An algorithm for extremal eigenvectors computation of Hermitian matrices and its FPGA implementation

We consider the problem of implementing an algorithm for the extraction of leading eigenvectors of a small Hermitian matrix on field-programmable gate array (FPGA). The evolution of FPGAs can now handle increasingly bandwidth problems or larger in size. Jacobi algorithms are usually implemented in FPGA for real matrix size not exceeding 20*20. The increase in size or complex number problem may lead to use other algorithms such as Lanczos, which are rarely implemented on FPGA. Recently, it has been pointed out that the Lanczos method can efficiently address the extreme eigenvalues computation problem on FPGA, for medium size real matrices. This paper presents an algorithm for the extraction of extremal eigenvalues and corresponding eigenvectors for small Hermitian matrix using a high-level approach for the architecture synthesis.