The rising demand for low-power, on-device intelligence in edge computing is driving the development of neuromorphic systems that combine novel algorithms with emerging materials and device geometries. Memristive technologies, particularly those based on solution-processed mixed halide perovskites, offer promising routes toward energy-efficient, scalable alternatives to traditional CMOS architectures. In this presentation, we showcase two complementary efforts that advance the use of perovskite memristors in neuromorphic computation.

First, we introduce a large, image-based dataset comprising thousands of experimental current–voltage (I–V) curves from printable perovskite memristors. [1] A convolutional neural network (CNN) is trained on this dataset to recognize and classify seven distinct memristive switching behaviors. This machine learning (ML)-based approach not only automates the assessment of device quality—achieving up to 91% accuracy in binary switching performance classification—but also establishes a foundation for predictive modeling of optimal operation conditions.

Building on this, we address the challenge of hardware-efficient multiclass classification by proposing a novel Outcome-Driven One-vs-One (ODOvO) algorithm, implemented using optoelectronic perovskite memristors as synaptic elements.[2] The light modulation of synaptic weights, fed in our algorithm from experimental data, is a key enabling parameter that permits classification without modifying further applied electrical biases. By integrating the algorithmic advantages of One-vs-One and One-vs-Rest schemes, our method reduces synaptic resource requirements by at least 10× (only 196 synapses) and achieves competitive accuracy on benchmark datasets like MNIST—all while significantly lowering power and iteration costs through light-based modulation of synaptic weights. Consequently, our approach constitutes a feasible solution for neural networks where key priorities are the minimum energy consumption i.e., small iterations number, fast execution, and the low hardware requirements allowing experimental verification.

Together, these studies illustrate how the synergistic integration of device-level advances and algorithmic innovation can pave the way toward scalable, low-energy neuromorphic platforms. The approaches presented offer a path toward practical, experimentally verifiable AI systems that meet the demands of next-generation edge IoT intelligence.

Despite their relatively emerging degree of development, metal halide perovskite light-emitting diodes (PeLEDs) have reached outstanding brightness and radiative efficiency levels that roughly graze the maximum theoretical limits. Unfortunately, the complete understanding of their working principles and the photo-electrochemical mechanisms involved in the charge carrier injection/recombination dynamics are still a conundrum. Additionally, the strong ionic character of perovskites enables the migration of ions and the gradual formation of crystalline defects upon exposure to light and/or to an external electric field, which aggravates the complexity of these systems. In fact, these ionic processes are apparently coupled with those electrical involved in the generation of light and seem to be also connected with the widely reported limited long-term stability of the devices. Here, I will discuss the exploitation of a new methodology based on the combination of three frequency-domain modulated techniques, i.e. impedance spectroscopy (IS), voltage-modulated electroluminescence spectroscopy (VMELS) and current-modulated electroluminescence spectroscopy (CMELS), aimed at extracting values of characteristic thermodynamic constants and at reaching a full understanding of the PeLEDs technology. We propose a new theoretical model and an equivalent circuit that unifies these three techniques, which consider both the non-radiative and radiative contributions, as a powerful tool for the advanced characterization of any light-emitting device, being especially useful for the study of perovskite-based optoelectronic devices due to their inherent complexity. Particularly important is the deconvolution of the electrical, optical and ionic processes that are involved in the current-to-photon conversion, heat generation and/or degradation of the light-emitting material/device, as well as the elucidation of how all these phenomena mutually interact.

Hyperspectral spectral photoluminescence (HPL) imaging is a powerful, contactless, and spatially-resolved technique widely employed to characterize solar cells at various stages of their development and operation. However, this method is time-consuming due to the sequential acquisition across multiple wavelengths and the subsequent analysis of large, high-dimensional data cubes. We propose an advanced framework for HPL data analysis in perovskite solar cells. We first investigate the application of deep learning (DL) techniques to significantly reduce both acquisition and processing times, paving the way for faster and more scalable solar cell analyses, and then explore the correlation between electrical and optical characteristics of the cells.

We perform a DL-based local analysis of HPL data cubes to extract maps of Urbach energy (Eu), quasi-Fermi level splitting (QFLS) and bandgap energy (Eg). A multilayer perceptron (MLP) is trained on synthetic data generated using the generalized Planck’s law, combined with a logistic absorptance model, and simulated additive noise. The performances of the MLP are compared to non-linear least squares (NLLS) fitting applied to HPL spectra, following the approach introduced by Laot et al. [1].

Our results show that, while the MLP does not significantly increase the prediction error, it substantially reduces the computation time. The average absolute relative error (ARE) across more than 13 000 spectra - between the NLLS fitting results and the MLP prediction - is of 1.76% for Eu, 0.07% for Eg and 0.12% for QFLS. The total prediction time of for all three parameters across 13 000 voxels is under 1 second using MLP, compared to approximately 10 minutes using NLLS fitting.

We investigate how increasing the wavelength step size - thereby reducing the number of sampling points in the spectrum - impacts the prediction accuracy on the three variables of interest (Eg, Eu, QFLS). This approach enables a significant reduction in acquisition time, which not only accelerates the characterization process but also minimizes the degradation of perovskite materials that may occur during prolonged measurements. Furthermore, we apply transfer learning methods to adapt DL models trained on densely sampled spectra for use with sparsely sampled spectra, thereby reducing the training time required for the new models.

Using the theoretical framework proposed by Kirchartz et al. [2] we compute the local photocurrent, recombination current and open-circuit voltage, under the approximation that the quantum efficiency is equivalent to the previously calculated absorptance of the cells. We finally correlate the optical characteristics of the cells with their electrical parameters by analyzing a dataset of more than 100 p-i-n halide perovskite solar cells.

We explore how local parameter heterogeneity influences the global electrical figures of merit.

The extraction of photogenerated charge carriers and the creation of a photovoltage are essential functions of any solar cell. These processes do not occur instantaneously but have specific time constants, such as one associated with the increase of the externally measured open circuit voltage after a brief light pulse and another related to its decrease. In this presentation, I will introduce a method for analyzing transient photovoltage measurements at various bias light intensities by combining the rise and decay times of the photovoltage. I will demonstrate how to distinguish between recombination and extraction using transient photovoltage measurements, along with an analysis approach based on determining the eigenvalues of a 2 × 2 matrix. The model yields two time constants (the inverse eigenvalues), one for the voltage rise and one for its decay after the pulse. These time constants can be experimentally measured as a function of light intensity. By comparing the model with experimental data, I can derive a time constant for recombination and another for charge extraction, with the ratio of these time constants directly correlating with solar cell efficiency.¹ This general approach is applicable in situations where the Fermi level splitting within the solar cell absorber can be approximated by a single value, and its depth dependence can be ignored. In such cases, it can be applied not only to transient photovoltage but to any small signal optoelectronic technique in the time or frequency domain.²

Despite the rapid development of perovskite solar cells several challenges remain. A deeper understanding of the main losses, and how to mitigate them, is needed to make targeted improvements possible. In this talk, I will discuss our drift-diffusion modelling approach to shed new light on these fascinating solar cell materials.

Drift-diffusion techniques make it possible to connect and explain the fundamental generation, transport and extraction processes to macroscopic device performance. If done right, one can also do the opposite: By reverse-engineering current-voltage measurements on actual solar cells, one can identify the limiting factors. As an example, we show how current-voltage data and electroluminescence measurements help us identify the voltage losses in a series of co-evaporated FACsPbIBrCl perovskite solar cells with organic transport layers.

The physical processes underlying the impedance response of perovskite solar cells are not well understood. Typically, the low-frequency peak in such impedance spectra is attributed to ion dynamics, while the high-frequency peak is associated with electronic processes.

We introduce a new formula which enables us to directly derive the ion diffusion coefficient from the impedance response of perovskite solar cells. The validity of this formula is confirmed through extensive drift-diffusion simulations.

Upon demonstrating this, we determine the ion diffusion coefficients of a MAPI and a FAMAPI solar cells. The obtained diffusion coefficients are consistent with previously reported values from other characterization techniques. The advantage of this method is that it facilitates the precise, rapid, and straightforward determination of the ion diffusion coefficients.

Next, we introduce an experimental method for identification of the limiting aspect of perovskite solar cells under operating conditions in terms of recombination losses. We illuminate a bifacial cell from both sides separately with either red, blue or white light, each absorbed differently in the cell depending on the position in the device. Using the fill factor from the device characteristics for each case taking into account the direction of illumination we are able to accurately identify which part of the cell is limiting the performance. We show that this holds for many typical perovskite solar cells using drift-diffusion simulations. Finally, we issue a protocol to determine the dominant recombination channel under operating conditions in full device configuration.

Outdoor stability testing under natural sunlight provides the most relevant test of solar cell stability under operational conditions [1]. Understanding perovskite-based solar cells’ (PSCs’) recovery properties under natural diurnal light-dark cycling can point to methods to extend its lifetime [2, 3]. We studied the effect of climate conditions on perovskite solar cell lifetime, which showed that outdoor T80 is climate dependent with the ambient temperature being the dominant factor [4]. Based on this understanding, we designed an agro-photovoltaic system, where semi-transparent PSCs were coupled to a photobioreactor held at a constant temperature. Cultivation of photosynthetic microorganisms is a sustainable approach for producing feed, food and high-value compounds, and excess light-caused photosaturation and photoinhibition can be limited by light filtration by the semi-transparent PSCs. The synergy between photovoltaics – improved outdoor stability of PSCs - and photosynthesis – with improved yields with the proper light filtration - in this system will be presented [Gupta, R.K., M. B. Maung, V. Dubovsky, G. Ziskind, N. Kamenaya and I. Visoly-Fisher, in preparation].

Emerging photovoltaic technologies are commonly processed using solution deposition methods. Layers deposited in this way often suffer from non-uniformities in thickness and composition, resulting in locally varied cell performance. Such inhomogeneities and defects can be visualized by various imaging techniques, such as photo- and electroluminescence (PL/EL), dark and illuminated lock-in thermography, as well as optical imaging. Such non-uniformities can be challenging for upscaling emerging solar cells, as they result in performance losses and may lead to local hot-spots, which are the origin of degradation.

In this contribution, we first present EL images of carbon-based perovskite solar cells with a mesoporous layer stack which exhibit locally varying temporal evolutions. The scan-rate dependent current-voltage characteristics of perovskite solar cells and the temporal evolution of the EL signal are generally associated with the presence of mobile ions in the cell.[1] To analyse the transient EL images, we set up a device model in the drift-diffusion simulation software Setfos which quantitatively reproduces a set of steady-state and transient measurements. By employing this model, we are able to attribute the inhomogeneities in EL intensity to spatially varying ion densities. We further show how the mobile ion density influences the reverse bias breakdown behaviour in perovskite solar cells due to the strongly varying potential at layer interfaces, which facilitates tunnelling current.[2] Reverse bias conditions are imposed on shaded sub cells in modules or can be induced by current mismatch situations in monolithically stacked tandem devices. Non-uniformities caused by a spatially varying ion density and consequently reverse bias breakdown voltages result in strongly varying (reverse) bias potentials, inducing current (and temperature) hot-spots.[2]

The effect of reverse bias stressing on perovskite solar cells is further assessed by a combination of characterization techniques. To this end, cells are stressed below their breakdown voltage while the EL or PL signal is recorded. During regular intermittent measurements, current-voltage scans, impedance measurements, as well as (forward) electroluminescence images are taken to assess the underlying degradation mechanisms induced by reverse bias conditions.

Understanding the physics and origins of degradation mechanisms in solar cells is a challenging task. As a result, research often concentrates on a few model systems, relying on extensive and often costly characterization to track how material properties evolve over time.

This approach is particularly limiting for technologies like organic and perovskite solar cells, which can use a wide variety of materials. To unlock their full potential, accelerated characterization methods are needed. Without these, building quantitative structure-property relationships in the hope of designing bespoke materials for targeted applications (such as indoor, semi-transparent, and agrivoltaics) will fall short.

In this talk, I will present how modeling and machine learning (ML) can be coupled with high-throughput data from accelerated aging experiments to address this challenge. I will introduce optimPV, a fully open-source framework that integrates multiple physical modeling tools (e.g., PDE solvers, optical simulations, drift-diffusion modeling) with ML-based optimization methods (including Bayesian optimization, Bayesian inference, and genetic algorithms).

I will show how optimPV enables the identification of the dominant degradation process and the extraction of key material parameters, such as charge carrier mobilities and recombination rates. The effectiveness of this framework will be illustrated through a large-scale degradation study on organic solar cells, featuring 25 donor-acceptor blends processed under varied conditions. This analysis revealed key degradation trends and identified problematic material combinations to avoid in future formulations.

Lastly, I will discuss how the framework can be extended to other systems and case studies, incorporating a range of material types (organic and perovskite), physical models (PDE-based and drift-diffusion), and experimental techniques, including: (i) transient absorption spectroscopy, (ii) transient photoluminescence and microwave conductivity, and (iii) light-intensity-dependent current–voltage measurements.

Perovskite solar cells (PSCs) are promising candidates to reach the market to complement the current offer of photovoltaic cells although for such a thing they still must overcome some challenges such as long-term stability. The hole transporting layer (HTL) is a crucial component in n-i-p PSC, since it must favor an adequate movement of charges and protect the perovskite layer from environmental conditions. In this sense, the commonly used HTL, spiro-OMeTAD, does not provide PSCs with sufficient stability and is too expensive. Cheaper molecular materials such as metallophthalocyanines are proving to be a good alternative, as they provide greater stability.[1]

In this communication, we will present novel ZnPcs and CuPcs monomers [2] (see as examples Figure 1), among others, as efficient, stable, and low cost HTMs in PSCs. The MPcs are substituted with functional groups that possess a very good solubility in a wide range of organic solvents, adequate HOMO LUMO levels and their photovoltaics performance as high stable solution processing in a wide range of perovskite solar cell devices.

Ensuring high performance and long-term stability of perovskite solar cells (PSCs) is essential for their transition to large-scale industrial deployment. However, device-level quality of PSCs is often limited by microscopic inhomogeneities within the active perovskite layer - such as local variations in composition,  structure, and optoelectronic response - which can adversely affect both device performance and stability.

Identifying and understanding these local defects is critical for improving device reliability. Given the complex composition and polycrystalline nature of perovskite materials, imaging techniques with high spatial resolutions are essential for identifying and characterizing such local variations [1,2].

In this talk, I will discuss how advanced imaging and spectroscopic techniques - ranging from ultraviolet photoemission to infrared scanning probe microscopies - can shed light on the microscopic origins of performance losses and degradation in PSCs. These insights provide a deeper understanding of the detrimental roles of microscopic inhomogeneities in shaping macroscopic device behavior of PSCs, informing strategies to improve their performance and stability.

Since the emergence of the Perovskite Solar Cells (PSC) in 2009[1], significant progresses have been made, with high-efficiency devices reaching up to an outstanding 27% [2] in just 16 years. However, despite this quick development in a relatively short time frame, one of the main limiting factors hindering the industrialization of these photovoltaic technologies is their limited operational stability.

To mitigate the fast degradation of these photovoltaic devices, several strategies has been explored, with encapsulation emerging as one of the most effective[3]. These methods play a vital role in delaying the rapid degradation of the perovskite photovoltaic devices protecting them from environmental factors as heat, moisture and oxygen. However, ensuring the long-term reliability of these devices not only require of stable materials and hermetic encapsulations, but also robust testing protocols to evaluate the inner stability of the device and the effectiveness of the encapsulation protocols of the devices.

In this context, standardized stressing tests -performed both in outdoor and indoor conditions- are key[4,5]. Outdoor tests track devices under real operation conditions, while during indoor tests the degradation processes are accelerated by exposing devices to controlled stressing factors, such as elevated temperatures, continuous illumination or light and dark cycles[5]. However, reliable stability data remains scarce in the literature. This is, in part due to the lack of well-equipped testing facilities with the infrastructure needed to perform these tests.

One notable location for outdoor tests is Valencia, which offers extreme climatic conditions- high humidity, high temperatures and a large number of sunny days, particularly during the summer time- making it ideal for outdoor stress testing. Moreover, the University of Valencia and in particular the MOED group has become an established group in this regard, having experience in vacuum-processed perovskite solar cells, in encapsulation of PSC, and in conducting indoor and outdoor tests.

Beyond infrastructure, several technical issues must be addressed to ensure reliable stability data. These include the right measurement of the light, the correct calculation of the active area of the devices, and the gathering of representative and consistent data from each pixel or device over time. Post-treatment of the data is also essential, not only to condense the information but also to reveal correlations between the stability data with variables that would otherwise remain hidden.

This presentation will highlight the results and the insights gained over the past year by the MOED group on encapsulation methods, stability testing under various conditions, and advanced data post-processing methods for analysing stability data.

Photoluminescence imaging techniques are commonly used to investigate the optoelectronic and transport properties of halide perovskite absorbers and devices [1]. However, obtaining precise spatial maps of key physical parameters—such as carrier lifetime, quasi-Fermi level splitting (QFLS), and bandgap energy (Eg)—requires a high local signal-to-noise ratio (SNR), typically achieved through long acquisition times. Prolonged measurements can compromise data integrity due to changes occurring during acquisition. This limitation is particularly critical in operando experiments conducted under elevated humidity or temperature, where shortened acquisition times are essential to track dynamic changes in the material’s properties in real time.
To address this challenge, we previously demonstrated a denoising strategy based on Total Variation Regularization (TVR) [2], enabling the extraction of high-quality lifetime images from quickly acquired, noisy time-resolved fluorescence imaging (TR-FLIM) datasets [3]. In this work, we present a significant advancement by applying a tailored version of the Noise2Noise (N2N) algorithm [4] to denoise multidimensional datasets. The main advantage of N2N lies in its unsupervised learning framework, which makes it particularly well-suited to complex, real-world image denoising situations compared to TVR.
Using this approach, we performed in-situ TR-FLIM data acquisitions on halide perovskite thin films—specifically triple-cation compositions—under controlled humidity conditions (relative humidity XX%). This allowed, for the first time, micrometer-scale tracking of local carrier lifetime degradation during environmental exposure. The reduced acquisition time made possible by N2N denoising enabled the resolution of spatial heterogeneity in perovskite degradation.
In conclusion, coupling advanced PL imaging techniques with unsupervised denoising approaches like N2N opens new avenues for accelerated, high-resolution characterization of halide perovskites. This methodology not only deepens our understanding of material stability under realistic conditions but also holds broader potential for studying other beam-sensitive materials and for developing fast, reliable imaging workflows for operando and accelerated experiments. 

Hybrid perovskite absorbers are poised to become a crucial part of next-generation photovoltaics (PVs) for mitigating the impact of energy production on the climate. In one projected application, two-terminal silicon-perovskite tandem PVs, a high-bandgap perovskite absorber is coated on top of a silicon absorber (with intermediate layers) - achieving power conversion efficiencies superior to the Shockly-Queisser limit of single-junction silicon PVs. One of the major advantages of the two-terminal tandem architecture is that the perovskite deposition step can be integrated conveniently into the high-throughout production of mono-crystalline silicon modules that currently dominate the global market. In practice, these modules are assembled from individual silicon wafers in batch-to-batch operation. Consequently, for seamless integration, the perovskite absorber must be deposited onto a surface of silicon-wafer size, which are typically from about 20 cm to 30 cm wide. There are two principal routes of perovskite absorber deposition: Solvent-free vapor deposition and coating or printing from solution. While the first route offers advantages in homogeneity and compatibility with textured silicon substrates, the second route is more cost-effective and less operationally complex. However, solution processing of perovskite thin films is challenging to control on the targeted substrate scales for high-throughput coating techniques such as slot-die coating and spray coating (spin coating is most likely not an option for commercial PVs due to excessive material waste and throughput limitations). Therefore, there is a great merit in detailed investigation of how to deposit perovskite thin films homogeneously on substrates of typical silicon-wafer sizes.

For perovskite deposition from solution, not only the homogeneity of the coated wet films, but also the homogeneity of the drying process comes into play. This is because perovskite crystallization is highly drying-rate dependent. Typical perovskite precursor solutions require very high drying rates at moderate temperatures, which is why slot-nozzles purged by pressurized air or nitrogen, so-called “air knives” are often employed for drying. It is common knowledge that the mass transport under these air knives is highly inhomogeneous. That is to say that a liquid film under an air knife will dry much faster directly under the nozzle opening than on the edges of the film farther away from the nozzle. To counter-balance this inhomogeneity, technologists typically move the wet solution film under the air knife linearly. Still, the part of the substrate that experiences the air stream first, will dry differently than the part of the film situated at the center of the substrate. In turn, the part of the film that passes under the air knife last will have different drying dynamics from the other two parts mentioned before. Conclusively, it is challenging to homogenize drying dynamics of thin films moving under an air knife. In this talk, we investigate how this problem can be addressed by a systematic parameter variation during drying - the available paramters being the air flow velocity, the air knife distance, slot-width and mounting angle as well as the movement speed of the air knife. The investigation is implemented by a) developing a suitable and computationally efficient drying model of a high number of positions on the substrate b) defining how a homogeneous drying process is characterized and c) optimizing the available parameter space as a function of time for obtaining homogeneous drying. The presented results showcase the high potential and prediction power of employing drying models to homogenize perovskite drying. Further, the dependence of the controllability of drying on the homogeneity of the coated wet film as well as the time and film composition at crystallization onset is demonstrated.

The experimental research carried out by the DELFO group can be summarised in three main areas. The first area focuses on the study of the degradation mechanisms and performance optimization of perovskite-based optoelectronic devices under various conditions. First, we investigate the impact of ultraviolet (UV) light on perovskite solar cells with the structure ITO/PTAA/CsMAFAPbIBr/PCBM/BCP/Ag. Devices were exposed to continuous 385 nm UV light at intensities ranging from 1.5 to 30 mW/cm² in an inert N₂ atmosphere. Periodic J–V measurements reveal that UV exposure primarily affects the short-circuit current (JSC), while the open-circuit voltage (VOC) remains stable. The efficiency loss is attributed to reduced charge extraction, with a logarithmic trend observed in T80 versus UV power density. High-speed J–V scans suggest that JSC degradation is mainly driven by UV-induced ionic migration. Second, we compare inverted (p-i-n) perovskite solar cells using NiOX as the hole transport layer, with and without a self-assembled monolayer (Me-PACz). SAM-modified devices show improved VOC and efficiency due to reduced interfacial recombination, as confirmed by temperature-dependent electrical characterization. However, long-term outdoor testing reveals that SAM-based minimodules suffer from greater JSC degradation, likely due to SAM deterioration. These findings contribute to a deeper understanding of stability and performance in perovskite devices.Finally, we present the development of MAPbI₃-based memristors, focusing on the influence of buffer layers, perovskite thickness, and electrode materials. Optimized devices demonstrate excellent resistive switching behavior, with high retention (>2×10⁵ s), endurance (30,000 cycles), and ON/OFF ratios (~10⁶), making them promising candidates for non-volatile memory applications.

The study of perovskite solar cells degradation is a complex issue due to the multitude of phenomena that can contribute to it. Recently, we have obtained new insights using a model that combines several electronic and ionic processes, that can produce capacitive and inductive response in different circumstances.¹ These models are very successful to describe huge memory effects and hysteresis in perovskite memristors, by the combination of different techniques: current-voltage scan, time transients, and impedance spectroscopy.^2-4 Here we show the changes of impedance spectroscopy and time transients as a diagnosis of hysteresis, time constants and evolution of degradation in the perovskite solar cells. Normally the degradation can produce two main impacts, decrease of charge collection (lowering photocurrent) or increase of recombination (lowering photovoltage). We need to find dynamical signatures of the phenomena causing these effects to discover the physical reasons for the devaluated performance.