A study of a wire–wireless hybrid NoC architecture with an energy-proportional multicast scheme for energy efficiency☆
Introduction
Cache coherence among processing elements (PEs) in a multiprocessor system-on-chip (MPSoC) or chip multiprocessor (CMP) architecture is a fundamental problem that dominates the multiprocessing performance as well as energy efficiency. During the last decade, network-on-chip (NoC) technology had emerged as communication backbones to enable a high degree of integration in CMP [1]. Different from bus-based systems, PEs communicate with each other in NoC by sending data packets across an on-chip network instead of driving voltage signals across a dedicated bus [2]. Despite the traditional planar metal NoC architecture having better performance than the bus-based one, it is still limited by high latency and large power consumption due to the parasitic RC on the metal lines [3]. When viable millimeter-wave (mm-Wave) antennae and a transceiver technology integrated on chip are introduced [4], the wireless interconnects network-on-chip (WiNoC) architecture is investigated to enhance the multiprocessing performance by reducing the multi-hop links [5], [6], [7].
Traditionally an efficient way of packet transmission from one PE to multiple PEs (e.g., cache coherence among PEs on CMP) is to perform multicast transmission [8], [9], [10], [11], [12], [13], [14], [15]. However, these studies of multicast transmission are all based on wired interconnection and need a multicast routing protocol to transmit multicast packets. These packets are usually transmitted to their destinations involving multiple hops. This result leads to a longer transmission delay and costs much energy [16], [17], [18].
The naturally broadcast property of WiNoC is promising to enhance the multicast transmission performance in CMP as compared with the 3-D topological NoC [19], the optical NoC [20], and the radio-frequency interconnects (RF-I) NoC [21] since the transmission only involves one time transmission (i.e., one-hop transmission). Besides, these NoC technologies (3-D NoC, optical NoC, and RF-I NoC) are limited by today’s semiconductor manufacture technologies [6]; thus, the complementary metal–oxide–semiconductor (CMOS) compatible WiNoC has its unique opportunities for implementation. Hence, only metal-wire and wireless interconnects can be easily implemented and massively produced by taking today’s CMOS technologies.
The implementation of wired NoC and WiNoC has different merits. First, the wired NoC requires a smaller area overhead (i.e., lower energy consumption and heat) than the WiNoC because each wireless router (WR) of WiNoC requires a large area to implement a wireless interface (WI) consisting of a transceiver, a receiver, and antennas. Secondly, WiNoC possesses a better opportunity of achieving lower latency and higher throughput than wired NoC in transmission if the multi-hop transmission can be achieved in one-hop.
By considering the optimization of transmission latency as well as energy efficiency, it is a good way to combine these two kinds of NoC architectures to form a wire–wireless hybrid NoC (WHNoC) in CMP [22]. Deb et al. [7] proposed a WHNoC architecture where all PEs are divided into multiple subnets. Each subnet uses a star-ring topology to connect inside PEs. The interconnection among subnets uses hybrid wired and wireless links which are determined by using the principle of small-world graphs. The adopted protocol of the WiNoC is a token-passing protocol. Indeed, the token-passing protocol has several drawbacks. First, the token circulates among all WRs (i.e., round-robin scheduling) no matter the WRs need to transmit or not. Secondly, token circulation will lead to a long access delay if the number of WRs is large. Thirdly, there is a risk of token losing and will encounter a token re-election overhead. Fourthly, token-passing protocol is not easy to implement priority access.
The p-persistent carrier sense multiple access (CSMA) had been proven that the performance can be well managed in a finite population [23], [24], [25]. Seo et al. [23] had shown that the throughput upper-bound (or mean access delay lower-bound) can be achieved by an optimal persistent probability p from the number of backlogged terminals (i.e., WRs in our paper) in a finite population CSMA system (i.e., a fixed number of WRs on a chip). Thus, the aforementioned drawbacks of the token-passing protocol and the achievements of [23], [24] motivate us to adopt the p-persistent CSMA in a fully connected WiNoC architecture to achieve NoC energy efficiency since the number of PEs on CMP is fixed.
Although CMP technologies nowadays have better performance than their predecessors, they still consume a lot of energy [26]. In this paper, we focus on the energy efficiency of NoC and study a WHNoC architecture in which all subordinate PEs of a subnet connect to a WI-equipped hub (i.e., WRs) by wires to form a star topology and WRs connect with each other in a fully connected topology, called sWHNoC, as shown in Fig. 1(a). To greatly reduce the energy consumption an energy-propositional multicast scheme (EMS) is proposed to support this kind of sWHNoC. The EMS uses a power-gating (PG) technique to temporarily power off WRs when they are not involved in the multicast transmission. To the best knowledge of the authors, we are the first to propose the slotted p-persistent CSMA for sWHNoC with EMS in the CMP.
The rest of the paper is organized as follows. Section 2 introduces the hybrid architecture and the mechanism of EMS with PG technique. The energy consumption of EMS is analyzed in Section 3. A performance comparison between CSMA protocol and token-passing protocol is presented in Section 4. Section 5 evaluates the energy efficiency of EMS with simulation and analysis results. Finally, some conclusions are given in Section 6.
Section snippets
System model
The sWHNoC is a hierarchical architecture that contains two levels: the bottom-level (i.e., the intra-subnet) and top-level (i.e., the inter-subnet). The sWHNoC is partitioned into S subnets where each subnet occupies one WR and uses the star topology to connect N PEs belonging to the subnet. The subnet constructs the bottom-level. The top-level is constructed by the WRs that are inserted in subnets as the central hubs for wireless transmission to connect all the subnets.
Energy consumption analysis
In this section, we mainly focus on the energy consumption caused by multicast transmission in the sWHNoC with EMS. Multicast transmission comprises intra-subnet communication and inter-subnet communication. In intra-subnet, data collision can be avoid by using arbiter and virtual channels. In inter-subnet, the optimal slotted p-persistent CSMA is adopted [23], [25].
Fig. 3 illustrates the wireless channel status of the p-persistent CSMA whose time axis is slotted and one slot duration is a
Performance metrics of CSMA and token-passing protocols
In this section, we compare the performance of CSMA and token-passing protocols. To study the performance metrics, we use simulation to compare the CSMA protocol with the token-passing protocol in terms of area cost, energy consumption, and MAC access delay. Both CSMA and token-passing protocols use the time division multiplexing (TDM) scheme to access the wireless medium. The time is divided into slots. A token is circulated among WRs in sequence for obtaining the medium access right in the
Experimental results
To illustrate the energy efficiency of EMS, a case of NoC, , is examined. Suppose it has S subnets and every subnet contains N PEs (i.e., the subnet size is N). The WI component considered in this study is a mm-Wave transceiver with body-enabled techniques [4], and is well studied with a metal zigzag antenna on an mm-Wave NoC in [7] due to its good property of providing a wide bandwidth as well as low power consumption on-chip. The power consumption per bit of transmitter, receiver, and
Conclusion
In this paper, a sWHNoC architecture with EMS for energy efficiency was studied. We proposed the p-persistent CSMA for the fully connected WiNoC. To the best knowledge of the authors, the p-persistent CSMA for fully connected WiNoC is proposed and studied in literature first time. Our study indicated the following results:
- •
The hybrid star-topology subnet and fully connected WiNoC architecture is suitable for energy efficiency in NoC.
- •
The optimal energy efficiency can be achieved by adjusting the
Peng Dai received the B.S. and M.A. degrees in the department of Microelectronics from Tianjin University, Tianjin, China, in June 2012 and January 2015 respectively. His research focuses on the design and implementation of VLSI, Mixed-Signal integrated circuit. Currently, he works at Spreadtrum Communications Ltd.
References (38)
- et al.
Multi-hop communications on wireless network-on-chip using optimized phased-array antennas
Comput Electr Eng
(2013) - et al.
On an efficient NoC multicasting scheme in support of multiple applications running on irregular sub-networks
J Microprocess Microsyst
(2011) - et al.
A hybrid packet-circuit switched router for optical network on chip
Comput Electr Eng
(2013) - et al.
A high level power model for Network-on-Chip (NoC) router
Comput Electr Eng
(2009) - et al.
Networks on chips: a new SoC paradigm
Computers
(2002) - Seongmoo H, Asanovic K. Replacing global wires with an on-chip network: a power analysis. In: Proc IEEE ISLPED’2005,...
- Yu X, Sah SP, Deb S, Pande PP, Belzer B, Deukhyoun H. A wideband body-enabled millimeter-wave transceiver for wireless...
- et al.
Scalable hybrid wireless network-on-chip architectures for multicore systems
IEEE Trans Comput
(2011) - et al.
Wireless NoC as interconnection backbone for multicore chips: promises and challenges
IEEE J Emer Sel Top Circ Syst
(2012) - et al.
Design of an energy efficient CMOS compatible NoC architecture with millimeter-wave wireless interconnects
IEEE Trans Comput
(2013)
Custom networks-on-chip architectures with multicast routing
IEEE Trans VLSI Syst
Cost-effective power-aware core testing in NoCs based on a new unicast-based multicast scheme
IEEE Trans Comput Aid Des Int Circ Syst
Adaptive and deadlock-free tree-based multicast routing for networks-on-chip
IEEE Trans VLSI
New theory for deadlock-free multicast routing in wormhole-switched virtual-channelless networks-on-chip
IEEE Trans Parallel Distrib Syst
Path-based partitioning methods for 3d networks-on-chip with minimal adaptive routing
IEEE Trans Comput
Path-based multicast routing for 2D and 3D mesh networks
Cited by (18)
A survey and taxonomy of congestion control mechanisms in wireless network on chip
2020, Journal of Systems ArchitectureCitation Excerpt :However, these techniques are flexible and scalable. However, these techniques are flexible and scalable [37,38]. Without loss of generality, to obtain higher performance and efficiency in WiNoCs, dynamic MAC strategies needed to be fully compatible with physical layer constraints, routing algorithm, flow control, traffic load characteristics, and network performance objectives in the wireless on-chip communication paradigm.
CPCA: An efficient wireless routing algorithm in WiNoC for cross path congestion awareness
2019, IntegrationCitation Excerpt :As network congestion in WiNoC can greatly degrade system performance, alleviating congestion becomes a key issue of the WiNoC [13]. Dai et al. [21] propose a communication scheme based on p-persistent Carrier Sense Multiple Access (CSMA). Each wireless interface (WI) performs as channel monitor.
A low-power wireless-assisted multiple network-on-chip
2018, Microprocessors and MicrosystemsCitation Excerpt :Localizing power adaptation is crucial for implementing power management technique in NoC. To reach this goal, power should be delivered to each network component according to its local traffic load [8,51]. Despite its intuitive simplicity, this condition is difficult to be satisfied since network traffic distribution is both temporal and spatial.
A novel hierarchical architecture for Wireless Network-on-Chip
2018, Journal of Parallel and Distributed ComputingAnalytical Model for Performance Evaluation of Token-Passing-Based WiNoCs
2023, IEEE Design and Test
Peng Dai received the B.S. and M.A. degrees in the department of Microelectronics from Tianjin University, Tianjin, China, in June 2012 and January 2015 respectively. His research focuses on the design and implementation of VLSI, Mixed-Signal integrated circuit. Currently, he works at Spreadtrum Communications Ltd.
Jenhui Chen received the B.S. and Ph.D. degrees in the department of Computer Science and Information Engineering (CSIE), Tamkang University, Taipei, Taiwan in January 2003. He is a professor in the department of CSIE, College of Engineering, Chang Gung University. His main research interests include design, analysis, and implementation of communication protocols, wireless networks, cloud computing, big data, augmented reality, SoC, and NoC.
Yiqiang Zhao is a professor at the School of Electronic Information Engineering, Tianjin University. His primary research interests are mixed-signal integrated circuit and system, VLSI imaging system, information security.
Yen-Han Lai received the B.S. degree in the department of mathematics, National Taitung University, Taitung, Taiwan, and M.S. degree in the department of CSIE, Chang Gung University, Taoyuan, Taiwan, in 2009 and 2013 respectively. He is currently a Ph.D. student in the department of CSIE, Chang Gung University. His main research focuses on wireless communications and network-on-chip.
- ☆
Reviews processed and recommended for publication to the Editor-in-Chief by Associate Editor Dr. M. Daneshtalab.
- 1
This work was supported in part by the Ministry of Science and Technology, Taiwan, R.O.C., under Contract MOST 103-2221-E-182-042.