Signal integrity experts have taught us to design power distribution systems that can support all devices on the board with a low low impedance over a wide frequency range.
We all know, trust or believe that is important. But we soon realize that achieving low impedance from DC to infinity is impossible.
So how can we work with a specification like this one:
Maybe not the best example, but the thing worth noting is:
- No mention of frequency range for the supply voltage requirement(1)
- The datasheet contains no information on the worst case current consumption rate of change
Note 1: I recognize that this is “recommended” operating conditions, but given the fact that there are no other “guranteed to work if you do this” data available this becomes a de’facto requirement for the designer.
Unknown frequency range
So what we have is a device, pulling in current in an unknown frequency range, and we have to keep the voltage within the +/- 4-5%.
Usually this is done with a combination of a power regulator and a set of bypass capacitors. The power regulator is limited in frequency by the frequency response of the feedback loop.
And capacitors work like this: If you pull out current, the voltage drops. This is dominated by 3 effects:
- Limited capacity. Pulling out 1A in 1s will remove 1 Q (Coulomb). The capacitor voltage will drop accoding to V = Q / C
- Non-zero ESR. Pulling out 1A from a 0.1 ohm ESR capacitor will lower the voltage 0.1V as given by V = I * R
- Non-zero ESL. Changing the current through a capacitor will change the voltage as given by V = L * di/dt
Impossible, so what do we do?
Without knowing di/dt, the design problem posed is impossible to solve. So what do we do?
Do something reasonably good and cross our fingers? Or what is your solution? Feel free to comment below.
Three partly working solutions
Currently I have only 3 solutions to this fundamental problem.
- Keep pounding on the chip manufacturers to give board designers the data they really need
- Use evaluation boards to measure one device (often difficult to do), add a safety margin and cross my fingers
- Look at the chip package and try to figure out what frequency response the package offers the die with respect to power. Then use this to set an upper limit for where we have a chance to do something at the board level – and cross my fingers that the manufacturer did what the needed to do for the higher frequencies.
None of these solutions are really good. Crossing fingers in engineering is never “the right solution”.
A better solution
I think a better solution would be for the chip manufacturers, to simply specify a max frequency along with the voltage requirements.
Would that be sufficient? Or what do you think? Voice your opinion below.
Realise that it inherently impossible to elicitate all necessary requirements up front to perform a waterfall project.
The most appropriate way to proceed would in my opinion be to spend reasonably short time to gather requirements and to create a suitable design to evaluate wherether it will work reasonably well in the application. The requirements to that will be different depending on the solution is applied on a space vehicle or a cellphone.
So, my recommandation is to enter a project cycle based on experience and iterations while taking the goal into consideration. At the end each iteration the status is evaluated to reveal conformance to requirements, written or not.