Presenting a Systematic Method for LU Decomposition of a Matrix with Linear Systolic Arrays

Systolic arrays are a set of simple processors that they have placed beside each other based on specific pattern. Their properties such as regularity, local communications, ability of parallel and pipeline processing have introduced them as a useful structure for executing computational algorithms with high speed. In this paper, we will present a systematic method for mapping LU decomposition of a matrix onto linear systolic arrays. The proposed method will use a Space-Time transformation so that space index of algorithm will be mapped onto linear systolic array. By using such transformation, we can determine time and location of each computation of algorithm on the processor array, respectively. The resulting structure has been compared with General Parameter Method and Dependency Method from < Processing Time >, < Number of Processor Element> and <Number of Processor Element * Processing Time > parameters point of view with several order of LU decomposition algorithm. The results of evaluation shows that use of this method will improve the <Number of Processor Element * Processing Time> parameter. This parameter is very important in VLSI designs, therefore it will bring about a trade off among computational time, occupied area and power consumption.