In the simplest configuration have a X-Y stage that you raster, along with Z axis control - generally these are all piezo controlled. Your signal is then acquired by maintaining a constant tunneling current while you raster, so you track the Z axis position as you raster.

This means that you're getting a "pseudo-height" map - if you had a surface with 2 types of atoms, both the same size, but with different tunneling barriers, you would see them appear to be different sizes.