STDP and STDP variations with memristors for spiking neuromorphic learning systems

Area of Science:

• Neuroscience
• Materials Science
• Computer Engineering

Background:

• Spike-Timing-Dependent-Plasticity (STDP) is crucial for synaptic learning in neural systems.
• Memristors offer a promising hardware platform for neuromorphic computing due to their analog memory capabilities.
• Current memristor-based STDP implementations often require global synchronization or separate learning/performing phases.

Purpose of the Study:

• To review methods for realizing asynchronous STDP using memristors as synapses.
• To enable memristive synapses to perform synaptic weight multiplications without global synchronization or phase separation.
• To investigate the implementation of STDP rules in memristor cross-bar architectures for artificial vision.

Main Methods:

• Reviewing existing literature on memristor-based STDP.
• Analyzing two memristor physics models: 'moving wall' and 'filament creation/annihilation'.
• Describing the implementation of pure timing-based and hybrid STDP rules in cross-bar architectures.

Main Results:

• Demonstrated memristor-based STDP realization without global synchronization or phase separation.
• Showcased the flexibility of memristors in implementing different STDP rules (timing-based and hybrid).
• Presented potential applications in artificial vision using memristor cross-bar arrays.

Conclusions:

• Asynchronous STDP is achievable with memristors, offering a more biologically plausible and efficient approach.
• Memristor technology can support diverse STDP learning rules and large-scale neural network implementations.
• These advancements pave the way for more sophisticated neuromorphic computing systems, particularly for vision tasks.