Modeling homosynaptic and heterosynaptic plasticity with a single neuromemristive synapse

Efforts are being made to design a brain-like intelligence due to its robustness, synaptic modification (i.e., learning and memory), analog synaptic multiplication, multi-state storage, ultra-low power consumption, and parallel computation. However, current bioelectronic and biomedical technologies have yet to fully replicate brain-like intelligence. In particular, devices that can emulate both homosynaptic and heterosynaptic plasticity remain extremely limited.

This work presents a neuromemristive synapse capable of replicating key biological features, including homosynaptic and heterosynaptic plasticity. The proposed synapse uses a memristor, a promising candidate for achieving bio-realistic features of synapse due to its low power consumption, multi-state operation, analog behavior, high data storage durability, and CMOS compatibility.

The artificial synapse is designed as a composite 1-port structure consisting of a memristor (M) and a controlled capacitor (CCon). The memristor is responsible for emulating synaptic plasticity in response to distinct brainwave patterns, while the capacitor modulates the discharge rate through the memristor. During the active input phase (synaptic potentiation), the composite 1-port charges up. During the inactive phase (synaptic depression), CCon governs the discharging of the memristor, enabling full or partial discharging through memory fading or homosynaptic and heterosynaptic depression pathways.

We designed the proposed synapse in SPICE and validated its bio-functionalities through various simulations. The proposed synapse demonstrates low power consumption and replicates key neurobiological processes for learning and memory such as heterosynaptic homeostasis, modular input specificity, associativity, and homosynaptic long-term and short-term potentiation and depression (LTP, LTD, STF, STD), along with memory fading effect (MFE), and strong stimulation (SST).

The proposed synapse bio-realistically mimics synaptic plasticity, neurotransmitter dynamics, and neuronal responses. Implemented using off-the-shelf components, it supports both volatile and non-volatile modes, making it suitable for CMOS integration. This enables advancements in spiking neural networks, brain function analysis, and scalable neuromorphic computing systems.