Branch prediction techniques are commonly used for speeding-up code execution. Modern microprocessors use predictors based on a set of finite automata predictors. This paper shows that finite automata branch predictors can be created by using evolutionary algorithms. These evolved predictors have better performance in predicting the code execution, which they have been trained for, than a standard 2-bit counter predictor.

Branch prediction techniques are commonly used for speeding-up code execution. Modern microprocessors use predictors based on a set of finite automata predictors. This paper shows that finite automata branch predictors can be created by using evolutionary algorithms. These evolved predictors have better performance in predicting the code execution, which they have been trained for, than a standard 2-bit counter predictor.