As power and latency bottlenecks grow, engineers are exploring neuromorphic chips to deliver low-energy, real-time AI at the edge of embedded and IoT systems.
Neuromorphic computing, or hardware modeled on the brain’s neurons and spikes, is rapidly moving from research labs into real-world engineering. In this edge-computing era, companies and researchers are building “brain-like” chips that process data only when events occur.
The result? Dramatically higher energy efficiency and much lower latency than traditional GPUs and CPUs. Spiking neural networks are radically different from conventional AI. Inspired by the brain’s event-driven “spiking” behavior, SNNs use asynchronous spikes instead of fixed numeric activations., “ offer high accuracy and easy-to-use tools but are computationally intensive and consume significant power. SNNs utilize bio-inspired, event-driven architectures that can be significantly more energy-efficient, but they rely on less mature training tools.”that Loihi-based neuromorphic chips can perform AI inference and optimization using 100 times less energy, at speeds 50 times faster, than conventional CPU/GPU systems. IBM’s TrueNorth chipease of use, whereas SNNs “can be significantly more energy-efficient” but require novel training methods . Researchers are closing this gap, but for now, spiking AI often slightly lags behind standard AI in raw accuracy. Brain-inspired SNNs process data in milliseconds with microjoules, not seconds with joules. Their event-driven “fire only when needed” paradigm delivers drastically lower latency and power, even if specialized algorithms and hardware are needed.The global race: From Loihi to Darwin Monkey Tech giants and nation-states are investing heavily in neuromorphic chips. In the US, Intel and IBM have led the way: IBM’s 2014 TrueNorth featured 1 million neurons, 256 million synapses, and drew just 65 mW while performing 58 Giga-synaptic ops/s. Intel’s labs followed with Loihi, Loihi 2, and large systems, such as Pohoiki Beach and Hala Point . Intelthose prototypes run 50× faster with 100× less energy than GPUs, and that packing 1152 Loihi2 chips into its Hala Point“orders of magnitude gains in speed and efficiency”. BrainChip and startups like SynSense are also pushing neuromorphic IP. BrainChip’s fully digital Akida SNN core is designed for edge AI in cars and cameras. On the other side, China is surging. Beijing’s tech plans pour tens of billions into AI hardware, with neuromorphic chips serving as a flagship. For example, China’sbuilt a rack-scale neuromorphic computer with over 2 billion neurons and 100 billion synapses, yet it draws only ~2,000 W of power. This system uses 960 custom neurosynaptic chips across 15 racks, rivaling Intel’s Hala Point in scale but at one-tenth the power. China’s government-backed initiatives complement this:Intel, IBM, and others are closely monitoring these developments. Intel’s Loihi and Loihi2 chips each model 1–2 million neurons with up to hundreds of millions of synapses, while TrueNorth’s million neurons operate with event-driven parallelism—figures dwarfed by the Darwin project’s billions. Yet, each approach offers unique strengths. IBM recently committed ~$30 million to a next-gen neuromorphic R&D effort . EU and Japan also fund neuromorphic research . The global race pits Western chip-makers and startups against Chinese state-backed ventures. Analysts predict neuromorphic will spread across all edge AI niches. With governments and industry investing heavily in brain-like processors, this is a leading front in the Industry 4.0 semiconductor arms race.Neuromorphic chips are not just lab curiosities—real systems are already demonstrating ultra-fast, low-power industrial performance. Consider autonomous drones and robots: at TU Delft, researchers built a fully neuromorphic flying drone. Its event-camera input and control network ran on an Intel Loihi chip on board, achieving processing speeds up to 64 times higher than a GPU, while consuming approximately three times less energy. As Neuroscience News highlighted, “the drone’s deep neural network processes data up to 64 times faster and consumes three times less energy than when running on a GPU.” These efficiency gains translate to sub-millisecond reaction times for obstacle avoidance and control, which are crucial for agile robots. Real-time vision and sensing are another strong suit. Event cameras and spiking processors excel at high-speed inspection. In manufacturing trials, neuromorphic vision counted parts on a conveyor and detected weld defects far faster than conventional cameras.used an event camera to count corn kernels on a fast-moving feeder line, demonstrating accurate item counting in a production setting. Others placed event cameras on welding robots, which allowed them to “visualize welding dynamics with superior temporal fidelity” and flag anomalies in real-time. In vibration-based monitoring, an event-based stereo systemmechanical vibrations with the same precision as costly laser vibrometers. In practice, these demos show neuromorphic AI achieving millisecond-range sensing and decision loops.developed an SNN on Loihi for pump vibration data, achieving 97% fault-detection accuracy with nearly zero false negatives. This was accomplished using only 0.0032 J per inference on Loihi, compared to 11.3 J on an x86 CPU. That’s a 1,000× power reduction for real-time anomaly detection at the sensor. Similarly, simple IoT nodes, such as smart cameras and microphones, can offload heavy AI tasks to tiny neuromorphic chips. BrainChipThe technical payoff highlighted above suggests a market explosion. Analysts project that the edge-AI and IoT markets will rapidly embrace neuromorphic hardware to solve looming power and latency bottlenecks. For one, the number of IoT “things” is skyrocketing:suggest about 29 billion connected devices by 2027, driven by smart cities, manufacturing, and consumer wearables. With each device requiring tiny, efficient AI, industry experts predict that on-device intelligence will become the dominant approach. Gartner analysts have suggested that the majority of new IoT products will incorporate local AI by the late 2020s, and ABI Researchthe neuromorphic computing market will swell from ~$28.5 million in 2024 to $1.32 billion by 2030 . Similarly, Fortune Business Insightsthat “AI PC” shipments will increase from approximately 50 million units in 2024 to over 167 million by 2027.how it “could make AI calculations significantly more energy-efficient and faster”. The Mercedes R&D team reports that a neuromorphic vision system could reduce autonomous-driving compute energy by ~90% compared to today’s stack. Startups like SynSense and BrainChip claim hundreds of millions in private funding, and analysts say the global chip giants are rapidly adding neuromorphic to their roadmaps. In summary, with tens of billions of dollars projected for AI edge spending, analysts foresee that a majority of new IoT devices will run on brain-inspired chips within a few years. The oft-repeatedis “70 percent of IoT by 2027,” which aligns with IoT growth forecasts, meaning neuromorphic could power edge AI in most devices soon. In summary, neuromorphic edge computing is transitioning from theory to practice in Industry 4.0. From autonomous drones to smart factories, brain-inspired chips are delivering milliseconds of response and milliwatts of power consumption.it, these systems represent an “order of magnitude” leap in efficiency. With governments and companies investing billions in this technology, neuromorphic hardware is poised to drive the next generation of industrial IoT, robotics, and intelligent automation.Srishti started out as an editor for academic journal articles before switching to reportage. With a keen interest in all things science, Srishti is particularly drawn to beats covering medicine, sustainable architecture, gene studies, and bioengineering. When she isn't elbows-deep in research for her next feature, Srishti enjoys reading contemporary fiction and chasing after her cats.
Darwin Monkey Deep Neural Networks Internet Of Things Loihi Neuromorphic Chips Neuromorphic Edge Computing Spiking Neural Networks Truenorth
United States Latest News, United States Headlines
Similar News:You can also read news stories similar to this one that we have collected from other news sources.
Rapper Young Bleed Dead at 51 Following Brain AneurysmRapper Young Bleed has died from complications of a brain aneurysm.
Read more »
Four Universities Launch Major New Investments In Brain ScienceOctober saw four universities — Rice, MIT, the University of Cincinnati & the University of California San Francisco — launch major initiatives in brain science.
Read more »
Rapper Young Bleed dead at 51 after brain aneurysmRapper Young Bleed died of a brain aneurysm Saturday in Las Vegas, following his collapse at an afterparty from a recent performance. He was 51.
Read more »
Camberley mum warns pregnant women not to rely on private scansMother-of-three Charlotte says her son's brain abnormalities were missed by private sonographers.
Read more »
Stop Overloading the Wrong Part of Your Brain at WorkTry this cognitive audit.
Read more »
Four ways to ignite our brain’s hidden potential.Kinesia paradoxa, a remarkable phenomenon, reminds us that when we feel stuck, the brain is often just waiting for the right cue to unlock a hidden pathway.
Read more »