Researchers at Northwestern University have developed a novel brain-inspired computing device that drastically reduces energy consumption while enabling instantaneous anomaly detection in artificial intelligence systems. Published July 10 in Nature Communications, the study outlines a hardware architecture that mimics the cerebellum’s reflexive processing, offering a direct alternative to traditional AI models that continuously analyze data streams regardless of relevance. Led by Mark C. Hersam, Walter P. Murphy Professor of Materials Science and Engineering at Northwestern, alongside collaborators from the Feinberg School of Medicine, Weinberg College, and the University of Illinois Chicago, the team engineered a memtransistor that integrates memory and computation within a single unit. Conventional processors waste significant power shuttling data between separate memory and processing units. The new device bypasses this bottleneck by adopting a cerebellar strategy: rather than processing every input, it remains largely idle until an unexpected event disrupts a balanced signal state. The hardware replicates the cerebellum’s neural circuitry through competing excitatory and inhibitory responses. Built from atomically thin molybdenum disulfide, the asymmetric transistor architecture allows researchers to toggle between synaptic modes by simply reversing the applied voltage. In normal operation, excitatory and inhibitory signals maintain equilibrium. When a novel stimulus occurs, the balance shifts, triggering an immediate alert. This mechanism enables the system to filter routine data and concentrate computational resources solely on anomalies. In proof-of-concept trials using electrocardiogram recordings, the device achieved over 98 percent accuracy in identifying irregular heartbeats. Remarkably, it flagged abnormalities within one-fifth of a heartbeat, operating with approximately 10,000 times fewer computational operations than conventional AI algorithms. The research demonstrates that cerebellum-inspired hardware can outperform traditional models in both speed and efficiency, making it suitable for always-on applications where power constraints and latency are critical. Potential deployment scenarios include wearable health monitors, autonomous vehicles, robotics, and cybersecurity infrastructure. By shifting AI architecture from continuous pattern recognition to event-driven novelty detection, the technology promises to eliminate the computational overhead currently required to monitor steady-state environments. Hersam noted that the team’s next objective involves replicating the cerebellum’s adaptive learning capabilities, allowing the hardware to progressively treat recurring anomalies as normal patterns. This advancement marks a significant step toward decentralized, low-power AI systems capable of real-time decision-making without reliance on centralized data centers. By successfully translating a specific biological neural pathway into functional hardware, the researchers have established a scalable foundation for energy-efficient intelligence across multiple industrial sectors.