|
|
Fault tolerant systems design automation
Lojda, Jakub ; Plíva, Zdeněk (oponent) ; Steininger, Andreas (oponent) ; Sekanina, Lukáš (vedoucí práce)
If a digital system is required to maintain a high level of reliability, it must withstand the presence of naturally-emerging failures. Many of such systems utilize Field Programmable Gate Arrays (FPGAs). One of the approaches to increase the system's reliability is the insertion of the so-called Fault Tolerance (FT) mechanisms. It is, however, a significant challenge to design systems to be FT. In this thesis, an approach is designed and researched, capable of automatically transforming an unhardened design into its FT version. The thesis emphasizes the generality of such a process, which allows for the reusability of the methods among various description formats, languages, and abstraction levels. This thesis describes the proposed method and its main aspects: the source code modification approaches, design strategies, and acceleration of FT parameters measurement. Last but not least, design flows that target the minimization of required measurements are proposed, which significantly accelerates the complete automated design of the FT system. Several cases were experimentally studied during the research presented in this thesis. Multiple circuits described in different languages were targeted with various reliability metrics to cover multiple scenarios. The first steps use a robot controller written in C++ as a target for evaluating the source code manipulations and the so-called critical bits representation of an FPGA design. After that, our C++ benchmark circuits were used instead of the robot controller. At first, a strategy based on the Multiple-choice Knapsack Problem (MCKP) was used to automatically select the most suitable hardening from available hardening schemes (e.g., Triple Modular Redundancy, or N-modular Redundancy). The proposed design strategy found a solution with 18% fewer critical bits while even lowering the design size overhead compared to the previous approach with the static allocation of FT mechanisms. After that, means of FT mechanism insertion were implemented for VHDL. VHDL benchmarks were also used with the MCKP strategy to find solutions with the best Median Time to Failure (a.k.a. t50). For the actual case study, circa 25% savings in the area were achieved compared to the reference design to which the FT mechanisms were assigned statically and manually. The method allows the user to constrain the available chip area and obtain the result optimal on reliability for this given area (under assumptions specified in the thesis). Also, system recovery was tested, which further improved the t50 results by 70%. Finally, a comprehensive case was studied on a real circuit, the FPGA reconfiguration controller. This presents a method of finding a Pareto-frontier of optimal designs considering multiple criteria (i.e., power consumption, size, and Mean Time to Failure - MTTF). The method exploits the principles of dynamic partial reconfiguration.
|
|
|
Fault tolerant systems design automation
Lojda, Jakub ; Plíva, Zdeněk (oponent) ; Steininger, Andreas (oponent) ; Sekanina, Lukáš (vedoucí práce)
If a digital system is required to maintain a high level of reliability, it must withstand the presence of naturally-emerging failures. Many of such systems utilize Field Programmable Gate Arrays (FPGAs). One of the approaches to increase the system's reliability is the insertion of the so-called Fault Tolerance (FT) mechanisms. It is, however, a significant challenge to design systems to be FT. In this thesis, an approach is designed and researched, capable of automatically transforming an unhardened design into its FT version. The thesis emphasizes the generality of such a process, which allows for the reusability of the methods among various description formats, languages, and abstraction levels. This thesis describes the proposed method and its main aspects: the source code modification approaches, design strategies, and acceleration of FT parameters measurement. Last but not least, design flows that target the minimization of required measurements are proposed, which significantly accelerates the complete automated design of the FT system. Several cases were experimentally studied during the research presented in this thesis. Multiple circuits described in different languages were targeted with various reliability metrics to cover multiple scenarios. The first steps use a robot controller written in C++ as a target for evaluating the source code manipulations and the so-called critical bits representation of an FPGA design. After that, our C++ benchmark circuits were used instead of the robot controller. At first, a strategy based on the Multiple-choice Knapsack Problem (MCKP) was used to automatically select the most suitable hardening from available hardening schemes (e.g., Triple Modular Redundancy, or N-modular Redundancy). The proposed design strategy found a solution with 18% fewer critical bits while even lowering the design size overhead compared to the previous approach with the static allocation of FT mechanisms. After that, means of FT mechanism insertion were implemented for VHDL. VHDL benchmarks were also used with the MCKP strategy to find solutions with the best Median Time to Failure (a.k.a. t50). For the actual case study, circa 25% savings in the area were achieved compared to the reference design to which the FT mechanisms were assigned statically and manually. The method allows the user to constrain the available chip area and obtain the result optimal on reliability for this given area (under assumptions specified in the thesis). Also, system recovery was tested, which further improved the t50 results by 70%. Finally, a comprehensive case was studied on a real circuit, the FPGA reconfiguration controller. This presents a method of finding a Pareto-frontier of optimal designs considering multiple criteria (i.e., power consumption, size, and Mean Time to Failure - MTTF). The method exploits the principles of dynamic partial reconfiguration.
|
|
|
Fault tolerant systems design automation
Lojda, Jakub ; Plíva, Zdeněk (oponent) ; Steininger, Andreas (oponent) ; Sekanina, Lukáš (vedoucí práce)
If a digital system is required to maintain a high level of reliability, it must withstand the presence of naturally-emerging failures. Many of such systems utilize Field Programmable Gate Arrays (FPGAs). One of the approaches to increase the system's reliability is the insertion of the so-called Fault Tolerance (FT) mechanisms. It is, however, a significant challenge to design systems to be FT. In this thesis, an approach is designed and researched, capable of automatically transforming an unhardened design into its FT version. The thesis emphasizes the generality of such a process, which allows for the reusability of the methods among various description formats, languages, and abstraction levels. This thesis describes the proposed method and its main aspects: the source code modification approaches, design strategies, and acceleration of FT parameters measurement. Last but not least, design flows that target the minimization of required measurements are proposed, which significantly accelerates the complete automated design of the FT system. Several cases were experimentally studied during the research presented in this thesis. Multiple circuits described in different languages were targeted with various reliability metrics to cover multiple scenarios. The first steps use a robot controller written in C++ as a target for evaluating the source code manipulations and the so-called critical bits representation of an FPGA design. After that, our C++ benchmark circuits were used instead of the robot controller. At first, a strategy based on the Multiple-choice Knapsack Problem (MCKP) was used to automatically select the most suitable hardening from available hardening schemes (e.g., Triple Modular Redundancy, or N-modular Redundancy). The proposed design strategy found a solution with 18% fewer critical bits while even lowering the design size overhead compared to the previous approach with the static allocation of FT mechanisms. After that, means of FT mechanism insertion were implemented for VHDL. VHDL benchmarks were also used with the MCKP strategy to find solutions with the best Median Time to Failure (a.k.a. t50). For the actual case study, circa 25% savings in the area were achieved compared to the reference design to which the FT mechanisms were assigned statically and manually. The method allows the user to constrain the available chip area and obtain the result optimal on reliability for this given area (under assumptions specified in the thesis). Also, system recovery was tested, which further improved the t50 results by 70%. Finally, a comprehensive case was studied on a real circuit, the FPGA reconfiguration controller. This presents a method of finding a Pareto-frontier of optimal designs considering multiple criteria (i.e., power consumption, size, and Mean Time to Failure - MTTF). The method exploits the principles of dynamic partial reconfiguration.
|
host ::
RSS.