TY - GEN
T1 - Mixed wire and surface-wave communication fabrics for decentralized on-chip multicasting
AU - Karkar, Ammar
AU - Tong, Kin Fai
AU - Mak, Terrence
AU - Yakovlev, Alex
N1 - Publisher Copyright:
© 2015 EDAA.
PY - 2015/4/22
Y1 - 2015/4/22
N2 - Network-on-chip (NoC) has emerged to tackle different on-chip challenges and has satisfied different demands in terms of high performance, economical and reliable interconnect implementation. However, a merely metal-based interconnect reaches performance bound with the relentless technology scaling. Especially, it displayed a bottleneck to meet the communication bandwidth demand for multicasting. This paper proposes a novel hybrid architecture, which improves the on-chip communication bandwidth significantly using mixed wires and surface wave interconnects (SWI) fabrics. In particular, the bandwidth of multicasting can be drastically improved. We introduce a decentralized arbitration method to fully utilize the slack-time scheduling with deadlock-free flow control. Evaluation results, based on a cycle-accurate and hardware-based simulation, demonstrate the effectiveness of the proposed architecture and methods. Compared to a wire-based NoC, the mixed fabric approach can achieve an improvement in power reduction and communication speed up to 63% and 12X, respectively. These results are achieved with almost negligible hardware overheads. This new paradigm efficiently addresses the emerged challenges for on-chip communications.
AB - Network-on-chip (NoC) has emerged to tackle different on-chip challenges and has satisfied different demands in terms of high performance, economical and reliable interconnect implementation. However, a merely metal-based interconnect reaches performance bound with the relentless technology scaling. Especially, it displayed a bottleneck to meet the communication bandwidth demand for multicasting. This paper proposes a novel hybrid architecture, which improves the on-chip communication bandwidth significantly using mixed wires and surface wave interconnects (SWI) fabrics. In particular, the bandwidth of multicasting can be drastically improved. We introduce a decentralized arbitration method to fully utilize the slack-time scheduling with deadlock-free flow control. Evaluation results, based on a cycle-accurate and hardware-based simulation, demonstrate the effectiveness of the proposed architecture and methods. Compared to a wire-based NoC, the mixed fabric approach can achieve an improvement in power reduction and communication speed up to 63% and 12X, respectively. These results are achieved with almost negligible hardware overheads. This new paradigm efficiently addresses the emerged challenges for on-chip communications.
UR - https://www.scopus.com/pages/publications/84945907473
U2 - 10.7873/date.2015.0276
DO - 10.7873/date.2015.0276
M3 - Conference contribution
AN - SCOPUS:84945907473
T3 - Proceedings -Design, Automation and Test in Europe, DATE
SP - 794
EP - 799
BT - Proceedings of the 2015 Design, Automation and Test in Europe Conference and Exhibition, DATE 2015
T2 - 2015 Design, Automation and Test in Europe Conference and Exhibition, DATE 2015
Y2 - 9 March 2015 through 13 March 2015
ER -