A multi-timescale synaptic weight based on ferroelectric hafnium zirconium oxide

Area of Science:

• Materials Science
• Neuroscience
• Computer Engineering

Background:

• Brain-inspired computing addresses the challenge of massive data generation in connected societies.
• Biological neural networks excel due to complex short- and long-term memory dynamics.
• Artificial synapses are crucial for replicating biological neural network performance.

Purpose of the Study:

• To engineer sub-µm-sized artificial synaptic weights.
• To mimic short- and long-term memory using ferroelectric and ionic effects.
• To develop a multi-timescale artificial synapse for neuromorphic computing.

Main Methods:

• Utilized ferroelectric space charge effect and oxidation state modulation.
• Employed a ferroelectric field-effect transistor with an oxide channel.
• Engineered multi-timescale behavior by combining fast ferroelectric and slow ionic processes.

Main Results:

• Achieved quasi-continuous resistance tuning over a factor of $$.
• Demonstrated fine-grained weight updates exceeding $$ resistance values.
• Device exhibits high endurance (> $$ cycles) and long retention (> $$ years).

Conclusions:

• The developed artificial synapse effectively mimics biological memory functions.
• The multi-timescale behavior is suitable for advanced neuromorphic and cognitive computing.
• The device shows promising characteristics for future brain-inspired computing applications.