About the Event
Neural networks, consisted of interconnected neurons and synapses, enable biological systems to learn, remember, and adapt much more efficiently than von Neumann based computers. Recently, it has been shown that “memristors” are ideal candidates for implementing artificial synapses. Memristors (resistor + memory) are two-terminal solid-state devices with adjustable resistance (conductance) that depends not only on external inputs but also on the past history. The properties of memristors provide the desired connectivity, network density, power consumption, and adjustable weights with memory for building artificial neural networks.
This work begins with the fabrication, electrical studies, and material characterizations of tungsten-oxide (WOX) based memristors. Due to the sub-stoichiometric nature of WOX films, analog resistive switching behavior is observed, which mimics the potentiation and depression processes in biological synapses. Important synaptic functions have been demonstrated in these nanoscale electronic devices, for example, timing-dependent plasticity, rate-dependent plasticity, memory enhancement effect, sliding threshold effect, and heterosynapticity. Finally, integration of WOX memristors and CMOS technology are carried out, facilitating the realization of hybrid memristor/CMOS neural network systems.