Measuring and predicting radiation-induced soft errors in microprocessors

Semiconductor electronic devices are vulnerable to radiation which upsets the state of memory cells. The upsets, which are generally not permanent, are labeled as soft errors. Measuring and predicting the soft error rate is important in designing reliable systems for the terrestrial or space radiation environment. Inexpensive, but powerful, Raspberry Pi™ computers without radiation hardening, mitigation, or testing have found their way onto the International Space Station and into satellites. In this research, these microprocessor-based systems were measured for radiation effects of accelerated, high-energy neutrons at the Los Alamos Neutron Science Center at the request of NASA’s Jet Propulsion Laboratory. High energy neutrons are the product of galactic cosmic rays entering the earth’s atmosphere. The accelerated measurements can be scaled to the terrestrial or at-altitude neutron flux to predict soft error rates. Neutron results are also an indicator of system vulnerability to other radiation sources. Radiation testing is rarely done with an operating system due to the additional errors observed and the complexity of analysis. However, the commercial off-the-shelf hardware appearing in space requires an operating system – often Linux. This research incorporated testing with Linux and benchmarks modified to provide additional visibility into Linux detected errors. From the measurements collected for benchmarks, a new methodology for prediction of soft error rate for other software is proposed. Radiation cross section per cache level memory access is proposed as an alternative to the traditional cross section per memory bit. The memory access information is easily collected from the hardware performance counters present on high-end microprocessors. A Linux utility, perf, reports performance counter results without modification of the user code. This new method is hypothesized to better account for the variability of software working set, data lifetime, and memory latency. More accurate prediction is demonstrated on fourteen benchmarks measured in the neutron beam on the Raspberry Pi™ 3B+ with its ARM® Cortex®-A53 core.

Name thesis:degree_name

Doctor of Philosophy

Level thesis:degree_level

Doctoral

Discipline thesis:degree_discipline

Electrical and Computer Engineering

Grantor

The University of Texas at Austin

Year dc:date.issued

2024

Author dc:creator

• Corley, Charles Joseph

Advisor dc:contributor.advisor

• Swartzlander, Earl E., Jr., 1945-

Committee members dc:contributor.committeemember

• Heather M. Quinn
• Mark W. McDermott
• Nur A. Touba
• Andreas Gerstlauer

• Radiation testing
• Performance counters
• Accelerated neutron testing
• Raspberry Pi
• Microprocessor soft errors
• Soft error effects

Language dc:language.iso

English

Identifier URI

https://doi.org/10.26153/tsw/59424

OAI identifier oai:identifier

oai:repositories.lib.utexas.edu:2152/132080