Source-linked AI summary
Proton Conducting Graphene Oxide Coupled Neuron Transistors for Brain-Inspired Cognitive Systems
Changjin Wan, Liqiang Zhu, Yanghui Liu, Ping Feng, Zhaoping Liu, Hailiang Cao, Peng Xiao, Yi Shi, Qing Wan
TL;DR
The paper examines neural arithmetic in an artificial-neuron device using graphene oxide and electric-double-layer coupling. Experiments show rate-coded output decreases with increasing spike interval and indicate multiplicative input-output transformation, under Poisson-distributed presynaptic spikes.
Problem
The study addresses hardware implementation of neural arithmetic using an ionic/electronic hybrid artificial neuron.
Method
The approach uses graphene oxide films with ion migration and electrostatic coupling in an oxide-based electric-double-layer transistor.
Results
Neural output decreased from ~87 nA at d=10 ms to ~44 nA at d=200 ms when m=0.1 V, indicating multiplicative operation in the rate-coding scheme.
Takeaways & Limitations
The results demonstrate neural gain control through multiplicative arithmetic in an artificial-neuron device.
Takeaways & Limitations
The presynaptic spike train is modeled as Poisson distributed.
Abstract
from arXiv · showhide
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.
Experiment
GO films were prepared, characterized, and integrated into IZO neuron transistors with patterned Au electrodes. Electrical measurements were performed at room temperature and 50% relative humidity.
- GO preparation: Graphite oxide was oxidized, washed to pH ~5.0, and ultrasonically delaminated into a homogeneous GO suspension.The suspension was then used for film preparation.
- Neuron transistor fabrication: GO suspensions (~6 mg/mL) were spin-coated onto ITO glass and dried at 50 °C for 2 h.
- Neuron transistor fabrication: Patterned 100-nm Au source, drain, and gate electrodes were deposited through a nickel shadow mask by thermal methods.The gate electrode dimensions were 240 µm × 200 µm.
- Electrical characterization: GO capacitances and neuron-transistor electrical characteristics were measured using impedance analysis and a Keithley 4200 SCS system.Measurements were conducted at room temperature with 50% relative humidity.
S1. The experimental details of orientation tuning emulation.
Orientation tuning was emulated by moving a black-white grating panel across a photodetector and converting detected edges into presynaptic voltage pulses. The resulting spike count depended on orientation angle.
- Stimulus protocol: The panel orientation θ was varied between −90° and +90°, with the panel moved from side to side for each orientation.The grating pattern width and length were w and l, with w:l = 1:10.
- Stimulus protocol: A moving black-white grating generated a voltage pulse whenever a pattern edge crossed the photodetector coordinate origin.The pulse applied to the GO-gated IZO neuron was 0.5 V for 10 ms.
- Orientation-dependent spike count: For |θ|≤45°, ten spikes were triggered as the square panel moved from side to side.
- Orientation-dependent spike count: For |θ|>45°, spike number depended on 2·k/w, with k=|5w/tanθ|, and the study used frequencies from 10 to 50 Hz.The corresponding spike numbers were 2, 3, 4, 6, 9, and 10.
S2. Supplementary animations of orientation tuning experiments.
Supplementary animations illustrated grating motion and the associated generated spikes across six orientations. The tested angles ranged from 0° to 78.5° in absolute value.
- Animation set: Animations showed the square panel moving along the y axis at orientations of 0°, 48.2°, 60°, 70.5°, 75.5°, and 78.5°.
- Animation set: The animations also displayed spikes generated whenever the grating-pattern edge was detected.
S3. The protocol of Poisson-distributed presynaptic spike train.
Two presynaptic spike trains were applied to G1 and G2, each containing ten spikes with Poisson-distributed interspike intervals. The protocol varied λ from 20 to 300 ms and estimated the rate as 2000/λ spikes/s.
- Spike-train construction: Two presynaptic spike trains were applied to G1 and G2, and each train contained ten presynaptic spikes.Each spike was represented by a 0.5 V, 10 ms pulse.
- Spike-train construction: The interval Δt_ij between spikes was Poisson distributed, with λ and k denoting the expectation or variance and the number of intervals.
- Timing protocol: λ values of 20, 50, 100, 150, 200, and 300 ms were used, with k=10 for each spike train.
- Rate and implementation: The expected rate of the two spike trains was estimated as 2000/λ spikes/s, and random intervals were generated using MATLAB.Intervals were rounded to multiples of ten because the Keithley 4200 time resolution was approximately 10 ms.
S4. Experimental data for neural arithmetic based on rate coding scheme.
The artificial neuron’s neural input–output relationship is consistent with multiplicative arithmetic under temporal-correlated coding. Experimental curves show that output depends jointly on the interval between driving inputs and the modulatory voltage.
- Temporal-correlated coding: Neural output decreased from ~87 nA at d=10 ms to ~44 nA at d=200 ms when m=0.1 V.The output decreased gradually as the interval between the two driving inputs increased.
- Modulatory control: At ΔT=50 ms, neural I–O slopes were –79, –121 and –167 pA/ms for Vm=–0.1, 0 and 0.1 V, respectively.The slopes increased with modulatory voltage, showing modulation of the neural I–O relationship.
- Multiplicative arithmetic: Neural input–output relationships were identified as implementing a multiplicative operation.The relationship was expressed as a function of the input interval multiplied by a function of modulatory voltage.
- Empirical fitting: The I–O curves were fitted with an empirical function whose estimated parameters were A=216 nA, B=2.34 ms, C=0.47, D=32 nA and E=3.5 V^-1.The fitted form represented the dependence on both ΔT and Vm.