# TI Swift DCDC power supply design

#### Do you have a question? Post it now! No Registration Necessary

Translate This Thread From English to

•  Subject
• Author
• Posted on
I am primarily a digital engineer, but have some experience with the
analog world.  Heck, I started with electronics when digital was done
with transistors, not on a chip, cans on a board.  But I have not done
much with switching power supplies.  I am looking at using a TI Swift
part for a very, very small footprint power supply module.

TI has some very nice software to support these designs even if it has a
few rough edges.  But one problem I am having is in the parts list.  TI
makes available a list of many caps and inductors with all the required
information.  However Murata makes some ceramic caps that are smaller
for the values than anything in the TI parts list.  I can add them to
the list, but I am not clear on where to find the data.

The Swift designer parts list needs MinESR, MaxESR, ESL and Iripple.
Murata provides a graph of temperature rise vs. Iripple, so that is
covered.  They also provide graphs of ESR and Impedance vs. frequency.
This I am not clear on how to interpret.

First, I don't think that ESL is the same as Impedance.  I expect that
ESL is just Inductance.  So I guess I need to analyze the graph to get
the ESL value from the graph for impedance.  I seem to remember the
formulae
Il = (2pi f L)
and
Ic = 1/(2pi f C)

So at the resonance point, where Ic and Il are equal,

2pi f L = 1/(2pi f C)

L = 1/(4(pi)^^2 f^^2 C)

Is this what is being requested in the TI Swift software?

The min and max ESR have me a bit stumped.  I don't know if they want
the ESR at the capacitor resonance point, the frequency of the switcher
or some other point that corresponds to the loop analysis for the
filter.  I never got the classes for control theory.  So I guess I need
to sit down with a book or two someday to understand the whys and
wherefores of this sort of control loop.

Anyone worked with this software before and know how to add new
capacitors to the parts list?

--

Rick "rickman" Collins

snipped-for-privacy@XYarius.com
We've slightly trimmed the long signature. Click to see the full one.
Re: TI Swift DCDC power supply design

I don't see a response after several days. You can probably get
good responses if you ask this on sci.electronics.design, but I'll put
in my two cents here.

I know other companies that make switchers have such software. Look
at national.com - they've got an online simulator that runs spice -
I've never used it, but I think they would give a similar parts list
it's supposed to be a complete spice package. But I don't imagine
either of these would help with TI switchers, other than as giving

Let me check that from another direction: The formula for resonant
frequency is:
F = 1 / (2 pi sqrt (LC))
and solving for L gives what you got.

Yes, that sounds right to me.

Aww, just read Bob Pease's column titled "What's all this PID stuff
anyhow" and that's all you need to know. :)
But of course the idea of the design software for these chips is
that you shouldn't have to know control theory, they supposedly have
it all worked out, and if you use parts within the specs they give, it
"should" work right off the bat.
On the Murata parts, does the ESR really vary that much with
frequency? It seems to me you should use the value at the switching
frequency, as that's generally where the high currents happen that
would be most affected by the voltage drop across the ESR.
Looking at some typical Murata ceramic caps page, it looks like ESR
DOES vary a lot with frequency:
http://www.murata.com/cap/nproduct/cap12e.pdf
You say it wants min and max ESR - I'd use the switching frequency
ESR for max, and ESR at self-resonance for min. Since the switching
waveform is far from a sine wave, it will see all the ESR values in
that range.

Have you asked someone at TI? They have an incentive to help
designers with this - lots of others make switching regulator chips
(National, Linear Technology, surely several others) and if TI wants

-----

Re: TI Swift DCDC power supply design

...snip...

Thanks for the advice.  That's a good idea.  I'll look them up.

It has been awhile since I have used much from my AC circuits days.
Although I did not take this in college, I did have a little bit in high
school electronics.

I exchanged an email with TI support and they did not really answer the
question.  The one guy that always seems to answer my hard core switcher
design questions is Ed Walker.  He seems like a real good guy, but often
he does not quite understand what a newbie like me is asking.  Instead
he espoused on how capacitor makers don't spec their parts very well.

You would think that, but there is no clear indication (that I can see)
that a given design will work well as I vary the resistor that controls
the output voltage.  I believe that this resistor (or the output voltage
itself) has very little impact on the control loop, but it is not zero
impact.  The tool provides the gain and phase curves vs. freq for the
circuit, but then I am not familiar with how to analyze them other than
the zero intercept for one gives you the analysis value for the other.
Consider the gain when the phase is at zero and the phase when the gain
is at 0.  I can see the default values they chose for a limit so I guess
I will be ok as long as I stat with those.  But under some conditions I
see the gain/phase curves get rather curvy and I am concerned that this
might be a problem even if it does not go to zero.

Yes, the ESR in the graph is a "V" shape with capacitance dominating
below SRF and inductance dominating above.  The impedance curve has a
sharp "V" at SFR in addition to this overall V shape while the ESR does
not.  They seem to reach the same value at SRF with the inductance about
10x ESR at other freqs.  I don't have the URL for this data sheet and it
is a bit large to attach in a non-binary group.  But this seems to be a
typical capacitor curve to me.

Yes, that is the link.  cap04e is for the 22 uF in the 0805 package.  I
am actually finding some anomolous info in the TI library.  All the work
I have done with power decoupling showed me that smaller packages
typically have lower min ESR/impedance with little variation with
capacitance.  But the values used in the Swift tool show higher ESR for
the same value capacitance in smaller packages.  I don't know if this is
just a factor with these new, high value, very dense parts or if there
is something wrong with the data.

Yes, like I said I have spoken or emailed with Ed Walker several times
and he seems well intentioned, but it can be hard to get the answer to
the question you are asking.  My first try on this one did not work.

But while I was waiting for a reply from him I decided that I could try
margin testing.  Turns out you can set all your compensation component
values and draw new curves with the tool.  So I let the tool design the
circuit using a set of caps that have a higher ESR.  Then I lowered the
ESR value in the curve drawing tool.  That seemed to give workable
results, but when I redesigned the circuit with caps that match the
lower value, I get very different compensation components.  So I
repeated the test in the other direction.  If you adjust the output cap
ESR in this tool, it will not save the changes.  But if you change the
compensation components, you can save the changes.  However, when I do
that, the curves change slightly from what they were in the value
editing tool (which lets you see the curves at any time).

So I think I have a solution although I don't have high confidence in
the tool.  It also has some problems with printing and saving results.
I think TI needs a better windows programmer or two. :)

If anyone is more familiar with the tool and can talk switcher design to
a novice, I would appriciate the help, or even just a sanity check.

--

Rick "rickman" Collins

snipped-for-privacy@XYarius.com
We've slightly trimmed the long signature. Click to see the full one.
Re: TI Swift DCDC power supply design

In this datasheet, ESR declines monotonically. There is no 'V' to it.
Impedance is a V, which is typical.

By the way, I'm a little suprised at the ESR curve. This seems like a highly
non-ideal and non-linear capacitor. Also, if you look at graph 6, you will
see that the capacitance falls off pretty dramatically with applied DC
voltage, too. I don't know if this sheet is for the capacitors that the OP
is planning to use. But if it is, he should make sure that there will be
enough capacitance at the applied DC Voltage.

The switching controllers I've seen have all suggested tantalum or
electrolytic filter cap's, FWIW.

[snip]

This sounds somewhat plausible to me, too, except I might go below
switching frequency for the max, since we don't know what the
manufacturing spread is like. The graph is presumably for a typical
capacitor.

[snip]

Mac

Re: TI Swift DCDC power supply design

The data sheet simply does not show enough to include the rising ESR
above the SRF.  If you look closely you will see that it does start to
turn up above 600 MHz or so.

Hey, nothing in nature is perfect.  This not one of the caps being
considered.  Seems they all have roughly the same curve so when used at
3 to 5 volts, there is not much loss of capacitance.

Yes, but they have a high ESR.  Turns out that for the ripple spec the
capacitance is not nearly as important as the ESR at the higher
frequencies that are used in current switchers, ~1 MHz.  The capacitance
required is a function of the switching frequency and the current.  The
ESR typically limits the ripple.  With ceramic caps this will be very
low and for a 3 Amp supply with 700 kHz switching rate 4 x 22 uF (0805)
works very well, at least on paper.

--

Rick "rickman" Collins

snipped-for-privacy@XYarius.com
We've slightly trimmed the long signature. Click to see the full one.
Re: TI Swift DCDC power supply design

If you say so.  ;-)

And nothing made by man is perfect, either. Don't forget to check out the
temperature coefficient.

The other thing that bothers me is the high ESR at DC. In between
transitions, the capacitor really is at DC, in a sense. I think there may
be a good reason that most switchers are designed with tantalum. Kemet has
some very low ESR tantatlum caps (as low as 30 mOhm).

On the other hand, I went through the math with an ideal 88uF cap, and
assuming you were stepping down from 5V to 3.3V, you get only 16mV of
ripple. This is purely from the dV = Idt/C.

Then, with 3A current draining through a 0.03 Ohm tantalum cap, you get
90mV of droop. During the charge cycle, there will be even more current
flowing, so you will probably have over 200 mV of ripple with 30 mOhms. In
other words, I get what you are saying about most of the ripple being due
to ESR, not capacitance. ;-)

If you are bound and determined to go ceramic, I would urge you to
consider sticking with NPO/COG, or X7R dielectrics.

[snip]

best regards,
Mac

Re: TI Swift DCDC power supply design

Kemet have parts with ESR a lot better than 30 mohm, e.g.

T530X337M010ASE010 330 uF 10V  10 mohm

Regards,
Allan.

Re: TI Swift DCDC power supply design

Even better. Thanks.

Mac

Re: TI Swift DCDC power supply design

They are available from Digikey.

The Kemet parts are in a 'D' case with a non-standard height.  Sanyo
have equivalent ones in a standard D case, e.g. 6TPD330M (330uF, 6.3V,
10 mohm), but they seem a little harder to get.

Regards,
Allan.

Re: TI Swift DCDC power supply design

Thanks to all for the advice on the tantalum caps.  But this module is
very small and does not have the space for a D case tantalum.  I have
crammed two 0805 ceramics in that corner of the module and will live
with the results for this design.  I belive this will be a workable
solution giving about 100 mV of I/Cdt ripple and somewhere around 10 mV
of I*ESR ripple.

I am using a 10uF, 10V cap because I was taught to derate the caps for
any typical design.  I have also read about various modes of failure in
SMPS due to underspecing the voltage rating, although this is normally a
caution about tantalum caps.  I could double the capacitance by using a
22uF, 6.3V cap, but for now I can live with the higher ripple since no
logic will be running off the 5 volt rail directly.  Everything lives at
lower voltages with either a SMPS or an LDO providing conversion.

--

Rick "rickman" Collins

snipped-for-privacy@XYarius.com
We've slightly trimmed the long signature. Click to see the full one.
Re: TI Swift DCDC power supply design
supply design', on Sun, 14 Dec 2003:

That's not an ESR curve but an IMPEDANCE curve. the Equivalent Series
Resistance can't be capacitive or inductive.

Hardly surprising, because the residual impedance at the Series Resonant
Frequency IS the ESR.

--
Regards, John Woodgate, OOO - Own Opinions Only. http://www.jmwa.demon.co.uk
Interested in professional sound reinforcement and distribution? Then go to
We've slightly trimmed the long signature. Click to see the full one.
Re: TI Swift DCDC power supply design

Did you look at the graph?  They show an impedance curve as well as an
ESR curve.  Both have the V pattern with a sharper V at SRF for the
impedance.

--

Rick "rickman" Collins

snipped-for-privacy@XYarius.com
We've slightly trimmed the long signature. Click to see the full one.
Re: TI Swift DCDC power supply design
supply design', on Mon, 15 Dec 2003:

It wouldn't download when I looked before, but it does now. What I say
is incorrect is this:

The ESR **cannot** be capacitive or inductive; it is by definition a
pure resistance (but not necessarily independent of frequency, voltage,
temperature, state of the tide etc.).

But the ESR curve is definitely weird. I guess that there is a lot of
dielectric loss, which in a more realistic equivalent circuit appears as
a resistance *in parallel* with the capacitor. Reducing the components
to just a series R and C does produce behaviour like that.

If I had time, I'd Spice the impedance of a 1 uF with 1 ohm (? I don't
remember the precise ESR values at the SRF) in series and 15 kohms in
parallel. I think it might look a bit similar.
--
Regards, John Woodgate, OOO - Own Opinions Only. http://www.jmwa.demon.co.uk
Interested in professional sound reinforcement and distribution? Then go to
We've slightly trimmed the long signature. Click to see the full one.