Hardware-Accelerated Workload Characterization for Power Modeling and Fault Injection

During recent years the increasing introduction of system functionality into integrated devices resulted into several new problems for chip designers. First, high system-on-chip complexity combined with increased clock frequencies leads to power budget and thermal issues. Second, small semiconductor process structures are more sensitive to faults resulting from logic degradation and external radiation sources. Early testing and evaluation of hardware and software implementations have been enabled in the last years using hardware-accelerated emulation techniques. Such implementations rely on functional models of the target system, generated using specialized benchmark suites. These have been designed to accurately resemble typical application scenarios of the target implementation. Unfortunately, power and fault injection emulation accuracy is depending on a good coverage of the system´s logic, which is not guaranteed by typical benchmarks. Therefore, this paper proposes an exhaustive hardware accelerated methodology for the evaluation of such applications and generation of accurate emulation models. The behavior of standard benchmarks are investigated using an open available system-on-chip platform based case study.