ABSTRACT
To simulate a topological neural network handling weak signals via stochastic resonance (SR), it is necessary to introduce an inherent nonlinearity into nanoscale devices. We use the self-assembly method to successfully fabricate a phase-change quantum-dot string (PCQDS) crossing Pd/Nb:AlNO/AlNO/Nb:AlNO/Pd multilayer. The inherent nonlinearity of phase change couples with electron tunneling so that PCQDS responds to a long signal sequence in a modulated output style, in which the pulse pattern evolves to that enveloped by two sets of periodic wave characterized by neural action potential. We establish an SR mode consisting of several two-state systems in which dissipative tunneling is coupled to environment. Size oscillations owing to NbO QDs adaptively adjust barriers and wells, such that tunneling can be periodically modulated by either asymmetric energy or local temperature. When the external periodic signals are applied, the system first follows the forcing frequency. Subsequently, certain PCQDs oscillate independently and consecutively to produce complicated frequency and amplitude modulations.
ABSTRACT
Memristors based on niobium oxide have attracted wide interest due to their applications in artificial neural network. Phase-locking (PL) characteristics of NbOx newly explored roots in complex nonlinear dynamics and exhibits great potential in periodical signal handling because of its bioinspired nature. In this study, we fabricate a Pd/Nb/Ag-NbOx/Nb/Pd memristive structure and study the effects of Ag doping on the PL characteristics. We clearly see that the NbOx memristor responds to signal in three distinct patterns. For the low-frequency input, the memristor fires spike series and locks the phase near π2 during the positive period of sinusoidal input, and it encodes the input features by the spike number per period. For the middle-frequency input, the memristor fires one spike per period at a definite phase. The output has the same frequency as the input. The locked phase is proportional to the input frequency. For the high-frequency input, the memristor transforms high frequency to low frequency signal and locks at a definite phase higher than 2π. We combined the microstructural analysis and chaos dynamics to know that Ag doping will lower the activation energy, enhance the responding rate, and enhance thermal conductivity, which extends the locked frequency to 1.7 times the value of the undoped NbOx memristor. We also obtained the locked frequency of even 40 MHz after modulating the elementary circuit configuration according to the material modification and chaos model. Our study proposes to handle a signal with high efficiency resembling auditory sense and inspires a novel artificial intelligent computation prototype.