Saturday 15 August 2015

Fault Tolerance in NoC (Network-on-Chip)

   Among the favorite topics of research in the field of information technology is parallel processing which involves computer networks and network interfaces. In 2000, the field of NoC or Networks-on-Chip was designed because of the advent of the multi-core computer processing unit (CPU) and due to the predicted computing growth towards extremely incorporated design with interconnected networks. Several surveys have been conducted about NoC and have been published. One of the leading topics that involve NoC is the fault-tolerant design which is well-known in terms of its significance and interest of the experts in technology and computer system.
   In order to cope with the fast growing trend of the computation-intensive application and the need for low power yet high performance system, several computing resources in single chips have been invented[1]. This becomes possible because of the presence of the current VLSI technology which can support an extensive integration of transistors. The presence of interconnection among several computing resources such as CPU, DSP, specific IP’s and other resources in System –on-Chip, the interconnection process have become more challenging[2]. The continuous evolution of new application has become the reason of the invention of some new processor, memory and accelerator cores. Moreover, the fast development and improvement of computation and the complexity of communication have caused the invention of the scalability-centered paradigms. These are the reasons why network-on-Chip (NoC) is designed in order to be an alternative in providing high-performing communication system in the field of Systems-on-Chip (SoCs) with several cores incorporated in one silicon die[3].
   The Netwok-on-Chip is an effective architecture in the communication system under the System-on- Chip. It permits the incorporation of hefty quantity of computational blocks in a single circuit. NoC allows the reutilization of blocks which results to the provision of high degree parallelism and performance of the system in higher level. Accordingly, the rising demand for NoC is a result of the requirement to achieve higher bandwidth if the field of communication. As stipulated in Moorse law, the integration of the transistor can be doubled within the approximate eighteen months, however, challenges and the competition of the physical connection in Integrated Circuit (ICs) has caused severity in the delay of connections among the global wires[4]. The traditional method of connection namely, end-to-end, bus-based and other are not good solution to the problem regarding the wire delay in integrating IP cores with System-on-Chips because they do not scale well if there are more mechanism that are supplied into the system[5].  Moreover, there is the ad hoc solution that is of varying buses, crossbar and wiring that can attain mainly advantageous performance especially for specific application. However, the system cannot be recycled and it requires great costs for its development and verification.
  This is where NoC comes in because it has the potential of providing short circuits and can thereby reduce the delays in wire. The utilization of NoC replaces the traditional methods of interconnection and it has several advantages in terms of its performance and modularity. Furthermore, it was observed that the utilization of NoC among electrical properties have  optimized, while it has reduced the power consumption ten folds and increased the speed three times as compared with the system that utilizes the bus-based connection.



1. Chen, S.J., Lan, Y.C., Tsai, W.C, & Hu, Y. “Fault Tolerance in BiNoC”. Springer. 2012. 157-171.
2. Ibid., 158.
3. Radetzki, M., Feng, C., Zhao, X & Jantsch, A. “Methods for Fault Tolerance in Networks-on-  Chip”. ACM Computing Surveys. 2013. 46 (1). 1-38.
4. Ibid.
5. Ibid. 

No comments:

Post a Comment