Elsevier

Microprocessors and Microsystems

Volume 63, November 2018, Pages 190-198
Microprocessors and Microsystems

Wireless-assisted multiple network on chip using microring resonators

https://doi.org/10.1016/j.micpro.2018.09.006Get rights and content

Abstract

The multiple network-on-Chip (multi-NoC) architecture is an attractive solution to scale on-chip network bandwidth; however, its performance is influenced by its overall communication infrastructure. Incorporating wireless links attract the traffic and cause the power-gated components to stay in sleep state for a longer period of time. The presented research proposes a system generating discrete frequencies via a series of microring resonators to generate frequency channels with 200 MHz Full Width at Half Maximum (FWHM) and 2.5 GHz free spectral range (FSR) for wireless-assisted multi-NoC application. It also evaluates the proposed architecture tested with a 64-core processor with a mesh topology using synthetic traffic and the PARSEC benchmark. The results show that the average network power of the proposed architecture is 55.4% and 11.1% lower than traditional single-NoC and power-gating multi-NoC designs respectively. Furthermore, evaluation of network latency shows that the proposed architecture has an average reduction by 10.4% in comparison with power-gated multi-NoC design.

Introduction

Network-on-Chips (NoCs) are projected to be a scalable communication substrate for building multicore systems, which are expected to fulfil a large number of different threads concurrently to maximize system performance. The trend in processor development has been towards an ever increasing number of cores as processors with hundreds of cores (many-core) became possible [1]. However, by increasing the number of cores, NoC power consumption increases correspondingly. Thus, building energy proportional many-core processors has been an attractive solution for scaling network bandwidth [2].

In a single-NoC architecture, most of the routers need to be kept active to service a network load with only a few active flows, which significantly reduce the opportunities for power-gating. Because of frequent transitions between power states, which cause packets to wait for a router to wake up, power gating single-NoC has a significant performance penalty. Multi-NoC design [3] was introduced for it was more amenable for power gating than a single-NoC design. Das et al. characterized the power and performance of multi-NoC, and indicated that for high-bandwidth networks, multiple networks consume less dynamic power than single networks by assuming voltage scaling [4].

In spite of their upsides, traditional NoCs have got their own downsides such as high latency and high power consumption between far apart blocks. Experts have made great endeavors to solve the problem by using ultra-low-latency and low power express channels between long-distance nodes [5] and high radix networks [6]. Although these communication channels are notably more efficient than the conventional ones, they are still metal wires. An interesting alternative to the existing interconnected infrastructures is transmitting signals via wireless interconnections [7]. Intra-chip low latency communication can be achieved through wireless interconnects, where transmission of data is guided through on-chip transmission lines [8]. The wireless interconnects enable fast data transmission across longer distances on the chip, which leads to lower communication latency and power dissipation compared to traditional wireline NoCs [9].

Investigations in this area have set features of silicon integrated antennas operating in the millimeter (mm)-wave range of a few tens to one hundred gigahertz and it is now an accessible technology for intra- and inter-chip communication [10]. Moreover, excellent emission and absorption characteristics leading to antenna-like behavior in carbon nanotubes (CNTs) operating at optical frequencies have been observed [11]. These discoveries can open up new horizons for further investigations in wireless NoC (WiNoC) design with appropriate transceivers and on-chip antennas. On-chip wireless communication links not only reduce the latency and power consumption problems of traditional technologies but also remove complex interconnect routing and layout problems appearing in some of the alternative technologies [12]. Thus, such interconnects enable design of novel and efficient architectures which moderate the multi-hop communication of traditional NoCs to gain efficient performance. WiNoCs present unique opportunities and challenges, which should be explored to make them mainstream.

One possibility for the design of WiNoC is using UWB (ultra wide band) interconnections. The antenna that is used is a Meander type dipole antenna with data a transmission range of 1 mm [13]. Another possibility for the design of WiNoC is using mm-wave that uses a Zigzag antenna. In this method, the modulation scheme is Non-coherent on-off keying (OOK) with a data transmission range of 20 mm and the data rate is 16 Gbps as opposed to 1.16 for the previous method [10]. In the sub THz frequency range, no antenna is introduced to be used. In this category, the data rate is 320 Gbps which has the highest rate compared to other methods [14]. In the range of THz, the modulation scheme is the same as mm-wave range. In addition, the least amount of energy consumption has been in the range of THz where Multi-Walled Carbon Nanotube (MWCNT) antennas are used [11].

The main idea of this paper is to apply CNT antennas which operate at THz frequency, although a THz WiNoC shows better performance than mm-wave WiNoC and wired NoCs. It should be mentioned that it is not a CMOS compatible solution and the integration and reliability of CNT devices need further studies. The WiNoC paradigm is still in its initial stages. Of course, the replacement of existing interconnected infrastructure by THz WiNoC requires more investigation. Therefore, in this paper, we propose using the micro resonator device as an optical solution for wavelength generation. This paper proposes a system of microring resonators (MRRs) which generates an applicable frequency channel for wireless-assisted multi-NoC. An MRR consists of a single coupler and a microring resonator. Light of appropriate frequency is injected into the loop by the input waveguide. Over multiple round-trip, the intensity will build up due to constructive interference. Since only some frequencies resonate within the loop, it functions as a filter. Characterization of light inside an MRR system is investigated in [15]. Filtering certain wavelengths of light, the MRR has many interesting and effective applications. It shows an effective performance for generating mm-wave and micro wave and it can be an interesting tool to generate solitonic pulses needed in inter-satellite systems [16]. Amiri et al. have shown that MRR can be used to generate multiple wavelengths when a soliton pulse is propagated inside the system [17]. Recently, some MRR systems, which are able to generate solitonic pulse shapes and WDM channels, are proposed [18], [19]. Dense frequency channels can be produced when the soliton pulse is propagated within the nonlinear MRR system causing large bandwidth signals - which are applicable for wireless-assisted multi-NoC - to be achieved.

The researchers evaluate a 64-core processor with 8 × 8 mesh topology using synthetic traffic and 9 PARSEC applications. The main objective is to decrease the network power and latency through smart insertion of long-range wireless links.

Section snippets

Wireless-assisted multi-NoC

This section provides a brief background for wireless-assisted multi-NoC design followed by a detailed treatment of network power scaling and consumption issues of single, multiple, and wireless-assisted network designs, which are likely to motivate processor manufacturers to adopt a hybrid wireless multi-network design.

Selecting an appropriate topology is one of the most critical decisions in the design of an interconnected network. It bounds critical performance metrics such as the network's

Theoretical modeling of MRR

The proposed system consists of a series of ring resonators R1, R2, R3 and R4, as shown in Fig. 3. The input signal is inserted into the system via the input port. Here, a mathematical equation of the propagating input pulse inside the nonlinear ring system has been solved in order to show the nonlinear behavior of the output signals. When an optical Gaussian pulse is input into the nonlinear MRR, the large frequency channels of the output signals can be produced where the nonlinear behavior of

Results and discussion

In this section, we evaluate the proposed architecture based on simulation results. First, we present simulation results on the proposed system of discrete frequency generation. Then we present power and delay analysis of the proposed wireless connections. Finally, we evaluate the proposed wireless-assisted multi-NoC architecture, in terms of power efficiency and network latency, by showing various simulation results.

Conclusion

In this paper a system of MRR is presented to generate near 500 GHz frequency channels for wireless-assisted multi-NoC. Then, the researchers demonstrate how long-range wireless links improve both the power and latency profile of a multi-NoC chip. Utilizing long-range wireless links, we are able to improve energy proportionality by reducing power-gating overheads. The wireless links can attract a significant percentage of the overall traffic and accordingly decrease traffic density in wirelines

Ali Shahidinejad received the B.S. degree in computer hardware engineering from Islamic Azad University of Kashan, Iran in 2008, the M.S. degree in computer architecture from Islamic Azad University of Arak, Iran, in 2010 and Ph.D. degree in Computer Networks at the Universiti Teknologi Malaysia/RWTH Aachen University, Malaysia/Germany, in 2015. He joined the Department of Computer Engineering, Islamic Azad University of Qom, as an Assistant Professor. He is currently the head of department of

References (33)

  • A. Ganguly et al.

    Scalable hybrid wireless network-on-chip architectures for multicore systems

    IEEE Trans. Comput.

    (2011)
  • J.-J. Lin et al.

    Communication using antennas fabricated in silicon integrated circuits

    IEEE J. Solid-State Circuits

    (2007)
  • K. Kempa et al.

    Carbon nanotubes as optical antennae

    Adv. Mater.

    (2007)
  • S. Deb et al.

    Enhancing performance of network-on-chip architectures with millimeter-wave wireless interconnects

  • D. Zhao et al.

    SD-MAC: design and synthesis of a hardware-efficient collision-free QoS-aware MAC protocol for wireless network-on-chip

    IEEE Trans. Comput.

    (2008)
  • S.-B. Lee et al.

    A scalable micro wireless interconnect structure for CMPs

  • Ali Shahidinejad received the B.S. degree in computer hardware engineering from Islamic Azad University of Kashan, Iran in 2008, the M.S. degree in computer architecture from Islamic Azad University of Arak, Iran, in 2010 and Ph.D. degree in Computer Networks at the Universiti Teknologi Malaysia/RWTH Aachen University, Malaysia/Germany, in 2015. He joined the Department of Computer Engineering, Islamic Azad University of Qom, as an Assistant Professor. He is currently the head of department of higher educations in computer engineering at Islamic Azad University of Qom.

    His research interests include Quantum-dot Cellular Automata, Optical Wireless Communications, Micro Ring resonators, Network on Chip, Cloud Computing, Internet of Things and Network Security.

    Saeed Fathi received the B.S. degree in computer software engineering from Islamic Azad University of Qom, Iran, in 2018. He also has over 5 years of experience in network administration. His favorite subjects are IoT, Blockchain, Computer Networks and Network Security.

    View full text