A new architecture for accurate dot product of floating point numbers | IEEE Conference Publication | IEEE Xplore