Microstructure Controlled Organic Neuromorphic Electronics for Stochastic Update On-Chip Training

Synaptic transistors, which can mimic the functional properties of biological nerves, are gaining attention due to their promising applications in next-generation computing. Implementing the learning and memory functions of their biological counterparts is crucial for such technologies, requiring devices with long-term potentiation (LTP) and high endurance. Among the various types of synaptic transistors, ion-gel gated synaptic transistors (IGOSTs), which operate through ion doping and trapping within polymer semiconducting layers have been widely researched. Since ion behavior in the polymer semiconductor is a critical factor in determining the synaptic properties of these devices, understanding the relationship between polymer microstructure and synaptic performance is essential. Recent studies on IGOSTs based on poly(thienoisoindigo-naphthalene) (PTIIG-Np) and poly(3-hexylthiophene-2,5-diyl) (P3HT) have demonstrated enhanced LTP properties through crystallinity control.
In this study, we propose a new strategy to further enhance LTP and endurance by manipulating the packing structure of semiconducting polymers. A highly confined polymer film was successfully implemented, resulting in an improved retention time, τ₉₀ (the time at which the retention reaches 90% of its maximum value), of 2022.5 seconds. This advancement enabled successful loss convergence when solving a linear regression problem using uniform 8 × 8 arrays, marking the first demonstration of on-chip training with stochastic updates.