A Fast System-Level Design Methodology for Heterogeneous Multi-Core Processors Using Emerging Technologies

A fast and efficient system-level design methodology is developed and validated to evaluate and optimize processors using emerging technologies at the early design stage. It includes an updated empirical cycle per instruction model, a hierarchical memory model, and several multi-level interconnection network models. Multiple device- and system-level design parameters are simultaneously optimized to maximize the chip throughput for a given device technology and an architecture family under certain power, thermal and die size budgets. In the single-core processor analysis, a high-performance (HP) 25 mm ² FinFET processor can provide 26% more throughput than its planar HP complementary metal-oxide-semiconductor (CMOS) counterpart, and a low-power (LP) tunnel field-effect transistor (TFET) processor can offer more than 2X improvement in throughput compared with its LP FinFET counterpart at the 16 nm technology node. For various technology nodes, an accurate power-law relation is observed between the throughput and the die size area for a relatively small processor. In the multi-core processor analyses, multiple device- and system-level design parameters are obtained for both HP and LP applications with symmetric and asymmetric configurations. For a heterogeneous CMOS-TFET multi-core processor, about 45% throughput improvement and 50% energy reduction are observed compared with a FinFET processor at a 5 W power budget.