Evolutionary approximation of complex digital circuits

Circuit approximation has been developed in recent years as a viable method for constructing energy efficient electronic systems. An open problem is how to effectively obtain approximate circuits showing good compromises between key circuit parameters - the error, power consumption, area and delay.  The use of evolutionary algorithms in the task of circuit approximation has led to promising results; however, only relative simple circuit instances have been tackled because of the scalability problems of the evolutionary design method. We propose to replace the most time consuming part of the evolutionary design algorithm, i.e. the fitness calculation exponentially depending on the number of circuit inputs, by an equivalence checking algorithm operating over Binary Decision Diagrams (BDDs). Approximate circuits are evolved using Cartesian genetic programming which calls a  BDD solver to calculate the fitness value of candidate circuits. The method enables to obtain approximate circuits consisting of tens of inputs and hundreds of gates and showing desired trade-off between key circuit parameters.