A TSTDP memristive synapse based on a comprehensive mathematical model of memory-TFT threshold voltage shift

Area of Science:

• Neuroscience
• Materials Science
• Electrical Engineering

Background:

• Emerging nano-scale technologies like hydrogenated noncrystalline-silicon thin-film transistors (TFTs) and memristors enable low-cost, large-area fabrication and 3D integration.
• Existing synaptic learning rules, such as pair-based spike-timing-dependent plasticity (PSTDP), have limitations in explaining complex biological learning phenomena.

Purpose of the Study:

• To propose a mathematical model for memory-TFT threshold voltage shift due to gate bias instability.
• To develop a novel learning synapse device capable of realizing triplet-based spike-timing-dependent plasticity (TSTDP).

Main Methods:

• Developed a mathematical model for gate bias instability in memory-TFTs.
• Designed a novel learning synapse comprising a voltage/flux driven memristor and a common-source memory-TFT with a memristive load.
• Simulated various spike patterns, including different-frequency/timing spike pairs, triplets, and quadruplets, applied to the proposed device.

Main Results:

• The proposed mathematical model accurately explains threshold voltage shifts in memory-TFTs.
• The novel learning synapse successfully implements TSTDP, a more biologically realistic learning rule than PSTDP.
• Simulations demonstrated a close match between the device's behavior and experimental data from real biological synapses.

Conclusions:

• The developed mathematical model and novel learning synapse device offer a promising platform for advanced neuromorphic computing.
• The TSTDP implementation advances the realism of artificial synaptic learning, moving beyond simpler PSTDP models.
• The findings pave the way for more sophisticated and biologically plausible artificial intelligence systems.