Why a new core?
As SSDs get faster, controllers need more CPU cores. According to SK hynix, PCIe bandwidth doubles every generation, but controller CPU clocks and control capability can't keep up, and power and area are already at their limits.
What is the H-TPU?
The H-TPU (Hynix Transaction Processing Unit) is not an AI accelerator. It handles repetitive work such as read and write command processing, metadata management and buffer control. It uses a four-stage, in-order pipeline; it is not superscalar and does not use out-of-order units.
How it works
Storage control requires more than 100 detailed functions, and loading all of them onto a single CPU is not efficient. SK hynix splits these functions across dozens of small TPU clusters, with each H-TPU focused on a specific job. Error recovery, exception handling and storage-policy management stay on the main CPU. In other words, the H-TPU complements the CPU rather than replacing it. Cores that a given I/O doesn't need drop into a minimum-power mode.
Numbers and timing
Koo In-jae, the company's VP of system-on-chip design, outlined the plan this week at ISOCC 2026. The new cores are said to draw about 2.5% of a general-purpose CPU core's power and take up around 4% of its area. SK hynix reportedly plans to apply the H-TPU across its NAND product family, including next-gen enterprise SSD controllers.
