Konferenzen und Rahmenprogramm
An FPGA-based Neural Network Accelerator for Computational Storage Platforms
The nowadays requirements to process highly computation- and storage-demanding workloads is rapidly exceeding the traditional capabilities of server architectures. To this extent, the computational storage paradigm came into play. This approach revolves on the off-loading ability of the reliability/power consumption/performance management routines that are usually demanded in the storage components. In enterprise environments, this could be enabled by an FPGA architecture that shares the same interconnection fabric (usually PCIe) of the Solid State Drives constituting the server storage. In this presentation, we will show an example of an FPGA-based Neural Network Accelerator that relives the burden of the computation of a reliability management algorithm through an AI-approach, namely the Moving Read Reference, from the Solid State Drive controller. The results promisingly shows a low power consumption, and a very low latency that does not interfere with the normal SSD behavior.
--- Datum: 26.02.2020 Uhrzeit: 12:30 - 13:00 Uhr Ort: Conference Counter NCC Ost