Publikation

Approximate Hardware Generation Using Formal Techniques

Saman Fröhlich, Daniel Große, Rolf Drechsler

In: Sherief Reda, Muhammad Shafique. Approximate Circuits: Methodologies and CAD. Seiten 155-174 Approximate Circuits: Methodologies and CAD 2019.

Abstrakt

When it comes to the design of hardware for approximate computing, the exactness requirement between a specification of a circuit and its implementation is relaxed. In this chapter we present two different methods to generate approximate hardware for a given specification and its non-approximated implementation. We use formal techniques to guarantee that bounds for application specific error-metrics hold. The first method for approximate hardware generation is an exact BDD-based technique, which focuses on single-output functions. Due to the complexity of the problem, scalability is an issue. For this reason, we further present a heuristic approach, which uses Symbolic Computer Algebra to determine the error-metric. This approach is tailored for arithmetic circuits. We apply this method to Ripple-Carry-Adders and compare the results to state-of-the-art handcrafted approximate hardware.

Deutsches Forschungszentrum für Künstliche Intelligenz
German Research Center for Artificial Intelligence