An extended model to support detailed GPGPU reliability analysis

Abstract

General Purpose Graphics Processing Units (GPGPUs) have been used in the last decades as accelerators in high demanding data processing applications, such as multimedia processing and high-performance computing. Nowadays, these devices are becoming popular even in safety-critical applications, such as autonomous and semi-autonomous vehicles. However, these devices can suffer from the effects of transient faults, such as those produced by radiation effects. These effects can be represented in the system as Single Event Upsets (SEUs) and are able to generate intolerable application misbehaviors in safety critical environments. In this work, we extended the capabilities of an open-source VHDL GPGPU model (FlexGrip) in order to study and analyze in a much more detailed manner the effects of SEUs in some critical modules within a GPGPU. Simulation results showed that scheduler controller has different levels of SEU sensibility depending on the affected location. Moreover, a reduced number of execution units, in the GPGPU can decrease the system reliability.