Emerging materials for resistive switching memories: Prospects for enhanced sustainability and performance for targeted applications

The recent breakthroughs in large language models, such as GPT-4, and generally in modern artificial intelligence (AI) applications highlight the demand for implementing big data processing in an energy-efficient manner. Computing architectures based on Complementary-Metal–Oxide–Semiconductor (CMOS) technology inherently suffer from high power consumption that, coupled with unavoidable scaling down issues,¹ constitutes a hurdle toward enhancing further data processing capabilities at a low energy cost.² Resistive switching (RS) memories are emerging two-terminal (2T) non-volatile devices employed for brain-inspired, in-memory-computing while being compatible with CMOS integration.³ Crossbar arrays of multiple RS devices in a deep neural network configuration could be used to handle large datasets associated with Internet of things (IoT) applications.^4–6 These memristive crossbars offer a massive parallelism approach in computation, resulting in more efficient computing,⁷ overcoming the von-Neumann bottleneck where the memory and the processing unit are separated,^8,9 and leading to enhanced bandwidth¹⁰ with less data latency.^11,12 Besides data processing and storage, applications of resistive switching (RS) memories include data encryption and security,^13–17 neuromorphic vision based on photonic RS memories,^18,19 artificial synapses and neurons,^20,21 nociceptor,^22,23 Boolean logic,^24,25 classification tasks that include pattern and image recognition,^26–29 and neuromorphic sensory systems.²⁹ Optoelectronic RS memories are also capable of further reducing the power consumption in neuromorphic computing.^30,31 Extensive and ongoing research efforts have focused on improving the performance of current RS technologies by applying various material and device optimization strategies.^32–34 

Apart from the figures of merit that determine device performance, sustainability is another aspect that should be taken into account toward the commercialization of RS memories, addressing issues at the material, device processing, and system integration levels. Several types of materials have been evaluated in terms of RS properties and device performance. Among them are oxides exhibiting state-of-the-art switching speed in the ns range, while various other materials have been exploited, including organic materials,³⁵ 2D materials,³⁶ metal halide perovskites (MHPs),³⁷ and other bio-compatible materials.³⁸ Each of these materials exhibits different advantages and drawbacks, while investigating different materials and implementing heterogeneous integration of them could enable additional functionalities. In this regard, sustainable RS systems should be based on low temperature fabrication processes,³⁹ namely, below 150 °C, while the whole manufacturing should be as cheap as possible, minimizing both environmental and economic impacts. Solutions-based processing based on upscalable deposition techniques such as spray coating⁴⁰ or other advanced printing⁴¹ could be a solution to the required sustainability, at least for specific applications, while compatibility with flexible substrates is another advantage of solutions-based processing, such as for wearable devices.⁴² In most cases, state-of-the-art oxides require high processing temperatures as expensive and complex deposition techniques, such as chemical vapor deposition (CVD), are employed,⁴³ affecting the sustainable integration and upscaling process of RS devices. The environmental impact of the material precursors is also another parameter that should be considered, as many of these materials have very high levels of toxicity as in the case of MHP, where lead (Pb)⁴⁴ is used. RS systems based on environmentally friendly materials have been developed to address this issue.⁴⁵ 

In this Perspective, we investigate the sustainability aspects of available RS memory devices with a particular focus on printable, solutions-based devices. Solutions-processed RS memories have the potential for implementation in large-area devices using relatively cheap and printable techniques. First, we provide information about the fundamental properties and switching mechanisms of RS devices and the requirements for their commercial development. We focus on five main material categories: oxides, perovskites, organic materials, 2D materials, and eco-friendly, bio-compatible materials. We provide examples along with the advantages and limitations of each material technology, suggesting possible solutions that could enable sustainable, highly performing RS devices for neuromorphic computing. Finally, devices enabled by these sustainable materials are compared with conventional RS systems, and challenges toward commercialization are discussed.

Resistive switching memories are emerging memory elements for neuromorphic computing. The recent advancements in the field, along with the burst of artificial intelligence, make RS devices hold strong potential. A timeline illustration of some significant achievements in the field of RS devices is depicted in Fig. 1. At the device level, non-volatile RS memories [Fig. 2(b)] for practical applications should exhibit properties such as high ON/OFF ratios and extended durability. RS memories switch back and forth between a high resistance state (HRS) and a low resistance state (LRS), which enables information storage at different states. The switching speed or write/erase speed, which is the minimum time interval required for the device to switch between states, is another important figure of merit. Ideally, an ON/OFF ratio of >10 is needed, along with millions of endurance cycles with years of retention, and the write/erase speed should be in the ns range.⁴⁶ The cycling endurance is defined as the maximum write–read–erase–read a device can withstand without losing its ON/OFF ratio and without deterioration of either HRS or LRS. The state retention is a measure of the non-volatile properties of the device and is the maximum time that each HRS and LRS is maintained in the absence of an electrical stimulus to switch the device. These are the fundamental properties that a RS system should exhibit before its commercialization process. RS devices based on their switching mechanism are divided into (i) bipolar,⁴⁷ (ii) unipolar,⁴⁸ and (iii) Write-Once-Read-Many (WORM) [Fig. 2(c)] devices.⁴⁹ Bipolar devices require switching voltages of opposite polarity to switch between HRS and LRS, thus implementing the SET and RESET process, where SET is the transition from LRS to HRS and RESET is the reverse process. In the case of unipolar devices, both SET and RESET occur using the same voltage polarity. Finally, WORM RS memories can permanently store information as they transit from HRS to LRS while repeatedly reading the state of the resistance, and they remain in LRS⁵⁰ until the complete failure of the device.

Apart from the non-volatile RS devices, RS devices with a volatile I–V characteristic curve have been demonstrated [Fig. 2(a)]. Volatile switching or threshold switching is induced by transient ion diffusion. In this case, the device is initially in the HRS. When voltage is swept from 0 to positive or negative bias only (without changing polarity), the device reaches a threshold voltage and transits to the LRS. Then, during scanning voltage from positive values to zero, the device jumps back at the HRS; hence, volatile devices do not retain information unless a constant electric field is applied. This property either is inherent⁵¹ or can be modulated by controlling the maximum value of the applied compliance current (CC). A low compliance current will lead to the formation of weaker conductive paths or conductive filaments (CFs), while increasing its maximum value will allow the formation of stronger CF enabled by the associated stronger Joule heating effect. Overall, volatile memristors exhibit a transient volatile behavior resulting from the gradual filament rupture in the absence of an external electric field due to ions/carriers’ diffusion processes. The compliance current tuning method can be used to control the current flow in the device and, thus, tune the interplay between non-volatile and volatile switching.⁵² Volatile memristors are suitable to emulate neuronal behavior and specifically the biological spike firing process, where input electrical signals are processed and integrated, and when a threshold condition is satisfied, the device fires a new signal in a neighbor neuron termed action potential. This neuron model is defined as Integrate and Fire (LiF) and is the most biologically plausible neuron model.⁵³ This behavior can be used to emulate spiking neural networks (SNNs).

The switching mechanisms for the observation of RS are mainly categorized into two types, namely filamentary- and interface-mediated switching.⁵⁴ For filamentary switching, a conductive bridge is formed between the top and the bottom electrode when applying voltage to the top electrode, which sets the device in the LRS. CF can be formed by active metal cations such as Ag⁵⁵ or by vacancies in the active layer, such as vacancies of oxygen or halide ions in the case of oxides and perovskites RS, respectively. Vacancy-based CF can be confirmed using an inert top electrode as well.^56,57 Consequently, when applying a voltage of opposite polarity than the SET voltage, the filament is ruptured, and the device returns to the HRS. During filamentary switching, an initial current voltage step sweep from zero to a positive bias is usually required to set the device from the HRS to the LRS (termed formation process). The voltage for the formation process is often larger than the SET voltage. In some cases, RS behavior can be directly obtained without the electroforming step, and these devices are considered as forming-free.⁵⁸ For preventing a large amount of current flow in the device, causing permanent damage to the active layer or neighboring layers, a current limit is set as CC. In many cases, RS can occur abruptly, which is a common behavior in the case of filamentary switching. On the other hand, RS can occur progressively without any abrupt change. These two cases correspond to digital and analog behavior in RS memories, respectively, as shown in Fig. 3(a). Figure 3(b) represents the I–V characteristics for the cases of volatile, non-volatile, and interface switching with equivalent diagrams for the ion dynamics in each case, where metal and vacancy filaments are presented for the first two cases. For interface-mediated switching, controlling the Schottky barrier height at an interface (e.g., active layer/electrode) can induce RS behavior, as the device can transmit to the LRS by lowering the barrier height or at the HRS when opposite voltage is applied, and the barrier height increases, preventing charge carrier flow.

Novel material compounds and modifications focus on expanding the performance of these devices while studying the complex ion dynamics that could explain the principles of operation. On the device level, material choice is the first step. The cost aspect will affect the material choice, while the deposition method and scalability should be taken into consideration, as well as the ability to process them using solutions while avoiding high-temperatures and expensive deposition methods. On the previously discussed performance parameters, all of these figures should be maximized, while the switching speed should be as low as possible. However, the material choice can affect the final performance output, while limitations of each material further reduce performance, such as ambient lifetime stability. The device properties can also affect the targeted applications. For example, a memristive system that possesses analog behavior using electrical pulses can be implemented for artificial synapse emulation. Finally, sustainability is also another parameter that should be considered for the commercial establishment of resistive switching memories. Upscaling of these materials can play an essential part in this process, and printing methods on flexible substrates are a key requirement. Printed methods require careful optimization in terms of the manufacturing process, the deposition parameters, and the ink properties. Several types of materials have been chosen over the years for utilization in resistive switching memory devices. Some of them include oxides, perovskites, organic polymers, 2D materials, and bio-compatible, environmentally friendly materials [Fig. 4(a)]. In some cases, fully printed memristive devices have been demonstrated as well [Fig. 4(b)]. A detailed and thorough review of the recent developments in printing methods regarding resistive switching memories can be found in Ref. 61.

Oxide-based resistive switching (RS) materials possess unique properties such as cost-effectiveness, ease of formulation, and compatibility with CMOS that make them promising candidates for next-generation electronic devices.^62,63 Their compatibility with conventional microelectronics technologies is essential for the development of energy-efficient memory storage, neuromorphic computing, and other cutting-edge technologies.⁶⁴ Extensive research implemented so far suggests that the migration of oxygen vacancies inside the oxide materials significantly influences the conduction mechanism of memristors.⁶⁵ Some of the most researched materials for memristors, including oxides, where the RS phenomenon has been noticed are depicted in Fig. 5(a), while Fig. 5(b) shows the number of publications per year regarding oxide based RS memories. Transition metal oxides,^66,67 perovskite oxides,^68,69 and rare-earth metal oxides^70–72 are examples of these. TiO₂,⁷³ ZnO_x,⁷⁴ NiO_x,⁷⁵ TaO_x,⁷⁶ HfO_x,⁷⁷ AlO_x,⁷⁸ and CuO_x,⁷⁹ for instance, have been widely used as suitable active layers or switching materials in resistive random-access memory (RRAM) devices because of their simple chemical composition, ability to switch between several states, and compatibility with CMOS.^62,64,80

Conventional methods of synthesizing oxide-based memristors often involve complex and energy-intensive processes, which can contribute to environmental degradation and resource depletion.^81,82 To address these challenges, researchers are pioneering sustainable initiatives for the fabrication of oxide-based RS materials. These approaches encompass a spectrum of strategies, including the utilization of renewable resources, the reduction of hazardous materials, and the optimization of energy efficiency throughout the manufacturing process, which not only benefit the environment but also hold the potential to enhance the performance and reliability of oxide-based memristors.⁸³ 

Thin film deposition is an essential step in many processes, from electronics to energy storage and conversion. A multitude of techniques, including thermal oxidation,^84,85 e-beam evaporation,⁸⁶ atomic layer deposition (ALD),^87,88 chemical and physical vapor deposition (CVD and PVD),^89,90 pulsed laser deposition (PLD),^71,91 cold-pressing, radio frequency (RF) sputtering,^92,93 electroplating,⁹⁴ and printing.^86,95,96 All these techniques have different advantages and characteristics, and they all contribute to the synthesis and optimization of thin films. Out of these, PVD and ALD are the most used techniques for the fabrication of memristors.⁹⁷ In some cases, the electroplating technique is used, while it is considered less ecofriendly compared to PVD and less economical in terms of manufacturing compared to CVD.⁹⁸ 

Traditional binary metal oxides, such as TiO_x,⁷³ HfO_x,⁹⁹ and GdO_x,⁷¹ are frequently used to deposit using techniques such as ALD, PVD, and CVD. Among these, HfO_x and TaO_x stand out among them because of their promising properties, such as their high operating speed and longevity exceeding 10¹⁰ cycles.^100,101 Nevertheless, poor consistency and excessive power consumption are common problems with these oxides.¹⁰² As a solution to these challenges, Liu et al. presented a bismuth-doped SnO₂ memristor with an ITO/Bi:SnO₂/TiN structure. This device showed great promise for low-power memory applications with its low self-compliance current, switching voltage, and operating current.¹⁰³ Precise control over the formation and dissolution of conducting filaments (CFs) is necessary to improve the stability and homogeneity of memristive devices. Moreover, doping the active layer or designing multi-layered structures to modify the Schottky-type barrier profile across different resistance states is often necessary to maximize their performance.^104,105 When compared to undoped ZnO, some mixed oxides doped with Ti, Co, Ga, Al, and Cu in ZnO films have shown lower set voltages, an increased ON/OFF ratio, and improved RS behavior.^106–109 This approach aligns with sustainability goals by enhancing device performance without the need for additional materials or processes. Yu et al. employed a double-layer stacked HfO_x vertical RRAM with favorable switching capabilities using an economical fabrication technique. In addition, the integration of a TiON interfacial layer led to non-linear I–V curves, increasing the LRS resistance for unselected array cells at low applied bias. This study sets the foundation for implementing a 3D cross-point memory architecture with emerging oxide-based RRAM devices.⁹⁹ Similarly, Liu et al. created a bilayer configuration with a HfO₂ layer incorporated between the AZTO layer and the bottom contact with Ti/AZTO/Pt structure, resulting in a notable decrease in operating voltages as well as increased ON/OFF ratios and switching stability.¹¹⁰ Solution processing, characterized by its simplicity and accessibility, enables thin-film deposition from liquid-phase precursors onto substrates compared to vacuum-processed techniques.¹¹¹ Hydrothermal,¹¹² sol-gel,^113,114 and anodization^115,116 methods enable precise control over film thickness and morphology and are less complex and inexpensive.¹¹⁶ Among these, the sol-gel method stands out as the most commonly used. The sol-gel method is the most utilized of them. Kim and colleagues used a sol-gel-processed Y₂O₃–Al₂O₃ mixed oxide as the active channel layer in RRAM devices, with the goal of minimizing leakage current under the HRS condition by introducing an amorphous phase and reducing oxygen vacancy concentration. The YAl-50 film-based RRAM device demonstrated excellent endurance, retaining stability over ∼100 cycles with a high HRS/LRS ratio (>10⁵), and outstanding retention properties, maintaining uniform resistances for up to 10⁴ s without significant degradation.¹¹⁷ In contrast to methods such as PLD or RF, the sol-gel process is more prone to peeling and cracking despite its benefits. To overcome this problem, researchers have simplified the sol-gel process for fabricating titanium oxide (TiO_x) amorphous films in memristive devices. By doing away with the requirement for a dry nitrogen flow stage, this improved synthesis strategy lowers the process’s cost and complexity. This modified synthesis approach eliminates the need for a dry nitrogen flow step, thereby reducing both the cost and intricacy of the process.¹¹⁸ Metal oxide precursor solutions are primarily applied using spin coating, spray coating, dip coating, and printing techniques, with spin coating being the most used.¹¹⁹ Solution processing is mostly favored due to its versatility with a wide range of materials and substrates for both industrial and research applications. Due to its low cost, facile fabrication method, and great versatility, solution-based deposition procedures have become a promising technique during the past decade, showing great potential for future advancements in sustainable memristor technology.^120,121

Recently, the sol-gel spin coating technique has evolved as a promising technique to produce thin-film based devices with its capacity to achieve excellent uniformity across large areas and its simple operation.⁹⁵ Abdul Hadi et al. fabricated Cu/HfO₂/p++ Si memristive devices using a sol-gel spin coating technique and examined their potential for use as radiation detectors. These devices demonstrated high repeatability and ON/OFF ratios up to 10⁴, indicating their potential for radiation sensing with bipolar RS behavior dominated by electrochemical metallization (ECM). They also showed statistical fluctuation in electrical properties such as ON/OFF voltage and resistance due to variations in material and manufacture. The study suggests a feasible approach to achieve affordable and scalable memristor devices, implying their practical application in radiation detection systems.¹²² In a different study, Hu et al. developed TiO₂ thin films with an Al/TiO₂/FTO structure using the sol-gel and spin-coating techniques. These devices exhibited strong non-volatile bipolar memory switching behavior, boasting a high LRS/HRS current ratio surpassing 300 at 1.8 V, endurance over 100 cycles, and retention period exceeding 10⁴ s without substantial decay, attributed to the space charge limited conduction (SCLC) method.⁹⁵ 

Spray deposition produces more viable thin films, providing flexible, economical, and environmentally friendly material deposition onto substrates. This technique, due to its low material waste and compatibility with large-area deposition processes, makes it promising for a variety of sustainable applications, such as flexible electronics and solar cells.^123–125 Tarwal and co-workers developed a simple and inexpensive spray pyrolysis technique (SPT) to fabricate ZnO thin films with a nanoflake morphology. An excellent retention of 10⁴ s and a cycling stability up to 10³ cycles were among the better qualities displayed by the ZnO-based Al/ZnO/FTO based memory devices, which offered prospective applications in light-responsive memory, synaptic devices, and sensor components.¹²⁶ 

Researchers are actively exploring ways to develop flexible, affordable, vacuum-free, solution-processable, and environment-friendly memristors to achieve sustainability. In order to meet the needs of sustainable applications, these memristors are designed to deliver a high ON/OFF ratio, swift switching, low voltage operation, and improved storage stability.^127,128 Printing techniques will lead to the production of RS devices that are economical, environmentally friendly, and flexible. Remarkable foldability, lightweight design, stretchability, and affordability of flexible electronic devices are among its key characteristics. Screen, inkjet, and laser-induced printing methods are frequently employed for the fabrication of these devices.⁶⁴ 

Carlos and colleagues developed a combustion-based process to develop inkjet-printed aluminum oxide (AlO_x) RS devices using a combustion-based method, followed by low-temperature thermal annealing (150 °C). These devices were then subjected to deep ultraviolet (DUV) treatment. This method aims to harness low-temperature thermal annealing (150 °C). This approach seeks to maximize cost-effectiveness and facilitate large-scale manufacturing utilizing the abundant availability of Al₂O₃, thereby optimizing cost-effectiveness and enabling large-scale manufacturing. These devices exhibit bipolar RS behavior with excellent endurance, high reproducibility (95%), and retention time (10⁵ s), as well as an adequate ON/OFF ratio to achieve multilevel cell (MLC) operation with up to four states. The findings shed light on the electrical characteristics of printed metal oxide RS devices processed at low temperatures and underscore their potential for hardware security applications.¹²⁹ 

Awais et al. used electrohydrodynamic (EHD) printing on a polyimide (PI) substrate to fabricate a flexible Ag/ZrO₂/Ag memristor. The devices showed two orders of magnitude high OFF/ON ratio and were uniform and smooth when operating at low voltages of ±3 V. During HRS and LRS, they employed SCLC and trap charge limited current (TCLC) conduction mechanisms, respectively. The device also demonstrated 500 cycles of reversible switching, highlighting its viability for flexible electronics applications.⁹⁶ 

To enhance the performance of memristors, certain researchers have employed electrode engineering approaches to modify their functionality. Although platinum is commonly used, choosing the right electrode material is essential, particularly when looking for sustainability in active layer alternatives. Investigating substitutes for essential raw materials such as molybdenum and noble metals is essential to cut expenses, minimize resource dependency, and mitigate environmental impact during fabrication.⁴³ Pereira et al. used amorphous indium-gallium-zinc-oxide (a-IGZO) memristive devices with Mo contacts on both electrodes, with a structure MO/IGZO/Mo, showcasing good fabrication yield and scalability using a standard photolithography process. These devices displayed electroforming-free RS behavior and Schottky-diode-like properties, offering the potential to be used in brain-inspired computer applications and integration into affordable system-on-panel architectures using commercially available IGZO materials. In addition to low processing temperatures, this permits the utilization of unconventional substrates, such as plastic and paper, crucial for Internet-of-things (IoT) applications.¹³⁰ 

Ahn et al. fabricated Nb/NiO/Nb memristors using well-aligned Nb nano-pin array bottom electrodes. The devices were spin-coated with NiO sheets, which displayed unipolar RS behavior. The use of Nb nano-pin electrodes allows a reduction in forming, SET, and RESET voltages, ascribed to its field-enhancement factor.¹³¹ Tominov and collaborators examined the effects of nanoscale film thickness and electrode composition on RS in forming-free nanocrystalline ZnO films made using a cost-effective pulsed laser deposition method. They determined that a TiN/ZnO/W structure with a forming-free nanocrystalline ZnO thickness is optimal for memristive neuromorphic systems, exhibiting a high ON/OFF ratio and relatively low set and reset voltages. Moreover, using a top electrode increased the ON/OFF ratio and decreased switching voltages. These results offer insightful information for developing sustainable RS nanoscale devices for neuromorphic computing applications utilizing forming-free nanocrystalline ZnO oxide films.⁹¹ 

Electrode-free architectures offer a pathway toward more sustainable applications as an alternative to metallic electrodes in RS memory systems. Kumar and colleagues presented an electrode-free artificial synapse based on V₂O₅/ZnO for transparent neuromorphic computing, fabricated using thermal oxidation and RF sputtering methods. This structure has outstanding optical transparency, excellent stability across several cycles, and RS memory behavior as well as electrical and photonic synaptic activity. It also showed information storage, erasure, and learning-experience behavior successfully.⁹³ 

Although TiO_x, HfO_x, and AlO_x are primarily utilized in traditional RS devices, they suffer from high operation voltages leading to considerable energy consumption and insufficient reliability.¹³² To address these challenges, interfacial engineering has been explored as an alternative approach to improve the performance and reliability of memristors. Thin intermediate oxide layers (a few nanometers thick) are often formed at the metal/insulator interface, which helps to stabilize the switching characteristics of the device by improving the interaction between the metal and the oxide layer. The formation of these oxides depends on factors such as the oxygen affinity of the metal and the chemical stability of the oxide matrix. By controlling the composition and structure of these intermediate layers, memristor performance, including endurance and retention, can be optimized for reliable operation. Wang et al. explored the role of interfacial engineering in Ti/VO_x/ITO memristors fabricated at room temperature. They highlighted how the formation of TiO_x and V_o²⁺ near the Ti/VO_x interface significantly enhances the device’s performance. The V_o²⁺ lowers the barrier for oxygen ion migration, and the TiO_x layer induces a built-in electric field that accelerates the oxygen ion movement, resulting in devices with ultralow switching, a stable ON/OFF ratio, and extended data retention.¹³³ 

Improving the interface properties of the memristor devices by introducing an interface layer based on various materials, such as oxides, transition metal dichalcogenides (TMDs), or graphene, has also been explored to achieve reliability and long-term stability.¹³⁴ For example, Tada et al. inserted TiO_x and TaO_x films between the Cu electrode and TaSiO_y solid electrolyte. They observed that the barrier created by Ti protects Cu from oxidation and also acts as a solid electrolyte, resulting in bipolar switching behavior with a high ON/OFF ratio of 10⁶.¹³⁵ Similarly, Lee et al. introduced a 2D WSe₂ layer to regulate the oxygen vacancies in the Hf_xZr_1−xO₂ (HZO) oxide layer. The resulting Pt/WSe₂/Hf_xZr_1−xO₂ (HZO)/TiN memristor demonstrated bipolar RS with excellent reliability, good endurance (over 2.000 cycles), and retention (10⁴ s). I–V characteristics and an in-depth surface chemical analysis revealed that the WSe₂ layer effectively induced oxygen vacancies at the WSe₂/HZO interface. These vacancies are key to achieving stable resistive switching with minimal device variability.¹³⁶ 

In conclusion, sustainable synthesis techniques for oxide-based memristors appear to offer a potential path toward environmentally friendly electronic devices. Prospects for the future include investigating non-traditional substrates, improving solution-based deposition methods, and optimizing electrode materials. Sustainable practices could enhance environmentally friendly electrical technology while minimizing their impact on the environment and depletion of resources.

In recent years, a wide range of semiconductor materials, including metal oxides, two-dimensional (2D) materials, organic compounds, and biomaterials, have been investigated for the development of memristors. Silicon-based memristors are valued for their longevity, but they require relatively high operating voltages. Metal oxide memristors, on the other hand, offer superior thermal stability; however, the presence of inherent randomly distributed vacancy defects restricts their capacity to be widely used as memory devices. Challenges with complex preparation methods and restricted environmental stability also provide a hindrance to materials such as metal sulfides, MXenes, and organic electronic compounds.¹³⁷ As a result, there is a growing interest in investigating new materials, such as perovskites, especially halide perovskites (HPs), to overcome these obstacles and advance memristor technology further. Perovskites, because of their low formation energy, have a particular advantage over conventional metal oxide material in semiconductors.¹³⁸ This ensures compatibility with a variety of substrates, reduces energy, facilitates synthesis using ambient methods such as spin coating and printing, and integrates readily into various electronic systems.¹³⁸ Compared to other materials, perovskites are very promising for the fabrication of memristors due to their exceptional properties, which include high charge-carrier mobility, response speed, stability, low power consumption, cost-effectiveness, tunable bandgap, high ON/OFF ratio, and mechanical flexibility.¹³⁹ Perovskite memristors significantly contribute to neural networks, artificial intelligence, and emerging technologies such as artificial e-skin and vision, thanks to their ability to mimic the learning, memory, and signal processing functions of the such as system.¹⁴⁰ 

Perovskite materials, consisting of abundant and low-cost elements, such as lead, iodine, and methylammonium, help reduce the environmental effects associated with material extraction and processing. Significant advancements in the study of memristors based on perovskite materials have shown potential for future applications. These materials have been fabricated using a variety of procedures, ranging from conventional solution processing to advanced methods such as sol-gel processes, hydrothermal synthesis, high-energy ball milling, and RF magnetron sputtering.¹⁴¹ These methods enable the synthesis process to be flexible and controlled, resulting in perovskite materials with specific properties that are well suited for memristor applications. High-intensity ball milling and RF magnetron sputtering are two of these techniques that are seen to be less sustainable because they may use more energy, generate more waste, or be needed for particular, less environmentally favorable conditions.¹⁴² Other techniques such as precipitation, atomic layer deposition, and pulsed laser deposition can be employed for the fabrication of memristive devices.⁷⁹ Furthermore, in an era in which perovskite materials dominate cutting-edge photovoltaic technologies, the advent of devices that work as both solar energy harvesters and memristors presents a tremendous opportunity to reduce manufacturing costs and power consumption.^143–146 

Chen et al. developed a quasi-2D (BA)₂MA₄(Pb_0.5Sn_0.5)₅I₁₆ hybrid RS device. The perovskite was deposited on the glass/ITO/PEDOT:PSS substrate using the blade coating method, as illustrated in Fig. 6(a). This method is industrially compatible, enabling the potential for low-temperature (below 150 °C), solution-processed, and large-area perovskite memory devices for neuromorphic computing. The final ITO/PEDOT:PSS/(BA)₂MA₄(Pb_0.5Sn_0.5)₅I₁₆/PMMA/Au exhibited analog bipolar resistive switching behavior [Fig. 6(b)] with a good endurance of 2 × 10³ cycles, a state retention of 10⁵ s, and the I–V characteristics were maintained upon 3 months of storage in an inert atmosphere inside a glovebox. In addition, basic synaptic functionalities of spike-timing-dependent plasticity (STDP), paired-pulse facilitation (PPF), and short-term and long-term potentiation (STP and LTP) were realized. The identified mechanism responsible for the resistive switching is the ion migration along with charge trapping/de-trapping from bulk and surface traps.¹⁴⁷ Tang et al. developed a sustainable method in 2019 for fabricating a perovskite-like memristive device using sol-gel and coating techniques to deposit amorphous SrTiO₃ films. This device, featuring an Au/SrTiO₃/FTO structure, showed extraordinary endurance and a high switching ratio of 10², all while displaying great sustainability.¹⁴⁸ Wang et al. developed an innovative tri-cation organic–inorganic halide perovskite (OHP) optoelectronic coupling memristor. They used the hydrothermal synthesis method for the TiO₂ coating to manufacture the device with an FTO/TiO₂/Cs_0.05(FA_xMA_1−x)_0.95PbBr_yI_3−y/Al architecture. The device demonstrated notable characteristics, such as an exceptionally low power consumption of 10⁻⁹ W, a high light/dark current ratio of 10², and a very high HRS/LRS ratio of 10³.¹⁴⁹ Rogdakis et al. developed devices based on quadruple cation perovskite that are capable of both memristive and solar energy harvesting functionalities. The devices retained their photovoltaic performance and exhibited stable resistive switching characteristics using a structure of ITO/PTAA/perovskite/PC₆₀BM/BCP/Ag. With an endurance of 3 × 10³, they demonstrated a high ON/OFF ratio of up to 10⁵ and fast light-tunable switching cycles and a power conversion efficiency (PCE) of 17%. Potentially, these devices might produce a fully printable structure for memristive perovskite solar cells, which can be compatible with flexible substrates¹⁵⁰ and offer both light and electrical tunability at low manufacturing temperatures. The same device was used to emulate a variety of synaptic functionalities¹⁵¹ and for the development of a neural network with reduced power consumption due to the RS device optoelectronic response.³⁰ Furthermore, mixed dimensional heterostructures were employed to improve the device performance,¹⁵² along with replacing the expensive poly[bis(4-phenyl)(2,4,6-trimethylphenyl)amine (PTAA) hole transport layer (HTL) with a MeO-2PACz self-assembled monolayer to improve device performance and sustainability.¹⁵³ 

Perovskite thin film deposition is essential for sustainable applications since it offers eco-friendly alternatives to conventional methods. In order to produce sustainable perovskite memristors, spray coating, spin coating, and vacuum deposition techniques enable efficient and low-impact device fabrication. The spray method is a low-cost solution processing technology that may achieve desired film thicknesses by spraying conductive surfaces with perovskite thin thicknesses.¹³⁸ Spin coating is widely used to construct perovskite memristors since it is compatible with clean room lithography procedures. As solution-based approaches, spray and spin coating are applicable for creating flexible perovskite memristors that can function at room temperature. Vacuum deposition is preferred for producing ultrathin and layered perovskite films.^154,155 Chen et al. developed a high-performance neuromorphic computing system by employing all-vacuum deposition to fabricate a perovskite/metal oxide-based device. The device, featuring an ITO/CsPbI₂Br/MoO₃/Ag structure, demonstrated outstanding synaptic characteristics, as well as exceptional uniformity and stability across different devices and operational cycles.¹⁵⁶ 

The synthesis and deposition of the perovskite materials account for a large amount of the environmental impact. The utilization of eco- and human-toxic solvents such as N,N-dimethylformamide (DMF) and N-methyl-2-pyrrolidone (NMP) has notable long-term effects on the environment, especially when employed in industrial processes. Implementing recycling procedures and switching to environmentally friendly solvents such as dimethyl sulfoxide (DMSO) and gammabutyrolactone (GBL), as well as environmentally friendly co-solvents such as methylsulfonylmethane (MSM), will help alleviate this problem. In addition, a viable approach is to reduce the amount of hazardous antisolvents used in perovskite film formation processes, such as chlorobenzene (CB) and toluene, or to substitute them with less toxic alternatives, such as ethyl acetate (EA), 2-propanol (IPA), and anisole. Antisolvent-free techniques have been documented; however, their applicability is limited to specific types of perovskite synthesis. These approaches seek to preserve effective perovskite material synthesis while minimizing environmental effects.^156,157 Shaban et al. developed a probe-based resistive storage memory cell with CH₃NH₃PbI_3−xCl_x/FTO structure fabricated using a single-step solution spin coating method without the need for an antisolvent. The devices showed an endurance of 10⁴ cycles and a retention time of 2 × 10³ s.¹⁵⁸ 

Inorganic metal oxides are not appropriate for flexible substrates, as they typically need elevated processing temperatures and are often brittle. Nevertheless, organic and hybrid semiconducting materials offer flexibility under mechanical stress but are prone to degradation from exposure to oxygen and moisture, which might impair the functionality of the device. Therefore, flexible substrates must effectively block oxygen and water penetration to maintain stable and reliable long-term operation. Moreover, using flexible substrates becomes a more sustainable option when taking fragility, weight, and cost into account.¹⁵⁹ 

Liu et al. developed a flexible, transparent, high-efficiency photoelectric perovskite memristor by combining graphene oxide (GO) with CsPbBr₃ quantum dots (QDs). The devices have a structure of Ag/CsPbBr₃ QDs:GO/ITO and were made via the low-temperature spin coating method. When tested in both electric field and illumination, they showed a very high ON/OFF ratio of 1.4 × 10⁷. This ratio was significantly larger than that observed in dark conditions, surpassing by 1077 times. The migration of Ag⁺, Br⁻, and O²⁻ ions induced by bias and the Schottky barrier at the interface were identified as the prime causes of the resistive switching (RS) mechanism.¹⁶⁰ 

A flexible hybrid perovskite-based memristor with a PET/ITO/perovskite/Ag structure was presented by Patel and colleagues. They employed a screen-printed top electrode, offering a vacuum-free and simple fabrication technique. The devices displayed an excellent endurance characteristic, withstanding up to 2500 cycles in ambient conditions with a high ON/OFF ratio of 7 × 10³. These memristors hold promise for realizing brain-inspired computing applications.¹⁶¹ Ye et al. developed a non-toxic and stable high-performance flexible memristor using an AgBiI₄ perovskite on a flexible polyethylene naphthalate (PEN) substrate with an Ag/PMMA/AgBiI₄/ITO structure, fabricated via a low-temperature dynamic hot casting method. The device exhibits excellent electrical performance, including ultralow operating voltage (0.16 V), high ON/OFF ratio (10⁴), reversible resistive switching through pulse voltage operation (>700), and long data retention (>10⁴ s), along with good stability under repeated bending tests (>1000 cycles).¹⁶² 

Addressing environmental concerns also at the material level is crucial for future sustainable commercial applications beyond the synthesis and deposition of perovskites. Most halide perovskites contain lead (Pb), and the toxicity associated with lead or lead-based perovskites raises significant worries due to potential harm to humans, animals, and the environment.¹⁶³ Encapsulating and sealing devices effectively shields Pb-containing components, minimizing the risk of Pb leakage and exposure. Exploring Pb-free perovskites and perovskite-like materials, such as tin-based, double, or low-dimensional perovskites, can provide promising substitutes with favorable memristive properties. In addition, lead-free perovskites often show enhanced the stability and are less prone to degradation compared to those containing lead.¹⁶⁴ These factors are essential in ensuring stability and environmental friendliness of perovskite-based memristors.^165,166 An environmentally robust lead-free double perovskite Cs₂AgBiBr₆ memristor with an ITO/Cs₂AgBiBr₆/Au structure was developed by Cheng et al. By adapting to various CCs, the device demonstrated a remarkable ON/OFF current ratio of 10³ and accomplished multilayer storage. It is noteworthy that the memristive activity exhibited resilience even under extreme environmental factors, such as elevated humidity, temperatures reaching 180 °C, exposure to an alcohol burner flame, and gamma radiation. The robust Ag–Br bond and superior crystallinity of Cs₂AgBiBr₆ may be responsible for the device’s resistance to environmental stress.¹⁶⁷ 

In a different study, Ge and co-workers fabricated low-dimensional lead-free Cs₃Bi₂I₉ and CsBi₃I₁₀ perovskite-like films integrated into devices configured as Ag(AgO_x)/Cs₃Bi₂I₉/FTO and Ag(AgO_x)/CsBi₃I₁₀/FTO. These devices showcased an impressive ON/OFF ratio of about 10⁶ and bipolar resistive switching behavior with self-compliance features. Due to the weak Bi-halide bond and chemical interactions at the perovskite/Ag interface, their special feature allows them to operate at low voltages (∼0.1 V), suggesting potential for cost-effective and eco-friendly memory devices.¹⁶⁸ Other lead-free materials with a variety of structures, such as Ag/BA₂CsAgBiBr₇/Pt/Ti/SiO₂/Si with a high ON/OFF ratio,¹⁶⁹ Au/layer-Cs₃Sb₂I₉/ITO with an ultra-fast switching speed,¹⁷⁰ Ag/Cs₂AgInCl₆/ITO with high endurance and retention with a long-term speed,¹⁷⁰ and Ag/Cs₂AgInCl₆/ITO with high endurance and retention with long-term stability, have also been reported.¹⁷¹ 

The selection of electrode materials in memristor devices plays a critical role. While gold, silver, and aluminum are commonly used, integrating carbon electrodes into perovskite-based devices can be advantageous with the aid of ion migration in perovskites to enable switching behavior in the device, offering a promising avenue for sustainability. Carbon is a desirable substitute that is consistent with the eco-friendliness and resource efficiency of memristor technology because of its availability, lower cost, and recycling potential.^172,173 It is possible to fabricate conductive carbon electrodes using a variety of technologies, including screen printing, doctor blades, and spray methods that outperform high-temperature evaporation processes. Furthermore, the versatility and sustainability of carbon materials are demonstrated by their potential for wearable electronics and flexible sustainability.^173–175 Guillen and colleagues combined the advantages of thin-film perovskites with monocrystalline materials to fabricate the first thin single-crystal perovskite memristor. They attained a highly stable device with outstanding performance employing ITO and PTAA as bottom electrodes and graphite spray as a metal contact. The device demonstrated an ON/OFF ratio of 10 and an endurance of 10³ cycles.¹⁷⁶ 

Organic materials are a versatile platform for RS memories and are a well-established technology since organic electronics is a well matured technology spanning various sectors.^177,178 The solution-processing at low-temperatures and the high-compatibility with roll-to-roll^179,180 and printable deposition¹⁸¹ techniques on flexible substrates¹⁸² constitute organic materials as potential candidates for RS and neuromorphic computing applications.^183,184 A variety of organic materials are synthesized, including conjugated polymers,¹⁸⁵ organic semiconductors such as PEDOT:PSS,¹⁸⁶ and small molecules such as fullerenes.¹⁸⁷ 

Possible mechanisms that induce memory behavior in organic compounds include filamentary switching induced by active metals such as Ag or Cu through the electrochemical metallization mechanism (EM),^188,189 charge transfer (CT), charge trapping/de-trapping leading to SCLC,¹⁹⁰ or interface switching by Schottky barrier modulation.¹⁹⁰ Except for metallic filaments, conductive paths can be formed by redox reactions¹⁹¹ of organic molecules. The application of an external field in the top electrode can assist in moving electrons from electron-rich molecules, leaving behind holes and positively charged defects, thus reducing the bandgap. Therefore, the conductivity of the switching layer increases, thus setting the device in the LRS, and this process is reversible by opposing bias polarity, which sets the device in the HRS. This mechanism has been experimentally confirmed.^192,193 In addition, carbon filaments have been identified as the RS mechanism.¹⁹⁴ In the case of charge trapping/de-trapping, injected charge carriers from the top electrode can end up trapped at the organic material/electrode interface or inside the organic material due to impurities or electron-capturing organic elements. Eventually, when all traps are filled, the device switches from HRS to LRS, forming a continuous CF. The trapped charge carriers are then released from the traps when applying opposite electric field polarity, or if the same unipolar voltage is applied, will cause excess Coulomb repulsion between charge carriers in the active material and at the electrode interface,^195,196 resulting in breaking the CF between the two electrodes.¹⁹⁷ Another mechanism that governs RS in organic devices is charge transfer (CT). CT is observed in donor–acceptor systems, typically in organic small molecules.¹⁸⁷ The RS characteristics can be manipulated by introducing electron donating or accepting moieties. The presence of electron donors is likely to enhance conductivity and enable the transition from HRS to LRS. In some cases, it has been observed that due to the high dipole moment¹⁸⁷ in many of these molecules, the resulting RS behavior is WORM-like. Further optimization is required to determine the optimum donor/acceptor ratio.¹⁹⁸ Furthermore, the strength and orientation of donors/acceptors, the degree of conjugation, and the chain length¹⁹⁹ can also affect the RS characteristics. Other organic molecules undergo a so-called conformation transformation. In this case, planar conjugated polymers that are amorphous with disordered orientations switch to ordered π–π stacking configurations after the application of an external electric field, which leads to RS due to the HRS to LRS transition.²⁰⁰ 

Plenty of sustainable printable organic RS memories have been developed. Oh and colleagues developed a flexible polymer memristor on a planar and vertical configuration based on poly(vinyl alcohol) (PVA) as the active layer in a glass or PEN/ITO/PVA/Ag configuration. The PVA active layer was spin-coated on the rigid and flexible substrate. The device demonstrated good mechanical endurance under stress, and numerous synaptic learning rules were emulated (EPSC, SRDP, SNDP, LTP, LTD). The ECM mechanism based on Ag filaments was identified as the predominant mechanism of the RS device.²⁰¹ 

Novel deposition methods have been implemented beyond spin-coating. Sharma et al. presented a novel deposition method, namely, unidirectional floating-film transfer method, where a small amount of the active layer is dropped on a hydrophilic liquid substrate, which aids in the controlled spreading of the solution. This technique is compatible with flexible, large area device fabrication. The solvent gradually evaporates, and the final solid film is formed. The flexible device had an Al/PTB7/Al–AlO_x/PTB7/Al configuration, demonstrating a high ON/OFF ratio of 10⁵ with robust bending stability, a fast switching speed of 100 ns, and multiple resistance states. The authors demonstrated an increase in the ON/OFF ratio at 10⁸ by inserting five layers of PTB7. In addition, a crossbar array of 64 devices was demonstrated, and the conduction mechanism was identified as charge transfer via hopping.²⁰² Sajapin et al. fabricated an environmentally friendly RS device based on chitosan:polyaniline (CPA) ink, which was deposited using aerosol jet printing. The final device had an ON/OFF ratio of 10³.²⁰³ Strutwolf et al. developed a flexible, printed organic RS device based on PET/ITO/Poly(vinyl alcohol) (PVA): poly(methyl methacrylate) (PMMA) blend/Ag configuration. The organic active blend was deposited by the novel roll-to-roll, rotary in-line technique, namely, flexo-graphic printing. Ag metallic filaments were the main conduction mechanism.²⁰⁴ Wang and colleagues developed a volatile memristor based on a Zn-tetrakis (4-carboxyphenyl) porphyrin active layer (Zn-TCPP) with PMMA, and the final device had a Pt/Zn–TCPP:PMMA/Ag configuration. The RS device showed a very low switching voltage of 80 mV, and the device was stable after 10³ folding cycles. Metal Ag filaments were identified as the conduction mechanism, and the synaptic learning rules of EPSC, PPF, and LTP/LTD with and without bending were realized. Furthermore, the RS device was integrated with a piezoresistive sensor with carbonized cotton, which converts mechanical stimulus to electrical signals, demonstrating quick response times. An artificial neural network was used to recognize human gestures with 97.24% precision, and this number does not reduce below 80% upon bending. These results highlight the potential of organic materials for low-power biomimetic skin applications in neuromorphic computing.²⁰⁵ Kim et al. developed a flexible RS device on a PEN substrate with the organic material, namely, poly(vinyl cinnamate) (PVCi), as the active layer. The organic device in an array configuration can be implemented for neuromorphic computing applications [Fig. 7(a)]. The resulting flexible PEN/ITO/PVCi/Ag nanoparticles/Au device [Fig. 7(b)] showed a high ON/OFF ratio of 10⁴ operating as a volatile memristor. The RS device had a mechanical endurance of 10³ cycles upon 5 mm bending at both positive and negative bias, with a low-electric field required for switching. Furthermore, the I–V characteristics were recorded for several compliance current values [Fig. 7(c)], showcasing the potential of the device to exhibit multilevel behavior by generating multiple resistance states, which is a desired property for neuromorphic computing. The switching voltage variation is shown in Fig. 7(d) for numerous compliance currents, as variations limit the commercialization of resistive switching memory devices. According to Fig. 7(e), the device had a stable ON/OFF ratio after 3 × 10³ pulsed cycles. The migration of interfacial Ag nanoparticles was identified as the origin of the observed volatile behavior by forming transient conductive filaments, thus the diffusive dynamics. The synaptic learning rules of PPF, STDP, SRDP, STP, and EPSC were realized, and the STP behavior was a result of the transient dynamics of the conductive filament. Finally, a 6 × 6 crossbar array was designed for neuromorphic computing and specifically for solving the max-cut problem, which is related to optimizing the design of circuits.²⁰⁶ In addition to these materials, the utilization of polymer-based electrolytes can yield high-performing memristive devices. Yang et al. fabricated a resistive switching memory device based on polyethyleneimine (PEI) in the form of a solid electrolyte as the active layer. The PEI electrolyte solution was deposited through spin-coating. The obtained non-volatile, bipolar Pt/PEI/Ag device yielded a good ON/OFF ratio of 10⁵ while requiring low-electric fields for switching. The device was stable after 10³ write–read–erase–read pulsed cycles, while the state retention was 10⁵ s. In addition, the PEI-based device showed excellent thermal stability due to its stable molecular chain and the reduced glass transition temperature (T_g). As a result, the PEI-based memory device maintained stable resistive switching behavior for multiple cycles under heating in the 25–150 °C temperature range, showcasing the good thermal stability. On top of these characteristics, the device performance was evaluated on a flexible polyimide substrate using a crossbar architecture. The PEI memristive device showed good mechanical flexibility by maintaining its RS characteristics after 2 × 10³ bending cycles under a 0.25 mm bending radius. After the bending test, the RS performance was stable under heating at 150 °C. Finally, metallic Ag filaments were identified as the predominant switching mechanism.²⁰⁷ Organic-based memory elements can successfully emulate synaptic functions. For instance, Xie et al. proposed a three-terminal artificial synaptic non-volatile device based on the organic n-type donor–acceptor conjugated polymer, namely, polymer-naphthalene-1,4,5,8-tetracarboxylic-diimide-thiophene-vinyl-thiophene (NDI-gTVT) as a channel. The synthesized polymer-based device was used to emulate biological-like synapse functions such as EPSC, PPF, and LTP/LTD with energy consumption on the order of pJ. In addition, the device was employed in classification tasks, specifically pattern recognition of handwritten digits from the MNIST database, achieving a recognition accuracy of 94%.²⁰⁸ 

2D materials are also excellent candidates for RS applications and have been extensively studied in electronic devices.²⁰⁹ 2D materials provide several benefits: ability to choose from many types of compounds that include graphene and its derivatives,²¹⁰ transition metal dichalcogenides (TMDs),²¹¹ hexagonal boron nitride (h-BN),²¹² black phosphorus,^213,214 MXenes,³⁴ and ferroelectric materials such as CuInP₂S₆.²¹⁵ They are also compatible with solution-processed techniques for ink preparation and printable methods. Some typical techniques employed to prepare 2D material solution inks are liquid phase exfoliation²¹⁶ and electrochemical exfoliation.²¹⁷ The solution processability enables the production of lower-dimensionality inks such as nanosheets²¹⁸ and nanoparticles.²¹⁹ In addition, 2D materials offer mechanical flexibility and use of flexible substrates,^220,221 low-power consumption switching,²²² low-cost production, and applications in neuromorphic computing²²³ and atomically thin devices.²²⁴ In some instances, 2D materials have been utilized as electrodes in RS systems.^225,226

Memory effects are observed in 2D materials via several mechanisms, where filamentary switching is one of the main ones and originates either from metal filaments such as Ag^34,227 and Cu²²⁸ or by vacancies through the valence change mechanism, as they have been identified as sources of ion migration.^229,230 In TMD memristors, migration of sulfur vacancies (V_s) leads to conductive filament formation, setting the device in the LRS.²³¹ Solvent engineering of sulfur vacancies in MoS₂ enables the implementation of a synaptic memristive device.²³² In WS₂ nanosheet-based RS devices, experimental evidence suggests the formation of conductive filaments by both sulfur (V_s) and tungsten (W) vacancies.²³³ Oxygen vacancies can also form conductive filaments, as in the case of MXene-based RS memories.²³⁴ It has been experimentally shown that in monolayer TMD RS devices, Schottky barrier modulation in photo-responsive MoS₂ memristors regulates oxygen vacancy movement.²³⁵ 

Many solution-processed and sustainable RS devices have been developed lately using 2D materials in different device geometries. Zhu and colleagues have fabricated ink-based h-BN RS memories in an Ag/h-BN/Ag and Ag/h-BN/Pt configuration and deposited the h-BN ink and Ag top electrode on a glass substrate through inkjet printing. The resulting device demonstrated a high ON/OFF ratio of 10⁵ for the Ag/h-BN/Pt device operating at low voltage. Ag filaments were identified as the conduction mechanism. The authors exploited the high cycle-to-cycle variability and random telegraph noise behavior to develop a true random number generator for data encryption and hardware security applications. Peng et al. demonstrated a fully printed MoS₂ RS device with CVD graphene as the top electrode. The MoS₂ water-based ink was deposited on the flexible Kapton substrate using inkjet printing. Ag filaments were identified as the main mechanism.²³⁶ Huang et al. developed a flexible Pt/Au/Ti₃C₂T_x/Ag RS device where the MXene dispersion was spin coated on the substrate, where Ag filaments governed the RS characteristics. A tactile memristive-sensor neuron perception system was designed for sensory computing. A 3 × 3 array was implemented as a high-sensitivity and linearity pressure sensor system, and classification tasks were realized with high precision in handwritten digit recognition and human respiratory state classification upon pressure changes with accuracy of 93.21% and 97.26%, respectively.²³⁷ Moazzeni et al. have synthesized graphene oxide dispersion ink based on the modified Hummer’s method. The obtained ink was deposited on the glass/ITO substrates through the spray pyrolysis method, and the final device had an ITO/GO/Al structure. The authors showed the deposition method yielded reproducible results and multilevel behavior of five states, and the LTP/LTD method was demonstrated for synaptic weight change. Metal filaments were identified as conduction mechanisms.¹²⁵ Yan et al. developed a RS device based on solution-processed WS₂ nanosheets on alcohol solution in an Ag/ZrO₂/WS₂/Pt configuration with a fast switching speed of around 10 ns. The ZrO₂ layer was added to reduce switching variations and compared the performance with Ag/ZrO₂/Pt and Ag/WS₂/Pt devices. The ZrO₂–WS₂ showed the best performance with 10⁹ endurance cycles. The RS device operated on the formation/rupture of Ag filaments.²³⁸ Feng et al. demonstrated a solution processed RS device based on the low cost and low toxicity of 2D layered tin disulfide (SnS₂). The aqueous solution was prepared by spin coating on glass/ITO substrates with a final structure of ITO/SnS₂/(RF-sputtered) ITO. The memristive device with the incorporation of Ca²⁺ during the SnS₂ synthesis yielded a reliable RS device with 5 × 10³ endurance cycles and a retention time of 10⁴ s. Synaptic behavior by illustrating STP, LTP, PPF, and STDP was realized. Extensive characterization using energy dispersive x-ray spectroscopy (EDS) and XPS techniques revealed that the device switched due to the formation of filament-like moieties from Ca²⁺ flux between the top and bottom electrodes.²³⁹ 

Feng et al. fabricated a fully printable and flexible RS device based on solution-processed MoS₂ flakes as the active layer. The deposition of the ink was achieved through the aerosol-jet printing method [illustrated in Fig. 8(a)]. The MoS₂ dispersion ink was produced by the ultrasonic-assisted liquid phase exfoliated method (UALP) method [Fig. 8(b)]. The high-quality and resolution of this printing method are illustrated on the printed pattern shown in Fig. 8(c), while the array-patterned printed lines of both MoS₂ and Ag are shown in Fig. 8(d). The crystalline structure of the resulting deposited flakes is confirmed by Raman spectroscopy measurements in Fig. 8(e), while the morphological properties were examined through SEM and AFM measurements [Figs. 8(f) and 8(g)], respectively, revealing the presence of a uniform layer after several printing cycles. Current–voltage measurements [Fig. 9(a)] confirmed the existence of a threshold, volatile resistive switching at both positive and negative bias for multiple compliance current values. The device operates at low electric fields and possesses a very high ON/OFF ratio. The stability of the volatile device was evaluated for 100 current–voltage cycles from 0 to positive bias without any deterioration of the switching properties [Fig. 9(b)]. Increasing the compliance current above 100 μA [Fig. 9(c)] leads to a transformation from volatile to non-volatile device operation. These results corroborate the controllable operation of the device between the volatile and non-volatile modes. The state retention of the threshold switching device showed a stable ON/OFF ratio of 10⁹ for more than 10⁴ s [Fig. 9(d)]. As the next step, the current response of the device as a function of time was recorded for several compliance currents [Fig. 9(e)]. In this regard, multiple resistance states can be realized, confirming the multilevel data storage ability of the device. Finally, the crossbar array structured device can be used for artificial synapse emulation with the PPF, LTP/LTD, and SRDP synaptic rules being applied. The performance of the device was compared with the state-of-the-art flexible devices [Fig. 9(f)], exhibiting the lowest switching electric field with the highest ON/OF ratio among these devices. The proposed mechanism of operation is based on CF formed assisted by S vacancy exchange.²⁴⁰ 

Due to their intriguing qualities and electrical properties, two-dimensional (2D) materials including graphene, transition-metal dichalcogenides (TMDs), and black phosphorous (BP) based devices have attracted increasing attention recently. A variety of 2D materials have been explored in RS devices to achieve lower operation voltages with reduced SET and RESET voltages.²⁴¹ Kim et al. studied the effect of atomic-level interface changes on bipolar RS properties in Al–WO₃–Al ReRAM devices. Significant RS behavior changes were found when a monolayer of graphene or hexagonal boron nitride (hBN) was inserted between the top Al and WO₃ layers, in contrast to symmetric devices that did not undergo interface modifications. While the device containing hBN showed decreased performance, the Al(graphene)–WO₃–Al device showed increased stability and endurance in RS, indicating that the insertion of graphene and hBN alters the atomic arrangement near the interface.²⁴² Shin and Son explored the potential of graphene electrodes for transparent, flexible, and stretchable electronic devices. They used the RF sputtering approach to build polycrystalline NiO thin films over single-crystalline HOPG substrates. Because of the low work function of the graphene and HOPG electrodes, the resulting memory capacitor, which was made of graphene/NiO/HOPG, demonstrated typical RRAM behavior and showed promising electrical properties such as switching characteristics and low SET and RESET voltages in comparison to traditional Pt/NiO/Pt devices.²⁴³ 

Bio-compatible organic materials are another class of materials that can be exploited for RS device applications as they can be extracted from natural resources, such as plants and animals, thus eliminating the need to chemically synthesize complex organic compounds, such as polymers. Only slight modifications are necessary for biomaterials to be suitable for device fabrication, which is typically performed at low temperatures. The major advantages of this technology are the zero environmental footprint and the greatly reduced fabrication cost compared to technologies where the deposition of expensive materials using complex methods is required. Furthermore, these materials are compatible with flexible substrates for wearable applications and can be deposited by using environmentally friendly solvents, such as water, providing a link between green natural materials and electronics. In addition, bio-compatible materials can be recycled, which minimizes the environmental waste. These materials have also been utilized for neuromorphic computing applications.

Several mechanisms have been proposed to explain the RS mechanism behind the operation of bio-memristive devices. In agreement with other types of RS materials, CFs are formed by metal cations such as Ag²⁴⁴ and Cu²⁴⁵ in RS devices based on biomaterials. In addition, Mg²⁴⁶ and carbon²⁴⁷ induced filaments have been identified in biomaterial RS devices as well. Charge trapping/de-trapping has been identified as the RS mechanism in the Scindapsus aureus biomaterial.²⁴⁸ Beyond interfacial charge trapping/de-trapping, the same mechanism has been induced by oxidation/reduction in the silk fibroin RS system, which leads to filament formation.²⁴⁹ Interface modulation has also been observed in glucose-based memristive devices, resulting in conductive filament formation at the Al top electrode/glucose interface.²⁵⁰ Electrochemical reactions were proposed as the operation mechanism in ferritin protein RS devices. The spherical geometry of ferritin and its core structure are the primary causes of electron transfer. In this regard, the conductivity can be altered by applying voltage, which leads to the oxidation²⁵¹ of Fe(III) to Fe(II), thus enhancing the conductivity and inducing RS.²⁵² 

Many studies have proposed sustainable RS devices based on biomaterials. Abbas and co-workers have demonstrated a Cu²⁺ doped-salmon DNA RS device in a FTO/Cu doped-DNA/Pt configuration, where the aqueous DNA solution was spin-coated on the FTO substrate. The obtained device showed an ON/OFF ratio of 10³ and a high pulsed endurance of 10⁵ cycles, and the authors suggested the formation/rupture of Cu filaments as the origin of the observed RS.²⁵³ Sun et al. designed a flexible, plant-based corn starch memristive device. The active layer consisted of corn starch: polyvinylidene fluoride (PVDF) at a 10:1 ratio, and the incorporation of PVDF was to prevent the thin films from cracking during drying. The final device had an ITO/corn starch: PVDF/Ag configuration, and the authors provided evidence of Ag filament formation as the mechanism that governs RS. The RS memristor showed good mechanical stability under bending stress, and different current responses were recorded upon various bending angles; thus, the device can be used for logic operations and as a mechanical stress sensor for artificial skin applications, paving the way toward green, sustainable, and wearable electronics.²⁵⁴ Wang et al. developed an RS device based on Chitosan (CS):Multi Wall Carbon Nanotubes (MWCNTs) in a 1:1 ratio in a ITO/CS:MWCNTs/Al configuration. By varying the MWCNTs content, the authors observed a reduction in the ON/OFF ratio with increasing MWCNTs content. Oxygen vacancy filaments were proposed as the switching mechanism. In addition, the memristive device was integrated with a flexible tactile sensor that was used to convert the physical stimulation into electrical signals. The synaptic functionalities of EPSC, PPF, and SRDP were emulated, and some of these were observed for different amounts of pressure stimulation, realizing the potential for wearable artificial touch systems.²⁵⁵ Wang et al. developed Silkworm Hemolymph (SH): Au NPs RS memory based on a flexible PET/ITO/PMMA/SH:Au NPS/PMMA/Al [PMMA = Poly(methyl methacrylate)] structure. The prepared devices demonstrated a high ON/OFF ratio of >10⁵ with robust stability upon 10⁴ bending cycles. Finally, the synaptic learning rules of LTP/LTD and STDP were implemented.²⁵⁶ 

Raeis-Hosseini et al. developed a flexible device based on fish collagen as the active layer. The device is bio-degradable and water dissolvable, minimizing its environmental footprint. The proposed memristor can be implemented for a Physically Transient Electronic Device (PiTED), which can be realized by incorporating the memristive device along with an artificial synapse and sensors. This highlights the potential of novel, solution-processed memory elements for wearable health monitoring applications, as the device can store and analyze information for decision making [Fig. 10(a)]. The proposed device has a structure of PET/ITO/collagen/Mg RS [Figs. 10(b) and 10(c)] and is water-soluble. Collagen is a by-product of fish powder [Fig. 10(d)], and its chemical structure from a series of molecules is depicted in Fig. 10(e). The bio-compatible active layer aqueous solution was spin coated on the PET/ITO substrate, and the final device retained its ON/OFF ratio after 1000 switching cycles under 5 mm bending stress. The transmittance spectrum of the collagen thin film shown in Fig. 10(f) showcases the transparency of the active layer. Filamentary switching induced by Mg cations was proposed as the RS mechanism. In addition, the device’s current response can be altered both in steady-state and using pulses, showing successful synaptic weight modification, and the LTP/LTD synaptic rule was realized. A neural network for pattern recognition was designed, and the device demonstrated 99.2% accuracy after 200 epochs. Finally, the authors emphasized the washable transient characteristics of the device when drops of water fall into the device, highlighting the sustainability of the RS memory.²⁴⁶ 

Each material type analyzed here exhibits a variety of merits and drawbacks that should be carefully considered for evaluating which of these has the highest potential for commercial use. It is evident that oxide-based devices hold the highest promise so far as they have been widely studied for quite a long time since the experimental evidence of memristive behavior in terms of fabrication, optimization, and operation principles. Moreover, they offer fast switching speed on the order of ns,²⁵⁷ extended cycling endurance and stable retention, as well as heat-tolerant device operation. In addition, they have good ambient stability and have been extensively studied for neuromorphic computing applications, as they are also compatible with CMOS integration. Nevertheless, there are some limitations that should be mentioned. In most cases, oxides require high-complexity and expensive methods for deposition, limiting the potential for upscaling and the ability to use solution-processed methods despite recent efforts. On the other hand, other material types discussed here have demonstrated better compatibility and integration with solution-processed and printed deposition methods.

Apart from oxides, 2D materials are a very attractive candidate for resistive switching memory applications. Apart from using expensive deposition techniques (i.e., MBE and CVD), 2D materials have been thoroughly studied using solution-processed methods. Ink production methods such as liquid phase exfoliation can greatly reduce the cost and the processing temperatures while being compatible with upscaling printing methods and flexible substrates. These materials have demonstrated high performance with a high ON/OFF ratio, good retention, low-voltage operation, heat tolerance,²⁵⁸ and switching speed on the order of ns, competing with one of the oxides.²⁵⁷ While not entirely stable under ambient humidity, 2D materials are still less susceptible to humidity compared to other types of materials. Finally, these compounds are inevitably prone to switching variations and ink preparation, along with the fact that printed methods still require further optimization, and while they constitute rivals for oxides, they still lack in terms of performance.²⁵⁹ 

With performance close to 2D materials, perovskites also have a strong potential for resistive switching memory and neuromorphic computing applications. The hysteresis originating in their rich ion dynamics is a result of the observed memory behavior while offering composition- and dimensionality-dependent tunability. Furthermore, perovskites have an excellent optoelectronic response and can be solution-processed under low-temperatures, while they can be deposited through printing methods on flexible substrates. Memory devices based on perovskites have been realized to exhibit low-voltage operation, a high ON/OFF ratio similar to oxides and 2D materials, and good switching speed, even in the ns range,²⁶⁰ with potential applications in synapses and neuron emulation.^20,261,262 In contrast to all these advantages, halide perovskites are highly toxic due to Pb, although lead-free compounds can eliminate this issue. Notably, to mention, lead-free compounds with a high ON/OFF ratio have been realized with switching speeds reaching even ns²⁶³ while possessing synapse-like characteristics.²⁶⁴ The relatively poor ambient and heat stability of lead-free perovskites is a major drawback despite their promising properties. In addition, they suffer from relatively poor cycling endurance and from large device-to-device and batch-to-batch variations.

Organic materials can also be deposited through low-temperature, solution-processed, and printed methods on flexible substrates, offering the upscalability option. However, they lack performance compared to 2D materials or oxides²⁶⁵ while suffering from variability issues and low endurance, and are also prone to ambient air degradation. On top of this, there is a lack of understanding regarding the operational mechanism of organic memories, and additional research is needed to explain and control the switching behavior.²⁶⁶ 

Finally, biocompatible materials also possess some attractive properties. These materials are derived from earth-abundant resources while having zero environmental footprint for green and recyclable electronics and can be used for wearables and medical applications. These materials can be processed under low temperatures, reducing thus the fabrication cost. Despite all these advantages, their performance is lacking compared to other material technologies, while the compatibility with upscale methods is somehow questionable and not well studied. Moreover, these compounds suffer from low cycling endurance and are susceptible to humidity²⁶⁷ and heat. Finally, the explanation of the switching mechanism for these devices lacks understanding, and more research is required to provide more insights. On-going research efforts through advanced characterization techniques aim to improve the understanding of the operation mechanism of biomemristors.²⁶⁸ 

To summarize, while oxides still have a good chance of use in commercial applications due to their established use, high-performance, and good stability, 2D materials are closing the gap as they can reduce cost while maintaining high-performance. Perovskites also hold great potential as they possess good performance and a variety of emerging multi-functionalities; however, ambient stability is still an issue. Organic materials are also another option, as they can be used for solutions-based processing but are inferior in terms of performance, while the switching mechanism is still unclear. Finally, biocompatible materials offer the advantage of using earth-abundant materials but lack performance, reliability, and ambient stability, thus making them the least likely candidate for commercial use.

A universal problem for any material technology implemented in RS devices is the upscaling manufacturing process. Optimization of several parameters is necessary to achieve good RS performance, including deposition and achieving uniform films, especially in the case of printing methods. However, different strategies can be adopted per material category to obtain uniformly thin films. Switching variations is also a universal problem for all RS materials, and this is an issue that must be resolved for these materials to compete with evaporated inorganic oxides or silicon devices. Even for the case of oxides, solution processing can cause further switching variations, and it will be more challenging to precisely control the stoichiometry and the properties of the resulting films compared to high-precision techniques that have atom-level precision, such as ALD.²⁶⁹ The rest of the material types examined here have different degrees of ambient stability; however, all of them have some degree of degradation upon ambient atmosphere exposure, while 2D materials are the closest to oxides in terms of stability. For example, graphene derivatives exhibit relatively better chemical stability, while some 2D compounds are prone to oxidation and ambient moisture, which presents challenges for reliable RS device operation.^{231,270–272} In addition to these general issues that are common for all materials, each category presents additional challenges related to the material’s properties. While 2D materials are suitable for sustainable RS device fabrication using inks and printing methods, there are certain issues that must be resolved to establish their use for practical applications. Large-area thin film deposition using 2D material inks is still a challenging task, which can further induce variations during device operation. Additional optimization of both ink properties and deposition properties is required to obtain high-quality films with reproducible RS characteristics, especially in flexible substrates. In the case of biomaterials and organic materials, additional research is required to understand the fundamentals behind the switching mechanism. Many of the organic materials used, such as organic biomolecules or polymers, have complex structures, and the conduction mechanism might depend on the charge transfer mechanism that needs further interpretation. The thermal stability of biomaterials is another factor that should be considered, while upscalability of these compounds and compatibility with large-area techniques, such as industrial printing, have not been explored. Transitioning to fully printed devices is another factor that requires eliminating the thermal deposition of metal electrodes as the top contacts. Alternative solutions such as silver paste or carbon paste electrodes should be considered; however, this adds more complexity to device optimization. The research interest in perovskite-based RS devices is steadily growing, and while they possess many attractive properties, the relatively low cycling endurance and ambient stability are major challenges. Physical encapsulation with thin PMMA is a common practice to protect perovskites to minimize humidity exposure; however, more alternatives can be exploited. On the other hand, research focuses on alternate routes being exploited for bio-realistic neuromorphic computing. A considerable example is the use of fluid-based memristors, going beyond traditional solid-state memory devices. The major advantage of these systems is that they can successfully emulate various functionalities of chemical synapses such as potentiation/depression,²⁷³ paving the way toward bio-realistic neuromorphic computing.^274,275 Their operation relies on the intrinsic non-linear ion transport, which is the root cause of the observed hysteresis and resistive switching phenomena.²⁷⁶ In addition, the ionic conduction of these systems can be affected by factors such as the amplitude/frequency of the applied electrical voltage signal, the ion concentration, the type of salt employed, and the pH of the solution.²⁷⁷ These properties enable fluid memristors to emulate synaptic functions bio-realistically compared to solid-state memristive systems. Future work should focus on optimizing the deposition of materials by introducing novel methods that could enhance both ambient and thermal stability. Cost-effective fabrication under low-temperatures without high-complexity techniques is crucial for practical use. In addition, the active material should be as uniform as possible with good morphological properties, along with optimization of active material/electrode interfaces and interfaces with other buffer materials to maximize performance. These approaches should contribute to minimizing switching variations, which limit practical applications. Then, moving on to the performance essentials, all figures of merit should be as high as possible, except switching speed and energy consumption, which should be minimized. Due to the intriguing properties of these novel materials, they can be used for a variety of applications that slightly differ due to device structure and mode operation. To summarize, several properties and characteristics are essential, starting from the device fabrication level, then to the performance characteristics, and finally to the applications of these devices. These characteristics are depicted on the schematic roadmap presented in Fig. 11.

The emerging field of RS memories is constantly evolving, while novel materials and devices are under continuous development aiming at brain-inspired computing applications. In this Perspective, we highlighted the importance of material choice and manufacturing conditions as parameters to be considered for practical applications, apart from the performance aspect, which is equally important. Solutions-based processing for flexible devices using advanced printing methods enabled by sustainable inks and environmentally friendly precursor materials is the focus for several types of materials and device configurations. In this Perspective, we provided information on recent advancements in sustainable RS devices based on several material types. Furthermore, we discussed insights into the operation mechanisms of sustainable RS systems and presented the current challenges and required future improvements that are needed to enable real-life commercial products using memristive systems.

This research project is implemented under H.F.R.I.’s action “Basic Research Financing (Horizontal Support for all Sciences)” of the National Recovery and Resilience Plan “Greece 2.0” with financing from the European Union—NextGenerationEU (H.F.R.I. Project No. 81045).

The authors have no conflicts to disclose.

Michalis Loizos: Formal analysis (equal); Investigation (equal); Methodology (equal); Writing – original draft (equal). Konstantinos Rogdakis: Investigation (equal); Methodology (equal); Supervision (equal); Writing – review & editing (equal). Ashitha Paingott Parambil: Investigation (equal); Methodology (equal); Writing – original draft (equal). Monica Lira-Cantu: Methodology (equal); Supervision (equal); Writing – original draft (equal); Writing – review & editing (equal). Emmanuel Kymakis: Funding acquisition (equal); Supervision (equal); Writing – review & editing (equal).

Data sharing is not applicable to this article as no new data were created or analyzed in this study.