The research team led by Professor Yoo Ho-cheon of the School of Electronic Engineering at Hanyang University announced the development of ‘Screen-Gate Driven Analog Activation Transistors.’ These transistors can directly implement and regulate two types of non-linear activation functions—Sigmoid and Gaussian—within a single transistor. The study is recognized for coming up with a core technology for next-generation analog AI semiconductors, as it marks the world's first structure to implement activation function operations directly at the device level without digital circuits or complex CMOS blocks.

Existing 토토사이트 털림 semiconductors require complex CMOS circuits consisting of dozens to hundreds of transistors or digital arithmetic blocks to calculate activation functions, leading to high power consumption and large circuit areas. This has been a significant constr토토사이트 털림nt for edge 토토사이트 털림 semiconductors, where high integration and ultra-low power operation are essential. To solve the problem, the research team introduced a new two-gate structure with a screen gate inserted inside the transistor. This allows for free adjustment of the curve's slope, threshold point, width, and amplitude within a single device, enabling the direct generation of analog activation functions with a much simpler structure than previous methods.

Professor Yoo’s team first applied the SA-transistor, which implements characteristics of Sigmoid functions, to MRI image classification, boosting model accuracy from 77% to 84%. They also applied the GA-transistor, which implements Gaussian functions, to environmental time-series data, confirming an improvement in the coefficient of determination(R²) from 0.82 to 0.93. Furthermore, a hardware-based MLP(Multi-Layer Perceptron) system combining both devices recorded an accuracy of 96.7% in experiments using the IRIS dataset, proving the practicality of analog neural networks that perform activation function operations directly at the device level.

Professor Yoo Ho-cheon stated, "This research is an example showing that activation function operations, which are essential in conventional AI semiconductors, can be implemented directly in a single device. It holds great potential for various applications such as low-power, high-integration AI computing, neuromorphic systems, and edge devices."

This study was supported by the Information & Communication Technology Innovation Talent Nurturing(Nurturing Advanced Talent for 토토사이트 털림 Semiconductors) project of the Ministry of Science and ICT and the Institute of Information & Communications Technology Planning & Evaluation(IITP). The research results were published on November 24 in the international academic journal Advanced Materials (IF: 26.8, JCR Top 2.2%), a leading journal in the field of materials and electronic devices.

In the paper ‘Transistor-Level 토토사이트 털림 Functions via Two-Gate Designs: From Analog Sigmoid and Gaussian Control to Real-Time Hardware Demonstrations,’ Jo Joon-hyung (Integrated Master’s-Doctorate student in the Department of AI Semiconductor Engineering) served as the first author. Lee Won-woo, Han Young-min, and Yoo Young-woo participated as co-first authors. Professor Yoo Ho-cheon(Hanyang University), Professor Shin Won-joon(Sungkyunkwan University), and Professor Kim Young-joon(Gachon University) participated as corresponding authors.