Network-on-Chips with simple topologies are widely used due to their scalability and high bandwidth. The transmission latency increases greatly with the number of on-chip nodes. A NoC, called single-cycle multi-hop asynchronous repeated traversal (SMART), is proposed to solve the problem by bypassing intermediate routers. However, the bypass setup request of SMART requires additional pipeline stages and wires. In this paper, we present a NoC with rapid bypass channels that integrates the bypass information into each flit. In the proposed NoC, all the bypass requests are delivered along with flits at the same time reducing the transmission latency. Besides, the bypass request is unicasted in our design instead of broadcasting in SMART leading to a great reduction in wire overhead. We evaluate the NoC in four synthetic traffic patterns. The result shows that the latency of our proposed NoC is 63.54% less than the 1-cycle NoC. Compared to SMART, more than 80% wire overhead and 27% latency are reduced.

Embedded processors are widely used in various systems working on different tasks with different workloads. A more complex micro-architecture leads to better peak performance and worse power consumption. Shutting down the units designed for performance enhancement could improve energy efficiency in low-workload scenarios. In this paper, we evaluated the energy distribution in various embedded processors. According to the analysis, pipeline registers and the dynamic branch predictor, which are employed for better peak performance, have great impacts on energy efficiency. Thus, we proposed an ultra-low-power processor with variable micro-architecture. The processor is based on a 4-stage pipeline core with a Gshare branch predictor, and all units work in high-performance mode. In normal mode, the Gshare predictor is shut down and Always-Not-Taken prediction is used. In low-power mode, some of the pipeline registers are bypassed to avoid unnecessary energy dissipation and improve executing efficiency. A mode register (MR) is designed to indicate current working mode. Switching between different modes is controlled by the software. The proposed core is implemented in 40 nm technology and simulated with the traces of 17 benchmarks in Embench. The average amounts of power consumed by the respective modes are 41.7 W, 59.7 W and 71.1 W. The results show that normal mode (N-mode) and low-power mode (L-mode) consume 16.08% and 41.37% less power than high-performance mode (H-mode) on average. In best case scenarios, they could save 25.36% and 49.30% more power than H-mode. Considering the execution efficiency evaluated by instructions per cycle (IPC), the proposed processor consumes 7.78% or 51.57% less energy for each instruction than the baseline core. The area of the proposed processor is only 7.19% larger than the baseline core, and only 3.08% more power is consumed in H-mode.

No abstract

No abstract