Fractal Brownian motion (fBm)
Stack several layers of noise — each at double the frequency and half the amplitude — until the result has detail at every scale.
A single layer of Perlin or simplex noise is smooth and dull. Sum five or six octaves instead, each one finer and fainter than the last, and you get the self-similar roughness of mountain ranges, clouds and coastlines. The technique borrows its name and idea from Benoît Mandelbrot's fractional Brownian motion.
Two knobs set the character: lacunarity (how much the frequency grows per octave) and gain or persistence (how fast the amplitude falls off). Small changes separate a soft hill from a jagged cliff, which makes fBm the workhorse behind procedural heightmaps and terrain.