Spice solves a large set of equations to determine DC conditions. It uses an interative technique to superimpose AC variations. Convergence is a problem in many situations.
CMOS digital circuits don't need such generalized tools. Voltages swing from power to ground. Even if rise/fall time is inaccurate, circuit behavior is not. Errors are not cumulative, since voltages are recalibrated when they reach a limit.
The MOSFET model used is derived from solid-state physics. It is complex beyond comprehension (mine at least). Even the latest pragmatic model. This is not necessary. A simple empirical model of the IV curves is adequate for digital transistors.
Transistor temperature is not accounted for. This causes a huge error, swamping the accuracy of the elaborate model. Measured IV curves necessarily refer to a hot transistor. Its temperature may be 100 K above ambient, depending on the power dissipated. Current must be increased for cold transistors, perhaps 50%. Digital transistors are normally cold, since they heat only when switching.
SPICE works because it is usually limited to small circuits. Designs are extremely conservative. Ground truth is very hard to obtain.