TiN/Ti/HfO₂/TiN memristive devices for neuromorphic computing: from synaptic plasticity to stochastic resonance

Area of Science:

• Materials Science
• Neuroscience
• Computer Science

Background:

• Neuromorphic computing aims to replicate brain functionality using artificial systems.
• Memristive devices offer promising characteristics for building energy-efficient and high-performance neuromorphic hardware.
• Understanding and modeling synaptic plasticity is key to developing intelligent learning systems.

Purpose of the Study:

• To characterize TiN/Ti/HfO2/TiN memristive devices for neuromorphic applications.
• To model and analyze synaptic plasticity features, specifically spike-timing-dependent plasticity (STDP).
• To investigate the impact of device variability and stochastic resonance on network performance.

Main Methods:

• Fabrication and electrical characterization of TiN/Ti/HfO2/TiN memristive devices.
• Experimental measurement and analytical modeling of spike-timing-dependent plasticity (STDP).
• Implementation of the STDP model as a learning rule in a spiking neural network (SNN) for MNIST dataset recognition.
• Analysis of device variability and stochastic resonance effects on SNN accuracy.

Main Results:

• The TiN/Ti/HfO2/TiN devices exhibit behaviors mimicking biological synapses, including STDP.
• A model accurately reproduced the measured STDP data.
• The STDP-based SNN achieved recognition of the MNIST dataset, with variability influencing accuracy.
• Stochastic resonance was identified as a significant synaptic feature, highly dependent on noise characteristics.

Conclusions:

• TiN/Ti/HfO2/TiN memristive devices are suitable for neuromorphic computing due to their synaptic plasticity.
• The developed STDP model and its integration into SNNs show potential for learning and recognition tasks.
• Device variability and stochastic resonance are critical factors to consider for robust neuromorphic system design.