Extremely simple neural network and circuit design with multi-influence patterns in neurons for modeling sleep-wake cycles

Neuromorphic circuits that simulate the sleep cycle serve as a critical bridge between neuroscience and hardware implementation. To address the limitations of conventional circuits, which are often complex and lack mechanisms for biological plasticity and memory, this paper presents two highly compact memristor-transistor hybrid neural network circuits designed to simulate transitions between different sleep stages. The core innovations of this design are: 1) utilizing the non-volatile memory property of individual memristors to directly implement the retention of neuronal activation states in hardware, thereby simulating the persistence of sleep stages; and 2) employing a specific memristor connection topology to accurately simulate the indirect and direct inhibitory functions of gamma-aminobutyric acid (GABA)-ergic neurons in the rostromedial tegmental nucleus (RMTg) region on neurons in the laterodorsal tegmentum (LDT) and laterodorsal hypothalamus (LH) regions, using a minimal number of components. Based on this design, corresponding circuits are constructed and simulated using PSPICE. The results demonstrate that this memristive neural network can successfully simulate state transitions between wakefulness, non-rapid eye movement (NREM) sleep, and rapid eye movement (REM) sleep by modelling the activation of GABAergic neurons in the RMTg and their subsequent inhibitory effects in the LDT and LH regions. The designed circuits exhibit a high degree of functional alignment with the target biological neural networks. Furthermore, simulation analysis confirms that the circuit can not only represent normal sleep architecture but also, by adjusting the initial states of key memristors, quantitatively replicate and differentiate the sleep structure fragments between healthy elderly individuals and young adults.