Enhanced control of self-doping in halide perovskites for improved thermoelectric performance

Metal halide perovskites have emerged as promising photovoltaic materials, but, despite ultralow thermal conductivity, progress on developing them for thermoelectrics has been limited. Here, we report the thermoelectric properties of all-inorganic tin based perovskites with enhanced air stability. Fine tuning the thermoelectric properties of the films is achieved by self-doping through the oxidation of tin (ΙΙ) to tin (ΙV) in a thin surface-layer that transfers charge to the bulk. This separates the doping defects from the transport region, enabling enhanced electrical conductivity. We show that this arises due to a chlorine-rich surface layer that acts simultaneously as the source of free charges and a sacrificial layer protecting the bulk from oxidation. Moreover, we achieve a figure-of-merit (ZT) of 0.14 ± 0.01 when chlorine-doping and degree of the oxidation are optimised in tandem.

Optical images Supplementary Figure 1 (A) to (E), show that all the films formed by vacuum deposition present a "mirror-like" surface. Films formed from the co-evaporation method directly show a perovskite black phase, whereas the sequentially deposited and SLS films are red-brown (due to the coexistence of SnI2 and CsI layers in the film) but convert to a black phase upon baking. The B--CsSnI3 undergoes a significant phase transition to the Y-phase (yellow phase) when exposed to air. The Yphase will continuously change to a dark-green phase (Cs2SnI6) where Sn 2+ is totally oxidised to Sn 4+ .
The Cs2SnI6 films are semi-transparent and stable in air.

CsSnI3-xClx thin films.
We performed X-ray diffraction analysis of SLS 5% SnCl2 mixed halide CsSnI3-xClx thin films, which showed (Supplementary Figure 2 (a)) peaks at 25.02° and 29.15°, corresponding to (220) and (202) planes respectively of the orthorhombic B--CsSnI3 crystal structure. In fact, the mixed halide films processed by SLS show a similar crystal structure to undoped SLS CsSnI3 films (XRD presented in the main text). To investigate the difference between the pure CsSnI3 and mixed CsSnI3-xClx, we performed a slow X-ray diffraction scan from 20° to 30° at a rate of 1°/minute (Supplementary Figure   2 (b)). A peak at 23.00° is clearly observed, which is different to the typical peak at 22.80° of the CsSnCl3 perovskite (011) plane and is also shifted slightly with respect to the weak CsSnI3 perovskite (213) plane at 22.94°. 6 Supplementary Figure 3 (a-b) shows that the main change to the spectrum after exposure to air for 30 minutes is the (103) at 32.9°. The other peaks are reasonably unaffected.

Quantitative analysis of Cl states in mixed halide perovskite by X-ray photoelectron spectroscopy (XPS).
To investigate the Cl bonding environment in the films, we performed X-ray photoelectron spectroscopy (XPS) of Cl 2p in 0.5%, 1%, 3% and 5% SnCl2 mixed halide CsSnI3-xClx perovskite films.
The percentage we use refers to the mass of SnCl2 relative to SnI2 in our thin films before the baking step. The final atomic % of Cl in the film will be much lower. As shown in Supplementary Figure

Supplementary Note 4 UV-vis absorption spectra for air stability analysis.
We quantified the film stability in ambient air by following the quenching of the optical absorbance.
The sequentially deposited films presented a poor stability where the intensity of a degradation peak at 680 nm gradually increase after 100 minutes air exposure. For the co-evaporated films, there was no clear peak at 680nm after 500 minutes air exposure though the absorbance at 420 nm reduced to 32% of its original value. In the SLS films, the degradation peak at 680 nm presented from ~380 minutes, and, after 500 minutes, the absorbance at 420 nm had reduced to 41% of its original value. When SnCl2 was introduced to form the mixed halide CsSnI3-xClx, the large improvement in film stability evident from UV-vis absorption spectra that show no emergence of the degradation peak at 680 nm even after 500 minutes (Supplementary Figure 8 g and h). In Supplementary Figure 8(i), sequentially deposited films (red) show a poor stability with 60% reduction in absorption at 420 nm after 100 minutes air exposure. Mixed halide CsSnI3-xClx perovskite films with 5% SnCl2 show the best stability among all with just 3% quenching of the 420 nm peak after 100 minutes air exposure.

Supplementary Note 5 Sn oxidation states in CsSnI3-xClx perovskite films.
For the pristine CsSnI3 perovskite films, the binding energy of Sn 3d5/2 was observed at 485.8 eV. However
The reported Sn 3d core binding energy has shifts between Sn 0 , Sn 2+ and Sn 4+ oxidation states in the range 1 -1.8 eV (Sn 0 to Sn 2+ ) and 1.8 -2.5 eV (Sn 0 to Sn 4+ ), respectively. 8,9 The small shift of 0.4 eV from Sn 2+ to Sn 4+ makes quantitative analysis of oxidation in our films challenging, AES is employed to distinguish Sn 2+ and Sn 4+ states due to substantial spectral shifts. Supplementary Table 1 shows the AES peaks of Sn metal. 10,11 Upon oxidation of metallic Sn samples, Kövér found Sn MNN kinetic energy shift of ~3.8 eV, 5.8 eV for SnO and SnO2. 9 Lee observed small shifts ~2 eV for 0.4 monolayer (ML) and 3.4 ML Sn oxides attributed to the formation of SnO rather than SnO2. 12 Another indicator of Sn chemical state, the modified Auger parameter, is shown in Supplementary of Sn 0 (915 eV) 9,12 and Sn 4+ (911.2 eV) 8 it is clear that our films contain oxidised forms of Sn in a mix of Sn 2+ and Sn 4+ states. As the depth goes down to 10 nm, there is a reduction in the Sn 4+ character of the perovskite until something resembling pure Sn 2+ is reached. This combined with the spectral fitting in the main paper (Figure 4a,b) we can conclude that on the timescale of our thermoelectric measurements, the oxidation process only proceeds in the outer surface layer of our films, leaving the bulk in a pristine state.

Wiedemann-Franz law in CsSnI3-xClx perovskite semiconductor thin films and extraction of the Lorenz number.
The Thermal conductivity can therefore be deviated into two parts: lattice and electron contribution, as following: We can extract L and lattice thermal conductivity, , by plotting thermal conductivity versus electrical conductivity for a number of experimental measurements. To do this, after the first measurement of the samples, it was exposed to air for 3 minutes before re-measuring the electrical and thermal conductivity (σ3mins and κ3mins). Then the same sample was exposed to air for an additional 3 minutes exposure, before measuring the electrical and thermal conductivity again (σ6mins and κ6mins).
Thus, we obtained several values (σtime and κtime) and plot σtime vs. κtime. From the equation, = + , we see that the slope gives Lorentz number times temperature (L*T) and the intercept is lattice thermal conductivity.
In this situation, we must approximate that the lattice thermal conductivity does not change as the dopants are introduced. In fact, dopants are defects, and do influence the lattice structure and disorder. Consequently, quantification of the Lorenz number is challenging in many thermoelectric materials. In our case, CsSnI3 perovskites support an effective way to extract L because of the selfdoping process, which does not introduce extrinsic dopants. The Sn 4+ sites that act as self-dopants are shown in the main manuscript to be located at the top surface of the film only (in a layer < 10 nm thick), but provide free charges to the bulk (~300 nm thick). The lattice thermal conductivity in > 95 % of the film thickness is therefore unaffected by the doping process. As shown in Supplementary Figure 11

Thermoelectric property measurement details.
Electrical conductivity measurement is based on the van der Pauw method with four needlelike contact pads at the four corners of the material under test, as shown in Supplementary Figure 16 (a). Electrical conductivity can be calculated by the following equation: where is film thickness, R is resistance of the films with vertical and horizon directions (as pictured).
The thin-film heater and thermometer are located along the centre-line of the membrane, and the temperature gradient can be adjusted with the current in heater. The cold side temperature is taken as the chip temperature (measured underneath in the base holder), allowing the thermovoltage to be obtained and a Seebeck coefficient calculated. In-plane thermal conductivity is be measured by a transient 3-ω method. An alternating current ( ) = 0 × cos( ) is used to heat the stripe. Joule heating occurs at frequency 2ω due to the heating power, 2 = 0 2 R(1 + cos(2ωt))/2. As a result, the temperature oscillates at 2ω, and the temperature dependent electrical resistance also has a component at 2ω (R2ω). The temperature change of the heating stripe is measured with a lock-in amplifier which can extract the third harmonic of the voltage drop across the heating stripe (V3ω = 2 ). The thirdharmonic voltage captures the second-harmonic temperature rise (ΔT2ω) in the heater, which is a function of the thermal properties of the underlying materials. Considering in-plane thermal conductivity, by solving the two dimensional partial differential heat equation across the membrane with the given boundary conditions, the general expression for the amplitude of the 3ω oscillation V3ω is expressed as 16 : where is the temperature coefficient of resistance, is the total thickness of sample plus 100nm of Si3N4 membrane and 33 nm of Al2O3, is the length of the heater, is the width of the membrane, is the thermal relaxation time, = / , where is the mass density and is the specific heat capacity.
the ω becomes negligible and V3ω becomes constant. Equation 3 can be written as: (4) To get the thermal conductivity of the sample, the contribution of membrane thermal conductance should be removed. The sample thermal conductivity is given by Where is the membrane thermal conductivity, is the membrane thickness, is the sample thickness. The errors of electrical and thermal conductivity are dominated by the measurement of film thickness. The errors on the Seebeck coefficient come from the fitting error of thermal volatge vs.
temperature gradient data. Because the film thicknesses used in electrical and thermal conductivity are identical, they cancel out in the calculation of ZT, limiting the error on the final value.