Supergate, a specialized company in system semiconductor design, noted on the 22nd that it has secured a project for the 'chiplet-based low-power on-device artificial intelligence (AI) semiconductor technology development' promoted by the Ministry of Science and ICT.
This project aims to implement a system semiconductor platform optimized for chiplet-based low-power on-device technology, developing specialized AI semiconductors for small language models (sLLM) and ultra-low-power spiking neural network (SNN) accelerators. Supergate will serve as the lead organization overseeing this project over the next four years. The total project cost is 26.5 billion won.
Chiplet technology is a method that assembles multiple small chips (chiplets) into a single semiconductor package, allowing for changes in semiconductor configuration based on user demand. Supergate considers the introduction of chiplet technology into on-device AI semiconductors to be highly important from technical, economic, and strategic perspectives.
The host CPU of Supergate will perform a central control role for on-device AI semiconductors. It is expected to function as the core brain coordinating the entire system, especially in chiplet architecture or heterogeneous computing environments.
Shim Seung-pil, CEO of Supergate, stated, 'Through this national project, we plan to develop a host CPU used from edge to cloud based on Arm's Neoverse N2 in TSMC's 5nm process and establish a chiplet assembly lineup in collaboration with ASICLAND, an Arm Total Design (ATD) partner company.' He added, 'In two years, we plan to supply host chips with global competitiveness in the AI semiconductor industry.'