Azobenzene redox layer stabilizes organic synapses for neuromorphic computing

An organic electrochemical transistor containing a reversible azobenzene redox layer retained an enhanced conductance state for more than 800 seconds after electrical stimulation, according to a peer-reviewed study led by Joon Hak Oh at Seoul National University. The device, called an azobenzene-introduced lithium-ion gel electrolyte transistor, or AB-LGET, also exhibited enhanced paired-pulse facilitation and substantially improved long-term potentiation compared with a conventional lithium-ion gel electrolyte transistor. When the experimentally measured responses of the AB-LGET were incorporated into a simulated neural network for handwritten-digit recognition, the system reached a maximum accuracy of 93.02%.

The result addresses a persistent challenge in organic electrochemical transistors, or OECTs, being investigated as artificial synapses for neuromorphic computing. These devices couple electronic transport in an organic semiconductor with ionic motion in an electrolyte, allowing channel conductivity to be modulated at relatively low voltages. They can display behaviors analogous to biological synapses, including the strengthening and weakening of conductance through electrical pulses. Such coupled ionic and electronic operation is attractive for hardware intended to process information with greater efficiency and parallelism, but the stability of the programmed conductance state remains a central problem.

In the lithium-ion gel electrolyte transistor examined in the study, applying a negative presynaptic voltage drove TFSI− anions from the electrolyte into the poly(3-hexylthiophene), or P3HT, semiconductor channel. This electrochemical doping increased the conductivity of the P3HT and represented synaptic potentiation. At the same time, Li+ ions migrated toward the gate side of the device. The redistribution of these ions created a charge imbalance within the electrolyte. Once the external voltage was removed, the system tended to recover electrical neutrality by driving the TFSI− anions back out of the P3HT channel, producing spontaneous de-doping and a rapid loss of the potentiated state.

The Seoul National University team altered this relaxation process by incorporating a redox-active azobenzene layer into the transistor. Lithium ions arriving at the gate side can participate in a reversible lithiation reaction with the azobenzene. This Faradaic redox reaction supplies localized charge compensation and helps preserve electrical neutrality within the electrolyte. With less charge imbalance, the thermodynamic driving force for TFSI− back-diffusion is reduced. The electrochemically doped P3HT can therefore remain conductive for substantially longer after the stimulating voltage has been removed, extending the lifetime of the stored synaptic state.

The modified charge dynamics translated into several measurable changes in device behavior. Compared with the conventional lithium-ion gel electrolyte transistor, the AB-LGET showed enhanced paired-pulse facilitation and substantially improved long-term potentiation. Following repeated stimulation, the enhanced conductance state remained evident for more than 800 seconds after the electrical stimulus was removed. The AB-LGET also supported stable multistate conductance modulation and highly linear potentiation and depression characteristics. In an artificial neural network, predictable changes in synaptic weight are important because they can reduce the error associated with programmed weight updates.

The researchers examined the physical origin of the improved memory retention using grazing-incidence wide-angle X-ray scattering, or GIWAXS. This analysis was used to look for changes in the structure of the P3HT channel before and after electrical stimulation. The results indicated that structural signatures associated with molecular doping were more effectively retained within the crystalline regions of P3HT in the AB-LGET. The observations support the proposed mechanism in which azobenzene-mediated charge compensation helps stabilize the electrochemically doped state of the organic semiconductor, linking the device’s longer memory retention with the persistence of doping-related molecular structure.

To assess how the measured synaptic behavior might contribute to neuromorphic information processing, the team inserted experimentally obtained potentiation and depression characteristics into an artificial neural network simulation. The task involved recognition of handwritten digits from the Modified National Institute of Standards and Technology MNIST dataset. Under the simulation conditions, the AB-LGET-based system achieved a maximum recognition accuracy of 93.02%. This was not an accuracy measurement from a physical processor built with an array of AB-LGET devices; rather, it was a computational evaluation informed by responses measured from the experimental transistor.

The central innovation is that the researchers actively regulated ionic relaxation through a reversible reaction at an interface rather than simply attempting to restrict the movement of ions. The azobenzene layer acts as a chemical participant in the memory process. As Li+ is reversibly incorporated into that layer during stimulation, the reaction provides compensating charge and weakens the tendency for TFSI− to leave the doped P3HT channel. This strategy preserves the ion-electron coupling that gives OECTs their low-voltage operation and synaptic-like behavior while directly addressing the charge imbalance associated with spontaneous de-doping.

The reported work is a device-level demonstration based on a lithium-ion gel electrolyte and a P3HT semiconductor channel, together with a network simulation using the experimentally measured potentiation and depression behavior. The retained conductance beyond 800 seconds and the maximum simulated handwritten-digit accuracy of 93.02% provide evidence that redox-mediated electrical-neutrality control can improve both memory stability and synaptic modulation in this architecture. The study also broadens the design options for organic artificial synapses by showing that a molecularly programmed redox reaction can regulate an ionic process that otherwise erases the programmed state.

Overall, the findings establish reversible redox chemistry as a promising design principle for organic neuromorphic electronics. By combining charge compensation from azobenzene with the ionic and electronic transport of an OECT, the researchers created a device that maintained an enhanced conductance state for more than 800 seconds and supported stable, multistate synaptic changes. The result connects materials chemistry, semiconductor physics and machine-learning evaluation in a single device concept, demonstrating how control of charge balance can transform the retention behavior of an organic artificial synapse.

Subject of Research: Azobenzene-mediated charge compensation for stable organic electrochemical synapses

Article Title: Azobenzene-introduced organic electrochemical transistors with enhanced synaptic properties for neuromorphic computing

Article References: Azobenzene-introduced organic electrochemical transistors with enhanced synaptic properties for neuromorphic computing. Original publication (Bibliographic details unavailable.)

Image Credits: AI Generated

DOI: Not provided

Keywords: organic electrochemical transistors, azobenzene redox, neuromorphic computing, artificial synapses, long-term potentiation, ionic charge compensation, P3HT semiconductor, memory retention, MNIST recognition, Azobenzene-introduced, organic, electrochemical