Seong Jun Kang

​

Department of Materials Science and Engineering

Laboratory for Advanced Nano Technologies

Kyung Hee University

Highly efficient and accurate machine vision enabled by band-engineered oxide neuromorphic devices with spike-number-dependent plasticity

Research in Laboratory for Advanced Nano Technologies

Quantum-dots Light emitting diodes

Photodetectors

JCP 130, 094704 (2009)

Fabrication of high-performance Quantum-dots LEDs

• All solution processable and inorganic materials for QDs LEDs.
• Controllable layer-by-layer deposition of QDs for LEDs.

Interfacial electronic structures of LEDs

• Synthesis inorganic charge injection/transport materials.
• Interfacial electronic structures to engineer the charge transport.

Transparent photodetector (NIR/RGB/UV)

• Wide band gap semiconductor for the transparent active channel
• Controllable defect state via nanomaterials

In-sensor and in-memory optical neuromorphic sensor

• Fabrication of photodetectors for the next-generation image sensor.

Optical logic circuits

Outline

∙ Basic IDEA

- Engineering gap states for multiband electronic structure

​

∙ Expanding oxide semiconductors for transparent photosensors

- How can a wide bandgap semiconductor generate a photocurrent from photons with energy lower than its band gap?

- “Highly” and “Truly” transparent photosensors, invisible to the human eye

​

∙ Expanding to vision inspired optical synaptic transistors

- Appropriate electronic band structure for optical synaptic devices

- Pentachromatic-vision inspired optical synaptic transistors

​

∙ Advancing machine vision technology

- A brief insight into AI vision technology

- In-sensor & in-memory optical synaptic devices for enhancing vision systems and machine learning validation

Basic IDEA: Engineering gap states for optoelectronics

What happens to the gap states

when light is present?

Outline

∙ Basic IDEA

- Engineering gap states for multiband electronic structure

​

∙ Expanding oxide semiconductors for transparent photosensors

- How can a wide bandgap semiconductor generate a photocurrent from photons with energy lower than its band gap?

- “Highly” and “Truly” transparent photosensors, invisible to the human eye

​

∙ Expanding to vision inspired optical synaptic transistors

- Appropriate electronic band structure for optical synaptic devices

- Pentachromatic-vision inspired optical synaptic transistors

​

∙ Advancing machine vision technology

- A brief insight into AI vision technology

- In-sensor & in-memory optical synaptic devices for enhancing vision systems and machine learning validation

​

Interfacial electronic structure of highly transparent photosensor

Realize the multi-band structure in oxide semiconductors via inserting QDs

- Low-energy photon absorption is possible

IDEA

Experimental result

Applied Physics Letters, 106, 031112 (2015)

Wide band gap semiconductor + controllable band gap quantum dots

(for the transparent active channel) (for the light absorber)

Highly transparent photosensor

The photo responsivity shows a selective response to the 650 nm light source.

External quantum efficiency to the 650 nm light source was 1.25 × 10⁴.

Applied Physics Letters, 106, 031112 (2015)

“Truly transparent” optical materials capable of detecting low-energy photon

Improving the visible-light photoresponse of a ZnO phototransistor via solution processable Li-dopants

- Increased number of oxygen vacancies by Li dopants is the origin of the visible-light photocurrent generation

“Truly transparent” optical materials capable of detecting low-energy photon

Journal of Materials Chemistry C 10, 5590 (2022)

Improving the visible-light photoresponse of a ZnO phototransistor via solution processable Li-dopants

- Increased number of oxygen vacancies by Li dopants is the origin of the visible-light photocurrent generation

“Truly transparent” optical materials capable of detecting low-energy photon

Improving the visible-light photoresponse of a ZnO phototransistor via a thin Al₂O₃ layer

- Increased number of oxygen vacancies are the origin of the visible-light photocurrent generation

Journal of Materials Chemistry C 10, 5590 (2022)

Improving the low-energy photoresponse of a IGZO phototransistor via a thin Ag₂O layer

Truly transparent UV photosensor fabricated on glass substrate for healthcare applications

Demonstration of “Truly transparent” photosensor

Outline

∙ Basic IDEA

- Engineering gap states for multiband electronic structure

​

∙ Expanding oxide semiconductors for transparent photosensors

- How can a wide bandgap semiconductor generate a photocurrent from photons with energy lower than its band gap?

- “Highly” and “Truly” transparent photosensors, invisible to the human eye

​

∙ Expanding to vision inspired optical synaptic transistors

- Appropriate electronic band structure for optical synaptic devices

- Pentachromatic-vision inspired optical synaptic transistors

​

∙ Advancing machine vision technology

- A brief insight into AI vision technology

- In-sensor & in-memory optical synaptic devices for enhancing vision systems and machine learning validation

​

Scientific issues: Neuromorphic device for AI

Nature Computational Science 2, 10 (2022)

Better architecture for massively parallel processing of “electrical signal”

Scientific issues: What about “light signals”?

Traditional neuromorphic

devices for electrical signal

Processing ”light signal”

Ultra-Fast

Highly Accurate

Low-Power Device Operation

Massive Visual Information

Advanced Functional Materials (2024)

Appropriate electronic band structure for optical synaptic devices

Multi-band structure exhibits a persistent photocurrent effect, which demonstrates synaptic behavior in response to the optical signals.

Multi-band structure

Persistent photocurrent effect

Optical synaptic behaviors

Optical synaptic transistors: Solution process of Cd doped IGZO

Vertically diffused Cd doping process of IGZO

- Cd dopants were uniformly diffused into the IGZO film, and the film density was increased.

ACS Applied Electronic Materials 5, 6275 (2023)

Optoelectronic synaptic behavior of the device under the green and blue light stimulation

Excitatory postsynaptic current (EPSC) and paired-pulse facilitation (PPF)

ACS Applied Electronic Materials 5, 6275 (2023)

Optical synaptic transistors: Solution process of Cd doped IGZO

Initial, learning and forgetting process of the 3x3 array of Cd doped IGZO phototransistor under GB mode

ACS Applied Electronic Materials 5, 6275 (2023)

Optical synaptic transistors: Solution process of Cd doped IGZO

Pentachromatic vision inspired optical synaptic transistor

UV-Vis-NIR pentachromatic vision for vision inspired optical synaptic transistor

Control gap states at the mixed zone of Ag₂O and IGZO

Advanced Functional Materials 34, 2402222 (2024)

Pentachromatic vision inspired optical synaptic transistor

Excitatory postsynaptic current (EPSC) shows pentachromatic response

Paired-pulse facilitation (PPF) shows synaptic response

- Low working voltage & wide wavelength response (405 nm ~ 830 nm)

Advanced Functional Materials 34, 2402222 (2024)

Pentachromatic vision inspired optical synaptic transistor

Learning process

- Memorize 20 times input optical signals and measuring the intensity of light

Forgetting process

- Measuring retention signal during 50 seconds

Fast relearning process

- Recall the light signal via 3~4 times additional input signals

Advanced Functional Materials 34, 2402222 (2024)

Pentachromatic vision inspired optical synaptic transistor

The device is capable of processing more than 512 different intensities of optical signals

- 512 conduction states at the oxide semiconductor optical synaptic device

- via wide wavelength input light signal with a good linearity

​

Higher capacity of optical input signal processing!

​

More complex calculations are possible!

​

Advanced Functional Materials 34, 2402222 (2024)

Outline

∙ Basic IDEA

- Engineering gap states for multiband electronic structure

​

∙ Expanding oxide semiconductors for transparent photosensors

- How can a wide bandgap semiconductor generate a photocurrent from photons with energy lower than its band gap?

- “Highly” and “Truly” transparent photosensors, invisible to the human eye

​

∙ Expanding to vision inspired optical synaptic transistors

- Appropriate electronic band structure for optical synaptic devices

- Pentachromatic-vision inspired optical synaptic transistors

​

∙ Advancing machine vision technology

- A brief insight into AI vision technology

- In-sensor & in-memory optical synaptic devices for enhancing vision systems and machine learning validation

​

Scientific issues: Optical neuromorphic for massive visual data recognition

Optical synaptic device is a key technique for future visual information recognition systems

In reality, incidents still occur due to insufficient visual information processing.

A new technological paradigm is needed for next-generation visual information recognition system.

Optical synaptic image sensor array

Conventional imaging system vs. electrical synaptic image system vs. optical synaptic image system

· Conventional visual recognition system

· Electrical synaptic visual recognition system

· Optical synaptic visual recognition system

Post process

Post process

Small-scale neuromorphic processor

Small-scale electrical data process

Small input

data

ACS Nano 19, 13107 (2025)

Optical synaptic image sensor array

· Optical synaptic visual recognition system

Post process

Small-scale neuromorphic processor

Small-scale electrical data process

Small input

data

ACS Nano 19, 13107 (2025)

Conventional and electrical synaptic should process 80 images

Process 20 images

In-sensor & in-memory optical synaptic device

Image overlapping pre-process via in-sensor & in-memory device

- Need suppressed spike-timing-dependent plasticity (STDP) characteristics and increase spike-number-dependent plasticity (SNDP) to reduce processing image data

Comparing the signal process of the device with/without STDP characteristics

​

To enhance accuracy, it is important to maintain equal weight for the sequence image

STDP behavior

SNDP behavior

ACS Nano 19, 13107 (2025)

InGaSnO synaptic device to suppress STDP and increase SNDP

- Strong bonding energy between Sn and O & Slow recombination

- InGaZnO vs. InGaSnO

In-sensor & in-memory optical synaptic device

ACS Nano 19, 13107 (2025)

In-sensor & in-memory optical synaptic device

STDP behavior

SNDP behavior

ACS Nano 19, 13107 (2025)

Prove the concept via Machine Learning

- 28 x 28 pixels image data input to the machine learning (480,000 images)

- Reduce the training time according to increase the number of overlapped images without change the accuracy

In-sensor & in-memory optical synaptic device

ACS Nano 19, 13107 (2025)

Prove the concept via Machine Learning

- Image recognition via an ideal device (100% SNDP).

​

​

In-sensor & in-memory optical synaptic device

ACS Nano 19, 13107 (2025)

Prove the concept via Machine Learning

- Image recognition via a device with the STDP (IGZO TFTs).

​

​

In-sensor & in-memory optical synaptic device

ACS Nano 19, 13107 (2025)

Prove the concept via Machine Learning

- Image recognition via a device by the SNDP (InGaSnO TFTs).

​

​

In-sensor & in-memory optical synaptic device

ACS Nano 19, 13107 (2025)

Prove the concept via Machine Learning

- Compare the result of image recognition of ideal/InGaSnO/IGZO device, with the wrong information at the end

​

Device with the SNDP (InGaSnO) gives more accurate image recognition result

​

​

More accurate and fast image recognition system is possible with InGaSnO in-sensor devices!

​

Conclusion

ACS Nano 19, 13107 (2025)

Summary

∙ We have suggested an appropriate electronic band structure of oxide semiconductor for optical synaptic devices.

​

​

∙ We have demonstrated a pentachromatic vision inspired optical synaptic transistor with 512 conduction states to the light signal.

​

​

∙ More accurate and fast image recognition system is possible with SNDP in-sensor devices!

​

​

∙ We believe that optical synaptic devices could be a key technique for developing fast and accurate machine vision.

​

Additional information

Lab members

Funding

National Research Foundation of Korea, 중점연구소, Kyung Hee University

Contact information

​

E-mail:

junkang@khu.ac.kr

​

Homepage:

https://lant.khu.ac.kr

​