Polymer-Electrolyte-Gated Ultra-Low Energy Consumption Synaptic Transistors Based IGZO with Aluminum Nanoparticles

Rising demands for energy-efficient computation have driven the development of neuromorphic electronics, mimicking human brain synaptic functions while enabling information storage. Particularly, polymer electrolyte-gated transistors (EGTs) based metal-oxide have been proven pivotal in neuromorphic devices due to their low-voltage operation and wide electrochemical stability. However, the neuromorphic devices based on metal-oxides studied so far still face challenges with high power consumption compared to the energy efficiency of the human brain (5 fJ/spike). Additionally, when using only a polymer-electrolyte, the devices exhibit minimal long-term synaptic potentiation (LTSP) effects. This limitation poses a challenge in replicating key synaptic functionalities crucial for neuromorphic applications, as LTP is essential for mimicking learning and memory processes observed in biological systems. Therefore, we suggest a neuromorphic EGTs that introduces nanoparticles at the interface between the polymer electrolyte and the solution-processed IGZO channel layer.
In this study, aluminum nanoparticles were deposited using a thermal evaporator on the channel layer, and native oxide components subsequently formed on the surface of the nanoparticles. These aluminum oxide components bind with the lithium ions in the polymer electrolyte through coordination bonding. We controlled the amount of aluminum nanoparticles to induce cation channel trapping, enabling an expanded hysteresis window and a threshold voltage shift in the IGZO channel layer. Cation channel trapping in the dielectric layer could be controlled by the amount of aluminum nanoparticles, resulting in slower relaxation of cations, which leads to enlarged hysteresis loops and a shift in the threshold voltage of the IGZO channel layer.
A systematic investigation of the electrical and synaptic properties confirmed that the EDL and trap effect within the electrolytes are critical to the stable operation of the device and the realization of excellent synaptic functionalities. Based on the MNIST dataset, a neural recognition accuracy of 91.97% was achieved. The energy consumption was measured at 0.62 pJ/spike, demonstrating significantly lower energy consumption compared to existing metal-oxide-based synaptic devices. This research on ultra-low energy neuromorphic devices is expected to make significant contributions to the advancement of next-generation robotics, bioelectronics, and neuroscience technologies.