteresis of perovskite solar cells. J. Phys. Chem. C 121, 24389–24396(2017).

14. Chen, S. et al. Light illumination induced photoluminescence enhancementand quenching in lead halide perovskite. Sol. RRL 1, 1600001 (2017).

NATURE COMMUNICATIONS | DOI: 10.1038/s41467-018-07571-6 ARTICLE

NATURE COMMUNICATIONS | (2018) 9:5113 | DOI: 10.1038/s41467-018-07571-6 |www.nature.com/naturecommunications 7

15. Li, C. et al. Real-time observation of iodide ion migration inmethylammonium lead halide perovskites. Small 13, 1701711 (2017).

16. Yongbo, Y. et al. Electric-field-driven reversible conversion betweenmethylammonium lead triiodide perovskites and lead iodide at elevatedtemperatures. Adv. Energy Mater. 6, 1501803 (2016).

17. Heiser, T. & Weber, E. R. Transient ion-drift-induced capacitance signals insemiconductors. Phys. Rev. B 58, 3893–3903 (1998).

18. Zhang, H. et al. Dynamic interface charge governing the current–voltagehysteresis in perovskite solar cells. Phys. Chem. Chem. Phys. 17, 9613–9618(2015).

19. Wen, X. et al. Mobile charge-induced fluorescence intermittency inmethylammonium lead bromide perovskite. Nano Lett. 15, 4644–4649 (2015).

20. Li, C. et al. Emission enhancement and intermittency in polycrystallineorganolead halide perovskite films. Molecules 21, 1081 (2016).

21. Stranks, S. D. et al. Recombination kinetics in organic-inorganic perovskites:excitons, free charge, and subgap states. Phys. Rev. Appl. 2, 034007 (2014).

22. Zohar, A. et al. What is the mechanism of MAPbI3 p-doping by I2? Insightsfrom optoelectronic properties. ACS Energy Lett. 2, 2408–2414 (2017).

23. deQuilettes, D. W. et al. Photo-induced halide redistribution inorganic–inorganic perovskite films. Nat. Commun. 7, 11683 (2016).

24. Azpiroz, J. M., Mosconi, E., Bisquert, J. & De Angelis, F. Defect migration inmethylammonium lead iodide and its role in perovskite solar cell operation.Energy Environ. Sci. 8, 2118–2127 (2015).

25. Senocrate, A. et al. The nature of ion conduction in methylammoniumlead iodide: a multimethod approach. Angew. Chem. Int. Ed. 56, 7755–7759(2017).

26. Bandiello, E. et al. Influence of mobile ions on the electroluminescencecharacteristics of methylammonium lead iodide perovskite diodes. J. Mat.Chem. A 4, 18614–18620 (2016).

27. Birkhold, S. T. et al. Interplay of mobile ions and injected carriers createsrecombination centers in metal halide perovskites under bias. ACS EnergyLett. 3, 1279–1286 (2018).

28. Ye, T. et al. Ultra-high Seebeck coefficient and low thermal conductivity of acentimeter-sized perovskite single crystal acquired by a modified fast growthmethod. J. Mat. Chem. C 5, 1255–1260 (2017).

29. Rahimnejad, S., Kovalenko, A., Martí-Forés, S., Aranda, C. & Guerrero, A.Coordination chemistry dictates the structural defects in lead halideperovskites. ChemPhysChem 17, 2795 (2016).

30. Zarazua, I., Bisquert, J. & Garcia-Belmonte, G. Light-induced space-chargeaccumulation zone as photovoltaic mechanism in perovskite solar cells. J.Phys. Chem. Lett. 7, 525–528 (2016).

31. Gottesman, R. et al. Dynamic phenomena at perovskite/electron-selectivecontact interface as interpreted from photovoltage decays. Chem. 1, 776–789(2016).

32. Lenes, M. et al. Operating modes of sandwiched light-emitting electrochemicalcells. Adv. Func. Mater. 21, 1581–1586 (2011).

33. Kim, G. Y. et al. Large tunable photoeffect on ion conduction in halideperovskites and implications for photodecomposition. Nat. Mater. 17,445–449 (2018).

34. Zhou, W. et al. Light-Independent ionic transport in inorganic perovskite andultrastable Cs-based perovskite solar cells. J. Phys. Chem. Lett. 8, 4122–4128(2017).

35. Zhao, Y.-C. et al. Quantification of light-enhanced ionic transport in leadiodide perovskite thin films and its solar cell applications. Light Sci. Appl. 6,e16243 (2017).

36. Sun, Q. et al. Role of microstructure in oxygen induced photodegradation ofmethylammonium lead triiodide perovskite films. Adv. Energy Mater. 7,1700977 (2017).

37. Wei, L.-y, Ma, W., Lian, C. & Meng, S. Benign interfacial iodine vacancies inperovskite solar cells. J. Phys. Chem. C 121, 5905–5913 (2017).

38. Meggiolaro, D., Mosconi, E. & De Angelis, F. Modeling the interaction ofmolecular iodine with MAPbI3: a probe of lead-halide perovskites defectchemistry. ACS Energy Lett. 3, 447–451 (2018).

39. Minns, J. L., Zajdel, P., Chernyshov, D., van Beek, W. & Green, M. A.Structure and interstitial iodide migration in hybrid perovskitemethylammonium lead iodide. Nat. Commun. 8, 15152 (2017).

40. Eames, C. et al. Ionic transport in hybrid lead iodide perovskite solar cells.Nat. Commun. 6, 7497 (2015).

41. Birkhold, S. T. et al. Direct observation and quantitative analysis of mobilefrenkel defects in metal halide perovskites using scanning Kelvin probemicroscopy. J. Phys. Chem. C 122, 12633–12639 (2018).

42. Li, W. et al. Control of charge recombination in perovskites by oxidation stateof halide vacancy. J. Am. Chem. Soc. 140, 15753–15763 (2018).

43. Li, C. et al. Iodine migration and its effect on hysteresis in perovskite solarcells. Adv. Mater. 28, 2446–2454 (2016).

44. Yuan, Y. et al. Electric-field-driven reversible conversion betweenmethylammonium lead triiodide perovskites and lead iodide at elevatedtemperatures. Adv. Energy Mater. 6, 1501803 (2016).

45. Zhang, Y. et al. Reversible structural swell–shrink and recoverable opticalproperties in hybrid inorganic–organic perovskite. ACS Nano 10, 7031–7038(2016).

46. Wang, Q. et al. Qualifying composition dependent p and n self-doping inCH3NH3PbI3. Appl. Phys. Lett. 105, 163508 (2014).

AcknowledgementsWe acknowledge funding from MINECO of Spain under Project MAT2016-76892-C3-1-R. A.G. would like to thank the Spanish Ministerio de Economía y Competitividad for aRamón y Cajal Fellowship (RYC-2014-16809). C.L and S.H. gratefully acknowledge thefinancial support by the Bavarian State Ministry of Science, Research, and the Arts for theCollaborative Research Network “Solar Technologies go Hybrid” as well as the GermanResearch Foundation (DFG). Part of this research was undertaken on the SAXS/WAXSbeamline at the Australian Synchrotron, Victoria, Australia. Furthermore, we are verythankful to Shannon Yee, Bernd Kopera, and Marcus Retsch for their dedicated supportwith the Seebeck coefficient measurements. C.L. thanks Xudong Cao and Yu Zhong forthe GIWAXS, XPS, and XRD characterization, Dr. Thomas Unger, Dr. Fabian Panzer,and Dr. Julian Kahle for optoelectronic characterization and discussion.

Author contributionsC.L. and A.G. conducted the experiments and coordinated the experiments. J.B. proposedthe model and prepared the first draft of the manuscript. S.H. and J.B. supervised thework. All authors co-wrote, discussed, and commented on the paper.

Additional informationSupplementary Information accompanies this paper at https://doi.org/10.1038/s41467-018-07571-6.

Competing interests: The authors declare no competing interests.

Reprints and permission information is available online at http://npg.nature.com/reprintsandpermissions/

Publisher’s note: Springer Nature remains neutral with regard to jurisdictional claims inpublished maps and institutional affiliations.

Open Access This article is licensed under a Creative CommonsAttribution 4.0 International License, which permits use, sharing,

adaptation, distribution and reproduction in any medium or format, as long as you giveappropriate credit to the original author(s) and the source, provide a link to the CreativeCommons license, and indicate if changes were made. The images or other third partymaterial in this article are included in the article’s Creative Commons license, unlessindicated otherwise in a credit line to the material. If material is not included in thearticle’s Creative Commons license and your intended use is not permitted by statutoryregulation or exceeds the permitted use, you will need to obtain permission directly fromthe copyright holder. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/.

© The Author(s) 2018

ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/s41467-018-07571-6

8 NATURE COMMUNICATIONS | (2018) 9:5113 | DOI: 10.1038/s41467-018-07571-6 | www.nature.com/naturecommunications