Content area
The rise in development of open-source hardware and the demand for energy-efficient, high-performance computing have led to increasingly complex processor and accelerator designs. While open-source tools streamline the design process, verification remains challenging and costly, as it requires extensive testing to avoid costly post-silicon faults. Conventional formal verification methods are limited by scalability, slow software simulators are slow, and FPGA prototyping offer limited design visibility. To address these challenges, Condominium is introduced to couple the speed of FPGA emulation with the transparency of RTL simulation. By enabling non-intrusive emulation and cycle-accurate data collection, Condominium provides real-time instruction-level bug localization and fine-grained performance profiling, enabling agile system evaluation methods for hardware engineers. Additionally, Condominium facilitates the precise emulation of peripherals and system calls, bypassing the need for extensive RTL development. Furthermore, this dissertation introduces a novel high-fidelity hardware coverage metric for elevating the efficacy of modern coverage-guided hardware fuzzers. By providing an accurate representation of design exploration By incorporating the relative latency information of the cascaded coverpoints into the metric, this high-fidelity metric aims to provide an accurate representation of design exploration to coverage-guided fuzzers. Through specially designed coverage engines integrated into Condominium, this work enables FPGA acceleration of high-fidelity coverage, addressing the scalability and acceleration issues of previously proposed coverage-guided fuzzers. By providing an environment for accelerated and cycle-accurate hardware emulation and the fine-grained verification methodologies it enables, this dissertation aims to provide a framework that addresses common challenges in hardware verification and analysis and significantly reduces engineering time spent on design functional verification and performance optimization.