We’re quite used to seeing overclocking records, which are broken from time to time by professional overclockers who manage to crank a CPU’s operating frequency to levels well above the speed at which they run at the factory; however, these quantities are generally on the order of less than ten gigahertz.
On the other hand, the news of processors that have reached 350 GHz at room temperature and which have reached 500 GHz through cryonics have also surfaced, but how is this possible? To understand it, you have to “sink into the mud” to understand the concepts of switching and pointing.
Clock and switching are not the same
When we see that an overclocker has broken a speed record in a processor, what he has done (explained in a basic way) is to increase its operating frequency, usually by increasing the voltage.
This produces higher heat generation and that is why they usually use advanced cooling systems, such as liquid nitrogen, thus avoiding problems with the system. In addition to increase the operating frequency processor, the frequency of the base bus is also changed, so that at the same time the speed of other components that depend on it, such as RAM for example, is increased.
On the other hand, taking the example of the processor which operates at 500 GHz (it was an IBM experiment), a cryonics system was used for its cooling, but despite this at room temperature it already reached 350 GHz … approximately 3, 5 THz. In the experiment, they explained that they used a technology called BiCMOS SiGe for your transistors, but in any case, where is the catch? There is no real catch, but they refer to different things: in one case, they are talking about pointing, and in another of switching.
the pointing he’s talking about the clock rate at which the processor is operating, that is, the speed at which its transistors switch together. For example, when we say that a processor is operating at 5 GHz, we are actually saying that its transistors are capable of switching (switching between zeros and ones) 5,000,000,000 times per second.
On the other hand, the switching is the speed at which a transistor can switch and go from one state to another. This is the case with the IBM example we gave you previously, since it referred to the fact that the transistors used with BiCMOS SiGe technology were able to switch at a speed of 350 GHz at room temperature or 500 GHz with cryonics cooling; In other words, this means that the transistors used are capable of changing state 500,000,000,000 times in one second.
The difference is that when you talk about synchronization, you mean the switching speed of all simultaneously transistors, which is known as clock cycles. For its part, when we speak of switching, it refers to the switching speed of the individually transistorsTherefore, one speed is not comparable to the other.
In a modern microprocessor, many transistors operate at the same time and are connected to each other. These interconnections create delays and therefore the clock frequency must be limited otherwise errors and jitter will occur. In fact, one has to be very careful with the way the clock signal is distributed through the functional units of the chip so that it works homogeneously, because otherwise it is when there is instability, errors of calculation, “hang up”, etc. that we see several times when overclocking (this is why it is considered unstable overclocking).
Therefore, we have to be some “careful” when we see that we are talking about a processor that is operating at stratospheric speed like the IBM example we talked about, because in this case it is referring to the switching speed but not even close to the processor clock frequency.
Lithography is essential to determine the switching speed
The size of transistors is a fundamental aspect in determining their switching speed, and for this reason, it is important for manufacturers to update their manufacturing nodes using new structures such as FinFET, GAA, etc. in order to reduce the gate capacity and thus improve the switching capacity.
In principle, the delay time depends on several factors such as the capacitance the gate or the voltage and current used, but it also depends on the physical dimensions of the transistor since the switching speed also depends on the width, length and thickness of the logic gate. If these factors are reduced to switching from one lithograph to another (understanding that it moves to smaller transistors), the switching speed could be increased at the same time.
In other words and explained simply, the smaller the lithography of the manufacturing node in which the transistors are fabricated, the higher the switching speed can potentially be, which in turn can (but not necessarily) allow a frequency of higher processor clocking.
Como podéis ver, todos los parámetros influyen y están relacionados entre sí a la hora de determinar la velocidad, rendimiento y potencia de un chip, pero la litografía física y el nodo de fabricación que determina el tamaño de los transistores es una de las más significant for it.