Samsung · Filed Jan 2, 2026 · Published Jul 23, 2026 · verified — real USPTO data

New Samsung Patent Shifts Storage Work to Where Resources Are Available

Samsung has filed a patent for a storage system that can decide, in real time, whether the drive itself or the host computer should handle a complex chain of read operations, depending on who has spare capacity at that moment.

Samsung Patent: Storage Controller That Offloads Reads Adaptively — figure from US 2026/0211578 A1
Figure from the official USPTO publication.
See all 21 drawings from this filing ↓
Publication number US 2026/0211578 A1
Applicant Samsung Electronics Co., Ltd.
Filing date Jan 2, 2026
Publication date Jul 23, 2026
Inventors Hyungon Moon, Chanyoung Park, Minu Chung
CPC classification 711/154
Grant likelihood Medium
Examiner CENTRAL, DOCKET (Art Unit OPAP)
Status Docketed New Case - Ready for Examination (Feb 3, 2026)
Document 20 claims

How Samsung's adaptive storage offloading actually works

Imagine you ask a warehouse worker to fetch a box. Inside the box is a note telling them which other box to fetch next, and so on for several rounds. Normally, you (the manager) would have to read each note yourself and send the worker back each time. Samsung's patent describes a smarter division of labor: the worker can keep fetching boxes on their own as long as they have the energy, but if they get too tired, they hand the task back to you.

In storage terms, the 'manager' is the host computer's processor, and the 'worker' is the storage device. For workloads where each piece of data tells the system where to look next, a common pattern in databases and AI inference, this back-and-forth is extremely slow. Samsung's approach lets the storage device chain those reads together internally, skipping repeated round-trips to the host.

If the storage device is too busy, it signals the host with a special flag and hands control back. The system adapts on the fly, so neither side gets overwhelmed.

How the storage device decides to take over or hand back work

The patent describes a two-layer system involving a host device (the main server or computer) and a storage device (an SSD or similar drive with its own onboard processor).

The host contains an offloading manager that receives a request for a workload with a data-dependent read pattern (a sequence of reads where the address of each subsequent read is contained in the data from the previous one, common in pointer-chasing operations like graph traversal or database index lookups). Instead of fetching each address itself, the offloading manager sends the storage device an adaptive read request bundled with metadata describing how to compute the next address.

The storage device then makes a real-time decision:

  • If it has enough internal resources (compute, memory bandwidth, processing headroom), it reads the first data block, uses the metadata to calculate the next address, and performs the second read, all without involving the host.
  • If it does not have enough resources, it returns a response that includes a reverse offloading field set to an activation value, telling the host to take over the chain.

This dynamic handoff is the core invention. The system avoids committing the full workload to either side permanently, instead routing each step to whichever component is less loaded at that moment.

What this means for data-heavy workloads and server storage

Data-dependent read chains are a well-known bottleneck in database engines, AI recommendation systems, and graph-processing workloads. Every round-trip between the host CPU and the storage device adds latency. Computational storage (putting processing power directly on the drive) has been proposed as a fix for years, but the risk is overloading the drive's modest onboard processor. Samsung's adaptive handoff attempts to get the latency benefits without that risk.

For data-center operators running storage-heavy workloads, this kind of architecture could reduce CPU load and improve throughput without requiring faster storage hardware. Whether Samsung ships this as a feature in its enterprise SSD line or keeps it as an internal research patent is the open question.

Editorial take

This is a solid, specific engineering patent addressing a real and well-documented performance problem in computational storage. It is not flashy, but the adaptive handoff mechanism is a genuinely practical answer to the 'what if the drive is too busy?' objection that has dogged computational storage proposals for years. Data-center engineers will find it more interesting than general consumers will.

There are more where this came from

We read every patent application Big Tech publishes and send you the ones worth knowing. Plain English, free, every week.

The drawings

21 drawing sheets from US 2026/0211578 A1 · click any drawing to enlarge

Patent filing page

Source. Full patent text and figures from the official USPTO publication PDF.