FADU Unveils New AI Data Storage Technology

As artificial intelligence continues to evolve, data processing speed has become critical. South Korean company FADU, a developer of high-performance controllers for data centres, has announced a new type of SSD. The company’s goal is to offload the central processing unit (CPU) entirely by moving file system management directly onto the drive's firmware.

The Problem: Why is the Processor Overloaded?

In traditional server systems, when a database or AI algorithm requests a file, the operating system and block layers consume significant resources determining where specific data is located on the disk. According to FADU, this process—which even GPUDirect Storage cannot fully resolve—consumes approximately 30% of CPU resources.

The new technology, dubbed File-Semantic SSD (FS-SSD), removes this barrier. It utilizes the NFSv3 (Network File System) protocol, which is embedded directly into NVMe commands. As a result, a GPU or DPU (Data Processing Unit) can access files without the intervention of the host server's processor. This approach makes data transfer more efficient and reduces latency.

Technical Innovations and the Future

FADU’s controller integrates the file system with the flash memory management layer. Unlike traditional systems where two separate processes occur, FS-SSD uses a unified map that points directly from the file to the physical address. The company states that this method reduces write amplification by 63% because the garbage collection process is optimized.

Although initial tests have shown impressive results—460,000 operations per second—the technology is still in its early stages. FADU’s current tests are based on the PCIe 5.0 interface, while more powerful results based on PCIe 6.0 are scheduled for the first quarter of 2027.

What This Means for Georgia

Against the backdrop of developing data centres and cloud services in Georgia, such technologies will be important in the future. If FADU can fulfill its promise and bring a product to market by 2027 that reduces server power consumption and costs by 40-50%, it will establish a new standard in the field of data processing. However, the company must first prove that its proprietary software will be compatible with industry-wide standards.

At this stage, FS-SSD remains an ambitious project, with its real potential only to be revealed once samples are distributed to partners.