Date of Award
Master of Electrical Engineering (MEE)
Yong Jin Kim
The NEUROSim framework consists of a compiler, assembler, and cycle-accurate processor simulator to facilitate computer architecture research. This framework provides a core instruction set common to many applications and a simulated datapath capable of executing these instructions. However, the core contribution of NEUROSim is its exible and extensible design allowing for the addition of instructions and architecture changes which target aspecic application. The NEUROSim framework is presented through the analysis of many system design decisions including execution forwarding, control change detection, FPU configuration, loop unrolling, recursive functions, self modifying code, branch predictors, and cache architectures. To demonstrate its exible nature, the NEUROSim framework is applied to specific domains including a modulo instruction intended for use in encryption applications, a multiply accumulate instruction analyzed in the context of digital signal processing,
Taylor series expansion and lookup table instructions applied to mathematical expression approximation, and an atomic compare and swap instruction used for sorting.
McNeil, David Eric, "NEUROSim: Naturally Extensible, Unique RISC Operation Simulator" (2016). Graduate Theses - Electrical and Computer Engineering. 8.