A Parameterized VLSI Architecture for Real-Time Inversion of Square Matrices and Moore-Penrose Pseudoinversion of Rectangular Matrices
P S Vishnu, Subrahmanyam Mula · 2024
We present a universal VLSI architecture for computing the standard inverse of square matrices and the pseu-doinverse of rectangular matrices via QR decomposition (QRD) followed by back substitution. Using CORDIC and Newton-Raphson (NR) algorithms, our design avoids complex square-root and division operations. FPGA implementation results demonstrate that the architecture suits real-time applications and achieves near-floating-point accuracy while meeting resource and timing constraints across varying matrix dimensions.