Abstract

Heterogeneous System on Chips (SoCs) provide a robust platform that can flexibly handle different applications. Their small form factor and low power consumption for their performance have made SoCs an attractive option for space applications. However, SoCs are sensitive to the ionizing radiation found in space. We need to understand how an SoC will behave in the high radiation found in space. This requires us to perform radiation testing of the SoCs on Earth to understand their behavior. The radiation testing and analysis of a heterogeneous SoC is complex as it contains multiple different processing cores, memories, and other system resources. Each of these components has unique failure modes and behaviors that must be evaluated. To effectively use testing resources, it is necessary to test these components in parallel. However, this adds to the difficulty of the analysis of failures, as the cause of failure between the different components must be properly identified. This thesis demonstrates a testing approach for SoCs that evaluates all memories and processors in an SoC simultaneously. This allows testing time to be reduced as components are evaluated in parallel. Extensive register and log data is captured using multiple interfaces, including UARTs, JTAG, and PCIe, and used to diagnose and categorize failures. This allows an evaluation of all of the core components of an SoC. This method was developed targeting the AMD Versal ACAP device. The approach was refined over the course of nine radiation tests at multiple different facilities using both heavy ion and proton radiation. Using the method, the memories, processors, and key subsystems' radiation response was characterized. The failure rates and failure modes of each of the four processors were determined. From the failures observed in the Versal, we made several key observations. First, we learned the importance of ECC memory protection and scrubbing of processor memories for processor reliability. Second, we observed that a primary source of processor failures is system networks and interconnects rather than internal processor failures.

Degree

College and Department

Ira A. Fulton College of Engineering; Electrical and Computer Engineering

Rights

https://lib.byu.edu/about/copyright/