Networks-on-Chip Cortex Inspired Communication To Reduce Energy Consumption

Erwan Moréac (1) , Johann Laurent (1) , Pierre Bomel (1) , André Rossi (2) , Emmanuel Boutillon (3) , Andrea Mineo (4) , Maurizio Palesi (4)

1 Lab-STICC_UBS_CACS_MOCS
2 LERIA - Laboratoire d'Etudes et de Recherche en Informatique d'Angers
3 UBS - Université de Bretagne Sud
4 Unict - Università degli studi di Catania = University of Catania

Ultra-deep sub-micron technology is shifting the design paradigm from area optimization to power optimization. In the context of Network-on-Chip (NoC) based design, energy consumption due to data transfer among network nodes is no longer negligible. Starting from the observation that, among the two brain hemispheres around 1 out of 10^6 synapses are active at the same time, in this paper we propose to mimic such behavior to trade-off transmission throughput with energy cost per transmitted bit in NoC based systems. In order not to alleviate the system performance, low level control mechanisms are proposed to use the so called “Cortex Inspired Communication” (CIC). We propose a set of strategies aimed at improving the energy efficiency of the NoC and we assess them in terms of energy saving, silicon area overhead and impact on performance metrics, namely, communication delay and throughput. The experimental analysis, carried out on a cycle-accurate and bit-accurate NoC simulator and under traffic scenarios generated by real applications show up to 30% of energy saving with almost no impact on the processing throughput.

Domaines: Electronique; Architectures Matérielles [cs.AR]; Systèmes embarqués

Erwan Moréac, Johann Laurent, Pierre Bomel, André Rossi, Emmanuel Boutillon, et al.. Networks-on-Chip Cortex Inspired Communication To Reduce Energy Consumption. Design Automation Conference (DAC 2017), Jun 2017, Austin, TX, United States. , Design Automation Conference, DAC 2017. ⟨hal-01576029⟩