SEOUL, December 31 (AJP) – Researchers on the Korea Advanced Institute of Science and Technology have developed a next-generation synthetic intelligence semiconductor that integrates sensing, computing, and reminiscence right into a single chip. By mimicking the construction of the human eye and mind, the expertise permits gadgets to understand and course of data instantly with out the facility loss or delays present in conventional computer systems.
The Korea Advanced Institute of Science and Technology (KAIST) introduced on December 31, Wednesday, {that a} analysis staff led by Professor Chon Sang-hun from the School of Electrical Engineering offered six papers on the International Electron Devices Meeting (IEEE IEDM 2025) in San Francisco. The staff work on built-in imaginative and prescient sensors was chosen as a “Highlight Paper,” and their analysis on extremely dependable reminiscence obtained the “Top Ranked Student Paper” award.
Standard AI programs perform like a relay race: a digicam sensor takes an image, converts it into numbers, strikes these numbers to reminiscence for storage, and then sends them to a processor to be calculated. This fixed transferring of information again and forth makes use of a number of electrical energy and creates a “lag” in response time. The KAIST staff solved this by making a “full-stack” AI chip the place the sensor and the brain-like processing circuits are bodily stacked on high of one another in very skinny layers.
This “in-sensor” expertise permits the chip to “see and think” on the identical time. Instead of sending uncooked information elsewhere, the sensor itself performs the calculations. This mimics how the human nervous system processes primary visible options earlier than the knowledge even reaches the deep mind. By eliminating the necessity to transfer information between totally different chips, the system makes use of considerably much less energy and reacts a lot quicker, making it very best for real-time use in cellular gadgets, drones, and robots.
In addition to the sensing expertise, the staff improved the storage a part of the AI chip. They developed a brand new sort of reminiscence referred to as “Next-Generation NAND Flash” that makes use of a particular materials to retailer information extra reliably whereas utilizing decrease voltage. This ensures that the AI can retailer huge quantities of data safely for a very long time with out draining the battery.
Professor Chon Sang-hun acknowledged that this analysis proves that sensors, computation, and storage—which was designed individually—will be mixed into one single system and materials. He famous that this platform will probably be expanded to be used in “edge AI,” which refers to sensible gadgets that must course of advanced data domestically and instantly with out counting on a central server.
The analysis was supported by the Ministry of Science and ICT and the National Research Foundation of Korea. It was carried out in collaboration with Samsung Electronics, Kyungpook National University, and Hanyang University.
(Paper data)
Journal: IEEE IEDM 2025 Title: Monolithically Integrated Photodiode–Spiking Circuit for Neuromorphic Vision with In-Sensor Feature Extraction DOI: https://iedm25.mapyourshow.com/8_0/sessions/session-details.cfm?scheduleid=255
Title: A Highly Reliable Ferroelectric NAND Cell with Ultra-thin IGZO Charge Trap Layer; Trap Profile Engineering for Endurance and Retention Improvement DOI: https://iedm25.mapyourshow.com/8_0/sessions/session-details.cfm?scheduleid=124
Copyright ⓒ Aju Press All rights reserved.