Design and Accuracy Analysis of the Posit Number Representation–Decimal Conversion Algorithm

Abstract Posit is a number representation that has emerged in recent years and has higher accuracy and a dynamic number range than IEEE754 floating-point numbers. Since it has just emerged, it is not yet mature, and studies on the posit number representation are important for its widespread use in digital systems. In this study, a posit–decimal conversion algorithm is developed for the first time between the posit studies. In the developed algorithm, the precision and exponent length are parametric, and a design that can be flexibly adapted to various application requirements is presented. This flexibility allows the algorithm to be analyzed for accuracy and error at different bit lengths. The algorithm is evaluated on some activation functions, and its accuracy in computing these functions is analyzed in detail. The results obtained show that the algorithm provides remarkable accuracy even at low-bit lengths. An average accuracy of over 93% is achieved in 8-bit precision; this rate exceeded 99% in 16-bit. In 32-bit and higher bit lengths, the algorithm worked with 100% average accuracy.