Silicon Peninsula #014 — FADU moves file handling into the SSD

Friday, October 16, 2026 · 5-minute read

The tape

The storage question after this week's infrastructure demonstrations is where useful work happens, not simply how fast a link runs. Today's brief examines a Korean controller supplier's newly disclosed architecture and the evidence an infrastructure buyer should request next. Nasdaq's public SOX page returned data unavailable at our check, so this edition omits an unverified overnight close. Market source.

Signal 1 — The controller takes on file responsibilities

In its October 15 announcement, FADU describes a File-Semantic SSD that moves file lookup, permissions and file-to-address translation into firmware on its Gen6 controller. File requests travel inside NVMe commands, allowing accelerators or remote servers to access files without routing those operations through a host CPU. The company presents hostless storage and OEM/ODM reference designs as development directions. FADU announcement.

Why it matters: Our interpretation is that this changes the integration conversation. A buyer evaluating such a design should ask which responsibilities remain with the application, which move into device firmware, and how the boundary behaves during failure. Moving a task does not eliminate the need to manage it.

For a Korean supplier watchlist, that creates practical questions beyond a controller's headline speed. Who maintains the firmware? How are access rules updated? What happens when an accelerator, network connection or drive fails during an operation? A useful trial should test ordinary operation and recovery under the same application requirements. These are evaluation questions, not defects demonstrated in FADU's implementation.

Signal 2 — A faster demonstration needs an application test

FADU reports an FC6161 demonstration using a 16TB SSD, with performance up to 2.5 times its previous Gen5 controller. It also reports whole-drive power below 28W during active sequential writes and below 5W idle. Those are different operating conditions, not interchangeable power measurements. Company-reported demonstration.

Why it matters: A storage team should separate the interface generation, controller implementation and workload path when comparing results. The purchasing question is whether the proposed configuration completes the team's actual job within its latency, reliability and energy constraints. A bandwidth multiple alone cannot settle that question.

Our suggested evaluation would hold the dataset and application constant, document the host configuration, and measure completion time alongside total system energy. It would also record how much accelerator time is spent waiting for data. A drive-level saving may matter, but a system estimate needs the rest of the machine and the cost of integration. We have not run that experiment and do not claim the announcement provides its result.

Signal 3 — Trace the route from component to qualified system

FADU already describes a flexible commercial model in the OCP marketplace: customers can work with it on a complete SSD or a joint design. That existing listing concerns its older Gen5 platform and should not be read as certification or availability of the newly announced file-semantic product. OCP product listing.

Why it matters: Our commercial interpretation is that different routes create different evidence trails. A branded drive, a controller design win and an OEM reference design are distinct milestones. Analysts should identify which one a later announcement actually establishes before translating it into an adoption assumption.

For overseas equipment and infrastructure teams tracking Korean suppliers, the useful follow-up is specific: request a supported configuration, software integration requirements, qualification criteria and an identified delivery route. An independently reproducible workload would strengthen the technical case; a named deployment or disclosed repeat delivery would strengthen the commercial case. Neither should be inferred from an exhibition photograph.

Korea-only radar

No English exclusivity claim. FADU's announcement is itself in English. This edition adds an evaluation framework rather than claiming a translation lead. Its dated announcement and the older OCP listing describe different product contexts.

Today's number

2.5 times is FADU's reported upper performance comparison with its previous controller, not an independently verified application or whole-system energy improvement. Source.

Seven-day calendar

Through October 23, our editorial follow-up will look for reproducible workload conditions, partner reference designs and explicit qualification or delivery evidence. October 23 is our review date, not a supplier-promised launch. An unanswered question stays open in the ledger; it does not become a negative finding by default.

Silicon Peninsula reads Korean semiconductor filings and company sources, separates documented facts from interpretation, and records the next evidence to check. Reply with a company you want tracked. Our 30-day Korea Watchlist Pilot covers an agreed watchlist; scope is confirmed before payment.