Parallel programming of an ionic floating-gate memory array for scalable neuromorphic computing.

Area of Science:

• Materials Science
• Computer Engineering
• Neuroscience

Background:

• Conventional computing faces efficiency limitations, particularly in artificial neural networks.
• Neuromorphic computing aims to improve efficiency via parallel processing and specialized memory.
• Achieving efficient learning requires selective, linear weight updates and low read currents (<10 nA).

Purpose of the Study:

• To introduce a new ionic floating-gate memory array for efficient neuromorphic computing.
• To demonstrate selective and linear programming of artificial neural network weights.
• To achieve synaptic weight readout with currents below 10 nanoamperes.

Main Methods:

• Utilized a polymer redox transistor integrated with a conductive-bridge memory (CBM).
• Implemented parallel programming by overcoming the CBM bridging threshold voltage for selective weight updates.
• Achieved low read currents by diluting conductive polymer with an insulator.

Main Results:

• Demonstrated selective and linear programming of the redox transistor array in parallel.
• Achieved synaptic weight readout with currents below 10 nanoamperes.
• The memory array demonstrated high endurance (>1 billion write-read operations) and high frequency (>1 MHz).

Conclusions:

• The developed ionic floating-gate memory array meets critical requirements for efficient neuromorphic learning.
• This technology offers a pathway to surpass conventional computing efficiency in AI applications.
• The system shows promise for robust and high-speed neuromorphic hardware.