ReRAM-based Neo-Hebbian Synapses For Training Neuromorphic HW (IIT Madras, UCSB)

Researchers from IIT Madras and UC Santa Barbara have published a technical paper introducing a novel NeoHebbian artificial synapse using ReRAM devices for neuromorphic systems. This synapse incorporates two state variables: a neuron coupling weight and an "eligibility trace" for synaptic weight updates. The coupling weight is encoded in ReRAM conductance, while the "eligibility trace" is encoded in local temperature modulated by voltage pulses. The synapse was tested for tasks like temporal signal classification and reinforcement learning, showing promise for efficient implementation of advanced learning rules in neuromorphic hardware.