Substitution of Ionic Input to Voltage Modulation in Peptide-Based Bimodal Memristor

In recent times, artificial intelligence (AI) has been applied across numerous domains, including image recognition, video analysis, and autonomous driving, serving as a novel computing paradigm. These applications typically involve handling large volumes of unstructured data, demanding experience-based learning analogous to the human brain operation. However, traditional computing architectures, where memory and data processing units are distinct, result in significant time and energy inefficiencies due to the data transfer required between these units during experience-based learning.<br/><br/>To address this challenge, neuromorphic devices inspired by the biological synapse structure have emerged as alternatives to conventional CMOS devices, based on the fact that the human brain operates on a remarkably low power consumption of around 20W. Although they were fabricated based on the structure of actual synapses, there remains a gap between the reported neuromorphic devices and the actual mechanisms at neurotransmitter modulation in biological synapses. Therefore, our research team has developed a memristor using a tyrosine-rich peptide (TRP) as the resistive switching layer. The peptide film exhibits coupled proton and electron transfer due to the redox-active properties of Tyrosine, resulting in high proton conductivity under high humidity conditions. Using this phenomenon, a memristor device incorporating the TRP film as the resistive switching layer was fabricated and RS behavior driven by both electrical (voltage) and ionic (humidity) inputs was observed. Nevertheless, employing humidity as an ionic input presents several drawbacks; lack of the controllability compared to voltage input, making it challenging to update input values in real time and leading to lower reproducibility, which could potentially exacerbate the stochastic behavior previously identified as an issue in memristors.<br/><br/>In this work, the external humidity, which works as the ionic input of the memristor device, was substituted with voltage modulation applied to the Pd proton-conducting electrode placed adjacent to the memristor device. This approach enables set voltage reduction, analogous to those seen in high RH conditions. Additionally, voltage modulation through precisely controlled ionic input successfully reduced the coefficient of variation for set voltage distribution from over 0.5 in the humidity operation mode to below 0.3. Ultimately, proton insertion through the Pd electrode promotes a carrier-rich environment in the resistive switching layer, leading to more gradual switching behavior compared to the intrinsic Y7C memristor, resulting in a 24% enhancement in image recognition accuracy.