Task-Adaptive Neuromorphic Computing Using Reconfigurable Organic Neuristors with Tunable Plasticity and...

Area of Science:

• Neuroscience and Materials Science
• Development of novel organic neuristors for artificial intelligence applications.

Background:

• The brain's dynamic reconfiguration capability inspires energy-efficient learning and inferencing.
• Existing architectures often lack the flexibility for adaptive, low-power AI.

Purpose of the Study:

• To demonstrate an organic neuristor with tunable plasticity and reconfigurable logic-in-memory functions.
• To explore its application in energy-efficient reservoir computing and high-performance artificial intelligence.

Main Methods:

• Utilized a ferroelectric-electrolyte dielectric interface in an organic neuristor.
• Controlled interfacial interactions for tunable short-term and long-term plasticity.
• Implemented logic-in-memory operations for constructing binarized neural networks (BNNs).

Main Results:

• Achieved high recognition accuracy in reservoir computing: 90.6% for images and 97.7% for acoustic signals.
• Constructed BNN hardware circuits demonstrating excellent noise tolerance.
• Attained high BNN accuracies: 99.2% on MNIST and 86.4% on CIFAR-10 datasets.

Conclusions:

• The organic neuristor offers a unified neuron-synapse architecture for adaptive AI.
• Demonstrated potential for power-efficient edge computing and high-performance AI tasks.
• Highlights a pathway toward developing next-generation, energy-efficient artificial intelligence systems.