Neuromorphic computing with multi-memristive synapses

Area of Science:

• Neuromorphic Engineering
• Artificial Intelligence
• Materials Science

Background:

• Neuromorphic computing aims to create next-generation intelligent systems.
• Memristive devices are proposed for synaptic weights in artificial neural networks.
• Precise control of memristor conductance is a key challenge for network accuracy.

Purpose of the Study:

• To present a multi-memristive synaptic architecture for improved neuromorphic computing.
• To address the challenge of precise synaptic weight modulation in artificial neural networks.
• To demonstrate the effectiveness of the proposed architecture for both spiking and non-spiking neural networks.

Main Methods:

• Developed a multi-memristive synaptic architecture with a global counter-based arbitration scheme.
• Focused on phase change memory devices and developed a comprehensive device model.
• Conducted simulations for spiking and non-spiking neural networks and performed large-scale experiments.

Main Results:

• Demonstrated the effectiveness of the multi-memristive architecture via simulations.
• Achieved successful unsupervised learning of temporal correlations using a spiking neural network with over a million phase change memory devices.
• Showcased the potential for precise conductance modulation over a wide dynamic range.

Conclusions:

• The proposed multi-memristive synaptic architecture is a significant step towards large-scale, energy-efficient neuromorphic computing.
• The efficient arbitration scheme effectively manages device conductance for high network accuracy.
• Experimental validation confirms the viability of the approach for real-world applications in artificial intelligence.