Dynamic Branch Prediction Using Machine Learning

Microarchitectural prediction based on machine learning has received increasing attention in recent years. The common problem is that most of the Neural Network Based Branch Predictors (NNBBP) can achieve high accuracy; however, that achievement is not big enough to offset the cost of latency. As a result, the idea of using Neural Networks (NN) to perform branch prediction is not practically used compared to industrial designs. As microachitecture becomes more and more deeply pipelined, branch misprediction latency is the most important component of performance degradation. At present, there are several NNBBP have been proposed, and their algorithms and relative performance will be examined in this project. In addition to that, we have proposed our own dynamic branch predictor using Learning Vector Quantization (LVQ) to suggest another approach within the realm of machine learning.