AnoDe : A Log-based Self-Supervised Framework to Detect Scrubber Failures in SRAM-FPGA

• SRAM-FPGAs used in radiative environment are integrated with a scrubber to protect the configuration memory from radiation effects. Any malfunctions in the scrubber degrades the reliability of the system and can have catastrophic consequences in critical applications. Existing solutions for a reliable scrubber focus on masking or detecting the scrubber errors through redundant modules implemented in the FPGA. While these approaches improve the overall reliability, they are not completely radiation-proof owing to their implementation on FPGA. In order to improve the scrubber reliability, a complementary scheme external to the FPGA is required. Based on this consideration, we propose AnoDe, a failure detection framework running on a supervisory processor external to the FPGA board. AnoDe leverages the logs generated by the scrubber to detect failures in real-time using an Autoencoder network. AnoDe provides a self-supervised solution right from generating the labelled training data to dynamically adapting to the prevailing radiation conditions. We evaluated the effectiveness of our approach on a Xilinx Ultrascale+ MPSoC ZCU104 board using fault injection. The results demonstrated a detection performance comparable to that of a custom approach with an F1 score of 0.85 for single bit upsets and 0.93 for multi bit upsets. Overall, the proposed approach could reduce the scrubber SEU sensitivity from 6 % to 1 %.

Ämnesord

TEKNIK OCH TEKNOLOGIER  -- Elektroteknik och elektronik (hsv//swe)

ENGINEERING AND TECHNOLOGY  -- Electrical Engineering, Electronic Engineering, Information Engineering (hsv//eng)

Nyckelord

radiation

scrubber

anomaly

single event upsets

fault tolerance

clustering

autoencoders

dynamic thresholding

Publikations- och innehållstyp

ref (ämneskategori)

kon (ämneskategori)