Systolic array

Systolic array is the name for a pipe network of DPU ( Data Path Units ), usually in a matrix arrangement, are clocked through which data flows - as opposed to the Instruction Systolic Array, are passed through the commands. The term " systolic " is intended to compare the data flows through the array, with the blood circuit, wherein the clock is almost the pumping heart. A normal systolic array does not require any commands, since the operations in the DPU by handshake by the arrival of the data at each DPU input "transport - triggered" are triggered automatically. (There are also mixed forms with instruction - systolic shares. ) Systolic arrays were especially in the 1980s, a popular area of ​​research, especially by mathematicians. The former synthesis methods were therefore exclusively algebraic nature (or a linear projection), which is why the application of systolic arrays has been limited to applications with regular data dependencies, because these synthesis methods gave only uniform linear pipes. This limitation of the practical applicability was around 1995 by Rainer Kress canceled by the generalization of the systolic array to the " super- systolic array ", by replacing the algebraic methods by simulated annealing. This was non-uniform pipe networks of any shape possible, such as zigzag, spiral, with branches and infinity of any other forms. So here was coarse-grained Reconfigurable Computing sense. From the super- systolic array Xputer the paradigm was derived.