MFLP: Most Frequent Least Power Encoding

TL;DR

This paper introduces MFLP, a novel low power encoding technique for NoC that reduces switching activity and power consumption by assigning less frequent symbols to high probability data without relying on spatial redundancy.

Contribution

MFLP is a new encoding method that decreases power in NoC by using a tree-based structure to assign codewords with fewer ones to more frequent data, improving power efficiency and data compression.

Findings

Reduces switching activity by up to 45%

Decreases link power consumption by up to 46%

Lowers total power dissipation by up to 34.9%

Abstract

This paper has been withdrawn by the authors. In this paper, we propose a new low power coding technique by decreasing the number of switching activities on the buses which use transition signaling to transmit data. This approach dedicates the symbols with less ones to high probability data. MFLP unlike the most low power encoding does not rely on spatial redundancy. Due to this superiority, MFLP is unique in power decreasing in the Network on Chip (NoC). Not only does this algorithm reduce the power consumption, but also it can compress the data. It offers a tradeoff to designers to choose between compression and power; that is, the more power consumption decrease we need, the less compression we earn. This coding uses tree based infrastructure in order to decrease the number of ones to reduce the switching activities, and the power consumption consequently. The proposed algorithm constructs the tree with this contribution that code words with less ones are allocated to more frequent data. The experimental results for the outside and inside of the NoC indicate that the proposed coding algorithm reduces the switching activity up to 30 and 45%, the link power consumption up to 35 and 46% and the total power dissipation up to 34.9 and 16% for the outside and inside of the NoC, respectively.