Artificial brain cells that fire in under five nanoseconds, targeting AI hardware that consumes up to 500 times less power than today’s systems, could bring brain-inspired computing closer to everyday devices. Researchers at Oakland University have designed an artificial neuron that combines a tiny magnetic device with a conventional transistor to mimic the way neurons in the human brain communicate. Rather than relying on increasingly complex digital circuits, the design uses the natural behavior of magnetic materials to process information, possibly creating the way for faster and far more energy-efficient AI chips. With the increased use of artificial intelligence, power demand has become a major challenge for the industry. Large AI models require enormous computing resources, driving up electricity use in data centers and making it difficult to run advanced AI directly on portable or edge devices. The researchers stated that the human brain remains the benchmark for efficient computing. It performs billions of operations while consuming only about 20 watts of power—roughly the energy needed to run a household light bulb. By comparison, modern AI systems require dramatically more energy. According to the team, today’s AI inference hardware can consume around 500 times more power than the brain to perform comparable tasks, creating a strong need for new hardware architectures. Instead of building increasingly larger processors, the researchers looked to biology for inspiration. ”Our goal is to make AI practical on everyday devices instead of depending on massive data centers,” the researchers said, adding that the brain’s efficiency comes from the way its neurons communicate through short electrical pulses rather than continuous computation. Tiny magnet acts like a neuron The new artificial neuron consists of a magnetic tunnel junction (MTJ), a nanoscale magnetic component paired with a standard transistor commonly found in mainstream electronics. As current flows through the device, the magnetic element naturally switches states and produces electrical spikes that resemble the firing behavior of biological neurons. Because the device relies on its intrinsic magnetic properties in place of complex digital logic, it requires far fewer circuit components. The researchers said the neuron responds in less than five nanoseconds, permitting information to be processed at extremely high speeds while keeping energy consumption low. The study pointed out that, unlike traditional AI chips that repeatedly move data between processors and memory, the magnetic neuron performs computation where the information is stored. This reduces unnecessary data transfers, one of the biggest contributors to AI’s power consumption. Instead of processing information solely as binary signals, the artificial neurons communicate through the timing of electrical spikes, similar to neurons in the brain. The researchers demonstrated this by connecting multiple magnetic neurons into a small neural network. They trained it to solve the exclusive OR (XOR) problem, a widely used benchmark for testing neural networks. The network successfully learned the task by adjusting spike timing instead of relying on conventional digital calculations. The team said this event-driven approach means the neurons remain largely inactive until new information arrives, helping reduce unnecessary energy use. Toward low-power AI hardware If demonstrated in hardware, the technology could enable energy-efficient AI processors for robots, autonomous systems, smart sensors, wearable electronics, and other edge devices where power is limited. Because the design combines conventional CMOS transistors with magnetic devices already used in commercial memory technologies, the researchers believe it could integrate more easily with present semiconductor manufacturing processes. The work is currently based on simulations, and the next step will be to fabricate and experimentally validate the magnetic neurons under real-world operating conditions. If successful, the approach could present a practical path toward AI hardware that is not only faster but also dramatically more energy efficient. The study was published in IEEE Transactions on Circuits and Systems. Get the latest in engineering, tech, space & science - delivered daily to your inbox.
US-made artificial neurons could fire under 5 nanoseconds with 500x less power
Full Article
Original Source
Read the full article at Interestingengineering →KhanList aggregates and links to publicly available news content. We do not host full articles from third-party sources. Always verify important information with original sources.