Memraybity Secures Key Technology to Overcome HBM Capacity Limitations

KOSDAQ-listed company Memraybity announced on December 15 that it has secured next-generation expandable memory technology capable of overcoming the GPU memory capacity limitations, which have been identified as a key bottleneck in artificial intelligence (AI) infrastructure.

This is part of its medium- to long-term growth strategy to address the rapid expansion of AI models and the surge in data processing demand.

Memraybity explained that it has acquired the exclusive licensing rights to the patented 'ByteFlash' technology developed by MemRay, a company specializing in GPU and storage interface and data path technologies, and has established a strategic partnership. Under the agreement, Memraybity will have exclusive rights to develop, manufacture, and sell products based on MemRay's patents in Korea and the United States, including ▲flash-based accelerators and computing devices containing them, and ▲memory controllers and computing devices containing them.

GPUs, which are the core computational units for AI operations, use high-bandwidth memory (HBM) for ultra-fast processing. HBM is considered the optimal memory in terms of processing speed. However, as AI models become larger, the memory capacity required by GPUs increases rapidly, and there are growing concerns that a structure centered solely on HBM leads to higher costs and structural limitations.

The patented technology from MemRay, now secured by Memraybity, is a next-generation expandable memory architecture designed to address these issues. It maintains the existing GPU memory structure while efficiently utilizing large-scale memory resources. The structure overcomes the limited capacity of the small but fast HBM by efficiently connecting and expanding external memory.

A company representative stated, "Leading domestic and international companies in the global AI market are considering the development of next-generation HBF (High Bandwidth Flexible Memory) as a future growth engine," and added, "MemRay's patented technology can serve as a technologically advanced alternative."

The representative further explained, "Assuming an HBF architecture, MemRay's 'GDS (GPU Direct Storage)' technology enables direct data transfer between storage devices and GPUs, while 'ByteFlash' forms the expandable memory area accessible by the GPU."

MemRay is a specialized original technology development company established in 2016 through an industry-academia collaboration between Yonsei University and KAIST.