Case:

For some applications, we have a payload to store the data for an amount of time while transferring the data to the bus slower than the data acquisition rate. So, there is no need of real time bridge to the bus. We have max rate of 24 Gbps of raw data, not including communication overhead, for short periods of time like 5 seconds. The main downstream interfaces to use, for example PCIe, Ethernet (1G/2.5G/10G) / SpaceWire / Custom SerDes, or direct solid-state storage such as NVMe.

Solution:

We will configure the onboard recorder based on the MPFS460T-1FCG1152I PolarFire SoC with the following architecture:

Ingest:

8 configurable high-speed SERDES inputs supporting CoaXPress (CXP-3 / CXP-6 / CXP-12) or custom serial protocols. Design point: 25 Gbps worst-case, comfortably covering your 23.1 Gbps raw payload rate.

Burst Buffer:

128 Gbit (16 GB) RAM with ECC (Total 144Gbit) with 12.8 GB/s sustained bandwidth. Your 5-second burst at 25 Gbps produces 125 Gbit — fits in the DDR4 buffer with 3 Gbit headroom. With write-through to storage active during capture, the buffer never fills completely.

Mass Storage:

• 2× U.2 NVMe (PCIe Gen2 x2 each, ~0.8 GB/s per drive)
• 4× U.2 SATA III (~0.5 GB/s per drive)
• Aggregate sustained write: ~3.5 GB/s

Each burst drains to persistent storage in roughly 5 seconds. Total capacity depends on SSD selection — e.g. 6× 2 TB gives ~12 TB (~768 full bursts).

Downstream Egress:

2× 10G, aggregate ~2.2 GB/s. A full burst reads out in approximately 6–10 seconds, and can overlap with the next capture. We will also expose extra 4 SerDes lanes for any other high-speed interfaces.

Mechanical:

Standalone aluminum enclosure. Micro-D connectors for command, telemetry, and power. Twinax on Micro-D or circular connectors for high-speed differential (CoaXPress, 10GigE).