Hanyang University’s WeDD Lab

On-Device AI: Gaussian–Sigmoid Transistors, Light-Driven Spikes, and Intelligent Risk Sensors

Presenter: Hocheon Yoo

Abstract: Emerging edge AI systems call for device-level approaches that are inherently low-power, secure, and capable of managing uncertainty. In this talk, I will share our recent exploratory efforts toward realizing on-device intelligence using custom-designed semiconductor devices. I will introduce three emerging device platforms for on-device AI: (1) probabilistic Gaussian–Sigmoid transistors, (2) photo-spike photodetectors, and (3) intelligent risk detection sensors.

Gaussian–Sigmoid transistors support analog activation and probabilistic inference by leveraging intrinsic device variability, providing a potential pathway for implementing Bayesian operations directly in hardware. Photo-spike photodetectors convert light fluctuations into asynchronous spike trains, functioning as both neuromorphic input interfaces and entropy sources for secure applications such as physical random number generation. Intelligent risk detection sensors based on UVC sensing detect early-stage fire events and enable high-level inference of fire types and potential causes through sensing behavioral pattern analysis.

While still in early stages, the combination of these platforms suggests a promising direction for hardware-embedded probabilistic learning, real-time sensing, and secure edge intelligence. This work aims to show how tuning the physics of emerging devices may open new opportunities for resilient, intelligent systems at the edge.

Speaker bio: Hocheon Yoo is an associate professor of electronic engineering at Hanyang University, South Korea, where he leads the We Design Devices (WeDD) Lab. His research focuses on the development of novel semiconductor devices for neuromorphic computing, secure hardware, and sensor-integrated AI systems. He pursues a material-device–application co-design approach, exploring how emerging materials and structures can directly enable new forms of intelligent behavior. He is currently the Korean principal investigator of a Korea–US joint project on foundational semiconductor technologies with Amit Trivedi at UIC (NRF–NSF, 2024–2027), and also leads a Korea–Canada academia–industry joint project with McGill University and 1-Material. His professional activities include:

• Vice Chair, Neuromorphic Device Standardization Roadmap, Korea Semiconductor Industry Association (2024–present)
• Co-Chair, MEMS & Sensor Systems Division, Korean Conference on Semiconductors
• Division Chair, Bio-Semiconductor Devices, Korean Society for Medical and Biological Engineering (2025–present)
• Secretary, Active-Matrix Devices Committee Korea Chapter, SID Korea (2024–present)

Sub-mV tunable photonic p-bits for probabilistic computing  

Presenter: Juhyung Seo

Abstract: Rising demands in optimization, inference, and machine learning expose limits of deterministic logic and von Neumann architecture. Quantum approaches promise speedups for specific tasks but remain limited by decoherence, scalability, and operating conditions, prompting classical alternatives. Probabilistic computing (p-computing) uses tunable randomness for exploration and inference, with the probabilistic bit (p-bit) fluctuating between 0 and 1 under input-controlled probability to enable invertible logic and optimization. We report a p-bit device in which stochastic generation and probability control are independent.

A back-to-back heterojunction provides photogeneration and voltage-tunable current asymmetry. Under illumination the dominant junction switches randomly, producing a binary bitstream. Biasing finely shifts only the mean value while maintaining the ideal Gaussian shape of the p-bit distribution. This is possible because the mechanisms for generating and controlling the p-bit are strictly separated. Using measured probability curves, we demonstrate invertible multiplication (Factorization), AND logic, and Max-Cut solving, achieving 0.81 in integer factorization and an average of 0.74 across four Max-Cut instances.

Speaker bio: Juhyung Seo is a Ph.D. student in the Department of Electronic Engineering at Hanyang University, where he conducts research in the We Design Devices (WeDD) Lab led by Hocheon Yoo. His research focuses on the development of next-generation semiconductor devices for neuromorphic computing, hardware-based random number generation, and probabilistic computing. In particular, he explores multi-structured transistors and photonic true random number generators (photonic TRNGs) to realize intelligent and secure semiconductor systems. He is currently supported by a national Ph.D. fellowship from the Korean government and is involved in international collaborations with the University of Illinois Chicago (UIC) and the National Institute for Materials Science (NIMS, Japan) on emerging semiconductor device research. His work has been published or is under review in leading journals such as Advanced Materials, Advanced Functional Materials, and Science Advances. Awarded the NRF PhD Fellowship (50 million KRW / 2 years) (2024).

Transistor-Level Activation Functions via Two-Gate Designs: From Analog Sigmoid and Gaussian Control to Real-Time Hardware Demonstrations  

Presenter: Youngmin Han

Abstract: Tunable analog activation functions are essential for energy-efficient artificial intelligence (AI) hardware. We present two transistor designs: the sigmoid-like activation function transistor (SA-transistor) and the Gaussian-like activation function transistor (GA-transistor), which implement analog sigmoid and Gaussian functions using a screen gate structure. In the SA-transistor, adjusting the screen gate voltage (VScreen-G) enables precise control of the sigmoid slope and saturation level. In the GA-transistor, the amplitude and standard deviation of the Gaussian response are tunable through the same mechanism. These transistors enable precise and continuous tuning of analog activation parameters such as slope, amplitude, and width at the device level. This controllability allows hardware-optimized neural computations tailored to specific tasks or datasets. Applied in real-world tasks, the SA-transistor improved lung magnetic resonance imaging (MRI) classification accuracy from 77% to 84%, and the GA-transistor raised time-series forecasting coefficient of determination (R2) from 0.82 to 0.93. Furthermore, by assembling these devices into a hardware-based multilayer perceptron (MLP), we demonstrated robust inference on the IRIS dataset with 96.7% overall accuracy. This system-level validation highlights that analog activation transistors can directly support neuromorphic accelerators without digital post-processing, reducing circuit complexity and power consumption while maintaining high classification fidelity.

Bio: Youngmin Han is a Ph.D. student in the We Design Devices (WeDD) Lab at Hanyang University, South Korea. His research focuses on next-generation semiconductor devices for AI computing, particularly activation-function transistors capable of implementing Gaussian, sigmoid, and anti-ambipolar nonlinearities at the device level. Also, he investigates how material interfaces, heterojunction transport, and hybrid organic–oxide architectures can be engineered to design and apply diverse device structures with tailored electrical characteristics, enabling new functionalities beyond those achievable with conventional transistor configurations. Youngmin Han also explores material device system co-design approaches to integrate these emerging devices into scalable AI hardware platforms.

• Published Advanced Materials, as well as works in Advanced Functional Materials, Advanced Science, and Nano letters
• Awarded the Korea Presidential Science Scholarship (2024)
• Awarded the NRF PhD Fellowship (50 million KRW / 2 years) (2025)

Faculty host: Amit R. Trivedi, amitrt@uic.edu