A Review of Posit Arithmetic for Energy-Efficient Computation: Methodologies, Applications, and Challenges

For many decades, IEEE floating-point formats are used as the golden numeric formats for many applications including signal processing, linear algebra evaluation, and other scientific computations. The IEEE formats can provide large dynamic range and high representation precision for applications. However, the formats themselves have some intrinsic problems, such as rounding errors and complex exception handling, which make them difficult to be fully verified. In addition, in recent years, with the emerging of machine learning and edge computing, energy-efficient computations become vital for digital applications. Under this circumstance, a more flexible numeric format which fits better to the data distribution of the applications is desired. Posit numeric format is proposed as an alternative to IEEE floating-point format. Posit format has a tapered number distribution and a much larger dynamic range than IEEE floating-point format which make it advantageous in many applications. In this chapter, a comprehensive review of posit arithmetic for energy-efficient computation is presented. We will start with the introduction of the posit numeric format and then go through all technology layers in a digital system including the applications, the developing tools, and the underlying hardware processors. Finally, design challenges in posit arithmetic units and hardware processors will be discussed to motivate more related research works.