If a chain reaction starts in an assembly of fuel rod, it may very rapidly develop into a process running out of control. Therefore, special rods are placed in between the fuel rods in order to control the reaction intensity. Such rods are made of materials strongly absorbing neutrons with energies in the range of a few eV. The most often used material are Cadmium metal, or compounds of Boron. The control rods can be moved up or down in guiding tubes.
In a normal procedure of starting up the reactor the control rods at first are fully lowered. Then, a small neutron source is placed in the middle of the fuel rods – nothing will start unless there are some “seed” neutrons in the system. Then, the control rods are slowly being raised, while the flux of neutrons in the moderator is all the time monitored. At some moment, the neutron flux starts increasing even if the control rods are no longer moving – it means that the reactor “has reached criticality”, it doesn’t need “seed neutrons” any more. The extra source may be thus removed. Raising the control rods may continue, until the rate of the fission reactions reaches a desired level. Then, a special mechanism move the rods slightly up the rate slows down, or slightly down, if the rate starts speeding up. Such an action is called a “negative feedback”; other examples of negative feedback you must know is the cruise control in a car: if the car’s speed starts decreasing because, for instance, the is going slightly uphill – the cruise control electronics slightly increases the pressure on the gas pedal. And vice versa, if the car starts accelerating because the road now goes slightly downhill, the automaton will release the pressure on the gas pedal.
The system of controlling the reactor power by moving the control rods is an “electro-mechanical feedback loop”. However, there is another negative feedback mechanism, we can call it “intrinsic”, because it does not need any electronics and electric motors. The thing is that the efficiency of the reactor moderator is usually slightly temperature-dependent: if the temperature goes up, the flux of neutrons with the lowest energies slightly decreases. So a small increase of the reactor power causes a small increase in the moderator’s temperature – and a small drop in the reactor power – which in turn lowers the moderator temperature – and so on.
But as long as the reactors works, thermal energy is generated inside, about 200 MeV for each fission event. One could argue: well, since energy is continuously released, the temperature can only increase – hence, the “power stabilizing thermal feedback effect will never work! This is certainly true. However, any reactor, if supposed to operate for a prolonged time, must be equipped with a cooling system removing the heat generated at a constant rate. And then the power stabilizing thermal feedback effect will indeed work.