RUI: Systolic Arrays with Sublinear Time Complexity

PI先前已经开发出用于矩阵乘法的立方脉动阵列。关于这种阵列的下列两个问题现在正在寻找答案：它们是否可以与通用处理器一起使用，以创建一个“矩阵计算机”，就像在“向量计算机”中使用“向量处理器”一样？这样的计算机能有效地解决其他各种线性代数问题吗？立方脉动阵列允许矩阵在时间上乘以矩阵大小的次线性。寻求进一步的脉动装置的设计，其将在时间上解决矩阵大小的次线性的进一步的代数问题。这项研究旨在适应计算问题的已知并行解决方案，以便在更容易实现的并行计算机上高效执行。这项研究的意义在于它有望为并行计算的各个方面提供更多的洞察力。

The PI has previously developed a cubic systolic array for matrix multiplication. Answers are now being sought for the following pair of questions concerning such arrays: Can they be used in conjunction with a general purpose processor to create a "matrix computer" in the same way that a "vector processor" is used in a "vector computer"? Could such a computer be used to efficiently solve various other linear algebra problems? Cubic systolic arrays allow matrix multiplication in time sublinear in the matrix size. Designs are sought for further systolic devices that will solve further algebra problems in time sublinear in the matrix size. This research is directed toward adaptation of known parallel solutions of computational problems for efficient execution on more readily implementable parallel computers. The significance of this research is in the additional insight it is expected to provide into aspects of parallel computation.