鞐瓣惮/靷绊暀
狠狠撸 Research 1000
| Lee Eun-kwang | Improving Image Generation Performance Using Noise | |||
| 鞛戩劚鞛� | 雽赌鞕疙檷氤挫劶韯� | 鞛戩劚鞚� | 2025-08-06 |
| 臁绊殞靾� | 72 | ||
| Lee Eun-kwang | Improving Image Generation Performance Using Noise | |||||
 |
雽赌鞕疙檷氤挫劶韯� |  |
2025-08-06 |  |
72 |
狠狠撸 and Hanyang University 鈥楧raw Attention鈥� with Counterintuitive Research: Improving Image Generation Performance Using Noise
- Development of Heterojunction-Based Probabilistic Control Transistor Device Published in <Advanced Materials>
A research team led by Professor Lee Eun-kwang from the Department of Chemical Engineering at 狠狠撸 and Professor Yoo Ho-chun from the Department of Convergence Electronic Engineering at Hanyang University has developed a heterojunction-based probabilistic control transistor device. Their work introduces a novel principle in which amplifying noise actually improves image generation accuracy.
The joint research team implemented a polymer-based heterojunction structure to realize a spontaneously formed negative transconductance (NTC) characteristic resulting from hole锛�electron injection imbalance. They also demonstrated for the first time that internal noise induces probabilistic doping and de-doping processes鈥�an essential mechanism for enhancing image generation accuracy.
This research is drawing attention as a counterintuitive approach emerging at a time when novel device studies that can efficiently implement stochasticity and nonlinearity at the hardware level are increasingly valued, especially with the rise of distributed computing paradigms like edge computing and the advancement of generative AI.
Much like synapses in the human brain, probabilistic transistors with inherent unpredictability are seen as promising next-generation devices capable of combining energy efficiency with functional diversity. However, traditional approaches to probabilistic devices typically focus on suppressing or avoiding noise, while devices that actively leverage noise in a controlled manner remain rare.
This study is particularly significant for driving a paradigm shift in which noise, traditionally considered an unwanted physical phenomenon, is repurposed as a valuable computational resource. If integrated into next-generation generative AI systems, edge devices, and neuromorphic sensors, this technology is expected to enable low-power, highly efficient computing architectures.
Professor Lee Eun-kwang commented, 鈥淭his research is a creative approach that simultaneously leverages the structural advantages of organic semiconductors and their nonlinear stochastic behavior, and it is expected to greatly accelerate the commercialization of noise-control-based hardware AI systems.鈥�
This research is being hailed as a groundbreaking breakthrough for next-generation nonlinear computing devices and generative AI hardware. The team鈥檚 findings were recently published in Advanced Materials (IF: 26.8, JCR Top 2.2%), a leading international journal in materials science and applied electronics, under the title 鈥楬eterojunction-driven stochasticity: BHJ noise-enhanced negative transconductance transistor in image generation.鈥�
The study was supported by the Individual Basic Science and Engineering Research Program funded by the Ministry of Science and ICT and the National Research Foundation of Korea, as well as the AI Semiconductor Core Talent Development Program (IITP) run by the Korea Institute for Advancement of Technology.