Modelling communication overhead for accessing local memories in hardware accelerators | IEEE Conference Publication | IEEE Xplore