BANDF AD
Fabless SSD and power management controller maker FADU is drawing back on its CXL switch investments, and thinks high-bandwidth flash excitement is misplaced.
The fabless Korean company has developed SSD controller chip hardware technology using parallel processing of pipeline stages and supplies server OEMs and hyperscalers with its technology. It has also developed power management integrated circuit (PMIC) technology to manage power more efficiently for NAND and DRAM devices, and is developing combined SSD controller/PMIC units.
BANDF AD
It recently announced a 2.5x revenue increase from KRW 10.1 billion ($690 million) in the third 2024 quarter to KRW 25.6 billion ($860 million) in the third 2025 quarter, driven by accelerating demand for SSDs used in AI data centers. FADU has recently won a series of large-scale contracts over four consecutive months, with expectations for a substantial increase in revenue next year.
During a press briefing about its earnings, reported on X by Korea’s Citrini Research analyst jukan05, co-founder and CTO Eyee Hyun Nam said FADU was developing a PCIe gen 6 SSD controller, and it will launch DRAM PMIC modules next year, which will drive more power-efficiency than competitors’ chips. He noted that competitors supply SSD controllers and PMICs as separate devices and not integrated in a unified package.
Like Panmnesia, FADU, via its EEUM subsidiary, has been developing a switch to link CXL 3.0 memory -sharing devices. But it is reducing investment in this effort as Nvidia’s NVlink technology, which provides memory pooling within a DGX/HGX pod, becomes stronger. The CXL market is not developing as fast as FADU, and others, had hoped.
It will develop its CXL switch in FPGA form up to the proof-of-concept stage, and then freeze it until the CXL memory sharing market becomes substantial, if it does.
BANDF AD
NAFADU CEO and co-founder Jihyo Lee discussed high-bandwidth flash (HBF), which is being promoted and standardized by Sandisk and SK Hynix. HBF is seen by the two as complementing a GPU’s local high-bandwidth memory (HBM). Nvidia, the largest HBM buyer in the market, and hence the most important potential HBF consumer, has not publicly said anything about HBF.
FADU suggested that there were three obstacles standing in the way of HBF. GPU accelerators can operate at higher temperatures than NAND can tolerate. The write endurance of TLC and QLC NAND is limited, very much so compared to DRAM, which would limit an HBF stack’s life, and there is no or limited compatibility between different types of NAND, making it hard for a NAND controller supplier, such as FADU, to find a large-enough market for an HBF controller.
