Yeah, traditional resistors are either made from doped semiconductor (which has modest resistivity, making reasonable values practical), either by thin necked diffusions (almost JFETs) or by long sinuous, average width tracks; or in actual resistive material (CrSi most often I think?) as a separate layer.
Diffusion is rather crude -- it depends closely on the dopant concentration and annealing time. The tolerance of a diffusion resistor might be -30/+50% or worse; but the resistors will be reasonably well matched between each other (better than 10%?).
Deposited resistors are quite good, I can't remember if I've read how good they are for initial tolerance. They are often laser trimmed, which means they have to be the final metal layer, on top of the existing metal layer. (Can't cheat by using it for the lone metal layer, even if you can handle the added resistance -- it can't be deposited on silicon because transition metals play dirty with silicon's bandgap.)
Metallization (aluminum or copper) isn't practical to use, as the resistivity is too low, even if the thickness is kept low. Would be practical for thin-film resistors on macroscale, but down on a chip, it just won't work out.
That's actually kind of a lie; obviously the point is, metallization would only practical for low resistances -- which you can actually see examples of, in power transistors like the outputs in TL431 and LM317 and etc. The power transistor is made of many BJT cells, connected in parallel through relatively long emitter contacts. The resistances tend to ballast the parallel array, improving current balance and SOA.
Probably worth noting that, just like copper on the PCB, the tempco is terrible, so while you can use it for small resistances, don't count on it being all that suitable. (The ratio of resistivity to inductivity of copper is also pretty poor on the PCB scale, making it even worse for "resistors" at AC!)
Tim