In this paper, we have investigated the static metrics and switching attributes of graphene nanoribbon field-effect transistors (GNR FETs) for scaling the channel length from 15 nm down to 2.5 nm and GNR width by approaching the ultimate vertical scaling of oxide thickness. We have simulated the double-gate GNR FET by solving a numerical quantum transport model based on selfconsistent solution of the 3D Poisson equation and 1D Schrödinger equation within the non-equilibrium Green's function formulism. The narrow armchair GNR, e.g. (7,0), improved the device robustness to shortchannel effects, leading to better OFF-state performance considering OFF-current, I_ON/I_OFF ratio, subthreshold swing, and drain-induced barrier-lowering. The wider armchair GNRs allow the scaling of channel length and supply voltage, resulting in better ON-state performance, such as the larger intrinsic cut-off frequency for the channel length below 7.5 nm at smaller gate voltage as well as smaller intrinsic gate-delay time with the constant slope for scaling the channel length and supply voltage. The wider armchair GNRs, e.g. (13,0), have smaller power-delay product for scaling the channel length and supply voltage, reaching to ~0.18 (fJ/μm).