NbOx synaptic devices for spike frequency dependent plasticity learning

Niobium oxide (NbOx) has been extensively studied for application in resistive random access memory (ReRAM) and threshold switching devices [1]. While most of the previous work in this area has studied non-volatile states in NbOx, we report observations on time-dependent volatile states in NbOx for applications as synaptic devices in neuromorphic computing. The dynamic nature of NbOx is critical as it allows short-term potentiation (STP) and short-term depression (STD) behavior, which can serve as the basis for developing recurrent neural networks (RNN) and spike frequency dependent plasticity (SFDP)-based learning algorithms [2]. SFDP-based learning mechanism in neuro-synaptic arrays has potential to offer an alternative Hebbian learning solution to the more commonly implemented spike-timing dependent plasticity (STDP) based learning schemes. Unlike STDP learning scheme which relies on the relative timing between post-and pre-synaptic neurons and requires complex timing circuitry [3], SFPD relies on the input spike rate into presynaptic neurons to condition the synaptic plasticity. Here we report our observation on the input spike-frequency dependent reconfigurable resistive states in NbOx and retention of these states on various time-scales.