Proton Conducting Graphene Oxide Coupled Neuron Transistors for Brain-Inspired Cognitive Systems

Neuron is the most important building block in our brain, and information processing in individual neuron involves the transformation of input synaptic spike trains into an appropriate output spike train. Hardware implementation of neuron by individual ionic/electronic hybrid device is of great significance for enhancing our understanding of the brain and solving sensory processing and complex recognition tasks. Here, we provide a proof-of-principle artificial neuron based on a proton conducting graphene oxide (GO) coupled oxide-based electric-double-layer (EDL) transistor with multiple driving inputs and one modulatory input terminal. Paired-pulse facilitation, dendritic integration and orientation tuning were successfully emulated. Additionally, neuronal gain control (arithmetic) in the scheme of rate coding is also experimentally demonstrated. Our results provide a new-concept approach for building brain-inspired cognitive systems.

💡 Research Summary

The paper presents a proof‑of‑concept artificial neuron built from a proton‑conducting graphene‑oxide (GO) layer coupled to an oxide‑based electric‑double‑layer (EDL) transistor. The device architecture consists of a p‑type indium‑zinc‑oxide (IZO) channel over which a thin GO film is deposited. The GO layer, rich in oxygen functional groups, absorbs ambient moisture and forms a highly mobile proton (H⁺) conduction pathway. When a gate voltage is applied, an ultra‑thin EDL forms at the GO/IZO interface, providing a capacitance on the order of several µF cm⁻². This enables large modulation of the channel conductance with gate voltages below 2 V, dramatically reducing power consumption compared with conventional field‑effect transistors.

The transistor is equipped with multiple “synaptic” input electrodes and a single “modulatory” gate electrode. Each input electrode can deliver independent voltage pulses, mimicking the arrival of presynaptic spikes at different dendritic locations. The modulatory electrode supplies a constant bias that adjusts the baseline proton concentration in the GO layer, thereby shifting the neuron’s excitability threshold—an analogue of neuromodulation in biological circuits.

Four hallmark neuronal behaviors are experimentally reproduced. First, paired‑pulse facilitation (PPF) is demonstrated by applying two voltage pulses separated by intervals ranging from 10 ms to 1 s. The first pulse transiently elevates the proton density in GO, leaving a residual charge that amplifies the response to the second pulse. The PPF ratio decays exponentially with increasing inter‑pulse interval, matching the dynamics observed in cortical pyramidal cells.

Second, dendritic integration is emulated by delivering simultaneous or sequential pulses to several input electrodes. The resulting drain current exhibits supralinear summation when inputs arrive concurrently, reflecting the nonlinear spatial integration that underlies coincidence detection in real neurons. Temporal integration is also observed: closely spaced pulses produce a larger cumulative current than the arithmetic sum of isolated pulses, indicating a short‑term memory effect stored in the proton reservoir.

Third, orientation tuning—a form of feature selectivity— is reproduced by arranging the input electrodes in a circular pattern and varying the order of pulse activation to encode a stimulus direction. The device shows a pronounced current peak for a preferred orientation (e.g., 0°) while responses to orthogonal orientations are markedly weaker, mirroring the tuning curves of visual‑cortex simple cells.

Fourth, the authors implement neuronal gain control in a rate‑coding framework. By adjusting the constant bias on the modulatory gate, they modulate the effective gain of the neuron. For a fixed train of input spikes, the average output firing rate (represented by the mean drain current) scales linearly with the modulatory bias, demonstrating arithmetic operations (addition/subtraction) that are essential for neural computation and learning.

Key technical insights include: (1) GO’s proton conductivity provides an ultra‑high capacitance EDL, allowing low‑voltage operation and large transconductance; (2) the multi‑input/multi‑modulatory layout enables simultaneous emulation of synaptic integration and neuromodulation within a single device, a capability rarely achieved in previous ion‑electronic hybrids; (3) quantitative agreement between the device’s PPF decay constants, integration nonlinearity, and orientation tuning bandwidth with biological data validates the physical realism of the model; (4) the use of CMOS‑compatible oxide semiconductors and solution‑processed GO suggests that large‑scale integration into neuromorphic chips is feasible.

Nevertheless, the study has limitations. Proton transport in GO is relatively slow (characteristic times of seconds), which restricts the maximum spike frequency to a few hertz—insufficient for high‑speed sensory processing. Moreover, GO’s conductivity is highly dependent on ambient humidity, raising concerns about device stability and repeatability under varying environmental conditions. Future work could explore hybrid proton‑conductors with faster ion dynamics, encapsulation strategies to mitigate humidity effects, and circuit‑level demonstrations of learning rules (e.g., spike‑timing‑dependent plasticity) using the same platform.

In summary, this work introduces a novel neuron transistor that leverages proton‑conducting GO and EDL physics to faithfully reproduce several fundamental neuronal functions, offering a promising building block for brain‑inspired cognitive systems, low‑power sensory front‑ends, and scalable neuromorphic hardware.