Improved goldschmidt algorithm for fast and energy-efficient fixed-point divider
Guilherme Paim, Pedro Marques, Eduardo Costa, Sérgio José Melo de Almeida, Sérgio Bampi · 2017
This paper proposes an improvement in the original Goldschmidt algorithm to produce a fast and energy-efficient fixed-point divider. Most of the work that deals with Goldschmidt algorithm uses a look-up table to generate the near optimal divisor term, and the input values are limited to a narrow range of values between 1 and 2. Our architecture solution is based on the suitable choice of the optimum denominator value that adjusts the first iteration of the algorithm, what contributes for fast convergence, with only three iterations for the error analysis. Moreover, in the proposed algorithm, the input values are defined by the Q7.8 format, what guarantees a wide range of values between -127.99609375 and +127.99609375. The principal results show that the proposed improved Goldschmidt architecture features low power dissipation and it can achieve high processing rates when compared to the original structure.