# Neuromorphic computing

> Source: https://aiwiki.ai/wiki/neuromorphic_computing
> Updated: 2026-06-09
> Categories: AI Hardware, AI Infrastructure, Neural Networks
> From AI Wiki (https://aiwiki.ai), a free encyclopedia of artificial intelligence. Quote with attribution.

# Neuromorphic Computing

Neuromorphic computing is an approach to designing computer hardware and software that draws direct inspiration from the structure and function of biological [neural networks](/wiki/neural_network). Unlike conventional processors built on the [von Neumann architecture](/wiki/von_neumann_architecture), which separate memory and processing into distinct units, neuromorphic systems integrate computation and memory at the same physical location, closely mirroring how neurons and synapses operate in the human brain. The goal is to achieve the brain's remarkable efficiency in pattern recognition, sensory processing, and adaptive learning while consuming only a fraction of the energy required by traditional computing hardware.

The term "neuromorphic" was coined by Carver Mead in 1990,[1] and the field has since grown into a major research area spanning chip design, algorithm development, and real-world deployment. Neuromorphic processors use [spiking neural networks](/wiki/spiking_neural_networks) (SNNs) as their primary computational model, processing information through discrete electrical pulses (spikes) rather than continuous numerical values. This event-driven paradigm enables neuromorphic chips to remain largely idle when no input is present, yielding dramatic power savings compared to [GPUs](/wiki/gpu_computing) and [TPUs](/wiki/tensor_processing_unit_tpu) that run continuous clock-driven computations.

## History

### Origins and Carver Mead

The intellectual foundations of neuromorphic computing trace back to the 1980s at the California Institute of Technology (Caltech), where Carver Mead, a professor of electrical engineering and computer science, began exploring how analog [VLSI](/wiki/vlsi) circuits could replicate biological neural systems. In 1989, Mead published *Analog VLSI and Neural Systems*, the first book to systematically apply analog VLSI design principles to neural computation.[2] The book demonstrated how the organizing principles of nervous systems could be realized in silicon integrated circuits, with working examples including analog chips that could see (silicon retina) and hear (silicon cochlea).

In 1990, Mead published the seminal paper "Neuromorphic Electronic Systems" in the *Proceedings of the IEEE* (Volume 78, Issue 10, pp. 1629-1636), formally introducing the term "neuromorphic" to describe VLSI systems containing electronic analog circuits that mimic neuro-biological architectures present in the nervous system.[1] This paper is widely cited as the founding document of the neuromorphic engineering field.

### Early Development (1990s-2000s)

Throughout the 1990s and early 2000s, neuromorphic research remained primarily academic. Researchers at institutions including Caltech, ETH Zurich, and the University of Manchester developed analog and mixed-signal circuits that emulated individual neurons and small networks. Key milestones during this period included the development of the address-event representation (AER) protocol for communicating spikes between chips and the creation of early neuromorphic vision sensors.[14]

The field gained momentum with the launch of DARPA's SyNAPSE (Systems of Neuromorphic Adaptive Plastic Scalable Electronics) program in 2008, which funded IBM and other organizations to develop brain-inspired computing architectures at scale.[3]

### Modern Era (2010s-Present)

The 2010s marked a turning point with the introduction of large-scale neuromorphic hardware. IBM unveiled [TrueNorth](/wiki/truenorth) in 2014, a chip containing one million spiking neurons.[3] Intel introduced [Loihi](/wiki/intel_loihi) in 2017,[4] followed by Loihi 2 in 2021.[16] The University of Manchester completed the million-core SpiNNaker machine in 2018.[17] Commercial products from companies like BrainChip began shipping in 2022, signaling the transition from research prototypes to deployable technology.[18]

In April 2024, Intel deployed Hala Point at Sandia National Laboratories, the world's largest neuromorphic system, containing 1.15 billion neurons across 1,152 Loihi 2 processors.[5] In August 2025, researchers at [Zhejiang University](/wiki/zhejiang_university) unveiled Darwin Monkey, a brain-inspired system built from 960 Darwin 3 chips that supports more than 2 billion spiking neurons, overtaking Hala Point as the largest neuromorphic system by neuron count.[19][20]

## Key Concepts

### Spiking Neural Networks (SNNs)

Spiking neural networks are the computational backbone of neuromorphic systems. Unlike traditional [artificial neural networks](/wiki/artificial_neural_networks) (ANNs) that use continuous activation values (such as those produced by [ReLU](/wiki/relu) or sigmoid functions), SNNs communicate through discrete spikes, binary events that occur at specific points in time. This is fundamentally closer to how biological neurons operate: a neuron accumulates incoming signals, and when its membrane potential crosses a threshold, it fires a spike that propagates to connected neurons.

The most commonly used neuron model in neuromorphic hardware is the leaky integrate-and-fire (LIF) model. In this model, a neuron's membrane potential increases with each incoming spike weighted by synaptic strength, gradually decays ("leaks") over time, and generates an output spike when the potential exceeds a threshold. After firing, the potential resets and the neuron enters a brief refractory period.[14]

SNNs encode information not just in spike rates (how frequently a neuron fires) but also in the precise timing of individual spikes. This temporal coding enables SNNs to represent time-varying signals naturally, making them well suited for processing audio, video, and other time-series data.

### Event-Driven Computation

Conventional processors operate on a synchronous clock, performing computations at every clock cycle regardless of whether there is meaningful work to do. Neuromorphic systems, by contrast, follow an event-driven (asynchronous) paradigm: computation occurs only when a spike arrives. If a region of the chip receives no input, it consumes nearly zero power.

This event-driven approach delivers two major benefits. First, it produces sparse computation, since only a small fraction of neurons are active at any given time in a typical SNN, most of the chip remains idle. Second, it enables low-latency responses, because spikes propagate through the network without waiting for a global clock signal, allowing the system to react to inputs with minimal delay.

### In-Memory Computing

One of the most significant bottlenecks in conventional computing is the "memory wall," the energy and time cost of moving data between separate memory and processing units. Neuromorphic architectures address this by co-locating computation and memory. Synaptic weights are stored directly at the processing site, eliminating the need for expensive data transfers.

Several approaches to in-memory computing are used in neuromorphic systems:

| Approach | Description | Examples |
|---|---|---|
| SRAM-based | Digital memory cells store synaptic weights adjacent to processing elements | [Intel Loihi](/wiki/intel_loihi), [IBM TrueNorth](/wiki/truenorth) |
| Memristor-based | Resistive memory devices whose conductance represents synaptic weight, enabling analog multiply-accumulate operations | Research prototypes from HP, University of Michigan |
| Phase-change memory | Non-volatile memory that uses phase transitions in chalcogenide materials to store analog synaptic weights | IBM research systems |
| Floating-gate transistors | Charge stored on floating gates represents long-term synaptic weights with analog precision | Intel's earlier analog neuromorphic circuits |

[Memristors](/wiki/memristor) are particularly promising for neuromorphic computing because they naturally exhibit properties analogous to biological synapses: their resistance changes based on the history of electrical signals passed through them, enabling a form of physical learning.[15]

### Synaptic Plasticity and On-Chip Learning

A defining feature of neuromorphic systems is their support for on-chip learning through synaptic plasticity rules. The most well-known is spike-timing-dependent plasticity (STDP), which strengthens or weakens synaptic connections based on the relative timing of pre-synaptic and post-synaptic spikes. If a pre-synaptic neuron fires just before the post-synaptic neuron, the connection is strengthened (long-term potentiation). If the order is reversed, the connection is weakened (long-term depression).

This local learning rule does not require a global error signal or backpropagation through the entire network, making it naturally suited to distributed, low-power hardware. Intel's Loihi chips support programmable on-chip learning rules, including STDP variants, enabling the network to adapt in real time without cloud connectivity.[4]

## Major Neuromorphic Hardware Platforms

### Intel Loihi and Loihi 2

Intel's Neuromorphic Computing Lab, led by Mike Davies, introduced the first Loihi chip in 2017. Fabricated on Intel's 14nm process, Loihi contains 128 neuromorphic cores, 3 x86 processor cores, and over 33 MB of on-chip SRAM. It supports up to 130,000 neurons and 130 million synapses across its 2.07 billion transistors.[4]

Loihi 2, announced in September 2021, represents a major architectural leap. Built on the Intel 4 process node, it packs up to 1 million neurons and 120 million synapses into 128 redesigned asynchronous neuron cores. Key improvements include up to 10x faster spike processing, a fully programmable neuron model, graded spikes (up to 32-bit resolution compared to 1-bit in the original), and a 50% reduction in die area.[16]

In April 2024, Intel unveiled Hala Point, the world's largest neuromorphic system.[5] Housed in a six-rack-unit data center chassis roughly the size of a microwave oven, it packages 1,152 Loihi 2 chips with 140,544 neuromorphic processing cores. The system supports 1.15 billion neurons and 128 billion synapses while consuming a maximum of 2,600 watts.[5] Applied to spiking neural network models at full capacity, Hala Point runs 20 times faster than real-time biological neural processing.[5]

### IBM TrueNorth

IBM's TrueNorth chip, published in the journal *Science* in August 2014, was one of the earliest large-scale neuromorphic processors. Developed under DARPA's SyNAPSE program and led by Dharmendra Modha, TrueNorth contains 4,096 neurosynaptic cores, each modeling 256 neurons and 65,536 synapses, for a total of 1 million neurons and 256 million synapses.[3]

Fabricated in Samsung's 28nm CMOS process, TrueNorth integrates 5.4 billion transistors while consuming only 65-70 mW during operation, yielding a power density roughly 1/10,000 that of conventional microprocessors.[3] The architecture uses a globally asynchronous, locally synchronous (GALS) design with event-driven communication between cores via hierarchical asynchronous routers.[7]

TrueNorth operates with a 1ms global timestep, updating all neurons and propagating spikes every millisecond. While it demonstrated remarkable energy efficiency, TrueNorth had limitations: its neuron model was relatively simple (no on-chip learning), and synaptic weights were limited to binary or ternary values.[7]

In 2023, IBM introduced NorthPole, a successor architecture that blends neuromorphic design principles with support for conventional [deep learning](/wiki/deep_learning) inference, targeting 8-bit integer precision workloads.[21] In the *Science* paper describing the chip, IBM reported that NorthPole achieved 25 times higher frames per second per watt than a GPU built on a comparable 12nm process on the ResNet-50 image classification benchmark.[21] At the IEEE High Performance Extreme Computing conference in September 2024, IBM researchers presented a 16-card NorthPole research server running a 3-billion-parameter language model derived from [IBM Granite](/wiki/ibm_granite) at under 1 millisecond per token, which IBM described as 46.9 times lower latency than the next most energy-efficient GPU it measured, with system throughput of 28,356 tokens per second.[22]

### BrainChip Akida

BrainChip, an Australian-American semiconductor company, developed the Akida neuromorphic processor specifically for commercial edge AI applications. The first-generation AKD1000 chip shipped in January 2022, making it one of the first commercially available neuromorphic processors.[18]

Akida employs event-based processing where computations are triggered only by new sensory input. Power is consumed only when a neuron's inputs exceed a threshold and generate an output event. The chip supports on-chip learning, allowing models to adapt in the field without cloud connectivity.[8]

The second-generation Akida platform adds support for 8-bit weights and activations, improved vision transformer acceleration, multi-pass sequential processing, and configurable local scratchpads for optimized memory access.[8] BrainChip also licenses its Akida technology as IP for integration into third-party SoCs, and has a partnership with ARM for edge deployment.

BrainChip provides the MetaTF development environment, a complete [machine learning](/wiki/machine_learning) framework for creating, training, and deploying neural networks on Akida hardware.[8]

In October 2024, BrainChip introduced Akida Pico, an ultra-low-power co-processor built on the second-generation Akida architecture that operates in the microwatt-to-milliwatt range, targeting always-on applications such as keyword spotting, wake-word detection, and presence detection in battery-powered wearables and consumer devices.[23] Akida Pico executes compact Temporal Event-based Neural Network (TENN) models compiled through the same MetaTF tool flow.[23]

### SynSense Speck and Xylo

SynSense, headquartered in Zurich, Switzerland, produces neuromorphic processors targeting ultra-low-power edge applications.

The Speck chip combines an event-driven vision sensor with a neuromorphic processor on a single die. It contains 328,000 spiking integrate-and-fire neurons organized across 9 configurable convolutional layers with pooling. The Speck 2f variant consumes approximately 1 mW during operation, achieving a neuron density of 10,900 neurons per square millimeter. It uses in-memory computing techniques for sparse, event-driven neural network computations.[10]

The Xylo chip targets audio and signal processing applications. It supports up to 1,000 leaky integrate-and-fire neurons with configurable synaptic and membrane time constants, 16 input channels and 8 output channels. Fabricated in a 28nm CMOS process on a 6.5 mm² die, Xylo achieves 219 microwatts idle power and just 93 microwatts dynamic inference power for audio classification.[9]

In February 2024, SynSense announced the acquisition of iniVation, the Zurich-based event-camera maker with which it had co-developed Speck, with the combined organization operating as the SynSense Group and positioning itself as an end-to-end neuromorphic sensing and processing company.[24]

### SpiNNaker (University of Manchester)

SpiNNaker (Spiking Neural Network Architecture) is a massively parallel computing platform designed by Steve Furber and the Advanced Processor Technologies (APT) research group at the University of Manchester. Unlike other neuromorphic platforms that use custom neuron circuits, SpiNNaker uses standard ARM processors to simulate spiking neurons in software, providing flexibility in the neuron models that can be implemented.[6]

The full-scale SpiNNaker machine, completed in 2018, contains 57,600 custom SpiNNaker chips, each with 18 ARM968 processor cores and 128 MB of mobile DDR SDRAM.[6] This totals 1,036,800 cores and over 7 TB of RAM, housed in 10 server racks and consuming approximately 100 kW. Each core can simulate about 1,000 neurons in real time, giving the system a total capacity of roughly one billion neurons.[17]

SpiNNaker was funded in part by the European Union's Human Brain Project and is used primarily for neuroscience research, specifically large-scale brain simulation.[17]

SpiNNaker 2, a next-generation system being developed at TU Dresden in collaboration with the University of Manchester, uses a new chip fabricated in GlobalFoundries' 22nm FD-SOI process.[11] Each SpiNNaker 2 chip contains 152 ARM Cortex-M cores with dedicated machine learning and neuromorphic accelerators, 19 MB of on-chip SRAM, and 2 GB of DRAM.[25] The full-scale SpiNNaker 2 system is projected to contain 5.2 million cores across 720 boards, achieving a 10x improvement in neural simulation capacity per watt over the original system through adaptive body biasing and near-threshold voltage operation.[11]

SpiNNaker 2 technology is being commercialized by SpiNNcloud Systems, a Dresden-based spin-off. In June 2025, SpiNNcloud announced that Sandia National Laboratories had deployed a SpiNNaker2 system whose server boards each carry 48 chips with 152 cores apiece; the installation can simulate between 150 and 180 million neurons and ranked among the five largest brain-inspired computing platforms at the time, supporting Sandia's research into energy-efficient computing for national security workloads.[25]

### BrainScaleS (Heidelberg University)

BrainScaleS, developed at Heidelberg University and likewise supported by the Human Brain Project, takes a different approach from the digital platforms above: its neuron and synapse circuits are implemented in analog electronics that physically emulate membrane dynamics in continuous time. Because the analog circuits use scaled-down time constants, the second-generation BrainScaleS-2 chip runs about 1,000 times faster than biological real time, compressing long learning experiments into seconds.[26] Each BrainScaleS-2 chip integrates 512 adaptive integrate-and-fire neurons and roughly 131,000 plastic synapses alongside embedded processors that implement programmable "hybrid" plasticity rules.[26]

### Tianjic (Tsinghua University)

Tianjic, developed by Luping Shi's team at [Tsinghua University](/wiki/tsinghua_university), is a hybrid chip that supports both spiking neural networks and conventional artificial neural networks on the same many-core fabric, allowing the two model families to be combined within a single application. The chip, which appeared on the cover of *Nature* in August 2019, contains 156 cores supporting about 40,000 neurons and 10 million synapses, and was demonstrated controlling an autonomous bicycle that performed object detection, tracking, voice-command recognition, obstacle avoidance, and balance control.[27]

### Darwin 3 and Darwin Monkey (Zhejiang University)

Zhejiang University's State Key Laboratory of Brain-Machine Intelligence develops the Darwin series of neuromorphic chips. Each third-generation Darwin 3 chip supports more than 2.35 million spiking neurons and over 100 million synapses.[19] In August 2025, the laboratory unveiled Darwin Monkey (also called Wukong), which the university describes as the world's first brain-inspired computer with more than 2 billion spiking neurons; the system distributes 960 Darwin 3 chips across 15 servers and consumes about 2,000 watts under typical operation.[19] Its neuron count approaches the scale of a macaque brain, and the machine runs a dedicated Darwin brain-inspired operating system that uses load-aware scheduling and dynamic time-slicing to share the hardware across workloads, including simulations of C. elegans, zebrafish, mouse, and macaque brains.[19] Independent coverage noted that Darwin Monkey's neuron count exceeds that of Intel's Hala Point, previously the largest neuromorphic computer.[20]

### Innatera Pulsar

Innatera, a Dutch neuromorphic semiconductor company, announced Pulsar in May 2025, describing it as the first mass-market neuromorphic microcontroller for the sensor edge.[28] Pulsar combines analog and digital spiking neural network blocks with a conventional convolutional neural network accelerator and a RISC-V processor core on a single chip, and the company claims up to 100 times lower latency and up to 500 times lower energy consumption than conventional edge AI processors on always-on sensing tasks.[28]

## Comparison with Conventional AI Hardware

Neuromorphic processors differ fundamentally from GPUs and TPUs in their architecture, computational model, and target workloads. The following table summarizes the key differences.

| Feature | Neuromorphic Chips | [GPUs](/wiki/gpu_computing) | [TPUs](/wiki/tensor_processing_unit_tpu) |
|---|---|---|---|
| **Computational model** | Spiking neural networks (event-driven) | Parallel floating-point operations (clock-driven) | Systolic array matrix multiplication (clock-driven) |
| **Memory architecture** | In-memory computing; synaptic weights stored at processing site | Separate DRAM (HBM/GDDR); data shuttled between memory and cores | Separate HBM; optimized data flow through systolic arrays |
| **Power consumption** | Milliwatts to low watts (e.g., Loihi 2: ~1 W per chip; Akida: sub-milliwatt) | Hundreds of watts (e.g., NVIDIA H100: 700 W TDP) | Hundreds of watts (e.g., Google TPU v5e: ~200 W per chip) |
| **Processing paradigm** | Asynchronous, event-driven; computes only when spikes arrive | Synchronous, clock-driven; computes every cycle | Synchronous, clock-driven; optimized for dense matrix ops |
| **Precision** | Binary spikes or low-bit graded spikes (1-32 bit) | FP16, BF16, FP32, FP64, INT8 | BF16, FP32, INT8 |
| **Training support** | On-chip learning (STDP, reward-modulated plasticity); limited backpropagation | Full backpropagation with automatic differentiation | Full backpropagation; optimized for large-batch training |
| **Latency** | Microsecond-scale for spike propagation | Millisecond-scale for batch inference | Millisecond-scale for batch inference |
| **Best suited for** | Edge inference, sensory processing, temporal data, always-on monitoring | Training and inference for [large language models](/wiki/large_language_model), [computer vision](/wiki/computer_vision), general-purpose AI | Large-scale training and inference for transformers, recommendation systems |
| **Software ecosystem** | Emerging (Lava, snnTorch, Norse, Brian2) | Mature (CUDA, [PyTorch](/wiki/pytorch), [TensorFlow](/wiki/tensorflow)) | Mature (JAX, TensorFlow, PyTorch/XLA) |
| **Commercial availability** | Limited (BrainChip Akida, SynSense Speck/Xylo, Intel Loihi via research program) | Widely available | Available via Google Cloud |

## Advantages of Neuromorphic Computing

### Energy Efficiency

The most compelling advantage of neuromorphic processors is their energy efficiency. Because computation occurs only in response to events, and because data does not need to travel between separate memory and processing units, neuromorphic chips consume dramatically less power than conventional hardware for appropriate workloads.

Specific benchmarks illustrate the scale of these savings:

- A 2025 study comparing Intel's Loihi 2 to an NVIDIA GTX 1080 GPU on image classification found that the neuromorphic system achieved a **99.5% reduction in energy consumption** and a 76.7% reduction in inference time.
- A 2024 IEEE case study reported that Loihi-based systems can perform sensor-fusion workloads using **100x less energy than a CPU** and **30x less energy than a GPU**.[29]
- An LLM adapted to run on Loihi 2 in 2025 matched the accuracy of a comparable GPU-based model while using roughly **half the energy**.[30]
- IBM's TrueNorth demonstrated a power density approximately 1/10,000 that of conventional processors.[3]

These efficiency gains are most pronounced for sparse, event-driven workloads. For dense matrix operations typical of [transformer](/wiki/transformer) training, GPUs and TPUs remain more efficient.

### Temporal Processing

Neuromorphic systems process time as a native dimension of computation. Because spikes carry timing information and neuron dynamics unfold over time (with membrane potentials rising, decaying, and triggering at specific moments), SNNs can naturally represent and process temporal patterns without the need for explicit time-windowing or recurrence mechanisms like [LSTMs](/wiki/lstm) or attention over sequence positions.

This makes neuromorphic hardware particularly effective for:

- **Audio processing:** Recognizing speech, environmental sounds, or music in streaming fashion
- **Motion detection:** Tracking objects using event-based cameras that output spikes only when pixels change
- **Biomedical signals:** Analyzing EEG, ECG, and EMG signals that are inherently temporal
- **Robotics:** Processing continuous sensory input for real-time motor control

### Adaptability and Online Learning

Neuromorphic chips that support on-chip learning can adapt to new data without retraining in the cloud. This is valuable for edge deployments where connectivity is intermittent or where privacy requirements prevent sending data off-device. BrainChip's Akida and Intel's Loihi both support forms of on-chip learning, enabling applications like personalized keyword recognition, anomaly detection in industrial settings, and adaptive robotics.

## Applications

### Edge AI and IoT

Neuromorphic processors are well suited for [edge computing](/wiki/edge_computing) deployments where power budgets are tight and real-time responses are essential. Their milliwatt-level power consumption enables always-on sensing without frequent battery replacement. Applications include smart home devices with always-on keyword detection, industrial IoT sensors that monitor equipment vibrations for predictive maintenance, environmental monitoring stations in remote locations, and wearable health monitors that continuously analyze biometric signals.

### Sensory Processing

The event-driven nature of neuromorphic computing aligns naturally with event-based sensors such as dynamic vision sensors (DVS), which output spikes only when pixels detect changes in brightness. This pairing eliminates the redundancy inherent in conventional frame-based cameras, which capture full frames at fixed intervals regardless of scene activity.

Neuromorphic sensory processing systems have demonstrated strong performance in high-speed object tracking, gesture recognition, optical flow estimation, and auditory scene analysis. Event cameras paired with neuromorphic processors can track objects at microsecond temporal resolution while consuming a fraction of the power of conventional vision systems.

### Robotics

Neuromorphic systems offer several advantages for [robotics](/wiki/robotics) applications: low-latency sensory processing for rapid reactions, energy efficiency for battery-powered mobile platforms, and the ability to learn and adapt on the fly. Research groups have demonstrated neuromorphic controllers for robotic arms, legged locomotion, drone navigation, and tactile manipulation.

Neuromorphic neuroprosthetics have also shown promise in restoring sensory feedback to amputees, where low-power spike-based processing of tactile sensor data enables real-time closed-loop control of prosthetic limbs.

### Autonomous Vehicles

Event-based cameras and neuromorphic processors are being investigated for [autonomous driving](/wiki/autonomous_driving) applications. The ability to detect changes in a visual scene at microsecond resolution, combined with extremely low latency, could improve reaction times for obstacle detection and avoidance. Several automotive companies and research labs are exploring neuromorphic sensor fusion, combining event cameras with lidar and radar data on neuromorphic hardware.

### Scientific Computing and Brain Simulation

Large-scale neuromorphic systems like SpiNNaker are used for computational neuroscience research, simulating biologically realistic neural circuits to study brain function. The European Union's Human Brain Project funded SpiNNaker specifically for this purpose, enabling researchers to run real-time simulations of brain regions with millions of neurons.

## Software Frameworks and Tools

The software ecosystem for neuromorphic computing, while less mature than the GPU-based [deep learning](/wiki/deep_learning) ecosystem, has grown substantially in recent years.

| Framework | Developer | Description |
|---|---|---|
| **Lava** | Intel | Open-source framework for developing neuromorphic applications; supports CPUs and Loihi hardware |
| **snnTorch** | Open source | Deep learning library for SNNs built on [PyTorch](/wiki/pytorch); focuses on gradient-based training with GPU acceleration |
| **Norse** | Open source | Extends PyTorch with bio-inspired primitives for event-driven spiking neural networks |
| **Brian2** | Open source | Python library for simulating SNNs; notable for flexible equation-based neuron model definitions |
| **NEST** | Open source | Simulator for large-scale spiking neural network models; used extensively in computational neuroscience |
| **Nengo** | Applied Brain Research | Neural engineering framework supporting both conventional and neuromorphic backends |
| **MetaTF** | BrainChip | Development environment for creating and deploying models on Akida hardware |

A notable interoperability effort is the Neuromorphic Intermediate Representation (NIR), which provides a unified format for exchanging SNN models across 7 simulators (Lava, Nengo, Norse, Rockpool, Sinabs, snnTorch, and Spyx) and 4 hardware platforms (Loihi 2, Speck, SpiNNaker 2, and Xylo).[12]

### Training SNNs

Training spiking neural networks presents unique challenges because the spike generation function is non-differentiable, meaning standard [backpropagation](/wiki/backpropagation) cannot be directly applied. Several approaches have been developed:

- **Surrogate gradients:** Replace the non-differentiable spike function with a smooth approximation during the backward pass, enabling end-to-end gradient-based training. This has become the most widely used technique.[13]
- **ANN-to-SNN conversion:** Train a conventional ANN and then convert it to an equivalent SNN by mapping activation values to spike rates. This leverages existing training infrastructure but may lose temporal coding advantages.
- **Local learning rules:** Use biologically inspired plasticity rules (STDP, reward-modulated learning) that operate locally at each synapse. These are natively supported by neuromorphic hardware but generally achieve lower accuracy than gradient-based methods on standard benchmarks.
- **Evolutionary and neuroevolution methods:** Use genetic algorithms to optimize SNN architectures and parameters without gradients.

## Challenges and Limitations

### Programming Model Complexity

Developing applications for neuromorphic hardware requires a fundamentally different programming paradigm than conventional [machine learning](/wiki/machine_learning). Programmers must think in terms of spike rates, membrane potentials, synaptic delays, and plasticity rules rather than tensor operations and loss functions. This steep learning curve limits adoption beyond specialized research groups.

### Training Difficulty

While surrogate gradient methods have made SNN training more accessible, training deep SNNs remains more computationally expensive and less stable than training equivalent ANNs.[13] Backpropagation through time for SNNs with long simulation durations is memory-intensive, and the discrete nature of spikes introduces optimization challenges. On standard benchmarks like [ImageNet](/wiki/imagenet), SNNs still lag behind state-of-the-art ANNs in accuracy, though the gap has been narrowing.

### Ecosystem Maturity

The software ecosystem for neuromorphic computing is fragmented compared to the GPU-based deep learning stack. There is no single dominant framework analogous to PyTorch or TensorFlow. Hardware-specific tools (Lava for Loihi, MetaTF for Akida) are not always interoperable, although initiatives like NIR are working to bridge this gap.[12] Pre-trained SNN models, datasets, and community resources are also far less abundant than their ANN counterparts.

### Scalability for Complex Tasks

Neuromorphic systems have demonstrated strong results on tasks like keyword spotting, gesture recognition, and anomaly detection, but scaling to the complexity of modern large language models or diffusion-based image generators remains an open problem. The first LLM running on neuromorphic hardware was demonstrated on Intel Loihi 2 in 2025, but it was a relatively small model compared to frontier systems.[30] That demonstration ran a 370-million-parameter MatMul-free model quantized for Loihi 2 with no reported loss of accuracy, achieving up to 3 times higher throughput at 2 times lower energy than a comparably sized transformer running on an edge GPU.[30]

### Limited Commercial Availability

As of 2025, commercially available neuromorphic hardware is still limited. Intel's Loihi chips are available primarily through the Intel Neuromorphic Research Community (INRC) for academic and research partners. BrainChip's Akida and SynSense's Speck/Xylo are commercially available but target niche edge applications. There is no neuromorphic equivalent to purchasing an NVIDIA GPU off the shelf for general-purpose AI workloads. The commercial picture nonetheless broadened during 2024 and 2025 with microwatt-class parts aimed at high-volume consumer designs, including BrainChip's Akida Pico and Innatera's Pulsar microcontroller.[23][28]

### Benchmarking and Standardization

The neuromorphic community lacks standardized benchmarks comparable to MLPerf for conventional AI hardware. Different chips use different neuron models, precision levels, and connectivity patterns, making direct performance comparisons difficult. Efforts to establish common benchmarks are underway, but no consensus standard has emerged. The most prominent effort is NeuroBench, a benchmark framework assembled by a collaboration of roughly 100 researchers and published in *Nature Communications* in February 2025; it defines a hardware-independent algorithm track and a hardware-dependent system track, covering tasks ranging from keyword few-shot class-incremental learning to event-camera object detection.[31]

## Current State and Future Outlook

As of early 2026, neuromorphic computing occupies a growing but still specialized niche within the broader AI hardware landscape. Market research forecasts project significant growth, with estimates suggesting the neuromorphic computing market could reach several billion dollars by 2030, driven by demand for energy-efficient edge AI.

Several trends are shaping the field's trajectory:

- **Convergence with conventional AI:** IBM's NorthPole and SpiNNaker 2's dual neuromorphic/DNN acceleration capabilities reflect a trend toward hybrid architectures that blend neuromorphic principles with support for standard deep learning workloads.[21]
- **Improved training methods:** Surrogate gradient techniques and ANN-to-SNN conversion pipelines are steadily closing the accuracy gap between SNNs and ANNs, making neuromorphic hardware more practical for a wider range of applications.
- **New memory technologies:** Advances in memristive devices, phase-change memory, and other emerging non-volatile memory technologies promise to deliver denser, more energy-efficient synaptic storage for future neuromorphic chips.[15]
- **Standardization efforts:** The NIR initiative and growing adoption of open-source frameworks like snnTorch and Norse are gradually building a more unified software ecosystem.[12]
- **Event-based sensors:** The increasing availability of event cameras (from companies like Prophesee and Samsung) and neuromorphic audio sensors creates a natural hardware ecosystem for end-to-end neuromorphic processing pipelines.

The human brain operates on roughly 20 watts while performing feats of perception, reasoning, and motor control that still exceed the capabilities of the most powerful AI systems running on megawatts of GPU clusters. Neuromorphic computing represents the most direct attempt to close that efficiency gap by building computing hardware that works the way biology does, processing information through spikes, learning through local plasticity, and consuming energy only when there is work to be done.

### 2024-2026 developments

The period from late 2024 through mid-2026 brought a cluster of milestones that moved the field from laboratory demonstrations toward larger systems, consumer-priced silicon, and record levels of investment.

| Date | Development |
|---|---|
| September 2024 | IBM researchers presented NorthPole results showing a 3-billion-parameter language model running at under 1 millisecond per token on a 16-card research server.[22] |
| October 2024 | BrainChip announced Akida Pico, a microwatt-class neuromorphic co-processor for always-on wearables.[23] |
| January 2025 | A 23-author review in *Nature*, "Neuromorphic computing at scale," argued that the field had reached a critical juncture at which it must scale up to compete with mainstream AI hardware, and laid out a research roadmap covering neuronal scaling, sparsity, plasticity, and software interoperability.[32] |
| February 2025 | NeuroBench, a community benchmark framework with roughly 100 contributing researchers, was published in *Nature Communications*.[31] |
| March 2025 | Researchers posted the first demonstration of a large language model (a 370-million-parameter MatMul-free architecture) running on Intel's Loihi 2.[30] |
| May 2025 | Innatera launched Pulsar, marketed as the first mass-market neuromorphic microcontroller for the sensor edge.[28] |
| June 2025 | Sandia National Laboratories deployed a SpiNNaker2 system from SpiNNcloud capable of simulating 150 to 180 million neurons.[25] |
| August 2025 | Zhejiang University unveiled Darwin Monkey, the first neuromorphic computer to exceed 2 billion neurons.[19][20] |
| December 2025 | Unconventional AI, a brain-inspired computing startup founded by Naveen Rao, raised a $475 million seed round at a $4.5 billion valuation, one of the largest seed financings on record.[33][34] |
| April 2026 | A Research and Markets analysis estimated that more than $5 billion in venture capital, M&A, and government funding flowed into non-GPU AI computing architectures between early 2024 and March 2026, while putting total commercial revenue across neuromorphic hardware vendors at only $5-10 million, underscoring the gap between investment and adoption.[35] |

The December 2025 financing of Unconventional AI illustrated renewed investor appetite for the field: the company was founded in late 2025 by [Naveen Rao](/wiki/naveen_rao), the former [Databricks](/wiki/databricks) AI head who previously founded Nervana Systems (acquired by Intel) and MosaicML (acquired by Databricks), and its seed round was co-led by [Andreessen Horowitz](/wiki/andreessen_horowitz) and [Lightspeed Venture Partners](/wiki/lightspeed_venture_partners) with participation from Jeff Bezos, [Lux Capital](/wiki/lux_capital), and DCVC.[33][34] Rao has said the company will draw on brain-inspired and analog computation to attack the energy costs of AI data centers.[34]

## See Also

- [Spiking neural networks](/wiki/spiking_neural_networks)
- [GPU computing](/wiki/gpu_computing)
- [Edge computing](/wiki/edge_computing)
- Intel
- [IBM](/wiki/ibm_ai)
- [Deep learning](/wiki/deep_learning)
- [Computer vision](/wiki/computer_vision)
- [Robotics](/wiki/robotics)

## References

1. Mead, C. (1990). "Neuromorphic Electronic Systems." *Proceedings of the IEEE*, 78(10), 1629-1636. DOI: 10.1109/5.58356.
2. Mead, C. (1989). *Analog VLSI and Neural Systems*. Addison-Wesley.
3. Merolla, P.A. et al. (2014). "A million spiking-neuron integrated circuit with a scalable communication network and interface." *Science*, 345(6197), 668-673.
4. Davies, M. et al. (2018). "Loihi: A Neuromorphic Manycore Processor with On-Chip Learning." *IEEE Micro*, 38(1), 82-99.
5. Intel Newsroom (2024). "Intel Builds World's Largest Neuromorphic System to Enable More Sustainable AI." https://newsroom.intel.com/artificial-intelligence/intel-builds-worlds-largest-neuromorphic-system-to-enable-more-sustainable-ai
6. Furber, S.B. et al. (2014). "The SpiNNaker Project." *Proceedings of the IEEE*, 102(5), 652-665.
7. Akopyan, F. et al. (2015). "TrueNorth: Design and Tool Flow of a 65 mW 1 Million Neuron Programmable Neurosynaptic Chip." *IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems*, 34(10), 1537-1557.
8. BrainChip (2022). "Akida Neuromorphic Processor." https://brainchip.com/ip/
9. SynSense (2023). "Xylo: Ultra-low power neuromorphic chip." https://www.synsense.ai/products/xylo/
10. SynSense (2023). "Speck: Event-Driven Neuromorphic Vision SoC." https://www.synsense.ai/products/speck-2/
11. Mayr, C. et al. (2019). "SpiNNaker 2: A 10 Million Core Processor System for Brain Simulation Research." arXiv:1911.02385.
12. Open Neuromorphic (2024). "Neuromorphic Intermediate Representation." Jens E. Pedersen et al. *Nature Communications*, 15, 8344.
13. Neftci, E.O. et al. (2019). "Surrogate Gradient Learning in Spiking Neural Networks." *IEEE Signal Processing Magazine*, 36(6), 51-63.
14. Schuman, C.D. et al. (2017). "A Survey of Neuromorphic Computing and Neural Networks in Hardware." arXiv:1705.06963.
15. TDK (2024). "Cutting AI's Power Consumption Down to 1/100 with Neuromorphic Devices Inspired by the Human Brain." https://www.tdk.com/en/featured_stories/entry_071-neuromorphic-devices.html
16. Intel Newsroom (2021). "Intel Advances Neuromorphic with Loihi 2, New Lava Software Framework and New Partners." https://www.intel.com/content/www/us/en/newsroom/news/intel-unveils-neuromorphic-loihi-2-lava-software.html
17. University of Manchester (2018). "'Human brain' supercomputer with 1 million processors switched on for first time." https://www.manchester.ac.uk/about/news/human-brain-supercomputer-with-1million-processors-switched-on-for-first-time/
18. BrainChip (2022). "BrainChip Achieves Full Commercialization of AKD1000." https://brainchip.com/brainchip-achieves-full-commercialization-akd1000/
19. Zhejiang University (2025). "World's first 2-billion-neuron brain-inspired computer unveiled by ZJU." https://www.zju.edu.cn/english/2025/0910/c19573a3079424/page.htm
20. Live Science (2025). "China's 'Darwin Monkey' is the world's largest brain-inspired supercomputer." https://www.livescience.com/technology/computing/chinas-darwin-monkey-is-the-worlds-largest-brain-inspired-supercomputer
21. Modha, D.S. et al. (2023). "Neural inference at the frontier of energy, space, and time." *Science*, 382(6668), 329-335.
22. IBM Research (2024). "IBM's NorthPole achieves new speed and efficiency milestones." https://research.ibm.com/blog/northpole-llm-inference-results
23. BrainChip (2024). "BrainChip Introduces Lowest-Power AI Acceleration Co-Processor." https://brainchip.com/brainchip-introduces-lowest-power-ai-acceleration-co-processor/
24. SynSense (2024). "SynSense and iniVation join forces to form leading neuromorphic technology provider." https://www.synsense.ai/synsense-and-inivation-join-forces-to-form-leading-neuromorphic-technology-provider/
25. HPCwire (2025). "Sandia Deploys SpiNNaker2 Neuromorphic System from SpiNNcloud." https://www.hpcwire.com/off-the-wire/sandia-deploys-spinnaker2-neuromorphic-system-from-spinncloud/
26. Pehle, C. et al. (2022). "The BrainScaleS-2 Accelerated Neuromorphic System With Hybrid Plasticity." *Frontiers in Neuroscience*, 16, 795876.
27. Pei, J. et al. (2019). "Towards artificial general intelligence with hybrid Tianjic chip architecture." *Nature*, 572, 106-111.
28. Innatera (2025). "Innatera Unveils Pulsar: The World's First Mass-Market Neuromorphic Microcontroller for the Sensor Edge." https://www.prnewswire.com/news-releases/innatera-unveils-pulsar-the-worlds-first-mass-market-neuromorphic-microcontroller-for-the-sensor-edge-302461286.html
29. Isik, M. et al. (2024). "Accelerating Sensor Fusion in Neuromorphic Computing: A Case Study on Loihi-2." 2024 IEEE High Performance Extreme Computing Conference. arXiv:2408.16096.
30. Abreu, S., Shrestha, S.B., Zhu, R.-J., Eshraghian, J. (2025). "Neuromorphic Principles for Efficient Large Language Models on Intel Loihi 2." arXiv:2503.18002.
31. Yik, J. et al. (2025). "The neurobench framework for benchmarking neuromorphic computing algorithms and systems." *Nature Communications*, 16, 1545.
32. Kudithipudi, D. et al. (2025). "Neuromorphic computing at scale." *Nature*, 637, 801-812.
33. TechCrunch (2025). "Unconventional AI confirms its massive $475M seed round." https://techcrunch.com/2025/12/09/unconventional-ai-confirms-its-massive-475m-seed-round/
34. Data Center Dynamics (2025). "Neuromorphic compute startup Unconventional AI raises $475m in seed funding." https://www.datacenterdynamics.com/en/news/neuromorphic-compute-startup-unconventional-ai-raises-475m-in-seed-funding/
35. Research and Markets (2026). "Neuromorphic and Brain-inspired AI Computing Hardware Market Research Report 2026: A Capital Allocation Framework for Four Competing Brain-Inspired Architectures 2025-2030." https://www.globenewswire.com/news-release/2026/04/28/3282902/28124/en/Neuromorphic-and-Brain-inspired-AI-Computing-Hardware-Market-Research-Report-2026-A-Capital-Allocation-Framework-for-Four-Competing-Brain-Inspired-Architectures-2025-2030.html