In many of today’s system-on-chip (SoC) designs, the cores are partitioned into multiple voltage and frequency islands (VFIs), and the global interconnect is implemented using a packetswitched network on chip (NoC). In such VFI-based designs, the benefits of 3-D integration in reducing the NoC power or delay are unclear, as a significant fraction of power is spent in link-level synchronization, and stacked designs may impose many synchronization boundaries. In this brief, we show the quantitative benefits of the 3-D technology on NoC power and delay values for such application-specific designs. We show a design flow for building application-specific NoCs for both 2-D and 3-D SoCs with multiple VFIs. We present a detailed case study of NoCs designed using the flow for a mobile platform. Our results show that power savings strongly depend on the number of VFIs used (up to 32% reduction). This motivates the need for an early architectural space exploration, as allowed by our flow. Our experiments also show that the reduction in delay is only marginal when moving from 2-D to 3-D systems (up to 11%), if both are designed efficiently.
Comparative Analysis of NoCs for Two-Dimensional Versus Three-Dimensional SoCs Supporting Multiple Voltage and Frequency Islands
BENINI, LUCA;
2010
Abstract
In many of today’s system-on-chip (SoC) designs, the cores are partitioned into multiple voltage and frequency islands (VFIs), and the global interconnect is implemented using a packetswitched network on chip (NoC). In such VFI-based designs, the benefits of 3-D integration in reducing the NoC power or delay are unclear, as a significant fraction of power is spent in link-level synchronization, and stacked designs may impose many synchronization boundaries. In this brief, we show the quantitative benefits of the 3-D technology on NoC power and delay values for such application-specific designs. We show a design flow for building application-specific NoCs for both 2-D and 3-D SoCs with multiple VFIs. We present a detailed case study of NoCs designed using the flow for a mobile platform. Our results show that power savings strongly depend on the number of VFIs used (up to 32% reduction). This motivates the need for an early architectural space exploration, as allowed by our flow. Our experiments also show that the reduction in delay is only marginal when moving from 2-D to 3-D systems (up to 11%), if both are designed efficiently.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
