Ultrafast synaptic events in a chalcogenide memristor

Area of Science:

• Materials Science
• Neuroscience
• Computer Engineering

Background:

• Neuromorphic computing aims to mimic the brain's efficiency, but electronic synapses face speed limitations.
• Plastic electronic synapses are crucial for developing compact and power-efficient neuromorphic chips.
• Current memristor-based synapses often exhibit slow synaptic event speeds due to material mechanisms.

Purpose of the Study:

• To investigate the intrinsic memristance of stoichiometric crystalline Ge2Sb2Te5 for electronic synapses.
• To demonstrate ultrafast synaptic plasticity using Ge2Sb2Te5 devices.
• To explore the potential of Ge2Sb2Te5 in developing next-generation neuromorphic computing systems.

Main Methods:

• Utilized stoichiometric crystalline Ge2Sb2Te5 to create memristive devices.
• Employed 30 ns potentiating/depressing electrical pulses to modulate device resistance states (synaptic weights).
• Demonstrated four forms of spike-timing-dependent plasticity (STDP) using programmed pre- and postsynaptic spiking pulse pairs.

Main Results:

• Revealed intrinsic memristance in Ge2Sb2Te5 originating from defect-related charge trapping and releasing.
• Achieved precise modulation of synaptic weights with 30 ns electrical pulses.
• Demonstrated STDP over time windows as short as 500 ns, significantly faster than biological synapses.

Conclusions:

• Stoichiometric crystalline Ge2Sb2Te5 exhibits intrinsic memristance suitable for ultrafast electronic synapses.
• The developed synaptic devices enable spike-timing-dependent plasticity at speeds orders of magnitude faster than the human brain.
• This research offers a pathway for creating ultrafast neuromorphic computing systems that surpass traditional Von Neumann architectures.