Neural networks composed of artificial neurons and synapses mimicking the human nervous system have received much attention because of their promising potential in future computing systems. In particular, spiking neural networks (SNNs), which are faster and more energy-efficient than conventional artificial neural networks, have recently been the focus of attention. However, because typical neural devices for SNNs are based on complementary metal-oxide-semiconductors that exhibit high consumption of power and require a large area, it is difficult to use them to implement a large-scale network. Thus, a new structure should be developed to overcome the typical problems that have been encountered and to emulate bio-realistic functions. This study proposes a versatile artificial neuron based on the bipolar electrochemical metallization threshold switch, which exhibits four requisite characteristics for a spiking neuron: all-or-nothing spiking, threshold-driven spiking, refractory period, and strength-modulated frequency. Furthermore, unique features such as an inhibitory postsynaptic potential and the bipolar switching characteristic for changing synaptic weight are realized. Additionally, by using a filament confinement technique, a high on/off ratio (≈6 × 107), a low threshold voltage (0.19 V), low variability (0.014), and endurance over 106 cycles are achieved. This research will serve as a stepping-stone for advanced large-scale neuromorphic systems.
- bipolar threshold switches
- electrochemical metallization
- filament confinement
- neuromorphic computing
- spiking neurons
ASJC Scopus subject areas
- Electronic, Optical and Magnetic Materials