Interesting tidbit regarding digital logic supply voltage

alpineranger

Senior member
Feb 3, 2001
701
0
76
I just learned this yesterday:

Propagation delay through a cmos inverter (and by extension, other typical cmos gates), is proportional to 1/(V_s - V_t) where V_s is the supply voltage and V_t is the threshold voltage for the transistors.

Why this might interest you:

Let's say you're overclocking your athlon, p4, or whatever. It's getting hot, so you turn down the supply voltage. Well, you've just raised the time required for the logic gates to switch between states, and at the same time, reduced the clock period - their margin or error for finishing the transition. If this is taken too far, then errors will result because the logic signas haven't had time to propagate property through the chip. Even if we weren't overclocking, we could turn the supply voltage down so close to the threshold voltage that the transition could literally take seconds.

I always wondered in the back of my head why you needed to turn up the voltage when overclocking, and now I've got actual equations to tell me. The intuition rests on the foundation of the cmos gate being a rather complicated RC low pass filter.
 

CTho9305

Elite Member
Jul 26, 2000
9,214
1
81
And over the past few days I've learned WHY that is :).
(can't post more now... maybe tonight)
 

rimshaker

Senior member
Dec 7, 2001
722
0
0
Propagation delay (PD) is proportional to many other factors combined. There isn't one single figure of merit that is consistently used when analyzing a single cell (inverter in this case). The expression you used is useful obviously when looking only at voltages. Just to give you a some other examples, PD is also proportional to (delta-V)/current. It is also proportional to the load capacitance and fan-out configuration.
 

alpineranger

Senior member
Feb 3, 2001
701
0
76
I'm aware that other factors are involved, but that there is such a direct relationship between supply voltage and delay was enlightening. I sort of figured one existed, but I wouldn't have bet on it before. Oh, and factors like load aren't really controllable by people who buy, let's say, a computer, when compared to supply voltage.