Memristive neural network for on-line learning and tracking with brain-inspired spike timing dependent plasticity

Area of Science:

• Neuroscience
• Computer Science
• Materials Science

Background:

• Brain-inspired computation offers a path toward humanlike cognitive abilities in machines.
• Replicating brain functions like high synaptic connectivity and plasticity requires advanced microelectronic technology.

Purpose of the Study:

• To demonstrate unsupervised learning and pattern tracking in a spiking neural network using memristive synapses.
• To investigate the efficacy of brain-inspired spike-timing-dependent plasticity (STDP) in memristive synapses.

Main Methods:

• Implemented a spiking neural network utilizing hybrid one-transistor/one-resistor (1T1R) memristive synapses.
• Employed spike-timing-dependent plasticity (STDP) for synaptic weight updates based on pre- and post-synaptic spike timing.
• Utilized two synaptic states (low and high resistance) for pattern recognition.

Main Results:

• Achieved unsupervised learning of static patterns.
• Demonstrated tracking of dynamic patterns up to 4x4 pixels.
• Showcased the capability of simple memristive synapses to perform complex learning tasks.

Conclusions:

• Memristive synapses with STDP are effective for unsupervised learning and pattern tracking.
• Scalable memristive neural networks hold promise for future intelligent hardware.
• This approach advances the development of cognitive computing technologies.