As traditional chip technology hits its physical limits, scientists are now exploring new ways to boost computing power. While recent innovations like finFETs and nanosheet FETs have helped extend the capabilities of CMOS technology, shrinking transistors down to just 3 nanometers has exposed major challenges. To move forward, researchers are turning to neuromorphic computing, a brain-inspired approach that could offer a smarter and more efficient path beyond current methods.
What is Neuromorphic Computing?
Neuromorphic computing takes inspiration from the brain to create smarter, more efficient hardware. Instead of copying the brain exactly, this approach focuses on matching its ability to handle complex or unclear data using very little power. Unlike traditional computers that need clear inputs and use a lot of energy, neuromorphic systems are designed to work more like the brain — fast, flexible, and energy-efficient.
Brain-Computer Interfaces (BCI): A parallel path
While neuromorphic computing draws inspiration from how the brain processes information, brain-computer interfaces (BCIs) aim to connect directly with the brain itself. In May 2023, Neuralink, a neurotechnology company co-founded by Elon Musk, became the first company to secure FDA approval for human trials of its brain implant, Link. This marked a milestone in the BCI industry.
(Explore Neuralink’s patenting activity in our report: Patent Snapshot: Neuralink)
In-memory computing: Addressing the bottleneck
A key feature of neuromorphic computing is its use of in-memory, or on-chip, computing—a major shift from traditional computer designs where memory and processing are separated. In conventional systems, data constantly moves between the processor and memory, causing delays and increased energy consumption, especially during machine learning tasks.
In comparison, the brain stores and processes information in the same location, and neuromorphic chips aim to mimic this structure. By placing memory closer to processing units, these systems reduce energy use and speed up tasks. This approach also helps overcome performance limits seen in standard chip architectures, making it ideal for more efficient AI applications.
2D Materials for synaptic devices
Two-dimensional (2D) materials, just one atom thick, are emerging as key enablers for next-generation neuromorphic hardware. Known for their exceptional mechanical strength, atomically smooth surfaces, and high crystallinity, these materials support dense 3D stacking while maintaining strong electronic performance at nanoscale dimensions. These traits make them well-suited for advanced computing architectures.
Another cutting-edge component in this field is the vertical cavity surface-emitting laser (VCSEL), which can rapidly emit light pulses to simulate neuronal firing. When integrated with spiking neural networks, VCSEL arrays enable tasks like image recognition.
Operation mechanisms
Researchers are uncovering a range of mechanisms that make two-dimensional (2D) materials strong contenders for next-generation memristors—key components in neuromorphic computing. Among the most widely studied is resistive switching, where a material’s resistance changes due to the formation and rupture of conductive filaments. These filaments form as metal ions migrate through the material, a process known as electrochemical metallization. A related mechanism, called the valence change mechanism, occurs when anions move within the material, leaving vacancies that alter resistance.
Another promising method is charge trapping and detrapping. In this approach, electrons become temporarily stored at internal interfaces in the material. When an electric field is applied, these trapped charges modulate the resistance, allowing the device to store and process information with improved energy efficiency.
Additional strategies include phase change and ferroelectric behavior. Phase change relies on toggling the material between crystalline (low resistance) and amorphous (high resistance) states through precise heating and cooling. Meanwhile, certain 2D materials such as α-In₂Se₃ exhibit ferroelectricity, maintaining electric polarization after voltage is removed. This property enables non-volatile memory and mimics synaptic activity by adjusting charge density in response to voltage.
In this article, we take a look at the landscape of neuromorphic computing, specifically the emerging role of 2D materials in enabling next-generation neuromorphic devices and how the field is moving beyond traditional chip designs.
2D Materials for Neuromorphic Computing: Patenting Activity
1Data extracted from patent publications that specifically mention graphene, hexagonal boron nitride (h-BN), or transition metal dichalcogenides (TMDs) like MoS₂ in connection with neuromorphic computing. This may not capture the full landscape of neuromorphic device innovations, particularly those that use alternative descriptors for 2D materials or focus on enabling components such as synaptic emulators or resistive switching elements without explicitly referencing material composition.
Global patent filings on 2D materials in neuromorphic computing have steadily risen since 2015, with a sharp surge beginning in 2019. This growth aligns with advances in AI, particularly machine learning and edge computing, that demand energy-efficient, brain-inspired hardware. Early recognition of neuromorphic computing’s potential, such as its inclusion in the World Economic Forum’s 2015 emerging tech list, helped fuel this momentum. Much of this activity centers around materials like graphene, hexagonal boron nitride (h-BN), and transition metal dichalcogenides (TMDs) like MoS₂, as identified through a targeted search of patent publications (see note 1).
Key milestones highlighted this tech’s real-world potential. In 2019, China unveiled a hybrid AI chip for autonomous bicycles, and in 2022, Tsinghua University developed a low-power neuromorphic chip for robotics. Academic research also surged, with studies such as the 2022 Roadmap on Neuromorphic Computing and Engineering and IEEE papers on memristor-based synapses highlighting how 2D materials like graphene and MoS₂ support miniaturized, low-power neuromorphic circuits.
Patent activity peaked in 2023, coinciding with the breakout year for generative AI. The rapid adoption of large language models spurred demand for new computing architectures, accelerating innovation in neuromorphic systems using emerging materials.
2D Materials for Neuromorphic Computing: Top Assignees
Patent ownership in 2D materials for neuromorphic computing is led by academic and research institutions, with Fudan University and Shaanxi German National Micro Semiconductor Technology among the most active assignees. Fudan’s strong activity is driven by its Institute of Science and Technology for Brain-inspired Intelligence (ISTBI), which leads interdisciplinary research in brain-inspired algorithms, neural modeling, and AI hardware.
Several other universities and research institutions, especially in China and South Korea, also show strong activity, underscoring regional investment in advanced materials. Notably, Samsung, in collaboration with Harvard, proposed a groundbreaking “copy-paste” approach to reverse engineer brain connectivity onto memory chips for neuromorphic computing. Merck KGaA, Darmstadt, Germany, through its Intermolecular unit, is accelerating analog AI hardware development by combining high-throughput materials experimentation with machine learning to optimize multi-element memory stacks.
2D Materials for Neuromorphic Computing: Top Jurisdictions
China leads in patent filings related to 2D materials for neuromorphic computing. This dominance is reflected in the top assignees, which are primarily Chinese institutions and companies. The United States and South Korea follow at a distance, with significantly fewer filings. Other jurisdictions such as India, Europe, and Taiwan show moderate levels of activity. The distribution of filings reflects varying levels of engagement across regions in the development and protection of innovations involving 2D materials in neuromorphic computing.
2D Materials for Neuromorphic Computing: Top Legal Representatives
Legal representation in 2D materials patent filings for neuromorphic computing is concentrated among a handful of key intellectual property firms, with a clear dominance by China-based agencies. Beijing Trusted Intellectual Property Agency LTD leads the field with the highest volume of filings, followed by Huazhong University’s own patent agency and several regional IP firms.
The presence of university-affiliated and regional agencies reflects strong domestic support structures for patent activity in China’s research institutions. While a few international firms such as Harness, Dickey & Pierce and Johnson IP appear on the list, the data highlights the largely localized nature of legal representation in this domain. Notably, individual attorneys like Tim Tingkang Xia and Soo-Yeul Lee also make the list.
2D Materials for Neuromorphic Computing: Top Technology Areas
Thermoelectric devices are leading innovation in neuromorphic computing. Devices classified under H10N, thermoelectric junctions of dissimilar materials, account for 25.1% of filings, reflecting a strong push for energy-efficient hardware in brain-inspired systems. Computational models (G06N) and semiconductor devices (H01L) followed, reflecting algorithmic advances in mimicking neural behavior and the importance of integrated hardware, respectively. Memory devices (H10B), static memory (G11C), and inorganic semiconductors (H10D) also saw significant activity, contributing to the core infrastructure of neuromorphic systems.
This growing activity across hardware and architecture classes reflects a deepening focus on efficient, brain-like processing. Among the standout innovations are patents using 2D materials to push the boundaries of memory performance, one of the key enablers of neuromorphic systems.
Next-gen memory architectures: 2D materials powering neuromorphic innovation
Recent patents showcase breakthroughs in low-power, high-density memory devices using MoS₂, graphene, and perovskites to emulate synaptic behavior.
2D flash memory for low power and high retention
U.S. Patent No. 11,800,705 addresses a key challenge in neuromorphic computing: overcoming power and performance limitations tied to the von Neumann architecture and sneak path issues in two-terminal synaptic devices. The patent describes a flash memory structure optimized for neuromorphic systems, leveraging a three-terminal design to separate read and write paths, effectively mitigating leakage currents without the need for selector devices.
The device uses 2D materials like MoS₂ and graphene to form the channel and floating gate, enabling low-power operation and improved charge retention, key for mimicking synaptic weights. Few-layered MoS₂ offers a ~1.2 eV bandgap and 4.3 eV electron affinity, while the floating gate may include graphene, graphene oxide, or carbon nanotubes.
A thin low-k tunneling insulator and a thicker high-k blocking layer (e.g., Al₂O₃, HfO₂) provide efficient charge control. Fabrication involves MOCVD, atomic layer lamination, and CVD, with Ti/Au electrodes added via e-beam lithography.
The patent was filed on November 30, 2021, and was granted on October 24, 2023 to Korea Institute of Science & Technology. The inventors are Joon Young Kwak, Eunpyo Park, Suyoun Lee, Inho Kim, Jong-Keuk Park, Jaewook Kim, Jongkil Park, and YeonJoo Jeong. Steven M. Rabin, Robert H. Berdo, William E. Curry et al. of Rabin & Berdo, P.C. represented the assignee during the application process.
Stacked 2D memristor enables scalable synaptic switching
Precise resistance control at the atomic scale is essential for advancing neuromorphic computing. Traditional memristors often rely on bulk switching materials that lack scalability and fine-tuned switching behavior.
U.S. Patent No. 11,374,171 presents a resistance switching structure built entirely from stacked two-dimensional materials, designed to enhance control over synaptic behavior while operating at low voltage. The invention addresses the need for highly integrated, low-power, and randomly accessible memory architectures. By leveraging 2D materials, the patent enables scalable and efficient memory cell designs suitable for neural networks and advanced computing systems.
The system includes upper and lower electrodes made from different materials, separated by a resistance change layer composed solely of two stacked 2D layers, such as MoS₂, WS₂, or black phosphorus. This double-layer exhibits bipolar switching characteristics, where resistance decreases with repeated positive voltage sweeps and increases with repeated negative sweeps. Defects, including grain boundaries and line-type imperfections, facilitate conductive filament formation, enabling stable, low-voltage (0.1–0.5 V) operation with consistent ohmic slopes in both high and low resistance states.
Filed on March 19, 2020, and granted on June 28, 2022, the patent was assigned jointly to Samsung Electronics Co., Ltd. and Harvard University. The inventors are Minhyun Lee, Dovran Amanov, Renjing Xu, Houk Jang, Haeryong Kim, Hyeonjin Shin, Yeonchoo Cho, and Donhee Ham. William J. Coughlin, John M. White, Andrew J. Telesz et al. from Harness, Dickey & Pierce, P.L.C. served as the legal representative during the filing process.
Perovskite device integrates sensing, memory, and neuromorphic functions
Achieving multi-functional optoelectronic performance in a single device remains a central goal in neuromorphic and photodetection technologies. Conventional devices often require separate components for light detection, memory, and switching, limiting integration and energy efficiency. U.S. Patent No. 11,676,772 presents a unified architecture that enables photodetection, optical memory, and neuromorphic functionality through the use of perovskite nanocrystals and charge-separating structures.
The patented device includes a perovskite nanocrystal (NC) layer composed of CsPbI₃, FAPbBr₃, or similar halide perovskites, paired with a carbonaceous or semiconducting charge-separating layer, such as graphene or carbon nanotubes, sandwiched between an insulating metal oxide layer and gate electrodes. This structure generates a photocurrent in response to light exposure and can operate across a broad wavelength range (200–1500 nm) with extremely high optical responsivity (up to 1.1×10⁹ A/W), enabling optical switching and memory behavior under varying voltages or light pulses.
The patent was filed on November 9, 2021, and granted on June 13, 2023. It was assigned to the Alliance for Sustainable Energy. The inventors are Jeffrey Lee Blackburn and Ji Hao. John C. Stolpa, Sam J. Barkley, Michael McIntyre et al. from Alliance for Sustainable Energy, LLC represented the assignee during the patent process.
Neuromorphic hardware roadmap
Neuromorphic hardware is reshaping AI and edge computing by mimicking the brain’s efficiency, using spiking neural networks (SNNs) and event-driven processing to deliver ultra-low-power, low-latency performance. As demand for brain-inspired computing surges, the global neuromorphic computing market is projected to grow at a staggering 89.7% CAGR, reaching USD 1.3 billion by 2030. This growth is fueled by rising adoption across edge devices, autonomous systems, and healthcare technologies.
Momentum began in the 2010s with efforts like IBM’s TrueNorth and the Human Brain Project, laying the foundation for today’s breakthroughs. Intel’s Hala Point, powered by Loihi 2, boasts 50× faster AI processing with 100× less energy use than traditional chips. Innatera’s T1 combines SNN, CNN, and RISC-V for real-time intelligence in sensor devices.
Edge-focused advances include BrainChip’s Akida, which supports on-chip learning, and Heidelberg’s BrainScaleS 2, simulating over 500 neurons and 200,000 synapses. SynSense introduced the Speck vision chip and compact Xylo, while Charlotte Frenkel’s ReckOn demonstrated real-time learning with minimal power via the e-prop algorithm. Scalable systems like SpiNNaker 2 (153 ARM cores) and DynapCNN push performance at low power.
As neuromorphic chips evolve, they are enabling breakthroughs in self-driving technologies, brain-computer interfaces, and smart edge devices, paving the way for a future of adaptive and energy-efficient computing.
