Reconfigurable MoS₂ Memtransistors for Continuous Learning in Spiking Neural Networks

Area of Science:

• Materials Science
• Computer Engineering
• Artificial Intelligence

Background:

• Current artificial intelligence (AI) and machine learning (ML) algorithms demand significant energy on conventional hardware.
• Neuromorphic architectures offer a promising alternative for energy-efficient computing.
• Memristive devices are key components for building neuromorphic systems.

Purpose of the Study:

• To introduce a novel memtransistor with gate-tunable dynamic learning behavior.
• To enhance the reconfigurability and learning capabilities of neuromorphic hardware.
• To explore continuous learning in neuromorphic computing.

Main Methods:

• Fabrication of memtransistors using monolayer MoS2 grown on sapphire.
• Utilizing gate pulses to modulate potentiation and depression, mimicking biological systems.
• Simulating spiking neural networks to evaluate learning capabilities.

Main Results:

• The MoS2 memtransistor exhibits gate-tunable dynamic learning behavior.
• Enhanced vertical field effect increases device response reconfigurability.
• Demonstrated diverse learning curves and simplified spike-timing-dependent plasticity.
• Enabled unsupervised and continuous learning in simulated neuromorphic networks.

Conclusions:

• Memtransistor reconfigurability offers unique hardware acceleration for energy-efficient AI and ML.
• The developed device facilitates advanced learning paradigms in neuromorphic computing.
• This work paves the way for more sustainable and powerful AI hardware.