Distributed Triple Modular Redundancy in Heterogeneous Multiprocessing Systems with C++

Abstract

Software used in safety-critical industries such as automotive, medical, and space applications should be safe and reliable. The systems used in such industries are subjected to hardware failures, human errors, or environmental/operational stress that may lead to software errors. Any errors in that software can result in functional failure and could have potentially catastrophic consequences such as personal injury, instrument damage, or even loss of life. So to improve the functional safety of the system, international safety standards such as IEC 61508 are framed which provide necessary tools, guidelines, and methods for developing safety-related systems. Triple Modular Redundancy (TMR) is a widely used safety mechanism that, when implemented in a certain way, the system will satisfy parts of the safety standard because of it. The focus of this thesis is to develop a Distributed Triple Modular Redundancy (DTMR) library in C++ that can be used in embedded software programs to improve the functional safety of the system. The safety mechanism developed is a C++ library that can be easily added to embedded software programs running on a heterogeneous system such as SoC FPGAs. The developed library implements the TMR concept in a heterogeneous environment. Parts of the library are able to be executed both on the hardcore processor in the SoC and on the softcore processor in the FPGA. The library masks any single fault/failure generated due to factors such as radiation, hardware failures, environmental stress and keeps the system operational. The Distributed Triple Modular Redundancy (DTMR) implementation successfully protects the user function improving the safety of the system. The developed DTMR library is evaluated by injecting errors into the software, which are errors simulated based on the real-world environment, such as silent data corruption and processor hang, which occur due to factors like radiation, temperature, and aging. The thesis also investigates the effectiveness and overhead costs of the developed DTMR library. The library is built in C++ and the entire hardware and software implementation is done on the PYNQ-Z2 FPGA board.