I. Core Research Abstract

• High-precision, uniform formation of carbon nanotube active channels via aerosol deposition.

Ultrasonic atomization is used to purify sc‑SWCNT ink, enabling aerosol direct writing of 110‑μm-wide channels and 1000‑μm-wide conductive networks. The carbon nanotube density reaches 40 tubes/μm², with a device on/off ratio of 5.7×10⁶, a mobility of 8.1 cm²/V·s, and a subthreshold swing as low as 93 mV/dec. The array devices exhibit excellent uniformity.

• 850 nm highly selective near-infrared photoresponse

The IHIC material exhibits an absorption peak at 830 nm, with energy levels that align perfectly with those of carbon nanotubes. Under 850‑nm pulsed illumination, the photocurrent reaches 76 nA—six times higher than that achieved under equivalent UV light intensity—enabling highly specific near-infrared signal detection while minimizing visible-light interference.

• Ultra-low-power bionic synaptic full-function emulation

A single 850‑nm optical pulse consumes as little as 56 aJ, far below the 1–10 fJ power consumption of biological synapses; it can replicate the full suite of bio‑neurobehavioral phenomena, including EPSCs, short‑term plasticity characterized by PPF, long‑term memory switching mediated by STP, spike‑timing‑dependent plasticity (STDP), and Pavlovian associative learning.

• Flexible, long-lasting, and stable air device

PI‑based devices with a curvature radius of 8/10/18 mm exhibit no significant performance degradation upon bending; after three months of storage in ambient air, the photocurrent decreases only slightly, demonstrating long-term operational stability.

• Edge-side Dynamic Trajectory Perception at Night

A 4×4 synaptic array simulates a drone‑based night‑vision scenario, leveraging 850 nm near‑infrared light signals to capture vehicle motion trajectories in real time. It can distinguish the depth of memory encoding based on stimulus duration and pulse count, thereby enabling image perception and dynamic tracking simulations.

II. Process Flow and Core Advantages of the Complete Aerosol Spray Device Manufacturing System

(1) Complete Optoelectronic Synaptic Printing Process

• PI flexible substrate with an aluminum gate deposited via thermal evaporation, and a 20-nm AlO₃ insulating dielectric layer formed by atomic layer deposition;
• Prepare purified sc-SWCNT toluene ink, perform aerosol ultrasonic atomization, and directly write transistor channels at a constant speed of 1.5 mm/s.
• Toluene washing is performed three times to remove excess impurities, followed by air annealing at 120°C to eliminate the solvent.
• Silver source–drain electrodes are fabricated by inkjet printing and then thermally cured at 120°C.
• Spin-coat the IHIC photosensitive ink, then perform low-temperature, short-duration annealing to complete the fabrication of the optoelectronic synaptic device.

(II) Five Key Advantages of Aerosol-Based Processes (Comparison with Lithography and Inkjet)

• High-Precision, Uniform Deposition of Carbon Nanotubes on Semiconductors

The aerosol plume can be focused and controlled, the carbon‑nanotube network exhibits uniform density, and large‑area array devices show extremely low electrical variability; by contrast, conventional inkjet printing often leads to carbon‑nanotube aggregation and nonuniform channel conductivity.

• Low-temperature flexibility ensures compatibility throughout, with no high-temperature damage.

The printing and annealing process can reach a maximum temperature of 120°C, making it compatible with ultra-thin flexible substrates such as PI. It does not compromise the toughness of the polymer substrate, enabling the fabrication of wearable neural electronics.

• Maskless Digital Fabrication of Micrometer‑Scale Precision Patterns

No photolithography mask is required; CAD enables free adjustment of channel dimensions and array layout, and a 4×4 synaptic array can be prototyped in just a few hours, facilitating rapid iteration of photosensitive material and carbon‑nanotube ratio formulations.

• Wide ink compatibility; a dedicated forming system for carbon-based materials.

By optimizing atomization parameters for high-purity semiconductor carbon nanotubes, the system is compatible with a wide range of two‑dimensional and one‑dimensional nano‑sensing inks, enabling a single platform to support multiple types of neural devices.

• Localized, site-specific deposition leads to substantial savings in nanomaterials.

Only the active region of the transistor is printed with carbon nanotubes, eliminating material waste on the blank substrate and significantly reducing the cost of high‑value, purified carbon‑nanotube consumables.

III. Five Key Technological Innovations

Innovation 1: A novel aerosol direct-writing sc-SWCNT flexible optoelectronic synaptic scheme

Innovation 2: IHIC / Near-Infrared Enhancement Mechanism via Energy-Level Matching of Carbon Nanotubes

Innovation 3: Ultra-Low-Power, Light-Controlled Synaptic Architecture of the Ya-Ajiao Class

Innovation 4: Fully Bionic Neural Plasticity Hardware Implementation

Innovation 5: Deployment of a Flexible, Night-Vision-Enabled Synaptic Array

IV. Existing Process Challenges and Corresponding Solutions for Overcoming Them

Challenge 1: Carbon nanotube inks tend to aggregate, resulting in poor transistor switching performance after printing.

Challenge 2: Most optoelectronic synapses are sensitive to ultraviolet and visible light, with only a weak response to 850 nm infrared light.

Challenge 3: Traditional lithography processes are complex and costly, and carbon nanotube devices have long iteration cycles.

Challenge 4: Synaptic devices typically consume power at the femtojoule level, limiting the battery life of lightweight devices.

Challenge 5: Bending of thin-film devices and rapid degradation of their optoelectronic performance during long-term storage.

V. Future Diversified Industrial Application Scenarios

1 Lightweight Unmanned Aerial Vehicle Night-Vision Perception System

2 Wearable Bionic Visual Sensor for Medical Applications

3. Electronic skin for robots in dark environments

4. Low-Power Security Night-Vision Chip for the Edge Side

5 Brain-Inspired Computing Laboratory Rapid Prototyping Platform

Figure 1. IHIC/SWCNT thin-film transistor devices and characterization of their material properties. a) Schematic of an aluminum‑gate SWCNT thin-film transistor device; b) Photograph of a 4×4 array of aluminum‑gate SWCNT thin-film transistors on a graphene substrate; c) Optical microscopy image of the aluminum‑gate SWCNT thin-film transistor; d) Scanning electron microscopy image of the SWCNT film deposited in the channel region; e) Chemical structure of IHIC; f) UV–vis–NIR absorption spectrum of IHIC in toluene and the spectrum of an IHIC film on a quartz substrate; g) Typical transfer characteristics of an SWCNT thin-film transistor before and after spin‑coating with IHIC; h) XPS spectra of the SWCNT film in the channel region before and after spin‑coating with IHIC, and i) the S 2p core-level XPS spectrum.

Figure 2. a) Schematic illustration of neural signal transmission in biological synapses and the SWCNT optoelectronic synaptic transistor; b) Transfer characteristics of an Al‑gate SWCNT optoelectronic synaptic transistor under dark conditions and under illumination at wavelengths ranging from 365 to 1050 nm; c) EPSC responses evoked by pulsed light at wavelengths of 660, 850, 940, and 1050 nm (3 mW·cm⁻², 1 s duration); d) Peak amplitudes of EPSCs induced by pulsed light at wavelengths from 365 to 1050 nm (3 mW·cm⁻², 1 s duration); e) Energy band diagrams of IHIC and single‑walled carbon nanotubes.

Figure 3. Photo‑synaptic characteristics of the IHIC/SWCNT thin‑film transistor under 850 nm light pulses (3 mW·cm⁻²): a) EPSC amplitude as a function of pulse number (1 to 50 pulses); b) EPSC amplitude versus pulse duration; c) EPSC amplitude versus pulse frequency; d) PPF behavior induced by two consecutive light pulses (ΔT = 1 s, VDS = −0.1 V, 3 mW·cm⁻²); e) PPF index as a function of the inter‑pulse interval; f) Associative learning–forgetting–relearning process under 850 nm light pulses; g) EPSC dependence on pulse number; h) EPSC dependence on pulse duration; i) EPSC dependence on pulse frequency for four different wavelengths.

Figure 4. Simulation of the role of visual neural computation in image learning and memory. a) The process by which the retina detects various light stimuli via photoreceptors; EPSCs are shown for: b) stimulation with a single NIR light pulse and c) stimulation with five NIR light pulses (wavelength 850 nm, intensity 3 mW·cm⁻²); d) the distribution of EPSCs across different numbers of stimulations and durations; e) a simulated rabbit image under varying numbers of NIR pulses, pulse durations, and a 20‑second image delay. Standard color‑chart values have been normalized to the maximum EPSC value.

Figure 5. Schematic diagram of behavioral simulation for trajectory tracking: a) Schematic illustration of a drone tracking a vehicle’s motion trajectory at night; b) Ten distinct conductance states of the same device; c) Real-time current‑state map of a synaptic transistor array displaying optical traces at a specific moment; d) Demonstration of tracking and recognition of a vehicle’s NIR‑light trajectory. The ten colors on the standard color chart correspond to normalized synaptic current values under ten different states.