Forums

polymer aluminum cap

Started by John Larkin September 17, 2012

I designed a 0.9 volt power supply using LTM8023 switcher bricks, and
used an aluminum bulk filter cap. It turns out to have too much ESR,
and I got a lot of output ripple, 0.6 volts p-p at 1/3 the switching
frequency, some weird sub-cycle oscillation. It only started
oscillating as my customer kept adding more and more code to the FPGA,
increasing core current. There are two fixes: increase the switching
frequency, and use a polymer aluminum cap. We'll do both.

LT Spice does model this pretty well. It does oscillate at higher load
currents than we see, but the effect is plainly there.

I got curious about the cap, and my customer is all worked up that we
have a good fix, so I measured a cap to make sure we're modeling it
right.

This is a United Chem-Con 180 uF 6 volt cap. I soldered a hardline
coax onto it as close as I could and TDR'd it:

https://dl.dropbox.com/u/53724080/Parts/Polymer_Test.JPG

(the second set of leads were added after the TDR test)

https://dl.dropbox.com/u/53724080/Parts/Polymer_TDR.JPG

This is astonishing. The inductive spike lasts under 100 ps. I
estimate the ESL at maybe 5 nH, some of which is the loop I made
soldering the coax to the cap leads.

And here's an ESR estimate, using a sort-of 4-wire connection. A 100
mA current pulse makes about a 1.5 mV step, which is 15 mohms.

https://dl.dropbox.com/u/53724080/Parts/Polymer_ESR.JPG


-- 

John Larkin                  Highland Technology Inc
www.highlandtechnology.com   jlarkin at highlandtechnology dot com   

Precision electronic instrumentation
Picosecond-resolution Digital Delay and Pulse generators
Custom timing and laser controllers
Photonics and fiberoptic TTL data links
VME  analog, thermocouple, LVDT, synchro, tachometer
Multichannel arbitrary waveform generators
On 9/17/2012 7:31 PM, John Larkin wrote:
> > > I designed a 0.9 volt power supply using LTM8023 switcher bricks, and > used an aluminum bulk filter cap. It turns out to have too much ESR, > and I got a lot of output ripple, 0.6 volts p-p at 1/3 the switching > frequency, some weird sub-cycle oscillation. It only started > oscillating as my customer kept adding more and more code to the FPGA, > increasing core current. There are two fixes: increase the switching > frequency, and use a polymer aluminum cap. We'll do both. > > LT Spice does model this pretty well. It does oscillate at higher load > currents than we see, but the effect is plainly there. > > I got curious about the cap, and my customer is all worked up that we > have a good fix, so I measured a cap to make sure we're modeling it > right. > > This is a United Chem-Con 180 uF 6 volt cap. I soldered a hardline > coax onto it as close as I could and TDR'd it: > > https://dl.dropbox.com/u/53724080/Parts/Polymer_Test.JPG > > (the second set of leads were added after the TDR test) > > https://dl.dropbox.com/u/53724080/Parts/Polymer_TDR.JPG > > This is astonishing. The inductive spike lasts under 100 ps. I > estimate the ESL at maybe 5 nH, some of which is the loop I made > soldering the coax to the cap leads. > > And here's an ESR estimate, using a sort-of 4-wire connection. A 100 > mA current pulse makes about a 1.5 mV step, which is 15 mohms. > > https://dl.dropbox.com/u/53724080/Parts/Polymer_ESR.JPG > >
You might want to solder some in series. It makes the measured values larger and provides a better estimate of the average component. Obviously divide by N. You didn't zoom in on the spike? I've seen both schools of thought, i.e. the person that like to flog the component, and the person that just builds the circuit and sees how it performs. I'm more inclined to do the latter since it is real life. Of course, if the component doesn't work well in the circuit, the person that did the flogging has a better idea of why.
On Mon, 17 Sep 2012 23:35:18 -0700, miso <miso@sushi.com> wrote:

>On 9/17/2012 7:31 PM, John Larkin wrote: >> >> >> I designed a 0.9 volt power supply using LTM8023 switcher bricks, and >> used an aluminum bulk filter cap. It turns out to have too much ESR, >> and I got a lot of output ripple, 0.6 volts p-p at 1/3 the switching >> frequency, some weird sub-cycle oscillation. It only started >> oscillating as my customer kept adding more and more code to the FPGA, >> increasing core current. There are two fixes: increase the switching >> frequency, and use a polymer aluminum cap. We'll do both. >> >> LT Spice does model this pretty well. It does oscillate at higher load >> currents than we see, but the effect is plainly there. >> >> I got curious about the cap, and my customer is all worked up that we >> have a good fix, so I measured a cap to make sure we're modeling it >> right. >> >> This is a United Chem-Con 180 uF 6 volt cap. I soldered a hardline >> coax onto it as close as I could and TDR'd it: >> >> https://dl.dropbox.com/u/53724080/Parts/Polymer_Test.JPG >> >> (the second set of leads were added after the TDR test) >> >> https://dl.dropbox.com/u/53724080/Parts/Polymer_TDR.JPG >> >> This is astonishing. The inductive spike lasts under 100 ps. I >> estimate the ESL at maybe 5 nH, some of which is the loop I made >> soldering the coax to the cap leads. >> >> And here's an ESR estimate, using a sort-of 4-wire connection. A 100 >> mA current pulse makes about a 1.5 mV step, which is 15 mohms. >> >> https://dl.dropbox.com/u/53724080/Parts/Polymer_ESR.JPG >> >> >You might want to solder some in series. It makes the measured values >larger and provides a better estimate of the average component. >Obviously divide by N. > >You didn't zoom in on the spike?
The scope is max'd out at 2 mV/cm.
> >I've seen both schools of thought, i.e. the person that like to flog the >component, and the person that just builds the circuit and sees how it >performs. I'm more inclined to do the latter since it is real life. Of >course, if the component doesn't work well in the circuit, the person >that did the flogging has a better idea of why.
I like to understand parts, especially their undocumented and destructive limits. The polymer caps seem to suddenly fail shorted at about 2x rated voltage, with no change in leakage current as a warning... like a film cap. -- John Larkin Highland Technology Inc www.highlandtechnology.com jlarkin at highlandtechnology dot com Precision electronic instrumentation Picosecond-resolution Digital Delay and Pulse generators Custom timing and laser controllers Photonics and fiberoptic TTL data links VME analog, thermocouple, LVDT, synchro, tachometer Multichannel arbitrary waveform generators
The quality of the test pulse is important. A common home brew device in 
a lab is a low accuracy but extremely clean pulser. Basically you make a 
very low inductance load resistor by paralleling up carbon comp 
resistors with strips of PCB. Then the load is pulsed with a power fet. 
This voltage step isn't all that accurate, but you can measure it. The 
idea is the pulse is very clean. Much cleaner than the commercial load 
pulsers.

In your case, it makes far more sense just to try the component in the 
actual circuit, but it you want to test components, you near a very low 
glitch test source.

On Tue, 18 Sep 2012 12:36:08 -0700, miso <miso@sushi.com> wrote:

>The quality of the test pulse is important. A common home brew device in >a lab is a low accuracy but extremely clean pulser. Basically you make a >very low inductance load resistor by paralleling up carbon comp >resistors with strips of PCB. Then the load is pulsed with a power fet. >This voltage step isn't all that accurate, but you can measure it. The >idea is the pulse is very clean. Much cleaner than the commercial load >pulsers. > >In your case, it makes far more sense just to try the component in the >actual circuit, but it you want to test components, you near a very low >glitch test source.
For ESR, I just used a B&K function generator. I set it to make a 5 volt pulse, which will make a 100 mA step. -- John Larkin Highland Technology, Inc jlarkin at highlandtechnology dot com http://www.highlandtechnology.com Precision electronic instrumentation Picosecond-resolution Digital Delay and Pulse generators Custom laser drivers and controllers Photonics and fiberoptic TTL data links VME thermocouple, LVDT, synchro acquisition and simulation
On 9/17/2012 9:31 PM, John Larkin wrote:
> > > I designed a 0.9 volt power supply using LTM8023 switcher bricks, and > used an aluminum bulk filter cap. It turns out to have too much ESR, > and I got a lot of output ripple, 0.6 volts p-p at 1/3 the switching > frequency, some weird sub-cycle oscillation. It only started > oscillating as my customer kept adding more and more code to the FPGA, > increasing core current. There are two fixes: increase the switching > frequency, and use a polymer aluminum cap. We'll do both. > > LT Spice does model this pretty well. It does oscillate at higher load > currents than we see, but the effect is plainly there. > > I got curious about the cap, and my customer is all worked up that we > have a good fix, so I measured a cap to make sure we're modeling it > right. > > This is a United Chem-Con 180 uF 6 volt cap. I soldered a hardline > coax onto it as close as I could and TDR'd it: > > https://dl.dropbox.com/u/53724080/Parts/Polymer_Test.JPG > > (the second set of leads were added after the TDR test) > > https://dl.dropbox.com/u/53724080/Parts/Polymer_TDR.JPG > > This is astonishing. The inductive spike lasts under 100 ps. I > estimate the ESL at maybe 5 nH, some of which is the loop I made > soldering the coax to the cap leads. > > And here's an ESR estimate, using a sort-of 4-wire connection. A 100 > mA current pulse makes about a 1.5 mV step, which is 15 mohms. > > https://dl.dropbox.com/u/53724080/Parts/Polymer_ESR.JPG
Thank you, John. Very informative and educational.
On 9/18/2012 12:59 PM, John Larkin wrote:
> On Tue, 18 Sep 2012 12:36:08 -0700, miso <miso@sushi.com> wrote: > >> The quality of the test pulse is important. A common home brew device in >> a lab is a low accuracy but extremely clean pulser. Basically you make a >> very low inductance load resistor by paralleling up carbon comp >> resistors with strips of PCB. Then the load is pulsed with a power fet. >> This voltage step isn't all that accurate, but you can measure it. The >> idea is the pulse is very clean. Much cleaner than the commercial load >> pulsers. >> >> In your case, it makes far more sense just to try the component in the >> actual circuit, but it you want to test components, you near a very low >> glitch test source. > > For ESR, I just used a B&K function generator. I set it to make a 5 > volt pulse, which will make a 100 mA step. > >
You really need that home brew pulser. A function generator, being a driver with feedback, is probably not putting a clean pulse into the cap. Most LDO test scope photos are done with the design I described. It is a good low tech (translate: cheap) scheme to put seriously fast edged into capacitors. Jim Williams has app notes on power pulse generators that are more complexe, but the scheme I mentioned is used by a few semis. The key is the low inductance resistor. I used a tape reel of free resistors.
On Tue, 18 Sep 2012 19:07:26 -0700, miso <miso@sushi.com> wrote:

>On 9/18/2012 12:59 PM, John Larkin wrote: >> On Tue, 18 Sep 2012 12:36:08 -0700, miso <miso@sushi.com> wrote: >> >>> The quality of the test pulse is important. A common home brew device in >>> a lab is a low accuracy but extremely clean pulser. Basically you make a >>> very low inductance load resistor by paralleling up carbon comp >>> resistors with strips of PCB. Then the load is pulsed with a power fet. >>> This voltage step isn't all that accurate, but you can measure it. The >>> idea is the pulse is very clean. Much cleaner than the commercial load >>> pulsers. >>> >>> In your case, it makes far more sense just to try the component in the >>> actual circuit, but it you want to test components, you near a very low >>> glitch test source. >> >> For ESR, I just used a B&K function generator. I set it to make a 5 >> volt pulse, which will make a 100 mA step. >> >> >You really need that home brew pulser. A function generator, being a >driver with feedback, is probably not putting a clean pulse into the >cap.
Why not? It's a pulse generator with a 50 ohm output impedance. The risetime is obviously adequate. Most LDO test scope photos are done with the design I described. It
>is a good low tech (translate: cheap) scheme to put seriously fast edged >into capacitors.
I could have used our P400 digital delay generator, but it wouldn't have made any difference. -- John Larkin Highland Technology Inc www.highlandtechnology.com jlarkin at highlandtechnology dot com Precision electronic instrumentation Picosecond-resolution Digital Delay and Pulse generators Custom timing and laser controllers Photonics and fiberoptic TTL data links VME analog, thermocouple, LVDT, synchro, tachometer Multichannel arbitrary waveform generators
On 9/18/2012 7:40 PM, John Larkin wrote:
> On Tue, 18 Sep 2012 19:07:26 -0700, miso <miso@sushi.com> wrote: > >> On 9/18/2012 12:59 PM, John Larkin wrote: >>> On Tue, 18 Sep 2012 12:36:08 -0700, miso <miso@sushi.com> wrote: >>> >>>> The quality of the test pulse is important. A common home brew device in >>>> a lab is a low accuracy but extremely clean pulser. Basically you make a >>>> very low inductance load resistor by paralleling up carbon comp >>>> resistors with strips of PCB. Then the load is pulsed with a power fet. >>>> This voltage step isn't all that accurate, but you can measure it. The >>>> idea is the pulse is very clean. Much cleaner than the commercial load >>>> pulsers. >>>> >>>> In your case, it makes far more sense just to try the component in the >>>> actual circuit, but it you want to test components, you near a very low >>>> glitch test source. >>> >>> For ESR, I just used a B&K function generator. I set it to make a 5 >>> volt pulse, which will make a 100 mA step. >>> >>> >> You really need that home brew pulser. A function generator, being a >> driver with feedback, is probably not putting a clean pulse into the >> cap. > > Why not? It's a pulse generator with a 50 ohm output impedance. The > risetime is obviously adequate. > > Most LDO test scope photos are done with the design I described. It >> is a good low tech (translate: cheap) scheme to put seriously fast edged >> into capacitors. > > I could have used our P400 digital delay generator, but it wouldn't > have made any difference. > >
The pulsers I've built are in the ohm range source impedance. You really need to whack the cap to evaluate it. In the case of LDOs, you need to insure that a blazing fast input won't make it latch or over-voltage. When the goal (mind set) is to sell something by the millions, you need to be very thorough in evaluation. If you don't want to evaluate the cap, just build the circuit. That at least gets the product out the door. Gee, it would have been nice if EDN provided the schematic.
> http://www.edn.com/design/other/4349554/EDN-Access--04-11-96-Load-pulser-tests-supply-s-transient-respons
The paper does describe building the power resistor.
> http://www.ti.com/general/docs/lit/getliterature.tsp?literatureNumber=snoa507&fileType=pdf
In my example, the fet is used as a source follower since the goal is to create a very low impedance pulse generator rather than a dynamic load. There may be commercial loads these days that do the same job, but these home brew pulsers are common in semi test labs. Nobody testing components wants the settling time of the source (which uses feedback for accuracy) to be part of the component test.
On Tue, 18 Sep 2012 23:52:55 -0700, miso <miso@sushi.com> wrote:

>On 9/18/2012 7:40 PM, John Larkin wrote: >> On Tue, 18 Sep 2012 19:07:26 -0700, miso <miso@sushi.com> wrote: >> >>> On 9/18/2012 12:59 PM, John Larkin wrote: >>>> On Tue, 18 Sep 2012 12:36:08 -0700, miso <miso@sushi.com> wrote: >>>> >>>>> The quality of the test pulse is important. A common home brew device in >>>>> a lab is a low accuracy but extremely clean pulser. Basically you make a >>>>> very low inductance load resistor by paralleling up carbon comp >>>>> resistors with strips of PCB. Then the load is pulsed with a power fet. >>>>> This voltage step isn't all that accurate, but you can measure it. The >>>>> idea is the pulse is very clean. Much cleaner than the commercial load >>>>> pulsers. >>>>> >>>>> In your case, it makes far more sense just to try the component in the >>>>> actual circuit, but it you want to test components, you near a very low >>>>> glitch test source. >>>> >>>> For ESR, I just used a B&K function generator. I set it to make a 5 >>>> volt pulse, which will make a 100 mA step. >>>> >>>> >>> You really need that home brew pulser. A function generator, being a >>> driver with feedback, is probably not putting a clean pulse into the >>> cap. >> >> Why not? It's a pulse generator with a 50 ohm output impedance. The >> risetime is obviously adequate. >> >> Most LDO test scope photos are done with the design I described. It >>> is a good low tech (translate: cheap) scheme to put seriously fast edged >>> into capacitors. >> >> I could have used our P400 digital delay generator, but it wouldn't >> have made any difference. >> >> > >The pulsers I've built are in the ohm range source impedance. You really >need to whack the cap to evaluate it.
Why? I measured 15 mohms to engineering accuracy, good enough for most purposes, and it only took a few minutes to set up. In the case of LDOs, you need to
>insure that a blazing fast input won't make it latch or over-voltage. >When the goal (mind set) is to sell something by the millions, you need >to be very thorough in evaluation. > >If you don't want to evaluate the cap, just build the circuit. That at >least gets the product out the door.
I did evaluate the cap, for ESL and ESR. I like to understand my parts and circuits.
> > > >Gee, it would have been nice if EDN provided the schematic. >> http://www.edn.com/design/other/4349554/EDN-Access--04-11-96-Load-pulser-tests-supply-s-transient-respons > >The paper does describe building the power resistor. > >> http://www.ti.com/general/docs/lit/getliterature.tsp?literatureNumber=snoa507&fileType=pdf > >In my example, the fet is used as a source follower since the goal is to >create a very low impedance pulse generator rather than a dynamic load.
I've done mosfet based pulsers when I wanted a lot of current, but I didn't need that to measure this ESR.
> >There may be commercial loads these days that do the same job, but these >home brew pulsers are common in semi test labs. Nobody testing >components wants the settling time of the source (which uses feedback >for accuracy) to be part of the component test.
The B&K makes a fast step with a 50 ohm source impedance. It sure behaves like an amplifier followed by a real 50 ohm resistor, which it probably is. -- John Larkin Highland Technology Inc www.highlandtechnology.com jlarkin at highlandtechnology dot com Precision electronic instrumentation Picosecond-resolution Digital Delay and Pulse generators Custom timing and laser controllers Photonics and fiberoptic TTL data links VME analog, thermocouple, LVDT, synchro, tachometer Multichannel arbitrary waveform generators