Date of Award
2016
Document Type
Open Access Dissertation
Department
Electrical Engineering
Sub-Department
College of Engineering and Computing
First Advisor
David Matolak
Abstract
The advent of integrated circuit (chip) multiprocessors (CMPs) combined with the continuous reduction in device physical size (technology scaling) to the sub-nanometer regime will result in an exponential increase in the number of processing cores that can be integrated within a single chip. Today’s CMPs already support tens to low hundreds of cores and both industry and academic roadmaps project that future chips will have thousands of cores. Therefore, while there are open questions on how to harness the computing power offered by CMPs, the design of power-efficient and compact on-chip interconnection networks that connects cores, caches and memory controllers has become imperative for sustaining the performance of CMPs.
As the limited scalability of bus-based networks degrades performance by reducing data rates and increasing latency, the Network-on-Chip (NoC) design paradigm has gained momentum, where a network of routers and links connects all the cores. However, power consumption of NoCs is a significant challenge that should be addressed to capitalize on the scaling advantages of multicores.
Also, improvements in metal wire characteristics will no longer satisfy the power and performance requirements of on-chip communication. One approach to continue the performance improvements is to integrate new emerging technologies into the electronic design flow such as wireless/RF technologies, since they provide unique advantages that make them desirable in a NoC environment. First, wireless technologies are ubiquitous and offer a wide range of options in communication, and there exists a vast body of knowledge for the design and implementation of wireless chipsets using RF-CMOS technology. Second, wireless communication, unlike wired transmission, can be omnidirectional, which can facilitate one-hop unicast, multicast, and broadcast communication that can result in a reduction in power consumption while allowing for faster communication. Third, wireless communication can increase the communication data rate by the combination of Frequency Division Multiplexing (FDM) and Time Division Multiplexing (TDM) (and in the future, potentially spatial division multiplexing (SDM)). Therefore, Wireless NoC (WiNoC) interconnects have recently emerged as a viable solution to mitigate power concerns in the short to medium term while still providing competitive performance metrics, i.e., low power consumption, tens of Gbps data rates, and minimal circuit area (or volume) within the chip. Worth noting is that wireless links are not envisioned as replacing all wired links, but rather as augmenting the wired interconnection network.
In this dissertation, we employ simulations in HFSS from Ansys® to present accurate wireless channel models for a realistic WiNoC environment. We investigate the performance of these models with different types of narrowband and wideband antennas. This entails estimation of the scattering parameters for the channels between multiple antenna elements in the WiNoC, from which we derive channel transfer functions and channel impulse responses. Using these results, we can estimate the throughput of the various WiNoC links, and this allows us to design effective multiple access (MA) schemes via FDM and TDM. For these MA schemes, we provide estimates of maximal throughput. To further the feasibility study, we investigate the performance of a simple binary transmission scheme--On-Off Keying (OOK)--through the resulting dispersive channels, which can facilitate one-hop unicast, multicast, and broadcast communication that can result in a reduction in power consumption while allowing for faster communication.
Our investigation of the performance of On-Off Keying modulation (OOK) also includes an analytical expression for bit error ratio (BER) that can be evaluated numerically. This enables us to provide the equalization requirements needed to achieve our target BERs. Finally, we provide recommendations for WiNoC design and future tasks related to this research.
Rights
© 2016, William Rayess
Recommended Citation
Rayess, W.(2016). Wireless Channel Modeling For Networks On Chips. (Doctoral dissertation). Retrieved from https://scholarcommons.sc.edu/etd/3920