Reducing power by optimizing the necessary precision/range of floating-point arithmetic | IEEE Journals & Magazine | IEEE Xplore