Decoding non-fullerene acceptors for organic solar cells: a theoretical structure–property–performance perspective Authors: Habib Ul Murtaza, Nimra Sultan, Muhammad Ramzan Saeed Ashraf Janjua Published: 2026 ABSTRACT Organic solar cells (OSCs) have become promising next-generation photovoltaic technologies because of their flexibility, solution processability, and light weight. However, the use of traditional fullerene-based acceptors has been hampered by poor absorption, poor tunability, and energy losses that require the fabrication of sophisticated materials like non-fullerene acceptors (NFAs). The purpose of this review is to provide multi-faceted theoretical insight into the structure–property–performance relationship in NFAs with emphasis on how electronic structure, optical behavior, charge transport and eventual device efficiency can be impacted under the influence of molecular design. It is analyzed systematically through theoretical and computational methods such as density functional theory (DFT), time-dependent DFT (TD-DFT), molecular dynamics simulations and machine learning (ML) methods. Such techniques are used to measure important descriptors including HOMO–LUMO energy levels, bandgap, reorganization energy, dipole moment, exciton binding energy and molecular packing. The work shows that optoelectronic properties can be accurately tuned using rational molecular engineering via donor–acceptor architectures, end-group modification, core extension and side-chain optimization. State-of-the-art NFAs, especially Y-series acceptors, feature narrow bandgaps, strong near-infrared absorption, lower energy loss and efficient charge separation which result in much higher power conversion efficiencies on the order of 20%. Morphology and intermolecular interactions are found to be key factors that determine dynamics of charge transport and recombination. The combination of theoretical knowledge and experimental results generates a coherent model of predicting and optimizing NFA performance. The review indicates that future multiscale and data-driven modeling strategies will be central to enhancing the design of next-generation NFAs that will lead to highly efficient, stable and commercially viable organic solar cells. Introduction OSCs have gained significant research attention as a new photovoltaic technology because of their inherent properties such as mechanical flexibility, lightweight assembly, semi-transparency, and solution-based fabrication form have been recognized. Over the last ten years, ongoing advances in molecular design and engineering of devices have dramatically increased the power conversion efficiencies (PCEs) of OSCs. The active layer of these devices is usually a donor–acceptor bulk heterojunction with the acceptor material playing a decisive role in exciton dissociation, charge transport and recombination dynamics. As a result, it is crucial to create understandable structure–property–performance links to develop high-efficiency OSCs rationally.1 OSCs make use of photoactive material in the form of conjugated polymers or small molecules that absorb light and produce excitons (pairs of electron holes). Together with an acceptor material these excitons dissociate at the donor–acceptor interfaces to give free charges which are then transported to electrodes. OSCs are attractive due to solution processing, mechanical flexibility and possible low production expenses that are appealing to scalable photovoltaics on flexible surfaces and building-scale built-in PV.2 Despite these benefits, the performance of OSCs was historically limited with relative to inorganic solar cells because of low charge mobilities, unfavorable morphology and low-light absorption. Current studies are on molecular design, morphology engineering, and device engineering to increase the PCE and stability to operate OSCs in competition with other renewable technologies.3 Another significant breakthrough in OSC technology was the concept of the bulk heterojunction (BHJ) as shown in Fig. 1. Electron-donating and electron-accepting materials are closely mixed in BHJ-based OSCs to create a nanoscale interpenetrating structure that provides the highest possible area of contact of the donors and acceptors and optimizes exciton dissociation efficiency. This architecture was much better in charge generation and collection than the previous bilayer devices. BHJ OSCs performance is also very sensitive to nanoscale morphology, phase separation and molecular packing sensitive and therefore, material selection and processing conditions play a vital role in the process of obtaining high power conversion efficiencies.4 The potential of OSCs to be utilized in low-cost, large-area and flexible solar energy is one of the most important benefits of this technology. The compatible fabrication methods are spin coating, slot-die coating, inkjet printing and roll-to-roll processing that can be used in scalable industrial manufacturing and is enabled by solution processability. Also, OSCs are mechanically flexible, semi-transparent and lightweight, and can be used to create new applications in wearable electronics, building-integrated photovoltaics, and portable power supplies. These distinguishing characteristics make OSCs an appealing alternative to use in cases where traditional silicon-based photovoltaics are not as appropriate.5 Although OPVs offer numerous benefits, they are not the perfect solution for all inorganic photovoltaic technologies. There are some basic problems that still hinder their commercialization. Organic semiconductors have relatively low dielectric constants, which leads to large exciton binding energies which generally require donor–acceptor heterojunctions for efficient charge separation, as opposed to a crystalline silicon solar cell. In addition, the efficiencies of state-of-the-art devices based on non-fullerene acceptors are still lower than the best commercially available silicon photovoltaic devices (20% and above). Long-term operational stability is an even greater challenge. Organic active layer degradation can be caused by oxygen, moisture, thermal stress and extended illumination resulting in morphological instability, loss of charge-carrier mobility and performance degradation over time. Therefore, increasing the durability of devices, environmental stability and operational lifetime are some of the most important unaddressed issues for commercializing OPV technologies.6 Schematic representation of a bulk heterojunction organic photovoltaic (BHJ-OPV) device and the basic photophysical processes occurring in the device: photon absorption, exciton generation, exciton diffusion, charge separation at the donor–acceptor interface, charge transport and charge collection at the electrodes. The description of BHJ solar cells is based on the original description of the BHJ solar cell and the latest reviews on OPVs. Modified and redrawn from ref. 5. Fullerene vs. non-fullerene acceptors (NFAs): limitations and advantage Fullerene-based acceptors including PC60BM and PC70BM were pivotal in the early development of OSCs because they are intrinsically strong electron acceptors with high electron affinity, efficient electron-transport characteristics, and an excellent ability to induce ultrafast charge separation from donor materials. The spherical geometry of fullerenes makes it possible for them to allow effective delocalization of electrons, and high electron mobility leading to repeatable device performance. However, fullerenes show low and broad absorption in the visible and infrared ranges resulting in reduced photocurrent efficiency. Furthermore, their unchangeable geometry renders them incapable of energy-level engineering, thus making it challenging to increase open-circuit voltage (VOC). In addition, fullerene derivatives are morphologically unstable and expensive to synthesize, which negatively impacts their long-term performance and commercialization.7 On the other hand, NFAs provide an alternative that can effectively counteract the inherent weaknesses associated with fullerene-based acceptors as illustrated by the representative molecular structures of PC61BM, PC71BM, ITIC and Y6 shown in Fig. 2 and summarized in Table 1. The advantage of NFAs lies in their molecular structure tunability allowing the manipulation of the absorbance spectrum, molecular frontier orbital energies, and molecular packing. The high and wide absorption of these materials in both visible and near-infrared ranges greatly increase light harvesting efficiency and short-circuit current density (JSC). NFAs allow for minimized energy losses due to optimal energy level alignment with greater VOC. Improved morphological and thermal stability leads to better operating lifetimes for NFA-based OSCs which makes them suitable for practical use.8 However, NFAs pose some difficulties that must be taken into account. Due to their complicated chemical structures, NFAs can be difficult to synthesize and show inconsistency between batches. Strong intermolecular interactions might cause over-aggregation if not regulated properly which affects morphology at the nano scale and charge transfer. Nevertheless, based on theoretical modeling and structure–performance relationships continuous improvement of NFA materials and fabrication techniques has brought significant improvements in efficiency. Thus, today NFAs have become the dominant materials in state-of-the-art OSCs and hold great potential for further progress.9 Representative molecular structures of fullerene-based acceptors (a and b) and well-known non-fullerene acceptors (c and d). The figure depicts the structural evolution from fullerene derivatives to much more tunable non-fullerene acceptors which can achieve more effective light harvesting, energy level engineering and better PV performance. Redrawn from ref. 8, 10 and 11. Table 1Fullerene vs. NFA acceptors in organic solar cellsFeaturesFullerene acceptorsNon-fullerene acceptorsReferencesRepresentative materialsPC61BM & PC71BMITIC & Y6 10 and 11Optical absorptionWeak, narrow visible absorption; minimal NIR contributionStrong, broad and tunable absorption extending to NIR 12 and 13Energy-level tunabilityLimited due to rigid spherical structureMolecular engineering enables HOMO/LUMO tuning and reduced offsets 14 and 15Voltage loss (ΔVloss)Larger non-radiative losses; requires ≥0.3 eV offsetNear-zero offset possible; suppressed non-radiative recombination 16 and 17Exciton dissociationDriven by large energetic driving forceEfficient even with small driving force due to delocalized CT states 18 and 19Morphology controlIsotropic aggregation; limited packing controlDirectional π–π stacking; optimized nanoscale phase separation 20 Charge transportGood electron mobility but limited structural tunabilityBalanced and tunable transport; improved carrier lifetimes 21 Efficiency rangeTypically ∼10–12%>18% in single-junction OSCs 22 and 23 Fullerene vs. NFA acceptors in organic solar cells NFAs rapid efficiency progress (state-of-the-art PCEs) The emergence of NFAs has completely changed the performance map of organic solar cells (OSCs). Before NFAs, OSCs based on standard versions of fullerene derivatives were constrained by low absorption and fixed energy levels limiting the highest achievable power conversion efficiencies (PCEs) to quite small numbers. As NFAs, especially fused-ring electron acceptors were introduced, OSCs started to experience considerable efficiency gains due to broad and tunable light absorption, frontier orbital energies that are tuned, minimized energy losses, and improved charge transport characteristics. Initial reports claimed NFA-based OSCs achieved efficiencies over 14% surpassing fullerene equivalents by approximately 2018 through the optimization of molecular design and donor–acceptor pairing schemes.24 The addition of the long-chain acceptors has been shown to play an important role in improving thermal stability of two other polymer solar cell (PSC) systems that are commonly studied. These results suggest that the polymerization of non-fullerene small-molecule acceptors is a promising and widely applicable approach toward enhancing the thermal stability of PSCs.25 Over the past decade, the power conversion efficiencies of organic solar cells have been radically and swiftly increased due to the use of non-fullerene acceptors. In place of the fullerene derivatives, NFAs have good and tunable light absorption, less energy loss and better morphological stability. Since the introduction of ITIC in 2015 that provided a PCE of 6.8% uninterrupted molecular innovations have resulted in continuous A–DA'D–A types acceptors like Y6 and its analogs.26 These materials have ultra-narrow bandgaps, broad near-infrared absorption, efficient transport of charge and low non-radiative recombination. Single-junction NFA-based OSCs have very quickly surpassed 14 percent efficiency and have recently reached state-of-the-art PCEs of nearly 18 percent, a new milestone in achieving high-performance organic photovoltaics through synergistic optimization of molecular design, donor–acceptor matching and device engineering.27 Recent work has taken the performance of NFA-based OSCs even further with a number of state-of-the-art devices now demonstrating PCEs of almost 19 percent, a significant breakthrough over previous organic photovoltaics. Progress in self-assembly control, side-chain engineering, and crystallinity enhancement provided devices with PCEs up to and including the 19.7% and even over 20% of some recent reports highlighting the efficiency of refined molecular structures and processing methods in reducing recombination losses and maximizing charge generation and movement.28 These developments combined illustrate that NFAs have facilitated an unprecedented increase in OSC efficiencies with initial reports of ∼6–7% in pioneering acceptors to current maximal values of 20% or higher in optimized devices. This swift development highlights the significance of structure–property–performance links in informing theoretical and experimental innovations that keep increasing the performance boundaries of organic solar photovoltaics.29 Need for computational and machine learning approaches The swift development of NFAs has opened up an immense chemical design space that renders trial-and-error methods of purely experimental design more and more ineffective and expensive. NFAs have very complicated structure–property–performance interactions in which minor changes in the planarity of the molecular backbone, substitution in the side chains, terminal group and intermolecular packing can have a drastic influence on optical absorption, energy level alignment, charge transport and recombination processes. Computational chemistry techniques, especially density functional theory (DFT) and time-dependent DFT (TD-DFT) have proven to be vital tools in investigating the electronic structure, excited-state properties and energy losses of NFAs on a molecular scale. These theoretical underpinnings are useful in designing rational molecular structures by providing quantitative correlations between chemical structure and photovoltaic activity.30,31 In addition to single–molecule properties the performance of NFA-based OSCs, that relies on the bulk heterojunction is tightly dictated by the nanoscale morphology and intermolecular interactions.32 Computer-based methods like molecular dynamics (MD) simulations, coarse-grained models and kinetic Monte Carlo simulations have given highly important information on molecular packing, phase separation, charge pathways and recombination processes, which are experimentally challenging to measure. These multiscale simulations fill this gap between molecular design and device physics enabling researchers to see how the molecular structure can affect mesoscale morphology and macroscopic photovoltaic characteristics like short-circuit current density and fill factor.33 As the reported NFA molecules and experimental data have grown exponentially, machine learning (ML) has become an effective complement to conventional computational algorithms. Algorithms based on ML make high-throughput screening and rapid prediction and identification of hidden structure–property–performance trends in large chemical libraries possible. The combination of quantum-chemical characteristics, morphology and experimental device data in the form of ML models can allow discovering high-performance NFA faster at the same time minimizing computational and experimental expenses. As a result, the synergistic benefits of computational modeling and machine learning in predictive design and optimization of next generation NFAs are becoming a fundamental approach to paradigm shift in organic photovoltaic studies toward data-driven research.34 Recent advances in artificial intelligence have further transformed materials discovery beyond conventional machine-learning-based screening. Modern AI-driven frameworks integrate deep learning, graph neural networks, generative models, reinforcement learning and autonomous optimization strategies to accelerate the design of functional materials with targeted properties. Unlike traditional forward-design approaches, inverse design methodologies enable researchers to specify desired optoelectronic characteristics and subsequently generate candidate molecular structures that satisfy those requirements. These emerging approaches can efficiently explore vast chemical spaces, identify hidden structure–property relationships and substantially reduce the time and cost associated with experimental trial-and-error discovery. As highlighted by Cheng et al. (2026), the integration of artificial intelligence with computational chemistry, high-throughput simulations and automated experimentation is reshaping the paradigm of materials research and providing powerful opportunities for the discovery of next-generation organic semiconductors and non-fullerene acceptors.35 Scope, novelty and structure of the review Organic solar cell (OSC) development has been revolutionized by the production of NFAs enabling single-junction power conversion efficiencies of around 17–19% and more with applications spaces extended beyond outdoor PVs into indoor and other specialized systems.36 The literature usually involves individual NFA classes (e.g. fused ring, non-fused, star-shaped, dimerized) or a device concept like ternary blends, layer-by-layer architectures and all small molecule or all polymer systems, but seldom combines a rigorous theoretical structure–property–performance framework across these families.37 NFAs will be discussed in terms of molecular topologies (fused and non-fused cores, A–D–A and analogous motifs, star-shaped and dimerized structures, macrocycles and polymeric acceptors), device configurations (binary and ternary blends, all small molecule, all polymer, indoor OSCs and layer-by-layer stacks) and operating environments (standard solar and low illuminance indoor environments). In this general context, the focus of the theoretical tools is on electron accepting material instead of donors or interfaces in an attempt to explain experimental structure–property relationships in these systems.38,39 This review is novel in three aspects. First, it constructs a single structure–property–performance platform of NFAs, relating molecular design variables like conjugation length, heteroatom incorporation and end group engineering to mesoscale variables like molecular packing, phase separation and charge–transport pathways and their effect on device parameters like open-circuit voltage, short-circuit current and fill factor and stability.40 This review is a comparative analysis of various NFA families as opposed to earlier studies, which primarily examine single NFA classes to extrapolate to predictive material development. Second, the computational and theoretical methods are combined to explain experimental trends in exciton diffusion, charge transport, and voltage losses in high-efficiency OSCs.41 Third, the discussion relates molecular-level understanding to emerging application demands such as low-light operation and long-term device stability which are essential to useful organic photovoltaic technologies.42 The review is structured to flow systematically through the molecular structure to device performance with each of the core sections being based upon theoretical descriptors and models that are directly correlated to experimental observables in NF based OSCs. NFA topology fused vs. non fused cores, A–D–A and related motifs, star-shaped and macrocyclic acceptors, DSMAs and polymeric acceptors is a first block that discusses theoretical predictions of optical gaps, transition dipoles and charge transport parameters as well as experimental device data in binary and ternary blends or all polymer architectures.43 A second block is centered on morphology and aggregation, synthesizing theoretical and experimental information on exciton diffusion, domain size, percolation and vertical phase separation in bulk heterojunction devices and in layer-by-layer devices including the importance of ternary components and additives in morphology control.44 A third block regards application driven performance measures as low-light efficiency in indoor OSCs, stability and morphology evolution and scalability in large-area or roll to roll processed modules and how theoretical descriptors can be used in the future to target these measures to novel NFA designs. The review finishes with a perspective section that identifies unresolved aspects of the theory such as multiscale simulations of the relationship between structure, property, performance and predictive design processes that combine quantum chemical simulations with morphology simulations and device modeling to define the future direction of NFAs on the way to commercially viable organic photovoltaics.45 Evolution and classification of non-fullerene acceptors Non-Fullerene acceptors will be classified and evolutionary classifications will be given. NFAs have been a gamechanger in the field of OPVs as they offer stronger absorption and tunability than fullerene acceptors, thereby allowing power conversion efficiencies to surpass 18%. Although the first PDI- and NDI-based acceptors showed promise of being aggregation-free and the ability to control the morphology, they were ultimately plagued by aggregation and poor morphology control. This resulted in the design of sophisticated small-molecule architectures such as A–D–A, A–π–D–π–A and Y-series acceptors and polymeric NFAs. Fused-ring electron acceptors (FREAs) have been among those with the highest efficiencies because of their rigid planar structure which makes it very efficient to absorb light and transport charges well. Their synthetic complexity has led to the development of other simpler and more scalable alternatives. The development of the NFA design is discussed in a critical way and the difference between small-molecule and polymeric systems is compared. Fused-ring and Y-series architectures are evaluated and the influence of molecular structure on the photovoltaic properties is discussed. Special attention is given to understanding structure–property–performance relations and how theory such as DFT calculations, excited state analysis and morphology modeling can aid in the development of next generation NFAs.46,47 Historical development of NFAs NFAs as a novel class of materials have emerged as a fundamental change in the field of OSCs because of the inherent limitations of fullerene-based acceptors. The fullerene derivatives like PC61BM and PC71BM were the most popular OSC for almost 20 years because of their beneficial electron mobility and charge separation properties. However, their absorption in the visible and near-infrared region was weak, electronic properties were not easily tunable, they were morphologically unstable and the production costs were relatively high, making further efficiency improvements increasingly difficult.48 These drawbacks created a demand for new acceptor materials that would be able to improve light harvesting at the same time offer rational molecular engineering. The first non-fullerene systems containing perylene diimide (PDI) and naphthalene diimide (NDI) showed that high electron affinity and low LUMO energies can be obtained without the presence of fullerene structures. However, these materials had the drawback of being highly planar with aromatic structures that tend to aggregate and form large phase-separated domains and resulted in inefficient exciton dissociation. As a result, despite the desirable electronic properties, modest efficiencies in the devices were achieved. This limitation highlighted an important design principle that has guided NFA development to date, namely that the optimal PV performance is achieved by balancing the molecular packing with the controlled nanoscale morphology and not by simply maximizing electron transport properties. Significant progress was made with the development of fused-ring acceptors like O-IDTBR which were used with P3HT to yield power conversion efficiencies (PCEs) of 6–8%. Unlike previous PDI- and NDI-based systems, the absorption of O-IDTBR was broader and it was more morphologically compatible with donor polymers suggesting the molecular architecture was able to be designed to optimize both optical and electronic properties. But the breakthrough was in the development of ITIC in 2015. The presence of strong visible-light absorption, tunable energy levels and favorable intermolecular interactions in the A–D–A framework used in ITIC proved to be more effective. Thus, ITIC-containing devices significantly surpassed previous NFAs, and led to the development of a molecular platform with many derivatives of the ITIC structure.49 Nevertheless, ITIC systems still suffered from energy losses and spectral limitations which limited their ultimate efficiency potential. In 2019, a new leap in the evolution of NFAs was observed with the appearance of Y-series acceptors. As opposed to the ITIC-type molecules, Y-series materials were designed according to the A–DA′D–A structure which provided for higher intramolecular charge transfer, longer absorption of near-infrared light to 900 nm and lower energy losses due to nonradiative recombination. The changes were not just incremental; they addressed a number of important efficiency bottlenecks present in previous generations of the NFA. Improved molecular packing and optimized energy-level alignment also played a role in the improved charge transport and collection efficiencies. Thus, the performance of devices quickly jumped from around 15% to more than 18% and finally beyond 20%, directly reflecting the direct translation of systematic molecular engineering to photovoltaic benefits. Perhaps most importantly, the history of NFAs shows that efficiency improvements have not been achieved by any single structural innovation but rather by evolved refinement of the absorption characteristics, energy loss management, molecular packing and morphological stability. As shown in Fig. 3 each generation brought in new design techniques which pushed the envelope of OSC performance based on NFA, as well as new challenges with regards to stability, scalability and commercialization.50 Non-fullerene acceptors (NFAs) for organic solar cells from 2000 to 2026. The timeline outlines important milestones in the molecular design innovations and associated power conversion efficiency (PCE) of PDI-containing acceptors, ITIC-type A–D–A structures, Y-series acceptors and the most recent high-efficiency NFA systems with PCE values greater than 21%. Only the data was taken from ref. 47–50, the figure was generated by the author. Small-molecule vs. polymer NFAs Based on the molecular structure, NFAs can be divided into three types namely small organic molecules, polymeric materials and inorganics. Small-molecule acceptors (SMAs) are the most widely used class for the development of high-performance organic solar cells because of their precisely defined molecular structures that are often in the molecular weight range of 500–2000 Da. They are well-controlled in their architecture allowing energy levels to be systematically tuned through core, end-group, and side-chain modifications leading to high light absorption coefficients and high charge generation efficiencies frequently greater than 105 M−1 cm−1. Perhaps more importantly, the structural uniformity of SMAs makes it easy to investigate the structure–property relationship and reduces the variations between different batches which is required for scaling up to manufacture devices. Optimized SMA-based systems can thus achieve PCEs of up to 20% under minimal driving-force conditions demonstrating that careful engineering of the packing and energy-level alignment of the molecules can significantly reduce charge recombination losses.51 However, SMAs have some drawbacks such as the formation of excessive crystals and morphological instability under long term operation for high efficiency. This restriction has encouraged growing interest in polymerized small-molecule acceptors (PSMAs) that have the good optoelectronic properties of SMAs but the mechanical strength of a conjugated polymer. Unlike the discrete SMA molecules, PSMAs are composed of a network of interconnected acceptor units resulting in an extended polymeric network as shown in Fig. 4b.52 The polymeric character of these materials leads to the increase of the stability of the material during operation as phase segregation is generally suppressed while film forming properties are improved. The advantage of this is that the molecular-weight distribution and structural heterogeneity are increased which may lead to lower charge mobility and to difficulties in precise energy-level control. Hence, PSMAs have been attractive pathways to viable devices but in the past the peak photovoltaic performance of PSMAs have lagged behind SMAs. This trade-off is clearly demonstrated by comparing recent SMA and PSMA systems. Recently, Wang et al. reported, polymerized ITIC-based acceptors (PG-IT and PG-IT2F) that resulted in a remarkable PCE of 17.24% when used in all-polymer solar cells based on PM6 as seen in Fig. 4a which shows that all of these materials are potential acceptors for all-polymer solar cells and that rational polymer design can significantly reduce the gap in performance between SMAs and PSMAs.53 In particular, the introduction of the π-bridges containing fluorinated side chains (IF) in PG-IT2F lowered the LUMO energy level and enhanced electron transport and the introduction of branched alkyl side chains improved miscibility of donor–acceptor and optimized blend morphology. The resulting HOMO/LUMO energy-level alignment of PM6, PY-IT2F and PG-IT2F is illustrated in Fig. 4c. However, benchmark SMA systems like Y6 and its derivatives still manage to achieve slightly better efficiencies usually from 18–19% for single-junction devices.54 This difference indicates that even with the superior molecular order and intermolecular packing of SMAs, they are still beneficial for charge transport and exciton dissociation. Overall, the available data suggests that SMAs continue to be the best devices for maximum device efficiency while PSMAs provide a promising approach to better morphological and operational stability. The development of polymer acceptors with electronic precision of SMAs, and the incorporation of the inherent stability benefits of polymeric structures, will be key to future advances. (a) Chemical structures of typical polymer-assisted NFAs. (b) The small molecule backbone of PY-IT and its modified polymerized versions, PG-IT and PG-IT2F with the change in a small molecule acceptor to polymerized acceptors. The design of the polymerization strategy and the fluorination of the π-bridge is used to control molecular packing, charge transport and energy levels. (c) π-bridge fluorination and branched alkyl chains lead to enhanced miscibility, mobility and energy levels. Modified and redrawn from ref. 53. Fused-ring electron acceptors (FREAs) The most successful non-fullerene electron acceptors are fused-ring electron acceptors (FREAs) featuring an electron-rich fused-ring core structure, electron-deficient terminal groups, and carefully designed side chains as a result of the D–A or A–DA′D–A molecular architectures. In addition to just structural description, the great development of FREAs is the covalent fusion of aromatic rings to rigidize the π-conjugated backbone which affects not only the structural planarity but also increases π–π stacking interactions and extends the length of effective π-conjugation and also decreases the structural reorganization energy-all of which together improve charge transport efficiency. Among the early representatives, the system ITIC is considered to be a benchmark system in which the indacenodithiophene (IDT) unit is electron rich and the 1,1-dicyanomethylene-3-indanone (IC) end groups are strong electron-withdrawing units to give a well-balanced intramolecular charge-transfer system.51 The absorption coefficient (typically >105 M−1 cm−1), the spectrum range and the flexibility in energy-level modulation of such architectures are significantly higher than in the other fullerene derivatives yet a subtle molecular modification is often enough to affect the performance. The modular structure of FREAs allows researchers to optimize systematically over three interconnected aspects: core extension, end-group modification and side-chain engineering.55 In particular, core engineering has been shown to be effective in extending the fused-ring size and the incorporation of more electron-rich or heteroatom-containing units (such as dithienothiophene, heptacyclic, dithienothiophenothiophene systems) which results in broadening of light absorption into the near-infrared region while simultaneously increasing the intramolecular charge delocalization. However, there is a clear trade-off between optoelectronic performance and processability for overly large cores as they may have worse solubility and film-forming properties. Regarding end-group functionalization, particularly halogenation (fluorination or chlorination) of IC molecules has been shown to be effective in lowering the LUMO energy level and enhancing non-covalent intermolecular interactions to achieve increased molecular ordering but may also increase the aggregation tendency and result in unfavorable phase separation if not carefully balanced. Side-chain engineering also offers an important control for tuning the solubility, crystallinity and blend morphology by varying the length, position and substitution pattern of the side chain which have a significant effect on the miscibility of donor–acceptor components and nanoscale morphologies formed.56 More recently, fused-core A–DA–D–A type Y-series acceptors with electron-deficient units such as benzothiadiazole or benzotriazole have been developed. Though this elegant architecture increases intramolecular charge transfer, widens absorption even up to 900 nm and shrinks the optical bandgap, more efficient exciton dissociation and charge separation is found to be strongly dependent on precise control of the morphology of the active layer.57 Y-series NFAs and next‐generation architectures An innovative subclass of the FREAs is the Y-series, having an architecture of A–DA′D–A, where two fused donor units flank an electron deficient benzothiadiazole core. The prototype of the Y6-type (also referred to as BTP-4F) was published in 2019, featuring 2,2-difluoroindane-1,3-dione terminal groups, thieno[3,2-b]thiophene linked fused rings, and a central acceptor unit, 5,5′-dioxide-benzothiadiazole as shown in Fig. 5. The molecular design leads to an extremely small optical band gap (∼1.33 eV) and a significant near-infrared absorption. As such, the first-generation power conversion efficiency (PCE) of PM6 : Y6-based OSCs is significantly higher at approximately 15.7% compared to previous non-fullerene acceptors. Further molecular engineering efforts such as side-chain relocation and heteroatom substitution along the conjugated backbone further improved the intermolecular interactions and charge transport with Y6-derivative devices achieving efficiencies between 18–18.3%. Recently, record single-junction PCE values of ∼19% have been reported (2022–2023) mainly due to optimized light harvesting and voltage losses in optimized Y6 derivatives.58 The unusual curvilinear molecular backbone of Y6 and the presence of fluorinated terminal groups favour the face-on π–π stacking configuration in solid-state. Experimental and theoretical studies indicate that Y6 thin films have a strong electronic coupling and dense intermolecular contacts which are desirable for efficient charge transport. Theoretical studies also suggest that there is an intramolecular excited state (i-EX) that can enable the dissociation of charge with low energy barrier, thus increasing the efficiency of exciton dissociation. The relatively low exciton binding energy of Y6 (∼0.1 eV) also makes it favorable to generate free charges even when the energetic offsets are small.59 In order to further fine-tune the electronic properties, a number of derivatives have been synthesized by systematic modification of end groups and by substitution of heteroatoms that includes Y7, Y18 and BTP-4Cl. In particular, the miscibility of blends is enhanced and the LUMO level slightly reduced by fluorination with chlorine in the terminal units such as in BTP-4Cl (PCEs of ∼18.6%). Likewise, the selenium analogues and fused rings with extended fused rings provide a red-shifted absorption and have shown efficiencies greater than 17% and short circuit current densities of about 28 mA cm−2. The asymmetric molecular design (ASMD) approach, introduced with asymmetric end groups (AS) or asymmetric side chains (SS) has also been used to achieve a VOC/JSC compromise, with devices achieving efficiencies greater than 17%.60 Although these progress has been made, there still exist several challenges for the Y-series materials such as complex synthesis, voltage losses due to narrow bandgap and morphology control in thick active layers. However, the Y-family is an important development of next-generation NFA design that combines molecular engineering and supramolecular packing to maximize the photovoltaic performance.61 The structure of non-fullerene acceptor Y6 includes a fused benzothiadiazole[2,1-b:3,4-b′] dithiophene (BTP) electron-deficient core, a backbone of conjugated thiophenes, branching alkyl side chains and fluorinated 2-(3-oxo-2,3-dihydro-1H-inden-1-ylidene) malononitrile (INCN-F) end groups. The donor–acceptor–donor configuration allows for good intramolecular charge transfer and for broad absorption and efficient charge transport properties in organic solar cells reaching high efficiencies. Modified and redrawn from ref. 58. Chemical diversity and design evolution NFAs have a rapidly growing chemical space. Simple A–D–A dyes initially dominated but various structures are currently investigated. In addition to planar A–D–A and Y-types star-shaped, and multi-branched NFAs have been reported (e.g. tetraphenyl benzene cores). Heteroatom variation is widely used, benzothiadiazole, thiadiazole or even quinoxaline cores have been fused to tune electron affinity. Asymmetric NFAs or unequal end groups or backbone units can create dipoles that facilitate charge separation. Non-fused acceptors (e.g. diketopyrrolopyrrole-based) are also present, but tend to have larger bandgaps. Linear vs. side-chain diversity branched to linear alkyls; fluorinated chains is under constant investigation to strike a balance between solubility and packing. In the example, shifting the 2-ethylhexyl branch on the backbone of Y6 increased crystallinity and facilitating 18.3% PCE.62 In the course of time, design trends can be observed: Reducing bandgaps: above 1.6 eV (ITIC) to around 1.3 eV (Y6) toward the optimum 1.4 eV. LUMO tuning: fluorination or halogenation decreases the LUMO (the opposite effect of donors, increasing VOC). Morphology control: fine-tuning D–A phase separation by adding more steric bulk or polymeric backbones.63 This chemical diversity enables customization to individual donor polymers or device designs (e.g. tandem cells). It also implies that new materials will be able to utilize complementary states of absorption or charge transfer. But with the constantly-growing design library it is difficult to determine which of the features is most useful. Complex fused systems can be degraded during production or difficult to produce which poses a concern to stability and manufacturability. One of the new strategies to direct this diversity is a systematic theory-informed approach (e.g. screening by DFT/MD). Finally learning the structure–property relationships of this large NFA palette is a key activity of the field.64 Molecular design principles of NFAs The tremendous progress in the performance of OSCs has been primarily contributed by the development of non-fullerene acceptors (NFAs). Donor–acceptor architectures, end-group engineering, core modification, side-chain optimization and control of molecular geometry are some of the key molecular design strategies. These methods can control electronic structure, optical absorption, intermolecular interactions and film morphology which affect the photovoltaic parameters including open-circuit voltage (VOC), the short-circuit current density (JSC), the fill factor (FF) and the power conversion efficiency (PCE). End-group functionalization with electron withdrawing groups such as halogens is a common approach to minimize the LUMO energy and increase the absorption towards the red end of the spectrum. The enhanced light harvesting and intermolecular interactions by core engineering strategies such as fused-ring extension or introduction of electron deficient units further enhance the properties of light harvesting. These strategies are often measured by the properties of their frontier orbital energies, optical bandgap, reorganization energy, and exciton binding energy.65 Donor–acceptor–donor (D–A–D) frameworks NFAs are frequently used in high-performance devices that typically utilize donor–acceptor architectures to facilitate intramolecular charge transfer and increase optical absorption. For instance, ITIC has a HOMO of −5.48 eV and a LUMO of −3.83 eV (bandgap 1.65 eV), while Y6 (BTP-4F) has a HOMO of ∼−5.65 eV and LUMO of ∼−4.10 eV (bandgap ∼1.55 eV) as shown in Table 2.66,67 These materials allow researchers to harvest visible and near-infrared photons efficiently while keeping favorable energy-level alignment with donor materials. The relationships between molecular design, electronic structure, charge transport and photovoltaic performance are summarized in Fig. 6. Achieving optimal balance between bandgap and energy-level offsets is crucial to the photovoltaic performance of D–A–D NFAs. Based on the above, the bandgap generally lies in the range of 1.4–1.6 eV which allows for a better balance of JSC and VOC thus achieving a higher PCE. Moreover, donor–acceptor coupling provides a facilitation of exciton binding energy reduction and subsequent charge separation by way of excitation state delocalization.68 Molecular packing also is important; a fused-ring structure increases the conjugation and the charge transport. The side-chain engineering is used to control aggregation and ensure proper morphology. It also shows that low reorganization energies for electrons and holes are beneficial for efficient charge transport and high FF as revealed in their computational studies. For instance, IDTT-based acceptors have recently been developed with lower reorganization energy than ITIC which is believed to have better ambipolar transport properties.68,69 Table 2The table below compares representative NFAs: e.g. ITIC versus ITIC-4F versus Y6, with their energy levels and record PCEsMoleculeHUMO (eV)LUMO (eV)Bandgap (eV)PCE (%) (device)ReferencesITIC (IDTT core)−5.48−3.831.65∼11–12 (with PM6) 70 ITIC-4F (difluoro endgroups)−5.63 (↓)−3.88 (↓)∼1.75≈13 (improved JSC) 71 Y6 (fused D core)−5.65−4.101.5515.7 (with PM6) 72 Y6-analog (Cl on core)−5.6−4.021.5817.6 (ternary cell) 73 The table below compares representative NFAs: e.g. ITIC versus ITIC-4F versus Y6, with their energy levels and record PCEs Schematic illustration of the D–A–D molecular design strategy that is usually employed in high-performance NFAs. The electron-rich donor units mainly impart the HOMO level while the electron deficient acceptor units mainly impart the LUMO level leading to efficient ICT. This electronic separation causes the optical bandgap to narrow, widens light absorption and helps to generate charges which in turn affects the photovoltaic parameters JSC, VOC and PCE. Examples are ITIC- and Y6-type acceptors which employ D-A-D interaction for achieving balanced energy-level alignment and efficient charge transport. Only the data was taken from ref. 67 and 68, the figure was generated by the author. End-group engineering (halogenation, electron-withdrawing groups) The electronic and optical properties of NFAs can be tuned by end-group engineering. The frontier energy levels, optical bandgaps, and light-harvesting properties are strongly affected by the electron-withdrawing strength of terminal groups. The end groups derived from INCN are especially effective, as they increase the electron delocalization and allow for efficient charge separation.74 Computational studies have indicated that increased electron-withdrawing groups tend to decrease the energy of the LUMO and the optical bandgap, resulting in wider absorption, and better photovoltaic properties. Halogenation, particularly fluorination and chlorination, has proved to be a useful method for modifying terminal groups. Substitution with fluorine enhances the electron affinity and has a relatively small steric effect that allows for favorable molecular conformations and increased electronic coupling. Theoretical studies of the derivatives of PZT indicated that the electron affinity and electrophilicity of the end groups containing fluorine is higher than of the non-halogenated ones, which leads to better electron–transport properties. However, chlorination has better electron withdrawing properties but at the same time can create extra steric hindrance that can impact molecular organization.75 The recent TD-DFT studies also show that exciton binding energies can be lowered by introducing halogenated end groups in order to enhance the exciton dissociation.76 With the success of high performance acceptors like Y6, it is critical to balance the electron withdrawing strength with steric effects. The halogen substitution also gives rise to more favorable intermolecular interactions via localized charge distribution, leading to efficient device operation as suggested by the MESP analysis.77 Core modification and π-conjugation extension Tailoring the optoelectronic properties of the NFAs by tuning the π-conjugated core is an effective approach. The larger the core the narrower the bandgap and the greater the extent of absorption into the near-infrared region.68 For instance, the introduction of extra thiophene or benzene building blocks into Y6-based compounds widens the light harvesting and increases photocurrent generation. Huang et al. developed non-covalently fused acceptors in which the intramolecular S⋯O interactions were found to effectively form a nine-ring fused structure. The absorption of these materials was as far as 900 nm and the power conversion efficiency (PCE) improved from 6.87% to 11.10%.78 Likewise, Zhan et al. found that the frontier orbital energy could be systematically tuned and efficiency over 13% could be achieved by using fused-ring donor units with eleven rings as shown in Table 3.79 In addition to the conjugation length, the chemical nature of the core significantly affects energy levels and excited-state characteristics. The insertion of electron deficient heterocycles into the core will decrease the energies of the frontier orbital and potentially decrease the singlet-triplet energy gap (ΔEST), thus decreasing energy losses and increasing charge generation. Computational studies of Y6 derivatives based on phenalene revealed a decrease in electron reorganization energies and increase in intermolecular interactions indicating better electron transport properties.80 Core substitution is possible to further optimize. Ge et al. showed that the molecular packing and blend morphology for methoxy substituted quinoxaline backbone were highly dependent on the substitution position. The superiority of the crystallinity and phase separation of the meta-methoxy derivative (Qx-B2) is reflected by the VOC of 0.88 V, the JSC of 25.97 mA cm−2, the FF of 77% and the PCE of 17.9%. Replacement of central heteroatoms, like oxygen or selenium atoms, is also a very effective method of tuning the electronic properties. In general, core extension and structural modification have been reported as two major strategies to optimize light absorption, energy levels and charge–transport properties for high-performance NFAs.81 Table 3Effects of core π-extensionAcceptor typeCore modificationAbsorption (soln. λmax)PCE (donor blend)ReferencesUnextended core (DC6-IC)2,5-Dihexylphenylene core623 nm (λmax in film)6.87% 30 68 Ladder-like core (DOC6-IC)2,5-Bis(alkyloxy)phenylene locked by S⋯O780 nm (λmax in film)11.1% 30 82 Fluorinated core (DOC2C6-2F)(Above + 2 F atoms)800 nm (λmax in film)13.24% 30 83 Extended fused core+Benzene/thiophene rings (e.g. NC-FREA)550–900 nm (broad)9–11% 730 84 Effects of core π-extension Side-chain engineering: solubility, morphology, packing The processability and morphology of the active layers of NFAs are strongly influenced by side-chain engineering. The presence of alkyl and alkoxy groups enhances solubility in organic solvents which helps create even thin films. The presence of too many bulky side chains, however, can negatively influence the packing and charge transport of the molecules, making a compromise between solubility and intermolecular interactions necessary. Linear and branched alkyl chains are effective in simultaneously achieving solution processability and solid-state organization for Y6-type acceptors. The electronic and steric effects of the side chains also vary with position, with para-alkoxy substituents being activated by conjugative effects and meta-alkyl substituents having predominantly inductive effects.85 Experimental results show the essential role of side-chain optimization in improving device performance. Shortening the linear alkyl chain on the thiophene unit and extending the branched chain on the pyrrole moiety in Y6-eC6-BO resulted in an improvement in crystallinity, donor–acceptor miscibility, and film morphology. Finally, the as-cast PM6 : Y6-eC6-BO devices showed a PCE of 17.33% while the PM6 : Y6 devices showed a PCE of about 15.7%. The same applies to Qx-based NFAs which were found to be very soluble in CHCl3 and chlorobenzene and to have good film-forming properties.86 In general, an ideal side-chain design should have a certain solubility level and morphology control but not block intermolecular interaction too significantly, which is essential for efficient charge generation and transport in OSCs.87 Planarity vs. steric effects The activity of NFAs relies on maintaining an acceptable balance between planarity and steric effects. Extended π-conjugation, reduced reorganization energy, and enhanced charge transport due to greater intermolecular interactions are the primary advantages of a planar backbone. However, excessive planarity can lead to increased aggregation and crystallization which can be detrimental to solubility and blend morphology. Therefore, controlled steric engineering is often used to control molecular packing and compatibility with the donor polymers.88–90 One method is to utilize non-covalent conformational locks, such as S⋯O, S⋯N, and H⋯F interactions especially for non-fused ring acceptors (NFRAs). Such interactions are enough to fix a near planar molecular structure and their absence often makes the structure more complicated to synthesize but not to the detriment of the optoelectronic properties.91 Computational investigations show that good intramolecular interactions can drive efficient electronic coupling similar to that seen for fused-ring analogues and can help to minimize conformational disorder. There are also steric features such as bulky side chains and judicious placement of substituents that can influence the molecular organization to further optimize the packing in thin films and reduce the amount of aggregation. Moderate steric hindrance is proposed to create nanoscale phase separation and interconnected transport pathways that facilitate exciton dissociation and charge transport. Thus, proper design of NFAs depends on the control of planarity and judicious steric control to achieve a proper molecular ordering and optimum device morphology.92 Structure–property relationships (computational insights) The tunable electronic structure and wide optical absorption are desirable properties of acceptors that are not met by fullerenes which is solved by the NFAs. Frontier orbital energies, optical gaps, charge-transfer excitations and charge-transport characteristics are commonly studied using computational methods, such as DFT and TD-DFT. The calculated HOMO and LUMO energies will vary from one study to another depending on the exchange-correlation functional, so it is important to emphasize relative trends rather than absolute values. Examples of commonly used functionals include B3LYP, CAM-B3LYP, PBE0, ωB97XD and M06-2X. These calculations have been pertinent to the design of NFAs (e.g., ITIC) which show energies that are favorable for many donor polymers and for fused-ring acceptors (e.g., Y6) which exhibit narrower bandgaps and superior IR absorption. Computational studies also show that tuning of the donor cores, π-bridges and terminal acceptor groups are effective strategies to control the electronic and optical properties of NFAs.93 These resulting structure–property relationships are discussed in the following sections. Electronic structure parameters (HOMO, LUMO, bandgap) Frontier molecular orbitals and bandgap of NFAs are strongly influencing the photovoltaic performance. While the absolute values of the HOMO and LUMO energies derived from DFT calculations vary from functional to functional and from basis set to basis set, the predicted structure–property trends are still valid. In general, narrowing the HOMO–LUMO gap widens and red-shifts the absorption which can improve the photocurrent generation and appropriate tuning of the HOMO and LUMO energies can improve the open-circuit voltage due to enhanced donor–acceptor energy alignment.94,95 The computational study has been shown that NFA energy levels can be conveniently controlled via molecular design. The LUMO level tends to decrease and the bandgap narrows with greater conjugation length and an increased electron withdrawing terminal groups. For instance, the calculated band gaps of phthalimide-based A–π–D–π–A acceptors range from 1.84 to 2.20 eV and the smallest bandgap is achieved for the most electron-withdrawing derivative. Likewise, fused-ring NFAs like IDIC and Y6 tend to have band gaps ranging from 1.3 to 1.9 eV, whereas the band gaps of simpler systems of A–D–A tend to be in the range of 1.4 to 2.2 eV.96 Representative DFT studies also indicate that frontier orbital energies and electronic properties can be greatly changed by structural modifications of the donor core or terminal acceptor units.97 Band alignment can also be affected by small structural variations. The electronic structures of quinoxaline-based NFAs with different halogen substituents (F, Cl and Br) is virtually identical in terms of HOMO and slightly different in terms of LUMO and thus, it is sensitive to the nature of the molecular substitution. Theoretical screening of NFAs is often based on an analysis of the frontier orbitals to predict the optical properties and photovoltaic potential. The examples of IDT-TC, IDT-IC and IDT-IC-Cl show that terminal acceptor engineering can be applied to optimize the HOMO/LUMO energies and bandgap and consequently the overall device performance, as summarized in the Fig. 7.98 Non-fullerene acceptor structures used in high performance OSCs are presented, such as representative A–D–A architectures. Systematic tuning of frontier orbital energies, optical band gaps, charge transfer characteristics and molecular packing behavior can be achieved by modulating the donor cores and/or π spacers and terminal acceptor groups. Modified and redrawn from ref. 95. Reorganization energy and charge transport Internal reorganization energy (λint): This is a structural relaxation that takes place when a molecule acquires or loses a charge and plays an important role in the charge transport in organic semiconductors. Marcus theory states that the processes of charge transfer are faster when the reorganization energies of both the electron (λe) and the hole (λh) are low, and that they are typically correlated with higher carrier mobility.93,99 The less the geometry changes when the molecule is charged, the better its transport efficiency. Many NFAs have recently been reported to have very low reorganization energies from DFT investigations. For instance, the A–D–A type acceptors have exhibited λe values from 0.0007 to 0.017 eV and λh values lower than 0.040 eV, which show good electron and hole transport properties.100,101 Likewise, the modified phthalimide-based NFAs had lower λ values than their reference structures, with BPDM4 having the lowest λe and λh and so being the most promising charge-transporting structure.99 Reorganization energy is an important theoretical parameter for the description of charge transport efficiency, though intermolecular electronic coupling and molecular packing also play important roles in charge transport in the solid state. Generally, decreasing λe and λh value is correlated with the better charge-transfer pathway, balance of carrier mobility, and better fill factor for the organic solar cells.102 Thus, the reorganization energy is an important parameter to study in order to understand the relationship between molecular structure and photovoltaic performance. Dipole moment and intramolecular charge transfer (ICT) Molecular dipole moments play an important role in determining the optoelectronic properties of NFAs especially their ICT behavior, excited-state dynamics and charge-generation efficiency. Ground state dipole moments obtained by DFT calculations give an indication of the amount of charge separation in the molecule and are strongly dependent on the donor–acceptor architecture. The moderate dipole moments (usually 2–6 D) in Y6-type NFAs allow for charge separation, mediated by the internal electric fields and ICT interactions that are more effective.103,104 Theoretical investigations of a truxene-BODIPY D–π–A system have demonstrated that the modulation of the donor–acceptor strength by π-spacer engineering can be used to enhance the dipole moments without compromising the favorable electronic energy level, thereby improving the photovoltaic performance.105,106 In recent years, theoretical and spectroscopic studies have shown that the effectiveness of charge transfer in aggregates not only depends on changes in dipole moment, but also on the molecular polarizability. It was shown that the aggregation-induced stabilization of charge-transfer states is mainly attributed to a decrease in polarizability changes (as opposed to mere increase in excited-state dipole moment) with the help of electro-absorption measurements and electronic-structure calculations. The results point to the importance of the dielectric response and polarizability tuning for reducing the energy losses and efficient exciton dissociation.107 The dipole moment effect in charge generation in neat NFA films has also been reported. In ITIC, combined experimental and theoretical studies indicated that the specific molecular packing configurations could result in the ultrafast intermolecular charge transfer which could facilitate the formation of free charges without the need for creating a donor–acceptor interface.108 In addition, the natural transition orbital analysis of Y6 assemblies showed that electron–hole separation is considerable and the excited state in neighboring molecules is delocalized, confirming the intrinsic ICT character of the materials. In summary, theoretical investigations suggest that a good balance between ground state dipole moments, significant charge redistribution in the excited states and an optimized polarizability all play a role in achieving high performance NFAs with efficient charge generation and low recombination losses.105,109 Frontier molecular orbital distribution Frontier molecular orbital (FMO) analysis is a valuable tool for understanding electron structure and charge transfer of non-fullerene acceptors (NFAs). Usually, DFT calculations reveal that the HOMO is spread over the π-conjugated bridge as well as the electron-rich, donor core while the LUMO is localized on the electron-poor terminal acceptor unit.110 The HOMO of the NFAs investigated is delocalized over the central donor segment and π-linkers, but the LUMO is mainly localized on the end acceptor groups.100,111 This spatial separation of frontier orbitals paves the way to efficient charge generation and separation upon photoexcitation by intramolecular charge transfer. The donor-centered HOMO–LUMO distribution is a hallmark of A–D–A type NFAs as shown in Fig. 8. Other NFA systems with similar localization patterns in the orbitals have been reported in literature in which the density of states (DOS) analyses reveal that the HOMO is localized in the donor/π-bridge parts while the LUMO is localized in the terminal acceptor groups. Further improvements in LUMO localization and further lowering of the energy gap are realized by structural modifications especially the introduction of stronger electron withdrawing end groups which allow for the improvement of the charge-transfer characteristics. In general, FMO visualization reveals that rational donor–acceptor molecular design is beneficial for the directionality of electron transfer and also can facilitate effective photovoltaic performance.112 Frontier molecular orbital distributions by DFT calculations of the reference non-fullerene acceptor (R) and its derivatives (BPDM1-BPDM7). The LUMO of the HOMO is mainly distributed over the terminal electron-accepting groups whereas the HOMO is mainly localized on the electron-rich donor core and π conjugated bridge. This spatial separation of frontier orbitals enhances the efficient intramolecular charge transfer (ICT) enhances exciton dissociation and improves the charge-generation performance in organic solar cells. Modified and redrawn from ref. 112. Role of molecular packing and aggregation In addition to intrinsic molecular properties, the solid-state packing of NFAs plays a significant role in the charge transport and device performance. The ordered NFA domains enable efficient intermolecular electronic coupling, decrease energetic disorder and foster pathways for charge transport. Previous studies have demonstrated that long range ordering of molecules plays an important role in determining the mobility of charge carriers and photovoltaic performance.113 The molecular packing can be drastically changed by structural engineering. For instance, NFAs based on quinoxaline are modified with bromine or chlorine to improve the intermolecular interactions and crystallinity and to achieve better phase morphology, charge transport and lower recombination losses.69 Similarly, side-chain modification can be used to effectively control aggregation behavior and film microstructure. The intermolecular organization may change with alterations in the architecture of alkyl chains, which in turn can influence charge–transport properties.114 A variety of computational methods such as DFT computation of molecular dimers and molecular dynamics simulations have been used to study packing-dependent parameters like electronic coupling and energetic disorder. Overall, both theoretical and experimental studies suggest that controlling molecular aggregation and ordering in the solid state is crucial to improving the charge mobility and filling factors in the OSC.81,115 Optical and excited-state properties NFAs are generally characterized by extremely intense, wide-band optical absorption and intermediate exciton binding energies due to their extended π-conjugation and donor–acceptor structures. Large molar extinction coefficients in the visible to near-IR range are attained with planar, linear backbones in which the atoms are highly polarizable (e.g. S, Se). Practically, NFAs such as Y6 absorb at wavelengths of the order of 600–930 nm, and solution-solid red-shifts are strong and the broadening is intense because of the intermolecular CT-Frenkel mixing. They have still high (few ×0.1 eV) exciton binding energies (Eb) but tend to be lower than fullerenes. The excited states are popularly studied using TD-DFT to predict frontier orbital energies and exciton character. Structural tuning (core fusion, end-group electron-withdrawal, side-chain engineering) is a systematic method of reducing optical gaps and increasing ε, increasing light-harvesting and device current.116 Lastly, exciton dissociation occurs via CT excitations intramolecular in aggregates or interfacial at donor acceptor junctions. NFAs in CT states tend to be below the local excitons and mix strongly to allow efficient dissociation even though the difference in energy is small. The TD-DFT is a complete picture of the relationship between molecular structure and optical properties, excitonic behavior and photovoltaic performance as follows. Absorption coefficient and spectral broadening NFAs exhibit high molar extinction coefficients compared to fullerene acceptors. For example, Y6 shows a log10 extinction >1.5 in the visible, far exceeding typical values for PC61BM. This is attributed to their planar fully conjugated backbone with polarizable heteroatoms. Large transition dipoles arise from extensive π–π* character indeed statistical TD-DFT studies correlate planarity, linearity and conjugation length with stronger absorption.116 NFAs cover a broad spectral range for instance, Y6 absorbs strongly from ∼600–930 nm overlapping much of the solar spectrum. High oscillator strengths (f) for the lowest excited states lead to intense peaks.117 The big absorption generates more excitons per incident photon and assists to drive the device photocurrent. Aggregation and mixing of CT-states cause spectral broadening and red-shifting to be strong in many NFAs in the solid state. An iconic case is Y6 vs. ITIC when passing solution to thin film the absorption peak of Y6 redshifts by about 180 meV and becomes much broader than the spectrum of ITIC does not change significantly. This has been ascribed not to the simple J or H-aggregation but to the creation of a manifold of hybrid Frenkel-CT states in the solid. Transient spectroscopy and theory demonstrate that Y6 films form mixed Frenkel-CT excitations, inter-molecular CT states are lower than the purely Frenkel excitons and they mix extensively. The high-density FE-CT states in Y6 provide very broad relaxation signatures and sub-ps absorption profile. NFAs do not only absorb strongly but also develop broad red-shifted bands in the solid state due to intermolecular interactions and CT mixing.118 Exciton binding energy In organic semiconductors a Frenkel exciton is the dominant photoexcitation. Exciton Eb are generally a few hundred meV because of low dielectric constants. Fullerene acceptors typically possess Eb ≳0.3 eV that would need similar energetic offsets to divide the exciton. NFAs have slightly smaller Eb due to their increased conjugation and dielectric screening. Remarked values of the state-of-the-art NFAs are around 0.2–0.3 eV.119 A recent study has reported that the majority of organic OSC semiconductors exhibit Eb greater than 300 meV and that a selenium-based Y6 (Y6Se) exhibited lower Eb than S-based Y6. It is possible to partially decrease Eb in NFAs that facilitate dissociation by molecular design. The quantitatively exciton binding is commonly approximated by the difference between the transport gap and optical gap. Design NFAs still exhibit large Eb typical values of around 0.3 eV in blends of ITIC or Y6 vs. around 0.5 eV in most donors. Experimentally combined studies propose e.g. PM6: PC6BM excitons with Eb = 0.22 eV PM6 : Y6 with Eb = 0.14 eV as indicated in Table 4. The rest of the binding suggests that the exciton dissociation has to be through interfaces or delocalized states. The low-offset OPVs have a small residual Eb implying that other effects such as exciton diffusion length and lifetime become significant since long-lived mobile excitons are able to travel to donor–acceptor junctions and develop CT states even in the presence of minimal driving force.120 NFAs handle moderate Eb (∼0.2–0.4 eV) smaller than fullerenes but meaningful reducing Eb (via screening or delocalization) remains also a primary approach to efficient charge creation. Table 4Optical and excitonic properties of representative NFAsNFA (type) λ max (solution) [nm] λ max (film) [nm] ε max [105 M−1 cm−1] E opt [eV] E b [eV]ReferencesITIC (IDTT)∼705∼722∼1.01.76∼0.3 119 ITIC-4F∼720∼740∼1.21.72∼0.3 121 Y6 (BTP-4F)∼730∼820>1.51.51∼0.2 105 Optical and excitonic properties of representative NFAs TD-DFT based excited-state analysis The main tool of studying NFA excited states is time-dependent DFT. Hybrid functionals and medium-size bases are typically studied e.g. the UV-vis spectrum of the acceptor Y6 was calculated with 6-31G (and Def2-TZVPP) being used in the hybrid functional A–DA′D–A and reproducing its spectrum. In one scenario the lowest two absorption peaks of Y6 (730 and 590 nm) were designated as HOMO to LUMO transitions with the highest oscillator strength is at 730 nm.122 A solvent model or dielectric is added to many works to simulate environment. Charge-transfer character range-separated functional are frequently used such as TD-CAM-B3LYP/6-31G(d) was used to examine a branched D–π–A molecule to correctly determine a CT-dominated S1 state. Validation experiments show that tuned hybrids (ωB97X-D, M06-2X) provide comparable excitation energies and prevent gross underestimation of CT transitions.123 Informed excited-state descriptors are also obtained in such TD-DFT calculations. The hole–electron distribution can be plotted using natural transition orbitals (NTOs). The calculated HOMO and LUMO in Y6 are −5.63 eV and −3.66 eV (gas-phase B3LYP) respectively with a gap of 1.97 eV as seen in Fig. 9. Both orbitals are delocalized around the central fused core and end groups with strong ICT. The S1 excited state is therefore mainly HOMO–LUMO in line with the computed 730 nm transition. The extent of CT character is further evaluated by the analysis of transition numbers and charge-transfer numbers. TD-DFT offers structure–spectral relationships (excitation energies, oscillator strengths and visualizations (orbitals/NTOs)) which can be used to analyze and interpret structure alongside spectral properties in NFAs.124 (a) Molecular structure of the benchmark non-fullerene acceptor Y6 and (b and c) are the results of TD-DFT calculations of the electron density distributions in the HOMO and LUMO states, respectively. Spatial separation and delocalization of frontier orbitals in an intramolecular charge transfer mechanism explains the excellent photovoltaic properties of Y6-based organic solar cells. Modified and redrawn from ref. 124. Effect of molecular structure on light-harvesting efficiency The structure of the NFA as it is detailed determines its bandgap, shape of absorption and therefore light-harvesting. Red-shifting the spectrum is caused by increasing conjugation and stronger electron-withdrawing end groups. Fluorinating ITIC to produce ITIC-4F causes the 0–0 peak of ITIC-4F to move to around 720 nm whereas an 800 nm absorption in Y6 is achieved by the use of a benzothiadiazole core in the core. Adding heteroatoms also decreases the gap the planarity is also important twisting side chains decrease the π-overlap and blue-shift absorption but keep the coplanarity at the same time as the linear alkyl chains. Planar rigid NFAs with minimal torsion produce the largest ε and widest spectra in general.116 The resulting structural effects are used to convert light-harvesting to the performance of the device. Reduced optical gaps provide the ability to harvest more solar photons that increase short-circuit current. A large ε minimizes the thickness of the absorber needed. The machine-learning analysis proves that NFAs that are linear with planar backbones and contain a high concentration of heteroatoms possess the largest εmax. In this way, the absorption spectrum and strength can be systematically tuned by rationally engineering the structure of NFA (core, end groups, side chains). Optimizing these parameters is a key focus of high JSC and high-efficiency OSCs design in practice.121 Charge-transfer states and exciton dissociation When excitons are produced, they have to dissociate through CT states at donor–acceptor interfaces. In an NFA crystal or aggregate CT excitons typically occur at lower energies than the pure Frenkel excitons. DFT-studies of Y6 dimers-crystals indicate the existence of two CT states which are lower than the two local exciton (LE) states. These LE and CT states are hybrids to create a manifold of mixed excitations which extends across molecules. The high FE-CT coupling (∼100 meV) means that an absorbed photon has the potential to directly reach inter-molecular CT character. At a donor–acceptor heterojunction the common image is used the singlet exciton of the acceptor diffuses to the interface and an electron or hole is transferred to the donor to create a bound CT exciton. The offset between LUMOs and LUMOs (HOMOs and HOMOs) is the driving force but NFAs can work with small offsets. In low-offset NFAs ultrafast measurements and theory show that charge generation yields are almost unity even with small energy differences. This has been explained by their crystalline packing and quadrupole moments that delocalize charge and form internal fields. Therefore, dissociation of even the low-energy excitons (offset energy of 0.3 eV) is possible. More importantly the binding energy of the CT intermediate (∼0.1 eV) is less than the original Frenkel exciton that could be thermally separated. Nevertheless, certain experiments note that a percentage of excitations relax to polaron-pair (bound CT) states instead of free carriers. The magnetic resonance of Y6 films concludes that the bound e–h pairs of adjacent molecules of the films are the primary contributors to post-excitation polaronic signals, rather than the free charges. In blends donor influence and energetic offsets contribute to separate these CT pairs. The combination of longer exciton diffusion lengths and lower exciton binding delocalized CT character results in the overall efficient dissociation of NFA-based OSCs.105,115 Morphology and interfacial effects NFAs have transformed organic solar cells by allowing them to absorb much more and to tune the energy. Their miscibility with polymer donors is high and requires a strict morphological control.125 The review examines how the exciton splitting, charge transport, recombination and device measures are governed by a combination of donor–acceptor mixing (χ parameter), crystallinity and phase separation, specific intermolecular locks (S⋯O, F⋯H, π–π) and interfacial energetics. We describe experimental and theoretical investigations of how Flory–Huggins miscibility is related to domain purity and fill factor and how the processing of crystallites tunes the size and coherence of crystallites (with examples of GIWAXS) and how non-covalent interactions are used to maintain planarity of molecules and alignment of energy. Lastly, multiscale simulations between DFT and coarse-graining and phase-field which seek to determine morphology–performance relations. The combined observations point to the need to optimize NFA-based BHJ morphology to achieve high charge production and low recombination losses.126 Donor–acceptor miscibility The Flory–Huggins interaction parameter χ is used to quantify the thermodynamic miscibility between donor and NFA, values of χ large positive values indicate phase separation whereas values of χ near zero indicate mixing. NFAs tend to be very compatible with polymers that give very mixed (low-χ) films. Ye et al. demonstrated that sufficient large x (strong demixing) is only correlated with large fill factor127 and Ghasemi et al. demonstrated high x systems in fact do form more kinetically stable morphologies (slower diffusion).126 Practically, low blends form small intermixed domains with large D–A interface and at the expense of trap sites, higher blends form larger pure domains with enhanced percolation. The binary mixture in PM6 : BTP-4F-C9 systems in which the PCE was 16.5% showed that the addition of a low concentration of harder acceptor Y7-BO increased the χ coarsened morphology and increased PCE to 17.3%. Comparative addition of a higher-χ NFA (ITCPTC) to PM6 : Y6, in a similar way, expanded domains and enhanced PCE by increasing 16.4% to 17.4% as indicated in Table 5. Fine-tuning miscibility is known to have a strong effect on domain purity and charge transport pathways.128 Table 5Donor–acceptor miscibility and device impact in recent studiesBlend (composition) χ (miscibility)Morphology/domainPCE (FF)ReferencesPM6 : BTP-4F-C9-20 (1 : 1) χ low (good misc.)Fine intermixing, small domains16.5% (≥0.60) 125 PM6 : BTP-4F-C9-20 : Y7-BO (1 : 1 : 0.1) χ↑ (Y7-BO higher)Enlarged domains, purer phases17.3% (↑FF) 125 PM6 : Y6 (1 : 1.2)Moderate χReference (mesoscale mixing)16.4% (ref) 125 PM6 : Y6 : ITCPTC (1 : 1.2 : 0.1) χ↑ (ITCPTC higher)Larger domains, better bicontinuity17.4% 125 PM6 : Y7-BO (1 : 1.2)Moderate χReference16.0% (ref) 125 PM6 : Y7-BO : BTP-S2 (1 : 1.2 : 0.2) χ↑ (BTP-S2 higher)Coarser BHJ, improved pathways18.16% 125 Donor–acceptor miscibility and device impact in recent studies Crystallinity and phase separation Molecular ordering in BHJ films is probed by GIWAXS. Donor and acceptor have distinct lattice planes typically the (100) lamellar spacing (∼15–20 Å) from side-chains and the (010) π–π stacking (∼3.4–3.6 Å) from the conjugated cores. Coherence lengths from Scherrer analysis quantify crystallite size. Processing strongly affects both thermal or solvent annealing and additives (DIO, CN etc.) can dramatically increase Dcoh. Small amounts of 1-chloronaphthalene (CN) roughly quadruple the PBDB-T lamellar Dcoh (face-on (100) stacking) from ∼4 to ∼18 nm and increase NFA lamellar coherence from ∼5 to ∼30 nm. The π–π coherence (∼3–4 nm) often saturates but its population intensity grows as polymer backbones straighten shown in Table 6. Longer Dcoh and face-on orientation improve mobility. Overly large crystals can reduce exciton splitting by enlarging pure domains. Thus, controlled crystallization is key to optimizing charge transport and recombination balance.129 Table 6IWAXS-derived crystallinity metrics for representative donor and NFA materials, with or without processingMaterial (orientation)Reflection (hkl)Out-of-plane vs. in-plane D coh (no additive) D coh (with 1% CN)ReferencesPBDB-T (lamellar, IP)(100)In-plane (face-on)∼4 nm∼18 nm 129 PBDB-T (π–π, OoP)(010)Out-of-plane (face-on)∼3 nm∼5 nm 129 NFA (π–π, OoP)(010)Out-of-plane∼4 nm∼4 nm 129 NFA (lamellar, IP)(100)In-plane (face-on)∼5 nm∼30 nm 129 IWAXS-derived crystallinity metrics for representative donor and NFA materials, with or without processing Non-covalent interactions (S⋯O, F⋯H, π–π) NFREAs frequently have non-covalent locks that make them planar and stack optimally. The TT-X-2F series depicts this. In the case of X = S (red dashed lines) intramolecular S 6 contacts flatten the backbone producing the largest possible overlap of the end-groups (blue). TT-S-2F indicates the greatest terminal-group overlap and exciton diffusion length with PCE of 15.29% significantly higher than TT-O-2F (PCE of 10.43%) with a twisted backbone. Likewise, planarity (PCE 12.23%) is improved by Se⋯O locks (X = Se) as illustrated in Fig. 10. Another design motif that adds –CF3 groups to enhance F-H bonding in the blend tightening increases the π–π stacking. The incorporation of CF3 into a 3DTBO : IEICO-4F mixture enhanced the π-stacking and increased the PCE which was 15.49% to 16.64%. These particular interactions (S⋯O, F⋯H, O⋯S, etc.) adjust the alignment of the molecular orbital and decrease the energetic disorder.130 Planar NFAs produce strong π–π stacking resulting in high intra-stack electronic coupling and hydrogen bonds or chalcogen contacts fix relative orientation between atoms to increase charge mobility and reduce recombination. The molecular structures and packing characteristics of acceptors with O, S and Se heteroatoms in the solid-state. The figure emphasizes that non-covalent interactions such as chalcogen bonding and conformational locking contribute to the planarity of molecules, increased π–π stacking, improved charge transport and increased photovoltaic performance in non-fullerene acceptor systems. Modified and redrawn from ref. 130. Impact on charge generation and recombination Exciton dissociation and charge recombination are directly dependent on morphology. The generation of charges should be efficient, i.e., the photo exciton must diffuse to a D–A interface within their diffusion length (∼10 nm). Exciton splitting can be enhanced with very fine mixing high D–A interface but charges can be trapped in disordered regions. On the other hand, crystalline domains of higher purity χ larger crystals promote charge transport at the expense of interface area. According to Ye et al. FF is maximized when χ is large enough to obtain strong phase separation resulting in high carrier mobility.127 High order of polymer NFA in an experiment tends to be associated with elevated FF. In PBDB-T NFA blends with CN additive that enhance molecular ordering enhanced FF as the carrier's percolate with ease. But too rough morphology may be a constraint to JSC. JSC peaked at 1% CN beyond this in the CN study of PBDB-T NFAs as the film had trap-assisted recombination because of the imbalanced mobilities. Recombination reducing the donor–acceptor bandgap offset is also influenced by non-covalent molecular design to suppress non-radiative CT-state losses and increase VOC. The formation of a ternary system with CF3-strengthened H-bonds enhanced the π-stacking leading to the enhancement of charge extraction and reduction in recombination.131 The tradeoff between morphology (domain purity, crystallinity and interface quality) and energetics is to minimize the geminate and bimolecular recombination to maximize the charge yield. Computational modeling of blend morphology OSC morphology has a multi-scale modeling and DFT gives molecular properties of individual NFAs. DFT is also restricted to single molecules or dimers however, atomistic MD is capable of simulating nano-scaled D–A mixtures and early-stage phase segregation with appropriate force-fields provided. Coarse-grained (CG) MD models simplified bead-spring models (e.g. MARTINI) to access larger volumes (hundreds of nm, 100 times shorter times) that form mesoscale domains but loses atomic detail. Continuum-phase-field models (e.g. Cahn–Hilliard equations) are models at the micron scale that simulate the process of solvent evaporation or phase evolution by annealing through the entire devices. Trade-offs Each level is optimal electronically with large scale, atomistic MD captures fine scale packing but is small time/size, CG large scale morphology with chemical specificity and continuum models are efficient but empirically-tuned as in Fig. 11. Such methods have been combined like back mapping CG morphologies to compute DFT to connect structural order to electronic couplings. These multiscale models are crucial in the prediction of the way molecular design and processing result in optimal BHJ architectures.132 Multiscale computational modeling framework for OSCs, assembling DFT, atomistic molecular dynamics (MD) simulations, coarse-grained simulations and continuum phase-field modeling. The hierarchy illustrates the relationship between properties at the molecular scale and the evolution of the mesoscale morphology and the predictions of device-level performance in BHJ architectures. Only the data was taken from ref. 132, the figure was generated by the author. Device performance relationships NFAs in organic solar cells have the benefit of tunable power levels and high-absorption that produces high VOC and JSC at a low loss. Planarity and side-chains Structural features fusion planarity and side-chains influence, frontier orbitals and packing thereby correlating with VOC, JSC, FF and PCE. Planar fused cores such as Y6 provide high JSC (>25 mA cm−2) and FF (∼75) with VOC 0.8 V substrate driving forces being established by energetic offsets (donor HOMO-acceptor LUMO gap vs. exciton binding). NFAs are generally able to minimize VOC loss (Eg–qVOC) to about 0.5–0.6 eV significantly lower than fullerenes.115 Loss channels are radiative/nonradiative recombination and trap-assisted both of which are enabled by molecular design. The chemical robustness is a stability factor with NFAs containing labile vinylene linkages (e.g. ITIC) deteriorating under light but rigid cores containing shielded bonds (e.g. IDTBR, fluorinated Y6 variants) being highly photostable. Recent efforts focus on the development of NFAs of high crystallinity and chemistries. These have been successful in operating with estimated lifetimes of decades.133 To examine these structure–property–performance relationships this review section provides an analysis of these relationships in detail summarizing the most important correlations with device metrics, energetic loss mechanisms and degradation pathways. Correlation of molecular properties with VOC, JSC, FF and PCE V OC and JSC are determined by frontier orbital energies HOMO/LUMO and optical gap. A bigger donor–acceptor energy gap (HOMOD–LUMOA) increases the maximum attainable VOC, and a reduced optical gap (Eopt) due to broad absorption enhances JSC. The VOC scales confirm machine-learning studies with the HOMOD–LUMOA difference.134 These levels of weaker electron-withdrawing end-groups on the acceptor are influenced by molecular design to increase its LUMO and enhance VOC. On the other hand, increase in Eopt reduced by prolonged conjugation leads to increased absorption shown in Fig. 12. NFAs have high tunable absorption compared to fullerenes,115e.g. PBDB-T-2F : Y6 cells have external quantum efficiency (EQE) ≈85% to 920 nm giving JSC ≈25–26 mA cm−2. The high JSC of narrow-gap NFAs is thus realized at a trade-off cost in terms of VOC. Schematic illustration of the basic photophysical processes in a non-fullerene acceptor (NFA) bulk-heterojunction organic solar cell. The diagram shows the photon absorption, exciton generation and diffusion, formation of charge-transfer (CT) states, free-charge generation, collection of free-charges as photocurrent (JSC), and other competitive radiative, non-radiative, geminate and non-geminate recombination pathways that together determine device efficiency. Only the data was taken from ref. 115 and 134, the figure was generated by the author. FF and charge transport is related to molecular packing and morphology. Planar, rigid NFAs are tight-fitting in 3D π-stacks that delocalize excitons and decrease Coulomb binding. This results in large internal quantum efficiency and equal electron–hole mobility that allow FF of approximately 70–80. In comparison disordered NFAs either due to twisting or poor ordering experience greater energetic disorder and reduced trapping to reduce FF and JSC. These trends are presented in Table 7 which compares literature representative NFAs it presents their optical gaps, energy levels and device metrics. The fused-core NFA O-IDTBR (Eopt ≈1.5 eV) has VOC ≈0.82–0.85 V and JSC ≈20 mA cm−2 whereas the lower-gap Y6 (Eopt ≈1.3 eV) has VOC ≈0.82–0.85 V. The correlations indicate that the predictability of molecular descriptors (energy levels, gap, packing) to device metrics (VOC, JSC, FF, PCE) exists. The greater the LUMOA or greater the HOMOD, the greater the VOC up to energy losses. Shallower absorption into NIR or narrower Eopt result in more photocurrent and higher JSC. Increased π–π stacking or crystallinity increased FF and less recombination. Optimization appropriate band gap and good packing together give high PCE.133 Table 7Selected NFA energy levels and device metrics. Values are typical for the given donor–acceptor pairing exact values depend on processing115,135,136NFA (donor) E opt (eV)LUMO (eV) V OC (V) J SC (mA cm−2)FF (%)PCE (%)Source (donor/NFA)ITIC1.6−3.90.70–0.8015–1865–7011–13PBDB-T/PTB7-Th + ITIC134O-IDTBR1.5−3.80.80–0.8518–2072–7513–15PTB7-Th/PBDB-T + O-IDTBR4Y6 (A–DA′D–A)1.30−4.0∼0.82∼2573–7715–17PBDB-T-2F + Y62BTP-e1.30−4.0∼0.8326–2875–8018–19PM6/PCE10 + BTP-eUnspecified newer NFA——>0.85>2875+>18Latest records (NREL) Selected NFA energy levels and device metrics. Values are typical for the given donor–acceptor pairing exact values depend on processing115,135,136 Loss mechanisms and energetic offsets Voltage loss in NFA cells combines radiative and non-radiative components. Conventional fullerene OSCs suffer large losses, NFAs can reduce this to ∼0.5–0.6 eV. The PBDB-T-2F : Y6 blend has an optical onset ∼1.33 eV and VOC ∼0.82 V giving ΔE ≈0.53 eV. Delocalized excitons in Y6 lead to very low non-radiative loss. Lowering the CT-state energy by increasing donor–acceptor offset increases ΔVOC, nr. Thus, there is a trade-off too small an offset yields insufficient driving force while too large an offset wastes energy as shown in Fig. 13. Modern NFAs mitigate this by negligible offset designs Y6 operates with only ∼0.05 eV HOMO difference yet still yields high IQE due to its favorable packing.137 Radiative losses obey the Shockley–Queisser formalism a high radiative efficiency raises the radiative VOC. NFAs with rigid structures and narrow gaps generally have lower radiative losses than disordered systems. Non-radiative losses stem from vibrational relaxation in CT states and trap recombination. Minimizing these requires suppressing molecular motions and defects. Highly planar NFAs have slower electron-phonon coupling improving EL quantum yield and VOC. Recent work shows golden rule optimizations new NFAs achieve ΔVOC, nr as low as 0.2–0.3 V.138JSC and FF losses arise from incomplete absorption and recombination if the NFA absorption does not cover the solar spectrum JSC is limited, hence NFAs are designed with complementary donors. Charge transfer losses can occur if exciton binding exceeds offset. Bimolecular non-geminate recombination and traps lower FF. The highly crystalline domains in modern NFA blends reduce trap densities yielding FF >75%.139 Loss channel magnitudes are direct outcomes of molecular properties rigidity and delocalization minimize voltage losses broad absorption maximizes JSC balanced transport minimizes FF losses. Schematic illustration of the energetic offsets and loss pathways in photovoltaics based on organic semiconductors. The radiative recombination, non-radiative recombination, trap-assisted loss and non-geminate charge recombination losses are the processes that reduce the achievable VOC, which is determined by the optical bandgap and the energy level of the charge-transfer (CT) state within the semiconductor. These losses can be minimized by optimizing the device performance through molecular delocalization, reduced energetic offsets and optimized morphology. Only the data was taken from ref. 137–139, the figure was generated by the author. Stability and degradation pathways Chemical structure and environmental factors are critical in the stability of NFA. Photo-oxidation or isomerization is common in NFAs that have labile bonds. Clarke et al. demonstrated that the ITIC based cells experience intramolecular twisting and rupture of bonds under illumination that leads to rapid burn in loss. NFAs such as O-IDTBR have insignificant burn-in as in Fig. 14a, O-IDTBR devices maintain over 90% of original PCE during long illumination periods and ITIC decreases by approximately half.135 Similar findings were made by Dryzhov et al. who found that PCBM-fullerene devices undergo photostability more than fullerene devices when subjected to the same conditions highlighting the fact that not all NFAs automatically perform better than fullerenes with respect to stability.136 The moral is that the key to intrinsic photostability is molecular rigidity and the lack of exposed linkers. There are also morphological degradation and interfacial degradation. The NFA crystallites can be induced to grow or phase-separate forming recombination sites by light and heat. Xu et al. compared the Y6 derivatives in the outdoors where the fluorinated end-groups and longer internal chains dramatically increased stability where the devices with the F-substituted Y6 retained approximately 70% of output after 60 days in severe weather conditions compared with only 40% in the non-fluorinated analog as in Fig. 14b. They found enhanced stability with less trap-state formation and inhibited photo-isomerization.140 Li et al. emphasize the principles of NFA degradation, photo-oxidation of conjugated backbones, morphological instability and interlayer-induced pathways. Molecular design, and device engineering are practical methods to stability. NFAs can be embedded in ternary blends or incorporate inert third elements to pin morphology. Extra intrinsic degradation is alleviated by encapsulation and interfacial passivation layers. Molecular side solutions including substituting the vinylene with less labile bonds steric blocking groups or sulfides/selenides have been suggested to enhance photostability. Modern NFA materials have maximized these factors and some Y-series acceptors have been found to exhibit insignificant burn-in over months to project multiyear operating lifetimes.141 The NFA molecular structure compared to its stability. (A) Under 1-sun illumination, cells based on O-IDTBR are more photostable than those based on ITIC. (B) Y6 acceptors with added fluorinated end-groups and long side-chains maintain ∼70% efficiency after 60 days in the outdoors compared with ∼40% for the non-fluorinated analog. Modified and redrawn from ref. 135. Machine learning in NFA design Machine learning (ML) has become a potent technique to speed up molecular design by identifying intricate correlations between chemical structure, material properties and device performance. Recent research points out that datasets and molecular descriptors quality are key factors influencing the success of ML models. Integrations of experimental measurements with computational data including DFT calculations have been made to create large curated datasets of more than 1000 NFAs. Descriptor strategies span between the standard fingerprints such as Morgan and MACCS to the advanced fingerprints such as graph-based embeddings and polymer-unit fingerprints (PUFp) and physics-informed features such as HOMO/LUMO energies dipole moments and optical band gaps. Various supervised ML algorithms have been used to design NFA such as random forests (RF), support vector regression (SVR), gradient boosting algorithms (e.g., XGBoost) and deep neural networks. The RF and XGBoost tree-based models have been shown to be strong predictive models with reported coefficients of determination (R2) near 0.85 when predicting PCE. In the meantime, deep learning models including graph neural networks and transformer-based models (e.g., abcBERT) are attracting attention due to their capability to capture complex molecular structures with competitive accuracy (e.g., MAE 1.78 to predict PCE). SHAP and LIME are becoming popular methods in identifying the main structural characteristics that affect device efficiency. In addition to performance prediction ML models are common in predicting basic electronic and optical characteristics such as HOMO/LUMO energy levels and band gaps. These properties play a key role in alignment of the donor–acceptor energy and efficiency of the devices. Complementary theoretical analyses indicate that structural changes like A–D–A type NFAs halogenation can red-shift absorption spectra and decrease exciton binding energy to increase photocurrent without a significant voltage loss. ML can now be used to screen many new NFA designs in a single run and generate designs. In modern workflows, molecular generation (e.g., BRICS fragmentation, variational autoencoders) and property-based filtering and ML prediction can be used together as in Fig. 15. Frameworks such as Deep Acceptor are based on generative models and graph neural networks and PCE predictors to find promising candidates in an efficient manner. These methods have searched through millions of combinations of donor and acceptor and have been able to propose new materials with proven high efficiencies experimentally.142–144 Data-driven workflow for designing high-performance non-fullerene acceptors (NFAs) for organic solar cells. (A) Collection of NFA structures and photovoltaic data. (B) Feature extraction using molecular descriptors/fingerprints (Morgan, PUFp, and graph-based representations). (C) Machine-learning model training (RF, SVR, XGBoost, and NN). (D) Prediction of key properties (PCE, HOMO, LUMO, and Eg). (E) Virtual screening based on ΔHOMO, ΔLUMO, Eg, and synthetic accessibility (SA score). (F) Selection of top candidate NFAs with high predicted photovoltaic performance and good synthetic feasibility. Adapted from ref. 142–144. Motivation for ML-assisted molecular discovery NFAs in bulk-heterojunction OSCs have the benefits of high PCE at low cost, flexibility and tunable optoelectronics. PCEs are being pushed to over 18% in particular non-fullerene acceptors. However, discovery is still slow as progress is based on thousands of syntheses and tests on the devices. According to Wu et al. experience-based screening is much needed in the development of OSC with PCE only being known after the torturous design that suggests massive costs in terms of trial and errors.142 On the same note, Sun et al. stress that NFAs provide good absorption and energetics alternative candidates that are even better are time consuming, expensive and challenging to find. Blind search is not feasible in the vast chemical space. Machine learning is a more realistic shortcut that learns latent associations between molecular structure and performance. It is capable of forecasting properties based on the structure that guides chemists in the right direction of promising scaffolds. Sun et al. note that the ML is capable of predicting material properties and shortlisting candidates by identifying patterns in extensive data sets that can provide the correct prediction without having to carry out the experiments exhaustively.145 Ahmed and Muhammad Sharif analyze numerous ML successes in OSCs and come to the conclusion that ML speeds up the process of discovery of new materials by searching big experimental/DFT datasets to predict PCE and energy levels. The development lifecycle is shortened by ML since it reduces exhaustive synthesis. ML models can be used to integrate theoretical understanding in such a way that predictions will capture both physics and data trends.146 ML is driven by the necessity to effectively search the huge chemical space using minimal experimentation resources that allow NFAs to be found quickly. Dataset generation and descriptor selection High-quality data form the foundation of reliable and robust machine learning (ML) models for non-fullerene acceptor (NFA) design. Most ML studies compile large NFA datasets from published literature, experimental databases and computational studies. Table 8 summarizes representative datasets employed in recent ML research, highlighting their size, source and composition. Dataset preparation is a critical step that involves data cleaning, standardization and aggregation to ensure consistency and reliability. For example, when multiple power conversion efficiency (PCE) values are reported for the same molecule, the highest experimentally validated value is often selected to minimize inconsistencies arising from variations in device fabrication and testing conditions. The predictive performance of ML models is highly dependent on the quality and balance of the underlying dataset. Consequently, researchers emphasize the use of curated datasets containing experimentally verified measurements to reduce bias and improve model generalizability. Although simulation-generated data, particularly from density functional theory (DFT) calculations, provide valuable molecular-level insights and can significantly expand the available data pool, ML models trained exclusively on simulated data often exhibit lower predictive robustness. This limitation arises because simulated datasets do not fully capture complex real-world factors such as morphology, charge transport, interfacial effects and processing conditions that strongly influence device performance. Therefore, integrating DFT-derived descriptors with experimentally measured PCE values generally results in more accurate and reliable predictions enabling the development of ML models that better reflect the practical performance of NFAs in organic solar cells.147 Molecular descriptors are also crucial in that they represent the input features of ML models. Initial methods used traditional cheminformatics descriptors including molecular fingerprints, topological indices and simple physicochemical properties. These descriptors are computationally inexpensive and can be useful in predicting. To capture more of the underlying physics a wide range of studies include quantum-chemical quantities such as HOMO/LUMO energy levels, dipole moments and Eg. More sophisticated means of engineering descriptors have been introduced. Three fragmentation-based techniques disperse molecules into functional units to more accurately describe structural motifs of interest to device performance. Polymer-unit fingerprints (PUFp) explicitly encode acceptor–donor sequences in conjugated systems whereas graph-based deep learning models e.g. BERT-like (e.g., abcBERT) are directly trained on the relationship between atom bonds without prior defined descriptors. Chemical accuracy and model simplicity often trade off using hybrid descriptor sets. Simple filters are used with calculated properties such as polarizability and synthetic accessibility which are then combined with workflows such as Deep Acceptor and then subjected to ML models. There are also features that are physically meaningful such as ΔHOMO and ΔLUMO which are typically included because they have a direct effect on exciton dissociation and charge transfer in OSCs. The objectives of an effective selection of descriptors are to include important structure–property correlations including energy level alignment, light absorption and charge transport and exclude unnecessary complexity that may result in overfitting. This equilibrium plays a critical role in the development of strong and predictive ML models to design NFA.8,148,149 Table 8Representative ML datasets for NFA design142,145,149Reference (year)Dataset/sourceCountFeatures/descriptorsSun et al. (2024)145Experimental NFAs from literature1027 NFAsGraph-based representations (abcBERT)4Liu et al. (2025)150Experimental NFAs (donor–acceptor units)1343 NFAsPolymer-unit fingerprint (PUFp)5Lee et al. (2026)149Lab OSC devices (D1 : D2 : A blends)1010 devices (A from 24 NFAs)Fragmentation-based descriptors (RDKit, STK)7Wu et al. (2020)142Literature D/A device pairs565 D/A combinationsDonor/acceptor structural descriptors9 Representative ML datasets for NFA design142,145,149 Common ML models (random forest, SVR, XGBoost, neural networks) A variety of ML algorithms have been applied to NFA property prediction that all have their own advantages and disadvantages regarding accuracy, interpretability and computational efficiency. RF has become one of the most popular models in the area that has been proven to have a strong predictive performance in various studies. RF demonstrated the highest predictive capability in the material design and screening processes in a detailed benchmark of five algorithms to a dataset of 1343 NFA materials. RF ensemble nature that combines the predictions of many decision trees has inherent overfitting resistance and inherent feature ranking features that are useful in interpreting structure–property relationships. RF models have been demonstrated to be used in conjunction with genetic algorithms to direct molecular design to produce the design with the desired PCE of up to 16.85% in novel donor NFA pairs following iterative optimization.151 Gradient boosting algorithms and specifically Extreme Gradient Boosting (XGBoost) have proved to have amazing predictive power of OSC properties. XGBoost models have been used to estimate PCE of OPV donor materials using structural features extracted using the Harvard Photovoltaic Dataset calculated by DFT with good performance when tuned to the correct hyperparameters. Comparative analysis has shown that gradient boosting algorithms tend to be more accurate than random forest in cases where there is enough training data and LightGBM and CatBoost have occasionally been found to be better in decision tree model comparison than XGBoost. The fact that XGBoost can deal with missing values makes its regularization properties useful to avoid overfitting as well as its ability to deal with sparse features makes it especially useful with NFA datasets that include a variety of molecular structures with different degrees of experimental characterization. Neural networks, both traditional feedforward networks and more advanced deep learning networks have continued to scale the limits of predictive performance of NFA properties. Convolutional neural networks (CNNs) have been used to predict donor–acceptor pairs with 42 molecular structure descriptors with Pearson correlation coefficients of 0.85 on training data and 0.83 on test data on PCE prediction significantly outperforming more traditional ML models like random forest and gradient boosting. In addition to basic prediction, more sophisticated frameworks that combine large-scale pretrained networks of graph neural networks (GNNs) with GPT-2 based reinforcement learning strategies have been created to create OPV molecules with possibly high PCE that show the potential of deep learning with generative models. Support Vector Regression (SVR) is also a useful model especially on small data sets where its kernel-based model can discover non-linear relationships without a lot of hyperparameter optimization although it tends to be worse in performance compared to ensemble models in the presence of large high-quality datasets.148,152 Property prediction (PCE, energy levels, absorption) In most of the studies, the main focus is on PCE prediction. ML models are trained to predict molecular descriptors to device efficiency usually through empirical models such as Scharber or directly to experimental PCE. AbcBERT model on hold out PCE data had an R2 of about 0.67. Lee et al. noted XGBoost models having R2 = +0.85 with PCE. Zhu et al. confirmed that ML predicted PCE of new NFAs agreed with experiment had a relative error within 3%.152 It is remarkable how accurate that is considering the variability of device data. ML predictions may be compared to experimental splits or through cross-validation. Components parameter models can also be used to indirectly predict component parameters such as VOC or JSC through intermediate calculations. Besides PCE, one of the priorities is the energy level prediction. Correct alignment of the bands of donor and acceptor is critical to OSC. A GNN was trained in Deep Acceptor pipeline to predict HOMO and LUMO of each NFA to calculate offsets against a fixed donor the predicted HOMO/LUMO are used. These offsets should have thresholds to provide efficient separation of charges. Direct target models can also target bandgap Eg. The calculation of Eg of NFAs by Patel et al. using DFT indicated that most low gap (1–3 eV) correlates with long absorption. In certain ML workflows the gap of the NFA is predicted or computed to filter candidates. The training can be fed with experimental correlations.153 The importance of optical absorption properties is also that they control the present generation. Although ML is not often used to predict the spectra, it can guess the trends through descriptors. Theoretical works demonstrate structure-absorption correlations, e.g. Patel et al. demonstrated that a bathochromic (red) shift of absorption and a decrease in exciton binding occurs when NFAs are halogenated.148 These alterations enhance light collection with minimal voltage loss. This is implicitly captured by ML models when the features include conjugation length or electron-donating/withdrawing motifs. Additional features such as computed oscillator strengths or simulated spectra are even used on some approaches. ML prediction of property involves PCE and its factors. Models generally model PCE directly but the success requires the correct modeling of underlying physics. Virtual screening and high-throughput design Virtual screening has risen as the practical climax of ML-assisted NFA design to allow the rapid screening of large chemical spaces to find the most promising compounds to be synthesized. The magnitude of such virtual screening programs is mind-boggling recent frameworks using CNN-based predictive models with transfer learning-generated molecular libraries have already produced over 260 000 donor molecules and 930 000 acceptor molecules with high-throughput screening thereafter finding over 12 000 donor–acceptor affinities predicted to achieve PCE above 19%.154 This is a paradigm shift to the point where hundreds of candidates are screened in the traditional experimental campaigns to the point where millions of virtual molecules are evaluated prior to any synthesis taking place. The most promising candidate out of such screening attained a predicted PCE of 19.20% depicting the amazing potential of ML-guided high-throughput design. Pipelines of generative design which combine several ML elements have shown to be especially effective in the discovery of NFA. An integrated method that uses both generative and predictive ML models and expert knowledge has been effectively applied to generate virtually screen and synthesize possible NFAs to high-efficiency OPVs. This closed-loop system optimized synthesized molecules by repeatedly repeating the synthesis of molecules to achieve synthetic feasibility, with diketopyrrolopyrrolopyrrole (DPP) core motifs being used to locally assemble candidates of NFA with high synthetic standards. A predictive ML model on the basis of PCE calculations was used to conduct virtual screening which was able to identify seven NFA candidates, all of which needed three or fewer synthesis steps. The predictive power of the integrated ML pipeline was validated by experimental validation of HOMO/LUMO measurements that resulted in calculated PCE values of between 6.7% and 11.8%.155 Another high-throughput approach utilizing genetic algorithms (GAs) with ML fitness functions is genetic algorithms that replicate natural evolution to search chemical space effectively. On unfused NFA discovery, GA methods have demonstrated speedups of up to 1.8 × 1012 over brute-force screening of 5426 new NFAs of which 1087 were predicted to have PCE over 18%. It was discovered that better performance was possible in the presence of a donor core and acceptor end groups with actionable design rules to guide the future development of NFA. The method of interpretability like SHAP (SHapley Additive exPlanations) analysis has become a part of high-throughput screening of the essential molecular substructures like the number of halogen substituents and polycyclic ring structures, which are strongly correlated with high PCE. Not only do these insights inform screening but also uncover core structure–property relationships which inform rational molecular design in general beyond the particular candidates identified.156 Integrated computational-machine learning frameworks OSCs design has been taking advantage of computational and ML design. The features in these pipelines are quantum-chemical descriptors calculated by either DFT or semi-empirical methods and used to train regression models that predict device metrics such as PCE. The use of advanced feature engineering and explainable ML (SHAP) allow quick-screening virtual NFAs. Zhu et al. showed ML models which informed the design of three new NFAs the experimental PCE of which compared within 3% to ML predictions.157 At the same time generative inverse-design techniques suggest new NFA chemistries to be optimized by ML-predicted properties. These methods are extremely fast discovery Greenstein et al. were able to find thousands of new acceptors with an evolutionary algorithm158 and Ramalingam et al. recently created seven DPP-based NFAs that are estimated to achieve 11.8% PCE.155 ML-augmented frameworks swap a small predictive uncertainty with orders-of-magnitude throughput and chemical space exploration pointers towards quicker data-driven OSC development. DFT-guided ML pipelines Modern NFA design DFT and other quantum techniques produce descriptors that include the HOMO/LUMO energies, optical gaps, dipole/quadrupole moments, reorganization energies and oscillator strengths of the electronic structure. These features are the input to the ML models predicting OSC performance. Information is usually summarized by reports of experimental devices or by virtual libraries. One-hot or fingerprint encodings, molecular graphs and even sub-unit contributions are all examples of feature engineering. The typical ML algorithms are random forests, gradient-boosted trees, kernel methods and neural nets that are trained to map descriptors to PCE or energy levels. Cross-validation or hold-out test sets are used to validate, e.g. Zhang et al. applied leave-one-out cross-validation to guarantee the generalization.159 Explainable AI methods are used to explain the models that Zhu et al. used SHAP to explain the important molecular fingerprint bits that affect PCE.157 The constraints of such pipelines are the small heterogeneous data; the overfitting and domain shifts are possible so it is recommended to clean the data and prevent inconsistencies. In fact, Zhang et al. introduced a data-cleaning step to eliminate inaccurate entries and designed sub-unit descriptors which reduced the errors of HOMO/LUMO predictions by up to 0.2–0.5 eV down to 0.06 eV as in Fig. 16.159 In a similar manner, semi-empirical shortcuts are able to produce descriptors much more quickly and yet equally accurate. The state-of-the-art DFT + ML pipelines focus on high-quality computed descriptors which are strong features selection and rigorous validation to predict NFA properties reliably. (a) Chemical structure of the benchmark non-fullerene acceptor (NFA) Y6 and (b) three sub-units of Y6 (donor-core, π-bridge and terminal acceptor) used by Zhang et al. to generate descriptors. The sub-unit-based machine-learning approach is more efficient in describing the local structural features which gives better prediction accuracy for frontier orbital energies (HOMO and LUMO). Redrawn from ref. 159. Comparison of traditional vs. ML-accelerated approaches Traditional computational predictions of NFA designs are long recognized as the gold standard of predicting electronic properties and guiding material development based on DFT and TD-DFT. These solutions have natural shortcomings when used to perform large-scale virtual screening. DFT which is based on a computationally expensive method is steeply dependent on the size of the system rendering complete search of the NFA chemical space impractical. More conventional techniques frequently demand a high level of quantum chemistry and special software such that access is restricted to experimenters. Single NFA molecule High level TD-DFT calculations are prohibitively expensive and take hours to days on high-performance computing clusters. Approaches accelerated with ML can make predictions about important properties of new molecules on the order of milliseconds to seconds on standard hardware with several orders of magnitude of speedup.160 A recent comparative study showed that a SISSO enhanced ML framework could predict PCE with experimentally grade accuracy (MAE = 0.43%, R2 = 0.92) with only 17 interpretable quantum-chemical features and 40 times faster screening than traditional methods. ML models can implicitly model the intricate structure–property relationships that are challenging to obtain using first principles such as subtle effects of molecular conformation, packing and device-level parameters. The methods of ML acceleration also have their limitations. The quality, size and diversity of the training set is inherently connected with the accuracy of the ML predictions. Weakly maintained or biased data may result in models that do not generalize outside the training distribution which may be missing new chemical spaces. ML models can be thought of as black boxes which is why it is difficult to extract physical insights unless using interpretability methods. A major challenge in quantum mechanical rigor and ML efficiency thus constitutes a major frontier requiring methods that are both scientifically interpretable and that surmount computational bottlenecks. Balanced solutions to solution would be hybrid frameworks which integrate DFT to produce high quality training data and ML to produce quick predictions. Recent contributions in physics-guided ML in which known physical laws and constraints are incorporated into the model architecture or training procedure are contributing to the interpretability and generalizability issues.161 Recent efforts to create open-source computational architectures that address key QM/ML interoperability issues like automated transfer of parameters between Gaussian, ORCA and ML potential engines further lower the multiscale simulation setup time and keep error margins below 0.1 eV as demonstrated on benchmark systems. The decision between the conventional and ML-accelerated approaches is determined by the particular research question. DFT is necessary to be able to benchmark and perform fundamental studies, and ML is performing high-throughput screening, inverse design and rapid property prediction with integrated frameworks that provide the best of both worlds.162 Future perspectives and conclusions Theoretical modeling, multiscale modeling and data-driven approaches will form the basis of future efforts to develop NFAs in organic solar cells. It is necessary to move beyond the single-molecule analysis of systems to frameworks that bridge quantum chemistry, morphology and device physics. Enhanced modeling of non-radiative losses, exciton diffusion and the movement of charges involves consideration of electronic coupling, disorder and realistic morphologies. Machine learning will hasten the process of finding effective NFA materials. It is essential to design NFAs which balance efficiency, stability and scalability. Although efficiencies have almost reached 20, there are still problems such as degradation, instability and complex synthesis. Theoretical knowledge about stability and energy alignment will be used to reduce losses and achieve efficient charge separation with high performance. The structure–property–performance correlations of NFAs are now understood more clearly including the roles of molecular planarity, donor–acceptor interactions and alignment of energies. Predictive design remains in need of coherent models which connect molecules to devices. By combining computational techniques, data-driven experiments and applications will speed up the creation of efficient, stable and commercially viable organic solar cells. Recent advances in computational chemistry, density functional theory (DFT), and machine learning have transformed the rational design of functional materials by enabling accurate prediction of structure–property relationships.163–167 These theoretical approaches provide valuable insights into electronic and physicochemical properties across diverse material systems and offer an efficient framework for accelerating the development of high-performance non-fullerene acceptors for organic solar cells.168–172 Conflicts of interest There are no conflicts to declare.