Source-linked AI summary
Artificial Synapse Network on Inorganic Proton Conductor for Neuromorphic Systems Applications
Li Qiang Zhu, Chang Jin Wan, Li Qiang Guo, Yi Shi, Qing Wan
TL;DR
Hardware implementation of neuromorphic computation requires artificial synapses that reproduce synaptic functions. This paper develops an in-plane oxide-based protonic/electronic synapse network and mimics paired-pulse facilitation, dynamic filtering, and spatiotemporal signal processing.
Problem
Developing artificial synapses remains a central challenge for neuromorphic computation.
Method
The authors fabricate a lateral in-plane-gate IZO synaptic transistor network with multiple gates on phosphorus-doped nanogranular SiO2.
Results
Proton migration produces strong lateral modulation, enabling paired-pulse facilitation, dynamic filtering, and spatiotemporal signal processing.
Takeaways & Limitations
Protonic/electronic coupling provides synaptic interconnections without intentional hard-wire connections in an artificial neural network.
Takeaways & Limitations
Scaling can compromise voltage level and noise margin, while the number of in-plane gates is not limited to two.
Abstract
from arXiv · showhide
The basic units in our brain are neurons and each neuron has more than 1000 synapse connections. Synapse is the basic structure for information transfer in an ever-changing manner, and short-term plasticity allows synapses to perform critical computational functions in neural circuits. Therefore the major challenge for the hardware implementation of neuromorphic computation is to develop artificial synapse. Here, in-plane oxide-based artificial synapse network coupled by proton neurotransmitters are self-assembled on glass substrates at room-temperature. A strong lateral modulation is observed due to the proton migration in the nanogranular phosphorus-doped SiO2 electrolyte films. Functional roles of short-term plasticity, including paired-pulse facilitation, dynamic filtering and spatiotemporally correlated signal processing are mimicked. Such in-plane oxide-based protonic/electronic hybrid synapse network proposed here are interesting for building future neuromorphic systems.
Results
The P-doped nanogranular SiO2 electrolyte forms nanoporous proton-migration channels and enables lateral IZO transistor modulation with high capacitance, proton conductivity, and ~10^6 on/off ratio. The resulting synaptic transistor reproduces temporal plasticity, dynamic filtering, and spatiotemporal EPSC summation with low and gate-tunable energy dissipation.
- Electrolyte and transistor properties: Nanopores measuring ~2.0 nm to 5.0 nm form effective channels for proton migration in the P-doped nanogranular SiO2 electrolyte.The SiO2 nanograins are tens of nanometers in size, producing high-density nanopores.
- Electrolyte and transistor properties: The electrolyte shows a maximum specific capacitance of ~3.0 μF/cm2 at 1.0 Hz and proton conductivity of ~10^-4 S/cm at room temperature.The capacitance arises from electric-double-layer formation, while proton accumulation produces strong lateral electrostatic coupling.
Discussion
The study demonstrates a lateral in-plane-gate IZO synaptic transistor network on P-doped SiO2, where proton migration produces strong electrostatic coupling and enables multiple presynaptic inputs without intentional hard-wire connections. The network mimics paired-pulse facilitation, dynamic filtering, and spatiotemporal signal processing for neuromorphic systems.
- Multiple-gate network: Multiple in-plane IZO gates and a self-assembled IZO channel form a simple artificial neural network without intentional hard-wire connections.The gates serve as presynaptic inputs, while the channel with source/drain electrodes serves as the postsynaptic output.
- Multiple-gate network: Proton migration triggered by gate pulses establishes interconnections between in-plane gates and the self-assembled IZO channel.The channel’s synaptic weight depends spatiotemporally on spikes applied to multiple presynaptic inputs.
- Device and coupling: A lateral in-plane-gate IZO synaptic transistor network was fabricated on a P-doped SiO2 electrolyte film, with strong lateral electrostatic coupling from proton migration.The electrolyte is described as nanogranular phosphorus-doped SiO2.
- Neuromorphic functions: The oxide-based synapse network successfully mimicked paired-pulse facilitation, dynamic filtering, and spatiotemporal signal processing.These functions motivate the network’s proposed use in future neuromorphic systems.
Methods
The study fabricated lateral coupled IZO-based synaptic transistors on glass at room temperature using P-doped nanogranular SiO2 films and patterned IZO electrodes. Self-assembled IZO channels were formed by controlled sputtering geometry, while FE-SEM and TEM characterized the electrolyte microstructure.
- Electrolyte-film deposition: P-doped nanogranular SiO2 films were deposited on glass substrates by PECVD using SiH4/PH3, O2, and Ar gases.The SiH4/PH3 mixture contained 95% SiH4 and 5% PH3.
- Electrolyte-film deposition: The PECVD process used 100 W RF power, ~30 Pa working pressure, and a film thickness of ~820 nm.Gas flow rates were 10 sccm for SiH4/PH3, 60 sccm for O2, and 60 sccm for Ar.
- IZO-device fabrication: IZO patterns were deposited by RF magnetron sputtering in pure Ar through a nickel shadow mask.The sputtering used 100 W RF power, 14 sccm Ar flow, and 0.5 Pa chamber pressure.
- IZO-device fabrication: When the pattern distance was 80μm, IZO nanoparticles formed a thin self-assembled channel between source and drain electrodes.The channel resulted from nanoparticle reflection at the mask edge and extension at low incident angle.
- Microstructure characterization: FE-SEM and TEM were used to characterize nanogranular P-doped SiO2 films deposited on polished Si (100) wafers and Cu TEM grids.These films were prepared separately for microstructure characterization.
Electrical characterizations of the P-doped SiO2 film and lateral-coupled IZO-based
The study characterized proton conductivity and frequency-dependent capacitance in P-doped SiO2 electrolyte films, and measured transistor and synaptic performance under approximately 50% relative humidity. Presynaptic spikes were applied to the in-plane gate, while postsynaptic output was read through a small drain voltage.
- Film characterization: P-doped SiO2 electrolyte films were characterized for proton conductivity and frequency-dependent capacitance using a Solartron 1260A Impedance Analyzer.Measurements were conducted in air at approximately 50% relative humidity.
- Synaptic measurement configuration: Presynaptic spikes were applied to the in-plane gate electrode, and postsynaptic output was measured with a small reading drain voltage on the IZO source/drain.This measurement configuration couples gate-applied presynaptic stimulation to drain-read postsynaptic output.
Figure Legends
The figure legends compare conventional and in-plane-gate oxide thin-film transistor configurations and depict a self-assembled IZO-based protonic/electronic hybrid synaptic transistor. In this architecture, in-plane gates act as pre-synapses and self-assembled IZO channels as post-synapses.
- Transistor architectures: The figure compares conventional bottom-gate thin-film transistors with in-plane-gate oxide thin-film transistors incorporating a bottom conductive layer.
- Device fabrication: A self-assembled IZO-based transistor is fabricated using a one-mask method.
- Synaptic transistor: The IZO-based protonic/electronic hybrid synaptic transistor assigns in-plane gates as pre-synapses and IZO patterns or the self-assembled IZO channel as the post-synapse.
nanogranular SiO2 electrolyte film and electrical performances for the lateral-gated
The section characterizes the P-doped nanogranular SiO2 electrolyte film and evaluates the lateral-coupled IZO transistor through capacitance, transfer, and output measurements.
- Film characterization: Cross-sectional SEM and TEM images characterize the P-doped nanogranular SiO2 films.The TEM image shows the films on a TEM Cu grid.
- Electrical characterization: Frequency-dependent specific capacitance is measured for the P-doped nanogranular SiO2 electrolyte film.The impedance spectroscopy characterization uses an in-plane structure.
- Transistor performance: The lateral-coupled transistor is evaluated using transfer curves measured at a constant Vds of 1 V.The transfer measurement plots Ids versus Vgs and includes a schematic of the lateral coupled transistor.
- Transistor performance: Output characteristics are measured for the lateral coupled transistor.The passage identifies the output-characteristics measurement but does not report numerical values.
and the excitatory post-synaptic current (EPSC) triggered by pre-synaptic spikes. (a)
The figure presents an IZO-based protonic/electronic hybrid synaptic transistor with mobile protons in nanogranular SiO2 and characterizes EPSC responses to presynaptic spikes. It further demonstrates paired-pulse facilitation, frequency-dependent EPSC gain, spatiotemporal signal integration, and a multiple-presynapse network scheme.
- EPSC response: The IZO-based protonic/electronic hybrid synaptic transistor uses nanogranular SiO2 with mobile protons and produces EPSC from a 0.3 V, 10 ms presynaptic spike.EPSC is measured at a constant Vds of 0.5 V.
- Paired-pulse facilitation: Paired presynaptic spikes separated by 200 ms generate first and second EPSCs whose amplitudes are denoted A1 and A2, with PPF defined as A2/A1.PPF index is plotted against the inter-spike interval Δtpre.
- Frequency-dependent response: Ten-spike stimulus trains at 1.0 Hz, 5.0 Hz, 10 Hz, 20 Hz, 25 Hz, 33Hz, and 50 Hz are used to record EPSCs and evaluate amplitude gain A10/A1.Each stimulus spike is 0.5V and 10ms.
- Spatiotemporal dynamic logic: An in-plane device uses two IZO gates as pre-synapses and measures post-synaptic current at a constant Vds of 0.5 V to simulate spatiotemporal dynamic logic.The two presynaptic spikes are 0.5 V, 20 ms and 1.0 V, 20 ms, respectively.
- Artificial neural network: The proposed artificial neural network combines multiple in-plane IZO gates as pre-synapses with a self-assembled IZO channel as the post-synapse.The scheme is described as an artificial neural network with multiple pre-synapses.
Figures
The figures compare the artificial synapse with prior thin-film transistors and biological synapses, while presenting its microstructure, lateral capacitance, plasticity behaviors, correlated spikes, and neural-network implementation.
- Figure 1 compares the previous thin-films transistor with the synaptic device.
- Figure 2 presents the microstructure and lateral specific capacitance of the P-doped material.
- Figure 3 compares the biological synapse with the IZO artificial synapse.
- Figure 4 presents paired-pulse facilitation and dynamic-filter behaviors of the lateral-gated synapse.
- Figures 5 and 6 address spatiotemporally correlated pre-synaptic spikes and an artificial neural network based on lateral-coupled IZO synaptic devices.