High-speed computation of unary functions

This paper presents an architecture for fast evaluation of unary functions such as reciprocal, square root and reciprocal square root. The theory behind the architecture has been presented in [1]. The paper shows the results of extensive simulation that have allowed us to implement the architecture with minimum chip count and maximum accuracy. The accuracy is about 8 % error rate in the LSB of the chosen representation (IEEE 32-bit floating point format). This architecture allows the computation of unary functions in less than 200 nsec.