Run-Time Mitigation of Power Budget Variations and Hardware Faults by Structural Adaptation of FPGA-Based Multi-Modal SoPC

Abstract

Systems for application domains like robotics, aerospace, defense, autonomous vehicles, etc. are usually developed on System-on-Programmable Chip (SoPC) platforms, capable of supporting several multi-modal computation-intensive tasks on their FPGAs. Since such systems are mostly autonomous and mobile, they have rechargeable power sources and therefore, varying power budgets. They may also develop hardware faults due to radiation, thermal cycling, aging, etc. Systems must be able to sustain the performance requirements of their multi-task multi-modal workload in the presence of variations in available power or occurrence of hardware faults. This paper presents an approach for mitigating power budget variations and hardware faults (transient and permanent) by run-time structural adaptation of the SoPC. The proposed method is based on dynamically allocating, relocating and re-integrating task-specific processing circuits inside the partially reconfigurable FPGA to accommodate the available power budget, satisfy tasks’ performances and hardware resource constraints, and/or to restore task functionality affected by hardware faults. The proposed method has been experimentally implemented on the ARM Cortex-A9 processor of Xilinx Zynq XC7Z020 FPGA. Results have shown that structural adaptation can be done in units of milliseconds since the worst-case decision-making process does not exceed the reconfiguration time of a partial bit-stream.

A Method for Run-Time Prediction of On-Chip Thermal Conditions in Dynamically Reconfigurable SOPCs

“…is mechanism can dynamically select a suitable system configuration, i.e., a set of ASP circuit variants for the active tasks such that the desired FPGA die temperature is maintained. e major advantage of this mechanism will be the following: once integrated with the decision-making mechanism proposed in [1], it will result in a complete multiobjective mechanism, which will allow autonomous mobile systems to adapt and sustain their dynamic workloads in presence of changing power budgets, system modes, temperature, and available hardware resources, at run-time! We have, therefore, observed the current literature from the same perspective; whether the thermal management methods proposed in recent research works can support multiobjective run-time adaptation in systems.…”

“…ere is a lot of work done in this area of workload management. However, with these existing solutions, it may not always be possible to satisfy the changing constraints on temperature, performance, power consumptions, and/or available resources along with dynamic changes in the workload [1]. is type of adaptivity is therefore not considered in this work.…”

“…e mode of operation assumes (a) required set of active tasks and their performance range, (b) available power budget, (c) die temperature limits, and (d) currently available hardware resources. For example, in a low power budget condition, ASP circuit variants of tasks which can operate at a lower frequency and occupy more resources as compared to the system configuration in high power budget condition can be configured [1].…”

“…erefore, the key aspect for making such systems capable of RTSA is a run-time decision-making mechanism that can find in runtime, the most efficient combination of ASP circuit variants of all active tasks that can fit in the available hardware resources, can provide acceptable power consumption, maintain the die temperature in the required range, and simultaneously satisfy the performance constraints of the tasks. e concept and initial experimental verification of the aforementioned multiobjective decision-making mechanism was done on the Xilinx Zynq 7000 family of FPGA devices [1,4]. e mechanism presented in [1,4] enables run-time adaptation to 2…”

“…e concept and initial experimental verification of the aforementioned multiobjective decision-making mechanism was done on the Xilinx Zynq 7000 family of FPGA devices [1,4]. e mechanism presented in [1,4] enables run-time adaptation to 2…”