Neuromorphic AI chips in 2026: overview
Neuromorphic AI chips are processors built to compute using principles inspired by biological neural systems. Many implement spiking neural networks, where events rather than continuous operations carry information. This can reduce unnecessary computation and support low-latency processing close to sensors, making the approach relevant to edge devices with tight power budgets. The main challenge is turning these potential efficiency gains into reliable system-level performance: algorithms, memory, communication networks and hardware must work together.
Recent work is shifting attention from individual neuron and circuit designs to deployment and co-design. In October 2026, Zhejiang University’s ConSwap and FlowMap addressed congestion-aware mapping and scheduling for concurrent spiking networks. August 2026 work explored synapse compression and neuron pruning, photonic SNN design, and CMOS-integrated transistor-based artificial neurons and synapses. Publication activity rose 26.7% in the last 12 months, while patent activity fell 38.8%, pointing to stronger research momentum than recent patenting.
- 1Mapping is a bottleneck: Zhejiang University’s ConSwap and FlowMap show that congestion and scheduling can limit large neuromorphic systems even when their compute cores are efficient.
- 2Co-design is widening: Recent work spans digital, photonic, resistive-memory and CMOS-plus-device designs, with algorithms increasingly tailored to the hardware they will run on.
- 3Use cases are broadening: Biomedical and wearable systems are gaining research momentum alongside established edge sensing, robotics and industrial monitoring applications.
Top organizations working on neuromorphic AI chips
Ranked by Linknovate activity score · last 24 monthsThe activity score adds up the relevance of every patent, publication, grant and news item linking an organization to neuromorphic AI chips. Showing 7 of 1,928 organizations.
| # | Organization | Approach | Signals (24 mo) | Momentum | Why it matters |
|---|---|---|---|---|---|
| 1 | 
Intel Corporation
Corporate · United States |
Digital event-driven ASICs | PAT 2 · PUB 19 GRANTS 1 · NEWS 334 |
-44% | Intel develops Loihi neuromorphic processors, using event-driven spiking neural networks for low-power research and edge-computing applications. |
| 2 | 
Zhejiang University
Research · China |
Digital event-driven ASICs | PAT 4 · PUB 191 NEWS 6 |
-14% | Recently, Zhejiang University researchers proposed congestion-aware SNN mapping and flow-aware scheduling methods for deploying concurrent networks on neuromorphic hardware. |
| 3 | 
Chinese Academy of Sciences CAS
Research · China |
Digital event-driven ASICs | PUB 242 · NEWS 23 | +52% | Recently, researchers associated with the Chinese Academy of Sciences developed a spatially fused, event-driven accelerator for spiking convolutional networks. Its work also addresses neuromorphic interconnects and edge-processing architectures. |
| 4 | 
IBM
Corporate · United States |
Digital event-driven ASICs | PAT 1 · PUB 8 NEWS 255 |
-41% | IBM’s TrueNorth research demonstrated a digital neuromorphic processor built from event-driven cores and on-chip memory. The company’s neuromorphic work explores brain-inspired computing architectures. |
| 5 | 
University of Chinese Academy of Sciences
Research · China |
Memristive and resistive-memory hardware | PUB 178 · NEWS 9 | +6% | Researchers affiliated with the University of Chinese Academy of Sciences study hardware for spiking networks, including in-memory accelerator co-design and memristive synapses. Their work spans processor architecture and device-level implementations. |
| 6 | 
University of Electronic Science and Technology of China
Research · China |
Digital event-driven ASICs | PUB 137 · NEWS 3 | -8% | Recently, researchers at the University of Electronic Science and Technology of China developed an energy-efficient architecture for on-chip SNN learning. Their work also covers event-driven accelerators and parallel spiking-neuron designs. |
| 7 | 
Peking University
Research · China |
Memristive and resistive-memory hardware | PUB 139 · NEWS 10 | +33% | Peking University research spans hardware for spiking networks, including ferroelectric tunnel-junction synapses and digital neuromorphic processors with on-chip learning. |
| 8 | ··Hidden organizationCompany | Approach | PAT 00 · PUB 00 GRANTS 0 · NEWS 00 | +00% | Available to registered users. |
| 9 | ··Hidden organizationCompany | Approach | PAT 00 · PUB 00 GRANTS 0 · NEWS 00 | +00% | Available to registered users. |
| 10 | ··Hidden organizationCompany | Approach | PAT 00 · PUB 00 GRANTS 0 · NEWS 00 | +00% | Available to registered users. |
Startups to watch and leading research centers
Startups to watch 29 total
Innatera NanosystemsNetherlandsInnatera develops ultra-low-power spiking-neural-network processors for edge devices. Its patent portfolio covers SNN encoding and adaptation methods for neuromorphic-chip deployment.
SynSenseChinaSynSense develops low-power neuromorphic processors for edge inference, combining event-driven computation with sensor processing. Published work has demonstrated sub-milliwatt epilepsy detection on a spiking-network edge processor.
SpiNNcloud Systems GmbHGermanySpiNNcloud commercializes large-scale, parallel neuromorphic computing based on SpiNNaker2. Its platform supports real-time spiking-network execution and research into event-based learning.- ··Hidden startupCountryAvailable to registered users, with funding and founding year.
- ··Hidden startupCountryAvailable to registered users, with funding and founding year.
Leading research institutions 1,650 total
- Chinese Academy of Sciences CASChinaRecently, researchers associated with the Chinese Academy of Sciences developed a spatially fused, event-driven accelerator for spiking convolutional networks. Its work also addresses neuromorphic interconnects and edge-processing architectures.
- Zhejiang UniversityChinaRecently, Zhejiang University researchers proposed congestion-aware SNN mapping and flow-aware scheduling methods for deploying concurrent networks on neuromorphic hardware.
- University of Chinese Academy of SciencesChinaResearchers affiliated with the University of Chinese Academy of Sciences study hardware for spiking networks, including in-memory accelerator co-design and memristive synapses. Their work spans processor architecture and device-level implementations.
- ··Hidden research institutionCountryAvailable to registered users, with their key researchers.
- ··Hidden research institutionCountryAvailable to registered users, with their key researchers.
Where neuromorphic AI chips innovation happens
Share of documents with at least one organization from each country.
Open the interactive map with all 1,928 organizationsMain technical approaches
Last 24 months| Approach | What it is | Documents | Leading organizations |
|---|---|---|---|
| Digital event-driven ASICs | Digital ASICs route and process spikes on specialized hardware; they offer programmable, scalable execution but depend on efficient mapping and interconnects. | 872 |
SNN Inc., Intel Corporation, Cc (Creative Conspiracy) |
| Analog and mixed-signal circuits | Analog and mixed-signal circuits implement neuron dynamics directly in electrical signals, potentially reducing energy but making precision, variability and programmability harder to manage. | 819 |
Cc (Creative Conspiracy), SNN Inc., 4 Front Inc. |
| Memristive and resistive-memory hardware | Memristive hardware combines memory and computation in device arrays, reducing data movement but facing challenges in device variability, endurance and integration. | 605 |
Cc (Creative Conspiracy), 4 Front Inc., SNN Inc. |
| Photonic hardware | Photonic systems use light to carry out neural operations, offering fast parallel processing but requiring effective integration, learning methods and electronic control. | 220 |
Intel Corporation, Xidian University, Nvidia |
| FPGA and reconfigurable implementations | FPGAs let developers adapt neuromorphic designs without fabricating a custom chip, but flexibility can come with limits in power, density and deployment complexity. | 378 |
Cc (Creative Conspiracy), Intel Corporation, SNN Inc. |
Emerging trends
Last 12 months vs previous 36 monthsTerms whose frequency in the last 12 months grew the most compared with the previous 36 months.
Funding and investment
Recent funding rounds
Funding indicators are broad: the data lists 343 grants totaling $687.8 million since 2020 and 24 funding-round entries totaling $22.6 billion. The entries include university and semiconductor financing, so they should not be read as neuromorphic-specific capital.
Patent landscape
Top patent holders · last 24 monthsRecent developments
Latest news, duplicates removed- Oct 2026Abstract Spiking neural networks (SNNs) are a promising computational paradigm for resource- and energy-constrained edge applications. However, their implementation on field-programmable gate arrays (FPGAs) remains a…Frontiers in Neuroscience | New and Recent Articles
- Aug 2026Currently, AI and edge computing are hitting a wall when it comes to power budgets, latency, and cloud dependence, hindering real-world deployments. This issue is complicated by the need to integrate a great deal of…Home | Electronic Design
- Aug 2026Abstract Energy consumption is becoming one of the main constraints for neural network inference on edge devices, where compute, memory, and power budgets are tight. A large number of works already study the efficiency…Frontiers in Neuroscience | New and Recent Articles
- May 2026Abstract This paper presents a neuromorphic processing system integrating a compressed sensing spiking neural network (CSSNN) designed for sparse signal classification. The proposed CSSNN combines data coding, data…Frontiers in Neuroscience | New and Recent Articles
- Month 20XXHidden developmentAvailable to registered users with alerts enabled.
Frequently asked questions about neuromorphic AI chips
What companies are working on neuromorphic AI chips?
Intel Corporation, IBM, Samsung, Innatera Nanosystems and SynSense are among the companies active in neuromorphic AI chips. Intel develops Loihi processors for event-driven spiking networks, while IBM’s TrueNorth work explored digital neuromorphic architecture. Innatera and SynSense focus on low-power processors for edge inference. SpiNNcloud Systems GmbH is commercializing large-scale parallel neuromorphic computing based on SpiNNaker2.
Which neuromorphic AI chip startups are most promising in 2026?
Innatera Nanosystems, SynSense and SpiNNcloud Systems GmbH are notable startup examples in 2026. Innatera develops low-power spiking processors for edge devices and holds 14 patents in the listed patent-holder data. SynSense combines event-driven computing with sensor processing, while SpiNNcloud commercializes parallel neuromorphic computing based on SpiNNaker2. Their different focuses span edge inference and larger-scale systems.
Which universities and research centers lead neuromorphic AI chip research?
The Chinese Academy of Sciences, Zhejiang University and the University of Chinese Academy of Sciences are prominent research organizations in the data. Recent work includes the Chinese Academy of Sciences’ event-driven accelerator for spiking convolutional networks and Zhejiang University’s mapping and scheduling methods. Other active groups include Tsinghua University, Peking University, Seoul National University and the Korea Advanced Institute of Science and Technology.
How much funding has neuromorphic AI chips received?
The funding indicators list 343 grants totaling $687.8 million since 2020, alongside 24 funding-round entries totaling $22.6 billion. These figures include broad university, infrastructure and semiconductor financing, so they are not a clean measure of capital dedicated only to neuromorphic AI chips. Treat them as directional funding context rather than a sector-specific total.
Who holds the most neuromorphic AI chip patents?
Innatera Nanosystems leads the listed patent holders with 14 patents. Tata Consultancy Services Ltd. has 9, while Samsung and SEOUL NATIONAL UNIVERSITY R&DB FOUNDATION each have 8. These figures identify organizations with the highest counts in the supplied patent-holder list; they do not establish the scope, validity or commercial importance of any individual patent.
Is neuromorphic AI chip patent activity growing?
Patent activity declined 38.8% in the last 12 months compared with the previous 12 months, while the 24-month window contains 195 patents. This contrasts with publication activity, which rose 26.7% in the last year. The figures suggest research output is strengthening while recent patenting has slowed, though they do not show whether individual technologies are closer to commercialization.
What are the latest developments in neuromorphic AI chips?
Zhejiang University’s ConSwap and FlowMap, published in October 2026, target congestion-aware mapping and scheduling for spiking networks. Other recent directions include synapse compression and dynamic neuron pruning, photonic SNN hardware, and transistor-based artificial neuron and synapse arrays. These developments address system scaling, efficient deployment and hardware-level implementation rather than focusing only on individual neuron designs.
What are the main approaches to neuromorphic AI chips?
The five main approaches are digital event-driven ASICs, analog and mixed-signal circuits, memristive and resistive-memory hardware, photonic hardware, and FPGA or reconfigurable implementations. ASICs specialize computation around spike events; analog circuits implement neuron dynamics directly. Memory-based designs aim to reduce data movement, photonics use light for processing, and FPGAs offer flexibility. Each approach trades off efficiency, scalability, precision or ease of deployment.
What are the applications of neuromorphic AI chips?
Applications include edge sensing and IoT, robotics and autonomous systems, industrial monitoring, automotive perception, and biomedical and wearable systems. Biomedical and wearable research grew 36.1% in the recent comparison period, while robotics and autonomous systems grew 1.9%. These applications suit event-driven processing where low latency or constrained power budgets matter, but practical deployment still depends on task performance and hardware fit.
Which countries lead neuromorphic AI chip activity?
The United States leads recorded document activity with 9,574 documents, followed by China with 8,109 and South Korea with 2,772. These figures reflect documents associated with each country, not market share or manufacturing output. The research landscape also includes substantial activity in the United Kingdom, Germany and India, indicating that development is distributed across several regions.