Research Of The VOQ Router And Link Fault Tolerant Technology In Network-on-Chip

The continuous development of integrated circuit makes the system-on-chip encounter insurmountable bottleneck problems, for the reason of good extensibility, high efficiency and reliability of communication advantages, Network-on-Chip(NoC) is becoming a new kind of interconnection architecture and widely studied by the researchers. As a critical component in NoC, the router is mainly in charge of the packet store-and-forward, which affects the performance of the whole system. Thus, it is especially important that propose a fault-tolerant scheme for the router. As the transmission of packets in NoC must go through the link and the link consumes most of the chip area, the fault-tolerant to the link is particularly important. The main ideas are as follows:(1) This thesis describes the development of integrated circuits, the research background of the NoC. Also, it introduces the topology of NoC, and the common fault tolerance method of NoC.(2) Buffer accounts for a large area in a router and traditional fault-tolerant methods for the buffer may lead to the head of blocking (HoL). Hence, it is necessary to design a fault-tolerant architecture, which cannot only avoid the HoL but also achieve fault-tolerance. In this thesis, we present the design of virtual output queue (VOQ) fault tolerant router which is loop-shared by adjacent virtual channels. The proposed fault-tolerant design can tolerate fault buffers effectively, and ensure the buffer utilization at the most extent when congestion occurs in the network, and even guarantee that the connectivity of the entire network in the case of all the input virtual channels breaking down. On that basis of the traditional VOQ router, this thesis obtains the looped-share of the adjacent virtual channels by adding a number of multiplexers and de-multiplexers. At the same time, extra bypass buses can guarantee that the router can still route packets in the case of all the input virtual channels break down. Meanwhile, this thesis designs a FIFO controller based on a double-ended queue, in which two sets of read/write pointers read or write packets no matter which virtual channels they are stored in.(3) To solve the problem of link failures, this thesis will divide the granularity of link failures into several levels, and propose an adaptive fault-tolerant method for the link. When there is a coarse-grained failure occurs in link, the direction of another link is reconfigured to transmit data packets. When there is a fine-grained failure occurs in link, the thesis combines the network status to reconfigure the direction of another link or to use its remaining intact bit lines to complete the transmission of data packets in many times. Experimental results show that, compared to the reconfiguration scheme of bidirectional link and the fault tolerant scheme of unidirectional link multiple transmissions, the delay of our scheme can be reduced by 27.2%and 34.2%, and the throughput can be increased by 21.7% and 35.3% at most. Thus, this scheme has obvious advantages and ensures the high-reliability when different link faults occur.