Reliable Network-on-Chip Using a Low Cost Unequal Error Protection Code

Low-overhead error detection for Networks-on-Chip

“…2. We denote concatenation by commas, e.g., data,p [1] represents the data with one parity bit appeared at the end. If pack=data,p[1] then we denote data=pack p [1].…”

“…We denote concatenation by commas, e.g., data,p [1] represents the data with one parity bit appeared at the end. If pack=data,p[1] then we denote data=pack p [1]. Note that since the encoder determines the routing path, this approach is applicable for transmission units that carry the source and destination addresses.…”

“…Note that since the encoder determines the routing path, this approach is applicable for transmission units that carry the source and destination addresses. In case the NoC employs wormhole routing [1], typically only the header flit carries these addresses. In such cases, our scheme can be used either for the entire packet (with checking at the destination node) or only for the header flit, which is the most important flit.…”

“…Modern device scaling results in deep sub micron noises, which cause interconnect errors to be more dominant and harder to predict [1,2,3,4,5,6,10,11], and also gives rise to new error sources [2,3]. The need for efficient low-power design techniques, along with aggressive voltage scaling and higher integration make interconnects even more susceptible to errors [1,6,11].…”

“…The need for efficient low-power design techniques, along with aggressive voltage scaling and higher integration make interconnects even more susceptible to errors [1,6,11]. In this paper, we focus on efficient solutions for interconnect reliability in the context of Networks-on-Chip (NoCs).…”