Ex) Article Title, Author, Keywords
Ex) Article Title, Author, Keywords
Abstract : Recently, researches on the various types of neuromorphic synaptic devices are attracting attention. In this paper, a synaptic device was fabricated and its characteristics were analyzed using a mesh-type platinum floating gate that mimics nanoparticles. Compared with single floating gates, mesh-type floating gates have a wider memory window and excellent electrical characteristics with improved operation speed and retention. Appropriate thickness conditions were set by checking the EOT (equivalent oxide thickness) and breakdown voltage of the tunnel oxide and control oxide using capacitance-voltage curves and a current-voltage curve. Excellent program and erase operation, synaptic weight, repeatability, reproducibility and memory window width were confirmed using the capacitance-voltage curves. It was compared whether the same performance was achieved even if part of the floating gate was damaged, suggesting the possibility of a synaptic device.
Abstract : Hamiltonian-based impurity solvers for dynamical mean-field theory (DMFT) approximate a continuous hybridization function with a finite set of effective bath orbitals to keep computational costs manageable. This process involves minimizing a cost function that quantifies the difference between the hybridization function of the continuous bath and that of a finite number of bath orbitals. However, as the number of effective bath orbitals increases, minimizing a multi-dimensional cost function becomes increasingly complex, and the computational expense of optimizing bath parameters escalates. To address these challenges, we employ a machine learning approach using supervised learning to replace computationally intensive tasks. We test various features and labels to identify efficient machine-learning models capable of bypassing the time-consuming bath fitting procedure.
Abstract : This study investigated pre-service elementary school teachers' preconceptions regarding the shape of light from the bulbs passing through a small hole by varying the shape of the hole (circular or triangular) and the type of light source (point, linear, or composite). The analysis revealed that the majority of participants, regardless of the number or shape of light sources, responded that the light passing through a circular hole would form a single circular shape on the screen, while the light passing through a triangular hole would form a single triangular shape on the screen. Additionally, 74 participants (71.2%) consistently responded with misconceptions across all the questions. They believed that the shape of light projected onto the screen was determined by the hole shape in the mask, or that the light spread out in a circular manner after passing through the small hole. Furthermore, some participants believed that as the number or size of bulbs increased, the lit area on the screen would also become larger. The findings could be used as fundamental data for developing educational courses and programs at teacher training universities (colleges) to correct the pre-service elementary school teachers' misconceptions regarding the rectilinear propagation of light.
Abstract : Unlike normal interference signals, chirped interference signals cannot be used to obtain the position of a reflector using the Fourier transform. Instead, convolution can extract information about the center wavenumber of the chirped interference signal, which can then be converted into specific position information. The positions and full-width at half maximum (FWHM) of the characteristic peaks resulting from the convolution are interpreted as the position and spatial resolution of the reflector. In this study, the chirped interference signals were acquired by varying the wavelength resolution of the spectrometer from 0.05 to 1.0 nm and the convolution of these signals was analyzed. As the wavelength resolution of the spectrometer was varied, the position of the characteristic peak remained similar, but the FHWM increased proportionally. The spectral resolution of the spectrometer did not have a significant effect on determining the position of the reflector, but higher spectral resolution resulted in better spatial resolution.
Abstract : In recent years, generative AI technology, especially large language models (LLMs), has garnered significant attention for its potential to transform education. This paper provides an overview of generative AI's development and examines its impact on education, focusing on the issue of `hallucinations' in LLMs. It explores the causes and proposes solutions such as finetuning, reasoning, iterative querying, and Retrieval-Augmented Generation (RAG). These methods aim to enhance the accuracy and reliability of AI responses. Examples of AI applications in education include real-time student query responses, personalized learning pathways, and assessment feedback. While these technologies promise to improve educational quality, they also raise concerns about biases and data privacy. This paper discusses strategies to effectively utilize generative AI in education, aiming to improve quality while minimizing negative impacts.
Aekyung Shin, Donggeul Hyun, Jeongwoo Park
New Phys.: Sae Mulli 2023; 73(1): 37-43
https://doi.org/10.3938/NPSM.73.37
Geon Park, Inseo Kim, Hojung Sun, Yongjei Lee, Kimoon Lee, JungYup Yang
New Phys.: Sae Mulli 2023; 73(1): 23-28
https://doi.org/10.3938/NPSM.73.23
Sangwoo Ha
New Phys.: Sae Mulli 2023; 73(9): 734-749
https://doi.org/10.3938/NPSM.73.734