Industry update — On 4 February 2026, AMD announced the Kintex UltraScale+ Gen 2 mid-range FPGA family, aimed at high-throughput imaging, professional video, test-and-measurement and industrial-vision systems — exactly the capture-heavy applications where an on-FPGA recorder has to stream a continuous data flow straight to disk without ever dropping a sample (AMD, 4 February 2026).
When an instrument, a sensor array or a video pipeline produces data faster than any host can absorb it, the storage stops being an afterthought and becomes the design. A SATA III recorder IP turns the FPGA itself into the logger: the fabric receives the stream, meters it into a solid-state disk over a 6 Gbit/s SATA link, and keeps writing — continuously, without a host CPU sitting in the byte-by-byte path. This guide explains what a SATA III recorder core does, why the SATA generation and the RAID topology set your sustained rate, how the recorder writes media a PC can read afterwards, and where the honest limits are.
The short answer
A SATA III recorder IP is a soft core that lets an FPGA log a continuous, high-rate data stream directly to one or more SATA solid-state disks, with no host CPU in the data path. Each SATA III link runs at 6 Gbit/s; Logic Design Solutions’ published figure for a single link is on the order of ~500 MBytes/s of sustained storage throughput, and the recorder scales beyond one disk by striping several in RAID0. It writes either raw blocks (maximum rate) or a FAT32 file system so the media reads back on a PC, and it targets any FPGA whose transceivers reach 6 Gbit/s — Kintex-7, Kintex UltraScale+, Virtex-7 and Altera Arria/Stratix among them. Size it from the sustained capture rate, not a burst datasheet number.
What a SATA III recorder IP is
A general-purpose computer records to disk through an operating system, a file-system driver and a storage stack — layers that add latency and, worse, jitter. For a recorder that must never miss a sample, those layers are the problem. A SATA recorder IP removes them: it implements the SATA host protocol and the write path as hardware inside the FPGA fabric, so the device that is already digitising the data also owns the disk. There is no soft CPU, no OS, and no software storage stack between the sensor and the SSD — just a deterministic hardware pipeline.
The recorder is a specialised sibling of the plain SATA Host IP. Where a bare host issues reads and writes on command, a recorder wraps that host in a streaming front end: it accepts a continuous data source, buffers it, meters it into the disk at line rate, and manages the on-disk layout so the capture is retrievable afterwards. In the LDS catalogue the same idea appears on the NVMe side as the NVMe Host Recorder IP; the SATA flavour is the right choice when SATA SSDs, ruggedised 2.5-inch drives or an existing SATA backplane are what the system already uses.
Why SATA III for continuous logging
SATA is a serial interface: one differential pair per direction, defined at three generations. Per the specification maintained by SATA-IO, the line rates are 1.5 Gbit/s (Gen1), 3.0 Gbit/s (Gen2) and 6.0 Gbit/s (Gen3, “SATA III”). SATA III is the fastest the standard offers, which is why a recorder that wants headroom runs its links at Gen3. Inside the FPGA, the LDS SATA host runs its system clock (Clock_sys) at 150 MHz for a 6 Gbit/s link (against 75 MHz at 3 Gbit/s and 37.5 MHz at 1.5 Gbit/s), and the 8b/10b encoding lives in the device transceiver.
Raw line rate is not usable rate. After 8b/10b encoding and protocol overhead, a single 6 Gbit/s link delivers less than its headline number, and the figure that matters for a recorder is sustained write throughput to real flash. LDS’s own published data point — from its SATA storage work — is on the order of ~500 MBytes/s per SATA III link. That is the honest planning number: fast enough for a large class of instrumentation, telemetry and video-capture workloads on one disk, and the unit you multiply when you need more.
This is not a niche academic idea. A January 2026 paper in Electronics describes a high-speed storage system built on exactly this pattern — a Kintex-7 FPGA as the control core streaming a PCM data flow into an mSATA disk over the SATA 3.0 bus — independent confirmation that SATA III on an FPGA is a proven route to continuous logging.
Have a capture rate and a target SSD in mind, and want to know whether one SATA III link or a RAID0 set covers it? An engineer who has shipped SATA recorders can size it against your real stream.
Inside the recorder data path
The recorder is more than a SATA host bolted to a FIFO. The LDS SATA recorder architecture adds the pieces a continuous logger actually needs:
- A choice of on-disk format — FAT32 or raw. Raw mode writes blocks straight to the disk for the highest, most deterministic rate; FAT32 lays the capture down as files a computer can mount and read directly.
- RAID0 across multiple disks. One SATA III link tops out around the ~500 MBytes/s figure above; to go faster, the recorder stripes the stream across several disks in RAID0, and the aggregate rate rises with the number of drives in the set.
- An Ethernet ingress option. The data to be recorded does not have to arrive from the fabric alone. The core can take a network feed through a 1 Gbit GMII or 10 Gbit XGMII/XAUI MAC, supports jumbo frames up to 9216 bytes, and speaks ATA-over-Ethernet (AoE) so a remote host can address the recorder’s disks over the network.
- A built-in Generator_Checker block. A hardware test-pattern generator and checker lets you validate the whole write-then-read path at full rate before trusting it with real data — the difference between “it compiled” and “it sustains”.
Under all of that sits the SATA host itself, with the machinery the protocol requires: 48-bit sector addressing, PIO and DMA transfers, and Native Command Queuing (NCQ). The DMA path is what makes streaming deterministic — the fabric moves data to the disk without a processor arbitrating each transfer.
Sizing for a sustained, drop-free rate
The only rate that matters for a recorder is the sustained one. Flash SSDs publish headline burst figures that they cannot hold once their internal cache fills and garbage collection kicks in; a recorder that is sized to the burst number will overflow its buffers and drop samples minutes into a run. Design instead from the disk’s sustained write rate and leave margin. The discipline is the same one we apply to NVMe recorders:
- Start from the source rate. Work out the true, continuous rate your digitiser or network feed produces — not its peak.
- Match it to sustained disk throughput. If a single SATA III link’s ~500 MBytes/s covers the source with headroom, one disk is enough. If it does not, add disks in RAID0 until the aggregate sustained rate clears the source rate with margin.
- Keep the buffers fed and the queues deep. Enough FIFO between the source and the disk absorbs the SSD’s momentary stalls; NCQ keeps commands outstanding so the drive is never idle.
- Prove it with the Generator_Checker. Run a full-rate capture-and-verify before the instrument is in the field, so a dropped sample shows up on the bench, not in the mission log.
None of this needs an invented benchmark. It needs the disk’s real sustained figure, the link’s ~500 MBytes/s planning number, and enough RAID0 width and buffering to keep margin between them.
Which FPGAs can host it
A SATA III recorder needs one thing from the silicon above all else: a transceiver that reaches 6 Gbit/s. That, not logic capacity, is what rules a device in or out — the host stack and recorder logic are a comparatively modest block. LDS ships the SATA Host IP across a wide range of AMD/Xilinx and Altera families with transceivers in that class:
| Vendor | Families with SATA III-class transceivers | Recorder fit |
|---|---|---|
| AMD / Xilinx | Kintex-7, Kintex UltraScale+, Virtex-7 (GTX/GTH), UltraScale+ (GTHE4), Artix-7, Zynq-7 | Mainstream choice for high-rate recorders; the newest Kintex UltraScale+ Gen 2 parts extend the mid-range further |
| Altera | Arria V/10 GX, Stratix IV GX, Cyclone V GX | Covered by the Altera SATA Host IP line for Altera-based systems |
The point of a portable core is that this stays a system decision. If your capture platform is already a Kintex UltraScale+ board, the recorder targets its GTH transceivers; the family-specific mechanics are covered in SATA Host IP on Xilinx Kintex UltraScale+. If the design is Altera-based, the same recorder concept maps onto Arria or Stratix transceivers instead. Confirm the specific part’s transceiver rating against the vendor documentation before committing — family membership is necessary but the individual variant decides.
FAT32, raw or exFAT: reading the media back
How the recorder lays data on the disk decides how you get it off. There are three practical modes, and the trade is rate against convenience:
| Mode | What it gives you | Constraints |
|---|---|---|
| Raw | The highest, most deterministic write rate — no file-system bookkeeping in the path | The media is a block image; you read it back with a matching tool, not by mounting it on a laptop |
| FAT32 | Files a PC can mount and read directly after a capture | Disk up to 2 TB, individual file up to 2 GB — long captures roll across multiple files |
| exFAT | PC-readable files without FAT32’s size ceilings — any disk size, large files | Layered via the exFAT file-system IP; suits removable, field-read media |
The rule of thumb: choose raw when nothing beats the capture rate and you control the read-back tool; choose FAT32 for simple PC-readable media within its 2 TB / 2 GB limits; and reach for the exFAT layer when you need OS-readable files but the FAT32 ceilings get in the way. Whether SATA is even the right interface, against an NVMe/PCIe recorder, is its own decision — we weigh it in SATA vs NVMe for FPGA storage.
What this article does not cover
To keep the numbers honest, a few things are deliberately out of scope:
- A guaranteed throughput figure for your system. Real sustained rate depends on the specific SSD, the RAID0 width, the buffering and the source — the ~500 MBytes/s-per-link figure is a planning anchor from LDS’s published SATA data, not a promise for an arbitrary drive.
- Per-device resource usage. LUT, BRAM and flip-flop counts depend on configuration and target part; we scope them per design rather than quote a universal number.
- Latency numbers. A streaming recorder is sized for sustained bandwidth and drop-free capture, not for a published latency figure — treat any single latency claim with suspicion.
- The digitiser and analog front end. This is about the storage path; getting the data into the fabric in the first place is a separate design.
Frequently asked questions
What is a SATA III recorder IP on FPGA?
It is a soft IP core that turns an FPGA into a continuous data logger: it implements the SATA host protocol and a streaming write path in the fabric, so the device captures a data stream and writes it straight to one or more SATA solid-state disks at 6 Gbit/s, with no host CPU or operating system in the data path.
How fast can a SATA III recorder log data?
Each SATA III link runs at 6 Gbit/s raw; after encoding and overhead, LDS’s published planning figure is on the order of ~500 MBytes/s of sustained storage throughput per link. To exceed that, the recorder stripes several disks in RAID0 and the aggregate sustained rate rises with the number of drives. Always size from the sustained rate, not a burst datasheet number.
Can the recorder write media a PC can read afterwards?
Yes. In FAT32 mode it writes files a computer can mount directly (disk up to 2 TB, file up to 2 GB); an exFAT file-system layer removes those size limits for large or removable media. For the highest, most deterministic rate you can instead write raw blocks and read them back with a matching tool.
Which FPGAs can run the SATA III recorder IP?
Any FPGA whose transceivers reach 6 Gbit/s. On AMD/Xilinx that includes Kintex-7, Kintex UltraScale+, Virtex-7 (GTX/GTH), UltraScale+ (GTHE4), Artix-7 and Zynq-7; on Altera it includes Arria V/10, Stratix IV and Cyclone V GX parts. Confirm the specific variant’s transceiver rating against the vendor documentation.
How do you keep a continuous capture from dropping samples?
Size the storage from the disk’s sustained write rate rather than its burst figure, add RAID0 disks until the aggregate clears the source rate with margin, keep enough FIFO buffering to absorb SSD stalls, and use Native Command Queuing to keep the drive busy. Validate the whole path at full rate with the built-in Generator_Checker before deployment.
Does the recorder need a CPU or operating system?
No. The SATA recorder runs entirely as a hardware data path in the FPGA fabric, which is exactly what makes it deterministic and jitter-free. A processor, where present, can supervise and configure the recorder but does not sit in the byte-by-byte write path.
Building a continuous high-rate recorder and want to skip writing the SATA host and streaming path from scratch? Tell us your sustained capture rate, target SSD and FPGA, and we will confirm the fit or scope the recorder IP.
- SATA-IO — Serial ATA specification: 1.5 / 3.0 / 6.0 Gbit/s generation line rates and 8b/10b encoding
- AMD — Kintex UltraScale+ Gen 2 mid-range FPGAs (announced 4 February 2026)
- Electronics 2026, 15(2), 452 — Design and Implementation of a High-Speed Storage System Based on SATA Interface (Kintex-7, SATA 3.0, mSATA), published 20 January 2026
- Logic Design Solutions — SATA Host IP, SATA recorder and related FPGA storage cores
