From left: KAIST PhD candidate Lee Ga-yeon, built-in MS-PhD pupil Jang Min-kyu, built-in MS-PhD pupil Kim Chang-hwan, Professor Seo Joon-gi, built-in MS-PhD pupil Heo Nam-uk, and postdoctoral researcher Wenxuan Zhu. Provided by KAIST
■ KAIST introduced on the twenty seventh {that a} analysis group led by Professor Joon-gi Seo within the Department of Chemical and Biomolecular Engineering, along with groups from Hanyang University and Rice University within the United States, has developed a producing expertise that exactly stacks a brand new semiconductor on prime of van der Waals supplies at a low temperature of 150°C with out harm. The analysis outcomes had been printed on July 24 (native time) within the worldwide journal Science Advances. The group used atomic layer deposition (ALD), a method that provides semiconductor precursors sequentially to construct thin films with atomic-level thickness, to make sure that tellurium (Te)-containing precursors type secure thin films. Using this method, they achieved epitaxial progress, through which the higher layer grows in an ordered method following the crystal association of the decrease layer, on numerous van der Waals supplies akin to tungsten diselenide (WSe2), molybdenum disulfide (MoS2), and rhenium diselenide (ReSe2). Conventional semiconductor stacking processes require excessive temperatures of a number of hundred levels Celsius, main to wreck to the underlying layer and interface deformation, however the brand new expertise grows high-quality thin films at 150°C, demonstrating potential for utility to next-generation synthetic intelligence (AI) and low-power semiconductor processes. The group additionally fabricated transistors and optoelectronic units utilizing the thin films, proving that the brand new manufacturing expertise will be utilized in precise machine fabrication.
■ KAIST introduced on the twenty seventh {that a} analysis group led by Professor Sungjin Ahn within the School of Computing has developed a synthetic intelligence (AI) world-model expertise that autonomously theorizes how the world works based mostly solely on observations. The research was chosen as an oral presentation at the forty third International Conference on Machine Learning (ICML 2026), held in Seoul on July 9, and acquired the Best Paper Award at the Constructive Learning Workshop. The group proposed “Learning-to-Theorize (L2T)”, through which an AI infers what guidelines operated between two states by trying solely at observations earlier than and after a change. The neural-network-based principle era mannequin implementing this concept, “NEO”, can autonomously be taught primary guidelines akin to rotation, translation, and coloring, and may clarify new mixtures of guidelines not seen throughout coaching by recombining beforehand discovered rules. Whereas typical world fashions give attention to predicting the subsequent scene, NEO focuses on organizing the causes of adjustments into executable guidelines. The group defined that this expertise might evolve right into a core basis for next-generation AI, together with clever robots, autonomous brokers, and AI methods that help scientific discovery.
■ Ulsan National Institute of Science and Technology (UNIST) introduced on the twenty sixth {that a} group led by Professor Hee-Cheon Park within the Department of Electrical and Electronic Engineering has developed a design automation expertise that mechanically optimizes the circuit structure and interconnection paths of memristor-based logic-in-memory semiconductors. The analysis outcomes had been printed on-line on June 2 in IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems (IEEE TCAD), a journal within the discipline of semiconductor design automation. Logic-in-Memory (LiM) is a semiconductor structure through which reminiscence handles each knowledge storage and logical operations, decreasing knowledge motion between the processor and reminiscence. The group devised an algorithm that mechanically identifies and optimizes the computation blocks that may carry out operations and the routes that may carry the ends in a construction the place memristor units are organized like a checkerboard. The new expertise arranges a number of operations to be carried out concurrently in each horizontal and vertical instructions, decreasing the whole variety of working steps by a median of 19.6% in contrast with state-of-the-art strategies. It may place circuits that typical strategies can not match inside strictly restricted reminiscence house, and is predicted for use for designing low-power, high-performance in-memory computing {hardware}.
■ Daegu Gyeongbuk Institute of Science and Technology (DGIST) introduced on the twenty seventh that it held the “2026 Kiturami Scholarship Award Ceremony” along with the Kiturami Cultural Foundation at the E1 Convention Hall on campus on the twenty fourth. At the occasion, 50 DGIST college students dedicated to their research and analysis acquired a complete of fifty million gained in scholarships and scholarship certificates. Since its institution in 1985, the Kiturami Cultural Foundation has continued its scholarship program for 41 years, supporting greater than 70,000 scholarship recipients nationwide, and the cumulative quantity of its social contributions by the Kiturami Cultural Foundation and the Kiturami Welfare Foundation has reached 61 billion gained. DGIST President Kun-Woo Lee, Kiturami Group Chairman Jin-Min Choi, and scholarship recipients attended the ceremony. DGIST plans to make use of this collaboration as a stepping stone to increase exchanges with Kiturami Group and to boost scholarship and welfare packages in order that college students can give attention to their research and analysis with out monetary hardship.
– doi.org/10.1126/sciadv.aef1430
– doi.org/10.48550/arXiv.2605.03413
– doi.org/10.1109/tcad.2026.3699619
Copyright ⓒ DongA Science. All rights reserved.