miso said:
miso wrote:
I'm dubious of those slides if they are actually implying they have
incorporated a switcher at 140MHz as a component on the MCM.
Regarding the regulators on the mobo, often using multiple phases is
cheaper. It all depends on component pricing. That is, using lower
power components and lots of them (i.e. phases) versus fewer
components that can take more power. Given how switchers about filter
caps, I rather have more phases. Using multiple phases reduces the
physical size of the individual components, making the board lower
profile, which helps air flow.
I never worked at a passive component company (just semis), but often
big means expensive due to yield reduction. For chips, the bigger the
die the greater the odds it hits a crystal defect site. [Really a
problem in bipolar due to vertical current flow, less so in CMOS). I'm
presuming for a filter cap, as it gets bigger there is a greater
chance of a defect as well, so more smaller components can have a
higher yield than one big component.
All that said, I don't aggravate myself over the number of phases in a
board I am purchasing. I have to assume whatever number of phases they
picked was out of performance and economics.
I don't see the 30MHz to 140MHz as being an outrageous figure.
This regulator is done in silicon, the article gives the impression
the inductors are integrated. And each cell contributes a small
portion of the needed power. So switching at a high
frequency isn't all that surprising.
I think the idea is overkill for desktops, but
that's probably not why Intel is doing this.
Paul
There is a bit more to a switchmode power supply than just switching.
When you talk high speed switcher, it is generally a few MHz, not in
the hundred MHz range. For stability, you are going to need a GBWP
over a GHz in the error amp, and comparators that switch in a few
nanoseconds. Since intel has SiGe technology, they can probably do
that, but nobody likes to switch fast since they tend to make low
efficiency converters. Plus why would you want all that noise
generated on your module, not to mention heat.
We had a small switcher at work (size of your thumb) that ran at 10MHz.
That's why I don't think of 30MHz as being that big a deal. And ours
was done years ago - as a custom design by our power supply group.
A group my company later sold off, leaving us without a custom
design capability.
Knowing Intel, they probably targeted an ordinary CMOS process
to make this thing. For reasons of economy. Nothing exotic. Or
at leas, a minimal number of additional process steps.
There are silicon processes, where you can get just about any
component you want. BICMOS processes used to be
the tech of choice - you could have caps, inductors, Schottky
diodes, just about anything you wanted, but not with large
values or anything. And that wasn't even intended as a mixed
analog digital environment or anything. But companies eventually
threw away that capability, because it took so many more
process steps. Still, while it lasted, it was pretty amazing stuff.
The cost of making the chips, forces us into the bland CMOS world.
It's my guess, that when Intel made those internal regulators,
it wanted something it could make on a regular CMOS line,
with only enough extra process steps to make the inductors say.
I think that's the reason it runs at 2.4V input - because it's
actually a not-so-old CMOS process, and if you go higher in voltage
than that, you'd have a problem with breakdown voltage. Otherwise,
if they had a choice, I don't think they picked 2.4V for fun.
If the technology was really good, they would be running it
at 12V, and avoiding the VRM on the motherboard to convert
12V to 2.4V.
Paul
