The exponential growth of data in various sectors (municipal traffic control, security surveillance, healthcare, industrial production) has led to an era of big data and artificial intelligence. The current reliance on cloud computing for data processing, while advantageous in terms of cost and security, leads to significant energy consumption due to the transmission of redundant data. This energy inefficiency poses a major challenge to the information technology industry. Low-power edge computing paradigms, implemented near the data source, are crucial for real-time processing and pre-screening, particularly in applications demanding fast responses, such as autonomous vehicle navigation. Memristors, with their CMOS compatibility, fast switching speeds, and low power potential, are ideal candidates for high-density information storage and in-memory computing. Their non-volatile resistance reconfiguration and simple two-terminal structure enable large-scale in-memory computations, enhancing efficiency and lowering energy consumption. Polymer memristors, in particular, offer advantages in flexible edge computing applications due to their lightweight nature. However, the poor fabrication yield and reliability of polymer memristors due to their structural inhomogeneity remain major obstacles to their widespread adoption. Molecular design and synthesis strategies, particularly the manipulation of macromolecule coplanarity and two-dimensional (2D) conjugation to enhance π–π stacking and crystallinity, offer potential solutions to improve the performance and yield of these devices.

Resistive switching in polymers has been studied since the 1970s, with mechanisms including charge trapping/detrapping, charge transfer, electrochemical redox reactions, conformational reconfigurations, and ion migration. However, the structural inhomogeneity of polymers often leads to localized resistive switching, resulting in low production yields and unreliable device performance. Previous research has shown that controlling the morphology and crystallinity of polymer thin films can improve device performance. The two-dimensional π-conjugation strategy has been successfully used to enhance the performance of polymer optoelectronic devices such as organic photovoltaics (OPVs) and light-emitting diodes (OLEDs) by improving the molecular planarity, packing ordering, and crystallinity. This strategy aims to promote effective charge transport through extended π-conjugation and improve the overall device performance.

This research designed and synthesized a 2D conjugated polymer, PBDTT-BQTPA, through a Stille reaction. The polymer incorporates redox-active triphenylamine moieties and coplanar bis(thiophene)-4,8-dihydrobenzo[1,2-b:4,5-b]dithiophene (BDTT) chromophores. The BDTT chromophores enhance the π–π stacking and crystallinity of the thin film, leading to homogeneous resistive switching. The synthesis was confirmed by spectroscopic analysis (NMR, GPC, UV-Vis, fluorescence), demonstrating good solubility and thermal stability. Cyclic voltammetry revealed reversible redox activity, primarily attributed to the triphenylamine units. The Au/polymer/ITO devices demonstrated bistable memristive switching behavior in both DC and pulse mode measurements. The switching speed was evaluated using pulse measurements, revealing response times of less than 32 ns. A high production yield (approximately 90%) and low device-to-device variation were observed. The microstructure and resistive switching were investigated using conductive atomic force microscopy (C-AFM) and grazing-incidence wide-angle X-ray scattering (GIWAXS). C-AFM showed homogeneous switching across the entire polymer layer even at the nanoscale. GIWAXS confirmed the semicrystalline nature of the PBDTT-BQTPA film with preferred face-on orientation and tight π–π stacking. The ability to perform in-memory Boolean logic operations, arithmetic operations (demonstrated with a 1-bit full adder), and neuromorphic computing tasks (using a binary neural network for handwritten digit recognition) was also evaluated. For the binary neural network simulation, the LeNet-5 model was employed, demonstrating high recognition accuracy. Molecular simulations using DFT and TD-DFT provided insights into the electronic structure and properties of the polymer. Device fabrication involved spin-coating the polymer solution onto ITO-coated substrates followed by thermal evaporation of Au top electrodes for larger devices. Nanoscale devices were fabricated using electron-beam lithography and lift-off techniques.

The key findings of this study include: 1. A record-high 90% production yield of polymer memristors was achieved using a novel 2D conjugated polymer, PBDTT-BQTPA. 2. The PBDTT-BQTPA memristors exhibit homogeneous resistive switching across the entire polymer layer, leading to significantly improved reliability. 3. The devices demonstrate fast switching speeds (≤32 ns) and low device-to-device variations (3.16–8.29%) in switching voltages and ON/OFF resistances. 4. The memristors exhibit excellent retention (over 10⁴ s) and endurance (over 10⁸ cycles) characteristics. 5. The devices are scalable to 100 nm, with power consumption as low as ~10⁻¹⁵ J/bit. 6. The PBDTT-BQTPA memristors are capable of performing both in-memory Boolean logic operations and arithmetic operations. 7. A binary neural network built using these memristors achieved a high recognition accuracy (99.23%) for handwritten digit recognition. The homogeneous resistive switching was attributed to the enhanced crystallinity and uniform microstructure of the polymer thin films, resulting from the 2D conjugation strategy. The nanoscale devices demonstrated comparable performance to the micrometer scale devices, highlighting the scalability of the technology.

This research addresses the challenge of low production yield and unreliable resistive switching in polymer memristors by introducing a 2D conjugation strategy. The high yield (90%) and homogenous switching behavior demonstrated by the PBDTT-BQTPA based memristors significantly advance the feasibility of polymer-based memristors for in-memory computing applications. The fast switching speed, low power consumption, and scalability to nanoscale dimensions further enhance their potential for use in low-power edge computing devices. The successful implementation of both Boolean logic and arithmetic operations highlights the versatility of these devices as fundamental building blocks for computer architectures. The demonstration of a high-accuracy binary neural network showcases the potential of these memristors for neuromorphic computing and artificial intelligence applications. The results of this study offer a significant advancement in the field of memristor technology, paving the way for more reliable and efficient computing systems.

This study successfully demonstrated a 2D conjugation strategy to achieve a record-high 90% production yield of polymer memristors. The resulting devices exhibited homogeneous resistive switching, high speed, low power consumption, and excellent scalability. These memristors successfully implemented in-memory logic, arithmetic, and neuromorphic computing operations, showcasing their potential for advanced computing paradigms. Future research could focus on exploring other redox-active moieties to further enhance the polymer's crystallinity and achieve even smaller device dimensions. Investigating alternative fabrication methods and exploring the integration of these memristors into larger-scale circuits could further strengthen their potential for practical applications.

While this study demonstrates a significant advancement in polymer memristor technology, some limitations exist. The study primarily focused on laboratory-scale fabrication and testing. Further research is needed to explore the long-term stability and reliability of the devices under various environmental conditions. The scalability to sub-10 nm dimensions, while theoretically suggested, remains to be experimentally verified. Moreover, a comprehensive comparison with existing state-of-the-art memristor technologies is warranted to fully assess the advantages and limitations of the developed polymer memristors.