Nanoarchitectonics for non-volatile ternary STDP synapse using anti-ferroelectric carbon nanotube devices

Anti-ferroelectrics (AFEs) offer unique properties such as double hysteresis window, high endurance, and low latency, making them appealing for neuromorphic computing architectures. This study introduces a novel, compact, and energy-efficient neuromorphic circuit for spike-timing-dependent plasticity (STDP) synapse using AFE field-effect transistors (AFeFETs). Although AFeFETs are volatile and unsuitable for storing synaptic weights, carbon nanotube field-effect transistors (CNTFETs) are employed with multi-threshold operation to achieve nonvolatility by shifting polarization-electric field (P-E) characteristics. By leveraging the tunable hysteresis characteristics and negative differential resistance (NDR) effect of AFeCNTFETs, a nonvolatile ternary weight storage latch is demonstrated to store STDP-governed synaptic weights. Simulation results show that this synapse improves power efficiency by 30% and saves 93% of energy compared to previous works while remaining immune to power outages. This efficient neuromorphic building block paves the way for enhanced neuromorphic learning architectures.