Open research questions in Advanced Memory and Neural Computing

What the literature leaves open

• The lack of explicit community structure in conventional SNNs, - the need for efficient SNN architectures, - the gap between biological neural systems and conventional SNNs.

  Community-aware sparse topology design for efficient spiking neural networks · 2026 · DOI

• investigating neuromorphic circuits that exploit tunable dual‐volatility within a single memristive device

  Two Coexisting Pathways to Volatility in Valence‐Change Memory Devices · 2026 · DOI

• Continual learning methods for SNNs have not been evaluated under realistic edge constraints (limited memory, online data streams, task boundaries unknown a priori); existing benchmarks use offline, class-incremental protocols with predefined task sequences rather than true online, streaming scenarios.

  Spiking Neural Networks with Continual Learning for Steering Angle Regression: A Sustainable AI Perspective · 2026 · DOI

• No study in this set measures or compares the energy efficiency gains of continual learning strategies for SNNs on actual edge hardware; energy benefits are reported only for baseline SNN inference or for continual learning in standard ANNs, not for the combined SNN+continual learning system on edge accelerators.

  Spiking Neural Networks with Continual Learning for Steering Angle Regression: A Sustainable AI Perspective · 2026 · DOI

• The physical origin of volatility in valence-change memory devices is not well understood. The role of electronic processes in the volatility of these devices is not clear.

  Two Coexisting Pathways to Volatility in Valence‐Change Memory Devices · 2026 · DOI

• The interaction between SNN architectural depth (residual connections, shortcut pathways) and catastrophic forgetting during continual learning remains unexplored; prior work addresses deep SNN training or forgetting mitigation separately, but not their joint effect in sequential task scenarios.

  XOResNet: Exclusive-OR meta-residuals facilitate deep spiking neural networks learning · 2026 · DOI

• Yet it remains unclear whether these influences arise from a single mechanism or reflect functionally distinct processes operating over different timescales.

  Working Memory Guides Perceptual Decisions Through Fast Capture and Slow Drift · 2026 · DOI

• Whether such a gradient can take a physical, causal form in biophysically detailed multi-compartment neuron models, and enable online, supervised learning, remains unclear.

  Online synaptic credit assignment in active dendrites · 2026 · DOI

• Our experiments on the OASIS-2 dataset reveal a significant limitation in current hybrid classical-quantum architectures, as they face difficulties converging when class images are highly similar, such as between moderate dementia and non-dementia classes of AD, which leads to gradient failure and optimization stagnation.

  CQ-CNN: A lightweight hybrid classical–quantum convolutional neural network for Alzheimer’s disease detection using 3D structural brain MRI · 2025 · DOI

• However, multifunctional ITO-based transistors combining memory, logic gates, and artificial synaptic behaviors are rarely reported.

  A five-terminal ITO transistor enabling memory, artificial synaptic behaviors, and logic operations · 2025 · DOI

• Negative differential resistance (NDR), a decrease in conductance with increasing potential, constitutes a new function from the perspective of time-dependent instead of steady-state nanoscale electrokinetic ion transport but remains unexplored in ionotronics to develop higher-order complexity and advanced capabilities.

  Negative Differential Resistance in Conical Nanopore Iontronic Memristors · 2024 · DOI

• However, OES at near‐infrared wavelengths have rarely been reported, seriously limiting the application in modern optical communication.

  VO2/MoO3 Heterojunctions Artificial Optoelectronic Synapse Devices for Near‐Infrared Optical Communication · 2024 · DOI

• These systems differ significantly from artificial neural networks as well as the brain, so the feasibility and utility of incorporating nonlinear elements have not been explored.

  Machine learning without a processor: Emergent learning in a nonlinear analog network · 2024 · DOI

• Additionally, most research utilizing organic ferroelectric materials has been focused on basic biological functions, and the impact on nonvolatile memory properties is still lacking.

  Controlling Long-Term Plasticity in Neuromorphic Computing Through Modulation of Ferroelectric Polarization · 2024 · DOI

• These advantages enable event cameras without being limited by the intensity of light, to perform better in challenging conditions compared to traditional cameras.

  MVT: Multi-Vision Transformer for Event-Based Small Target Detection · 2024 · DOI

• The paper does not mention any specific limitations of the study. However, it can be inferred that the study is limited to simulated experiments and does not involve real-world data or applications.

  Intrinsic stabilization of synaptic plasticity improves learning and robustness in artificial neural networks · 2026 · DOI

• The study is based on simulations, and practical implementation is required! The variability of the devices and the problem of scalability need to be addressed! The study does not provide a comprehensive comparison with other energy-efficient AI systems!

  Neuromorphic Computing Architectures Using Memristive Devices for Energy-Efficient Artificial Intelligence · 2026 · DOI

• The switching rate of memristive devices remains slow relative to silicon-based transistors, which may render the architecture inappropriate for AI use cases with very high processing needs, including high-definition video streaming or real-time data analytics in large-scale cloud environments.

  Neuromorphic Computing Architectures Using Memristive Devices for Energy-Efficient Artificial Intelligence · 2026 · DOI

• The development of scalable and reliable 2D material-based neuromorphic devices is a potential future research direction. The exploration of various mechanisms for artificial synaptic devices is a potential future research direction. The application of neuromorphic computing in AIoT systems and real-time information processing is a potential future research direction.

  Dedicated and Reconfigurable Artificial Neurons and Synapses based on Two-Dimensional Materials for Efficient Neuromorphic Application · 2026 · DOI

• The development of artificial neurons and synapses is key to implementing neuromorphic computing architecture. 2D materials have risen to prominence in recent years due to their atomic-scale thickness and rich tunable physicochemical properties. The paper aims to address the research gap in the development of dedicated artificial neuron and synapse devices based on 2D materials.

  Dedicated and Reconfigurable Artificial Neurons and Synapses based on Two-Dimensional Materials for Efficient Neuromorphic Application · 2026 · DOI

• Zero-power optoelectronic synapses mark a key inflection point in the evolution of neuromorphic electronics, providing a foundation for devices that perceive, learn, and adapt using light as their sole energy source to minimize power consumption during system operation. Recent advances have demonstrated that built-in potential and PTE effect can drive synaptic functions without external electrical bias, highlighting the potential for unprecedented energy efficiency. However, realizing fully autonomous, deformable, and intelligent systems requires overcoming multiple challenges, including photon management, bidirectional learning, operational stability, mechanical adaptability, and scalable system-level integration. Addressing these challenges demands cross-disciplinary efforts that connect materials chemistry, device physics, and circuit and architectural design, rather than focusing on isolated device improvements. To Na et al. Soft Sci. 2026, 6, 25 Page 17 of 22 address limited light utilization, future research should focus on enhancing light-matter interactions through advanced photon-management strategies, including high absorption photoactive materials and vertical structure. Insufficient adaptive learning capability may be overcome by designing wavelength-selective optoelectronic synapses, particularly through heterojunction engineering that enables controllable bidirectional charge carrier transport and reversible weight modulation without external electrical bias. Device instability and variability can be addressed by an encapsulation process using hydrophobic transparent materials to prevent environmental defects and by achieving a precise and uniform film morphology. Achieving system-level integration requires extending device-level advances toward array-based architectures, incorporating optical waveguides, passive optical interconnects, and wavelength-division to suppress crosstalk and enable scalable in-sensor computing. Developing intrinsically soft electronic materials and mechanically reliable interfaces will be as crucial as designing architectures that preserve neuromorphic fidelity under dynamically changing environmental and physiological conditions. Furthermore, integrating optical interconnects and scalable fabrication strategies will be essential for translating device-level concepts into practical systems capable of hardware-based image recognition and vector-matrix computation. In the end, the convergence of zero-power operation, mechanical compliance, and neuromorphic intelligence will define the next phase of wearable technology. Future systems are expected to evolve beyond discrete sensing modules into fully integrated, adaptive networks capable of continuous perception, energy regulation, and physiological interpretation within body-interfaced environments. By autonomously managing energy flow and decoding multimodal biosignals, these platforms could function as intelligent companions for long-term health monitoring and responsive therapeutic control. Ultimately, this shift toward zero-power, soft, and cognitively capable electronics will not only advance wearable systems but also redefine how human-machine interfaces enable sustainable and lifelong interaction with the body, environment, and technology. DECLARATIONS Authors’ contributions Conceived the topic: Na, M.; Park, J.; Sim, K. Original draft writing: Na, M.; Park, J.; Sim, K. Supervised and reviewed the manuscript: Sim, K.

  Recent advances in zero-power optoelectronic synapses with potential for wearable neuromorphic platforms · 2026 · DOI

• Limitations of ground-fabricated devices in terms of operating voltage and stability - Need for a new pathway for producing neuromorphic electronic devices in space

  Neuromorphic Electronics Beyond Earth: Opportunities in Microgravity Manufacturing · 2026 · DOI

• The energy efficiency bottleneck of conventional hardware. The need for novel chip architectures that offer both ultralow power consumption and high computational efficiency.

  Ambipolar Organic–Inorganic Heterostructure Transistor Array for Integrated Visual Information Processing · 2026 · DOI

• The need for sustainable and environmentally friendly artificial synaptic devices, - The lack of demonstration of critical synaptic functions in honey based memristive devices

  Honey-CNT memristive artificial synaptic device for sustainable neuromorphic computing system · 2026 · DOI

• Critical synaptic functions of the honey-CNT memristor, including spike-rate-dependent plasticity, spike voltage dependent plasticity, learn-forget-relearn, and supralinear spatial summation are revealed, which have not been reported by honey based memristive devices before.

  Honey-CNT memristive artificial synaptic device for sustainable neuromorphic computing system · 2026 · DOI

Most-cited papers in Advanced Memory and Neural Computing

• The missing memristor found · Nature · 2008 · 10,744 citations
• A million spiking-neuron integrated circuit with a scalable communication network and interface · Science · 2014 · 3,677 citations
• Training and operation of an integrated neuromorphic network based on metal-oxide memristors · Nature · 2015 · 2,806 citations
• Deep learning in spiking neural networks · Neural Networks · 2019 · 1,246 citations
• Opportunities for neuromorphic computing algorithms and applications · Nature Computational Science · 2022 · 1,067 citations
• A crossbar array of magnetoresistive memory devices for in-memory computing · Nature · 2022 · 625 citations
• Thousands of conductance levels in memristors integrated on CMOS · Nature · 2023 · 489 citations
• Hardware implementation of memristor-based artificial neural networks · Nature Communications · 2024 · 426 citations
• Neuromorphic functions with a polyelectrolyte-confined fluidic memristor · Science · 2023 · 382 citations
• Memristor‐Based Neuromorphic Chips · Advanced Materials · 2024 · 375 citations

Most recent work

• Maya-Meta P1: The Bhaya Quiescence Law and Buddhi S-Curve Determinism — Two Empirical Constants Confirmed Across 19 Substrates in Affective Neuromorphic Computing · Zenodo (CERN European Organization for Nuclear Research) · 2026
• Dual Memristor-Coupled Unidirectional Ring Neural Network: Abundant Hidden Firings and Application in Hardware Image Encryption · IEEE Internet of Things Journal · 2026
• High-temperature memristors enabled by interfacial engineering · Science · 2026
• Emerging CMOS compatible memristor for storage technology and neuromorphic computing applications · Chip · 2026
• Visible light-driven photoelectric synaptic transistors based on Ga2O3/IGZO heterostructure for neuromorphic computing · Applied Physics Letters · 2026
• Multi-scroll regulation and control of a programmable memristive Hopfield neural network with FPGA implementation · Science China Technological Sciences · 2026
• Shape deformation and electromagnetic induction affect wave stability in a memristive media · Chaos, Solitons & Fractals · 2026
• Neuromodulation enhances the capability and efficiency of spiking neural networks · bioRxiv · 2026
• A Unified Experience Replay Framework for Spiking Deep Reinforcement Learning · IEEE Transactions on Pattern Analysis and Machine Intelligence · 2026
• Ambipolar Organic–Inorganic Heterostructure Transistor Array for Integrated Visual Information Processing · Advanced Science · 2026

Related topics in Engineering