EvoApproxLib - a comprehensive collection of low-level approximate implementations of arithmetic operations

Approximate circuits, arithmetic circuits, low-power, accuracy

Approximate computing and approximate circuits attracted a significant attention of researchers as well as industry in recent years. In order to accelerate the approximate system design process and to support a fair benchmarking of approximation methods, we propose EvoApproxLib - a comprehensive collection of low-level approximate implementations of elementary arithmetic operations.  The collection includes hundreds of implementations of 8-, 12- and 16-bit approximate adders and multipliers showing unique trade-offs between several error metrics (such as the mean absolute error and the worst case error) and electrical parameters (such as delay, area on a chip and power consumption for a particular fabrication technology). Each implementation can be downloaded in various formats (C, Matlab, Verilog). These implementations can be used to create high quality approximate circuits or low-level software routines for approximate computing systems. EvoApproxLib was automatically generated using an automated circuit design system (based on Cartesian Genetic Programming) which is the original invention of the authors. EvoApproxLib extends the authors' results reported mainly in their paper which was awarded with the Best IP Award at DATE 2017. 

The implementation and all additional information are available at https://ehw.fit.vutbr.cz/evoapproxlib/ and https://github.com/ehw-fit/evoapproxlib/