Iron Device 將開發供機械手使用的端側 AI 運算模組
Iron Device to Develop On-Device AI Computing Module for Robotic Hands
Iron Device 將參與政府支持的研究項目，開發可在機械手內完成 AI 處理的低功耗端側運算模組。模組採用整合 CPU、GPU 和 NPU 的 SoC，目標是較 Nvidia Jetson Orin Nano 體積縮小約 59%、功耗效率提升至 3 倍。項目亦計劃把由收集感測數據至馬達控制的延遲降至 20 毫秒或以下，總預算約 5.25 billion won。
Iron Device is developing an on-device AI computing module for robotic hands. (Photo: Iron Device)
Fabless semiconductor company Iron Device said on Oct. 15 that it will participate in a government-backed research project to develop an on-device artificial intelligence (AI) computing module.
The company was selected as a joint research partner in a project led by the Ministry of Trade, Industry and Resources (MOTIR) to develop high-degree-of-freedom (DOF) robotic hands and precision manipulation and assembly control technologies based on on-device AI. Iron Device is participating in a consortium with Tesollo, Hanyang University's ERICA Industry-University Cooperation Foundation and the Korea Institute of Robotics & Technology Convergence. The project has a total budget of approximately 5.25 billion won and will run from August this year through the end of 2029.
Iron Device will develop a low-power computing module capable of handling all AI processing within a robotic hand. The module will take the form of a system-on-chip (SoC) integrating a central processing unit (CPU), graphics processing unit (GPU) and neural processing unit (NPU).
The company aims to reduce the module's footprint by approximately 59% compared with Nvidia's Jetson Orin Nano platform while tripling power efficiency, measured in tera operations per second per watt (TOPS/W).
The module will be integrated with robotic hands and existing sensor modules developed by project leader Tesollo. By unifying data pathways between modules, the project aims to reduce latency from sensor data collection to motor control to 20 milliseconds or less. Iron Device will also be responsible for simplifying and optimizing AI inference algorithms developed by Tesollo and Hanyang University. The optimized algorithms will be deployed on Iron Device's module.
The company said it plans to sell the completed module to domestic and overseas humanoid and industrial robot customers independently of the research project.
"We aim to reduce our customers' robot system costs by more than 25% through this platform," said Lee Young-sik, vice president at Iron Device.
來源：The Elec · thelec.net