Human mimetic sensory-interfaced neuromorphic devices and their training mechanism

Abstract Neuromorphic devices lead to artificial intelligence (AI) innovation, intricately engineered to mirror the architecture and cognitive functions of the brain. Their bioinspired design enables unparalleled efficiency, adaptability, and parallel processing, transcending the constraints of traditional computing. Among them, human mimetic sensory-interfaced neuromorphic (HMSIN) devices emerge as transformative breakthroughs, seamlessly integrating multiple sensory modalities such as vision, hearing, taste, smell, and touch. These devices grant machines human-like perception, allowing precise interpretation and response to stimuli. Despite notable advancements, HMSINs still face major hurdles in emulating the complexity of the human sensory system. Achieving stable performance, high selectivity, and real-time responsiveness remains a formidable task. These challenges are further exacerbated by environmental variability, signal interference, and material limitations. To overcome such limitations, researchers are exploring hybrid HMSIN sensors that combine multiple sensory mechanisms into single platforms to improve multimodal integration and robustness. Building on these efforts, recent advances in HMSINs research focus on enabling systems to adapt and learn by dynamically modulating their internal states in response to external cues, resulting in changes in conductance parameters. These changes generate electrical spikes that closely mimic neuronal firing. With repeated exposure, the devices progressively refine their response. This process encodes memory and enhances selectivity and sensitivity over time. This review explores HMSIN devices with an emphasis on fabrication materials, manufacturing strategies, and key performance factors such as learning and adaptation. We also examine brain-inspired mechanisms, including Hebbian learning and spike-timing-dependent plasticity (STDP), and their integration with neuromorphic hardware for intelligent perception and decision-making. Furthermore, we discuss the typical layered and interconnected architectures of HMSIN devices that closely mimic biological sensory systems. These architectures enable dynamic signal processing and adaptation. We also address current challenges and outline future directions for next generation bioelectronic applications. This exploration provides novel concepts for new neuromorphic devices that are operable as sensory-interfaced human brains.