跳到正文
The Elec· JEONG IL JOO·· 5 天前AI 評分24

Panmnesia 聯席行政總裁 Kwon Mi-ryeong 獲 IITP 院長嘉許

Panmnesia’s Kwon Mi-ryeong Receives IITP President’s Commendation

AI 導讀

Panmnesia 聯席行政總裁 Kwon Mi-ryeong 獲 IITP 院長嘉許,表揚她在 AI 數據中心互連技術開發、研究成果及提升韓國 ICT 能力方面的貢獻。Panmnesia 連續第二年入選優秀 ICT 研發成果展,並展示支援 AI 數據中心裝置互通的開放標準連結半導體。

正文

Panmnesia said on Nov. 13 that Kwon Mi-ryeong, its co-CEO, received a commendation from the president of the Institute of Information & Communications Technology Planning & Evaluation (IITP).

The award ceremony recognizing contributions to the dissemination of information and communications technology (ICT) achievements was held on Nov. 11 at the opening of the 2027 AI & ICT Technology Outlook Conference. Kwon was recognized for contributions to developing AI data center interconnect technology, generating research achievements and advancing South Korea’s ICT capabilities.

Panmnesia was selected for the event’s exhibition of outstanding ICT research and development (R&D) achievements for the second consecutive year. The company showcased open-standard-based link semiconductors that support communication between devices in AI data centers.

Panmnesia is developing interconnect technology that enables different devices, including neural processing units (NPUs), central processing units (CPUs) and memory, to operate within a single data center. The technology is based on open-standard interconnects such as Compute Express Link (CXL) and UALink.

Its modular artificial intelligence (AI) data center architecture, designed using the interconnect technology, separates resources traditionally tied to individual servers and connects multiple devices into a unified system. The architecture allows resources to be flexibly combined as modules according to AI workloads.

At the core of the architecture is Panmnesia’s PCI Express (PCIe) 6.4-CXL 3.2 Fusion Switch, which supports both PCIe and CXL on a single chip. The switch extends CPU memory expansion and sharing capabilities to the rack level through CXL.

Even devices located far apart can exchange data through memory read and write requests alone. This reduces unnecessary data copying and software intervention, lowers latency and enables an entire data center to operate as a single system, reducing total cost of ownership (TCO).

Panmnesia plans to develop all technology layers in-house, from intellectual property (IP) to chips, systems and software, to support sovereign AI data centers that are not dependent on any single company and contribute to sustainable AI infrastructure.

來源:The Elec · thelec.net