Electroluminescence transients and correlation with steady-state solar output in solution-prepared CH₃NH₃PbI₃ perovskite solar cells using different contact materials

Perovskite solar cells (PSC) are at the pinnacle of the efforts directed to the development of third generation photovoltaics, highly regarded as top candidates for near term massive impact on renewable energy production [1]. Despite the fast progress towards highly efficient and durable devices achieved by the scientific community, the complexity of PSC demands a diverse set of characterization methods and models to disentangle microscopic phenomena [2]. It is expected that a greater knowledge of the underlying mechanisms will pave the way to the development of new preparation methods and materials for production-ready devices [3].

A widespread tool in the characterization of solar cells in scientific laboratories as well as photovoltaic modules in industrial production is the electroluminescence (EL) imaging technique [4–6]. Using a simple setup composed by a CCD camera and a power source, EL imaging records the light emitted by the photovoltaic device in the dark upon biasing and current injection [6]. Owing to a correspondence between dark and illuminated characteristics, the features revealed in EL images are directly linked to the device under solar illumination [5], rendering EL imaging as a reliable and fast characterization technique for solar devices in general. In PSC, EL constitutes a valuable technique, already proven successful for the characterization of layer homogeneity and localization of degradation spots [7, 8]. However, the general use of EL in PSC requires careful inspection, since many PSC solar cell architectures show a time-dependent response, most widely revealed by the well-known rate-dependent hysteresis observed in current–voltage characterization [9]. The timescale of transient behavior in PSC spans from the microsecond to the hour range, with several different microscopic phenomena ruling each transient phase, from fast charge injection (in the order of ns), migration of fast ions (ms), interface charge/discharge (up to 10 s), and slow ion migration (1000 s) [10, 11]. Although screening EL as a fast tool, such long lasting transients may offer a complementary method to the characterization under solar operation. In previous investigations, EL transients were recorded in lead-halide PSC, showing a monotonous increase towards a stationary value in the minute scale [12], where others observed a non-monotonous evolution showing a maximum at timescales of up to 25 min, followed by a slower decrease [13]. In a recent paper [14], Wong et al employed EL transient characterization across solar cells prepared with different electron contacts, finding important differences in the different EL decay rates depending on the chosen electron contact materials. This allowed to estimate the amount of non-radiative recombination at the defect energy levels at the interfaces of each device introduced by each material. Such analyses can be expanded when incorporating other variables, for instance the relative EL intensity difference across samples, the link between transient EL and steady-state solar output characteristics, and the temperature dependence of EL.

In this contribution, we report the systematic EL transient characterization of methylammonium lead iodide (CH₃NH₃PbI₃ or MAPI for short) solar cells, focusing on the characteristic times, the EL intensity relative to the injected current, and the relationship between the EL characteristics and the open-circuit voltage under solar operation. We analyze in detail different shapes of the EL transients under several polarization conditions in solar cells using different electron as well as hole contacts, revealing the impact of interfacial recombination introduced by each contact material, and finding a correlation between solar cell output and EL transients across cells. Moreover, temperature dependent EL transients show a non-monotonous behavior which possibly reflects the occurrence of a change in crystalline phase near room temperature. The paper structure is as follows: section 2 describes the experimental methods employed during the preparation and characterization of the samples, detailing the procedure followed during EL recording. Section 3 presents the results, showing first a typical EL image time-sequence with the resulting EL transient curves defining characteristic transient parameters, outlining the possible origin of the observed behavior. The characteristic parameters are later linked to the solar output parameters and compared across devices with different contact materials. The last part of section 3 shows temperature dependent experiments, where the EL transients reveal a maximum that is associated to structural transitions.

2.1. Solar cell preparation

2.2. I–V characterization

2.3. EL imaging and processing

Solar cells with different performances were fabricated. Table 1 shows the forward– and reverse–scan output parameters of four MAPI-based solar cells measured under AM1.5G simulated light at room temperature. We purposely observed two devices with the same layer construction (namely Spiro-a and Spiro-b in table 1), using SpiroMeOTAD as HTL, but with different efficiencies in order to analyze the EL characteristics. Most likely, the smaller efficiency of sample Spiro-b is related to an increased light absorption and reduced electronic properties of the Spiro layer depending on its oxidation state [13, 15].

Figure 1 shows a sequence of images of EL emission of a MAPI solar cell with Spiro-MeOTAD as the HTL, upon 1.1 V forward bias polarization. A decay in brightness is observed after the maximum brightness is achieved at 4 s (EL_MAX), slowly reaching a stationary EL image at ca. 20 s. The EL images reveal a certain degree of defect inhomogeneity within the entire device area. The luminescence appears brighter initially at the frontiers of the device where typical border defects are expected. It later shows a brighter zone at the top of the device at t_MAX that may be due to the radial distribution (which corresponds to the vertical direction in figure 1) of the material defects that are inherent from the spin coating process.

Zoom In Zoom Out Reset image size

By summing up all the pixel counts of each image, we obtain the EL curve as a function of time as shown in figure 2 (black line, left y -axis), shown together with the injected forward current (red line, right y -axis). The EL signal increases towards EL_MAX from a starting value of about 25% of EL_MAX, followed by a much slower decrease towards a stationary value around 60 s. In contrast, the initial value of the injected current starts at nearly 80% of the steady state value reached at 53 s, while the EL signal continues to decrease. The timescale in the minute-range needed to reach the stationary current seems compatible with surface polarization models that take into account the migration and release rates of ions from/towards interfaces [16]. Field-induced structural changes in the crystal lattice of MAPI may also play a role in the slow dynamics, although to a smaller degree [17]. Since the amplitude of the variation of the current is clearly smaller than the amplitude of the EL transient, we consider that the observed rise and fall of EL is only linked to a strong shift in the balance between radiative and non-radiative recombination with time, disregarding the transient of the injected current from the analysis.

Zoom In Zoom Out Reset image size

Similar EL transients were observed previously in MAPI light emitting diodes having identical layer structure as our samples, and were understood by visible degradation effects upon current injection [13]. In our case, however, no appreciable degradation effects are seen after the EL transients, which are reproducible and reversible, as shown below. Therefore, it is much more likely that a change in the time-dependent radiative recombination is taking place in our measurements. Under this assumption, the observed peak-shaped EL transients must be ruled fundamentally by two counteracting mechanisms: a shift in the balance between non-radiative and radiative recombination during the first stages of biasing, and the generation of new defects that act as recombination centers with biasing time. Such defect generation upon polarization has been observed e.g. in MAPI films, where the vanishing of photoluminescence upon biasing in a coplanar contact geometry corresponds to the appearance of non-radiative recombination centers, conditioned by ion motion [18, 19]. We assume that this process of defect generation is present during the whole experiment. Contrarily, the factor that introduces the non-monotonous behavior in the EL is in our opinion the change in the band diagram. Within the typical 'wiggly band' [20] model for a MAPI solar cell, it is the central part of the perovskite layer which first absorbs the drop of the applied potential. As a consequence, the band bendings at the interfaces remain mainly unchanged, offering significant barriers for carrier injection (see scheme 3 in [20]). The barriers thus favor non-radiative recombination at the interfaces, leaving a certain initial level of radiative recombination in the perovskite. With the ongoing slow ion movement to counteract the applied potential, the band bendings at the interfaces are gradually lowered, decreasing the barriers to carrier injection. This implies that a larger portion of the injected carriers are allowed to reach the interior of the MAPI layer. Depending on the concentration of active recombination centers in the perovskite, this shift in the regions of higher recombination is capable of enhancing radiative recombination, with the resulting EL increase observed in our experiments. The aforementioned generation of defects with time counteracts the former process, leading to a reduction of charges available for radiative recombination. If defect generation is slow enough, then a peak in the EL must appear, as seen in our experiments. Such interplay of the involved stages must of course depend strongly on the interface defect concentration. Recently, EL transients in MAPI light-emitting diodes were found to show monotonous or non-monotonous behavior depending on the thickness of SnO₂ electron contact layer, highlighting the strong influence of interface properties on the observed dynamics [21]. Moreover, there are further processes that are believed to contribute to the observed EL decay, such as structural changes occurring in MAPI layers during polarization [17], which include lattice distortions and changes in the interface electronic properties. We believe that further investigations including detailed experiments and modeling are required to allow a better view of the dominating mechanisms.

Figures 3(a) and (b) shows different EL transients recorded at different bias voltages in two samples, Spiro-a and Spiro-b, prepared with the layer stack m-TiO₂/MAPI/Spiro-OMeTAD, where full lines correspond to measurements with increasing voltage, starting from the lowest voltage (1.1 V) to the highest voltage (1.3 V), while dashed lines correspond to biasing voltages in decreasing direction (with the same voltage values as in the increasing direction). Between each measurement, 4 min rest with no polarization were allowed for the cells to reach equilibrium. Figures 3(a) and (b) show that higher bias voltages deliver higher EL_MAX as well as higher stationary values, due to the higher injected current (figure S4 in the SI shows the relation between EL_MAX and applied voltage). After the bias ramp-up, the cells do not seem to undergo significant degradation, as the transients recorded with decreasing bias follow very closely the original transients (dashed lines).

Zoom In Zoom Out Reset image size

The applied bias not only affects the total EL value but it also shortens the transients, as shown in figure 4 where three characteristic times are plotted versus the applied voltage for the cells from figure 3. The times t_RISE50% and t_DEC50% correspond to the time that it takes the EL signal to rise and decay to 50% of its maximum value (EL_MAX), respectively. In addition, t_MAX is the transient peak time at EL_MAX. Since all three characteristic times tend to decrease with voltage, we conclude that biasing accelerates the involved fundamental processes mentioned above.

Zoom In Zoom Out Reset image size

The behavior observed so far is also found in cells with different contact materials on the electron and hole transport layers (ETL and HTL, respectively). Figures 5(a) and (b) shows the EL transients for cells with two different contacts: cell (a) is prepared with CuPc as the HTL and cell (b) contains a planar (or dense) TiO₂ as the ETL. The overall EL trend observed for the previous samples also holds here: the EL reaches a peak in around 10 s to later decay towards a stationary intensity within 30–60 s. The sample containing CuPc as the HTL shows the slowest and weakest EL emission.

Zoom In Zoom Out Reset image size

A comparison between all four samples is shown in terms of t_MAX in figure 6, showing that for a given polarization voltage, the slowest response is obtained by the sample with CuPc. We also notice that the sample with planar p-TiO₂ ETL shows intermediate times at comparable bias values. This suggests a stronger incidence of the HTL rather than the ETL on the EL transients, despite the large difference in effective interface area of the mesoporous compared to the planar TiO₂ layer.

Zoom In Zoom Out Reset image size

Figure 7(a) shows the correlation between the value of EL_MAX and t_MAX for all cases. The highest EL_MAX values always correspond to the shorter t_MAX. A better quantitative comparison between the different cells is given in figure 7(b), where we define the quantum yield Q_e as the quotient between EL signal and current density J according to

$$\begin{align} \newcommand{\e}{{\rm e}} \displaystyle {{Q}_{e}}(t)=EL(t)/J(t)\nonumber \end{align} \tag{ 1 }$$

in units of counts/mA. In figure 7(b), the value for Q_e(t_MAX) is obtained from EL_MAX and the current density J_MAX at t_MAX. Here, we notice that the cells with higher Q_e correspond to the cells with higher open circuit voltages (V_OC) in table 1 as expected. High radiative recombination corresponds to low defect (i.e. non-radiative) recombination for a given number of injected carriers such that it necessarily results in increased EL and high V_OC. An identical trend is found for the stationary value of Q_e at each voltage (see below). A further correlation between the transient response and the solar cell performance is possible by recalling the results of t_MAX from figure 6. It can be noticed that cells with higher V_OC presents lower t_MAX, i.e. faster EL transients. This suggests that the dynamics of the EL response is linked to the amount of non-radiative recombination under stationary conditions, provided t_MAX relates to the time needed to overcome non-radiative interface recombination, which depends on the concentration of interface defects.

Zoom In Zoom Out Reset image size

A further comparison is possible by relating the stationary quantum yield Q_e(∞) obtained with the values of EL and J at the end of the registered transients, and the steady state values of V_OC. According to the electronic reciprocity principle for non-ideal solar cells, steady state quantum yield and open circuit voltage are linearly related by [20]

$$\begin{align} \newcommand{\e}{{\rm e}} \displaystyle {{V}_{OC}}={{V}_{OC,rad}}+{{n}_{rad}}{{V}_{t}}\ln ({{Q}_{e}}),\nonumber \end{align} \tag{ 2 }$$

where V_t is the thermal voltage, n_rad is the radiative ideality factor, and V_OC,rad  =  1.33 V is the open-circuit voltage in the radiative limit for MAPI [21]. This equation assumes that the ratio between radiative to total recombination is identical under dark and illuminated characteristics, with the result that the same radiation is emitted from the device at a given voltage bias, regardless if it is a dark bias or V_OC. Applied to the devices studied here, figure 8 shows that Q_e(∞) correlates logarithmically to the open circuit voltage V_OC. The straight lines in figure 8 correspond to the fits of the values obtained from the m-TiO₂ cells at bias voltages V  =  1.15 V (open symbols and dashed line) and at V  =  1.25 V (filled symbols and full line), yielding n_rad  =  2.7 and n_rad  =  2.1, respectively. The solar cell based on p-TiO₂ cannot be included in the same fits as the m-TiO₂ cells because n_rad would change for a MAPI layer that is grown onto a different substrate since its crystallinity and optoelectronic properties are expected to depend on the underlayer material characteristics [22, 23].

Zoom In Zoom Out Reset image size

A similar validation of the optoelectronic reciprocity theory in MAPI solar cells was reported previously under stationary conditions by correlation of spectral quantum efficiency and EL [21]. The occurrence of radiative ideality factors higher than unity reflects the incidence of non-idealities, e.g. a high incidence of non-radiative recombination or the occurrence of radiative transitions between band tail energy states [20]. In amorphous Silicon solar cells, n_rad  >  2 have been obtained and also shown to vary with bias level [24].

Since the timescale of the observed EL transients is compatible with ionic diffusion and interfacial trapping/detrapping dynamics, the EL transients are expected to show significant sensitivity to temperature changes. Figure 9 shows the obtained EL transients recorded at a bias voltage of 1.3 V between 17 °C and 55 °C for a m-TiO₂/MAPI/Spiro-OMeTAD solar cell. The temperatures are suggested by the color of each curve, from cold (blue) to hot (red). The maxima are indicated for clarity, showing that the EL first increases with increasing temperature to later decrease again above 33 °C. Since the current at each part of the EL transients increases monotonously with temperature (see figure S5), it is the ratio of radiative to total recombination which must be changing with temperature.

Zoom In Zoom Out Reset image size

The temperature dependence of EL can be better examined in terms of the quantum yields Q_e(t_MAX) and Q_e(∞) shown in figure 10(a), displaying maxima at 29 °C and 33 °C, respectively. Figure 10(b) also presents the t_MAX values as function of temperature with a peak at 33 °C. Although more experiments are needed to understand the observed behavior, we suggest that the observed maxima between 29 °C–33 °C could be related to a phase change that occurs in MAPI at 37 °C, where the crystal lattice changes from tetragonal to cubic [25] (other authors report the same phase change to occur between 42 °C and 57 °C, see [26, 27]). Around the peak seen in figure 10(b), t_MAX decreases with temperature, possibly owing faster ion dynamics and faster interface charge trapping/detrapping at higher temperatures. The phase change could introduce a retardation of the involved mechanisms, originating the peak in t_MAX.

Zoom In Zoom Out Reset image size

EL measurements of solution-prepared one-step CH₃NH₃PbI₃ (MAPI) PSCs were recorded as a function of time, detecting a peak EL which depends on bias voltage, contact layer materials and temperature. The peaks are found at times in the 1 s–10 s range, while the whole transients last up to 60 s, suggesting that the involved mechanisms are likely governed by ion dynamics, e.g. originated in interface defect recombination and bulk defect generation by ion-vacancy generation. The characteristics of the transients are related to the open circuit voltage V_OC under solar illumination, showing that not only the stationary value but also the peak EL correlates with V_OC. The time required to reach the maximum EL holds an inverse relation to V_OC, possibly meaning that faster ion dynamics translates into better performing cells at bias voltages around or above V_OC. First temperature-dependent EL transient experiments show a maximum EL around 30 °C, close to the change of crystalline phase of MAPI at 37 °C. Further experiments in samples using different contact layers are planned to investigate if the observed maximum is strictly related to the phase change in MAPI or a consequence of interface properties. A possible technological implication of the observed non-monotonic behavior of EL with time and temperature is that the conventional use of EL imaging as a fast tool for diagnosing solar cells and modules requires caution when using MAPI devices prepared with traditional solution-processing techniques.