Connect with us

LM3478 design gets insanely hot

Discussion in 'General Electronics' started by Henrik [7182], Jul 29, 2008.

Scroll to continue with content
  1. Hello Experts,

    This is my first time designing a SEPIC type switchmode power supply with
    the LM3478MM from national as my controller.

    My design is pretty much the standard application as shown in the datasheet.

    The design is made to supply 24V out from a 15-32V input. Ouput current is
    1A.

    My problem is that my two inductors and the MOSFET gets insanely hot. Much
    more hot that I had anticipated. So I wonder if there is somewhere I have
    missed something vital.

    I am using CDRH127 series 22uH inductors from Sumida, which are rated at
    3.3Amp and has a DCR of 43mR

    I use a 39K resistor to obtain approx. 400kHz switch frequency.

    I use 7mR as Rsense and MBRS360 as my power diode.

    My Compensation-filter consists of 100nF and 1K.

    My Mos-Fet is NP55N055 from NEC which has 12.5mR on-resistance.

    Since I am a novice in the field of SMPS, this has me baffled quite a bit,
    and I hope that someone could lead me on right path.


    Thanking you all in advance.

    Best regards
    Henrik
     
  2. Joerg

    Joerg Guest

    IIRC those are 3.6A but could still be a bit marginal for this design.

    Time to take scope shots and post them. But better not on a.b.s.e.
    because some attorney general has shot that tool for most of us.

    Probe the gate directly at the FET, and source and drain. Those are the
    most important vitals for a switcher. I did a 6-9V to 0-100V adjustable
    SEPIC with the LM3478 a few years ago. Don't have scope plots but I kind
    of remember how it looked. Worked like a champ.
     
  3. neon

    neon

    1,325
    0
    Oct 21, 2006
    It cannot be insanely hot if it is stil working.
     
  4. Dear Joerg,

    Thank you for your valued reply. I have taken screenshots at the three
    mos-fet terminals. They are located here: www.impc.dk/files/lm3478_files.rar

    I hope that you will take a look and perhaps your past experience with the
    device could shed some light on my problems.

    All shots have been taken with 1uS for T/Div and the Gate has 2V/Div, the
    Drain is 10V/div and the Source is 500mV/Div

    Thanking you in advance.

    Best regards
    Henrik
     
  5. Joerg

    Joerg Guest

    Sorry but I cannot read this kind of format. Maybe a Linux format which
    I don't have. Can you post them in png, tiff, jpeg or some other more
    common format?
     
  6. legg

    legg Guest

    7mR seems like a pretty low value sense resistor for a 3A output and a
    350mV fault threshold. Shouldn't it be 15 or 30?

    RL
     
  7. Actually you may be right, I did not think this could have any effect on my
    heating issue, so I just picked somthing so low that I didn't expect to get
    any problems. Perhaps I am wrong? As I wrote earlier, I am still a novice in
    the field of SMPS and SEPIC in particular, but eager to learn ;-)

    Best regards
    Henrik
     
  8. Hello Joerg,

    Thank you for your time, and am sorry for the fileformat, I have now placed
    each file separately as JPEG for easier access:

    www.impc.dk/files/LM3478_fet_gate.jpg
    www.impc.dk/files/LM3478_fet_drain.jpg
    www.impc.dk/files/LM3478_fet_source.jpg
    www.impc.dk/files/schematic.jpg

    I hope they are viewable at your location.

    Thanking you in advance

    Best regards
    Henrik
     
  9. legg

    legg Guest

    Safest sense resistor values would be slightly higher (even double)
    the estimated optimum, to begin with.

    If it was large enough, you could use it to scope the current
    waveform, and post this as well.

    Try to scope a single cycle, or even the rising or falling edges -
    with surrounding noise. It helps with visual correlation of single
    trace plots, if the waveforms are not close to 50% duty cycle, as
    well.

    RL
     
  10. Joerg

    Joerg Guest


    The ramp up looks really sluggish. I don't know the data of this FET but
    maybe it's way too big for the LM3478 to drive. Looks like it may have
    too much Cgd.


    That must be one hell of a FET being able to hold all this down. Your
    inductor must drop almost a volt at peak.


    Way, way too much peak current. I don't know what you voltage levels
    are, no data in the image about that.

    Aha! Unless I am mistaken your current peaks exceed your inductor
    rating, big time. Isense trips around 165mV and with 7mohms you'd be at
    23amps by then. Try raising Rsense to something like 50mohm. You may
    have the inductor sitting in saturation and there it won't have any
    inductance to write home about, just resistive losses.

    BTW I wouldn't use an electrolytic for the SEPIC cap (C5). I'd consider
    a few ceramics in parallel. Electrolytics get stressed out there and
    might explode some day. Unless it's a really, really good one and can
    take the ripple at frequency. But if it gets warm to the touch don't use it.

    Yes, they are. Next time try if you can get a real screen shot from your
    scope, with the setting and all.

    Just FYI, before you release BOM and schematic some day: Replace LM347B
    with LM3478.
     
  11. Dear Joerg,

    Thank you very much for your replies. I will with a 50mR or so Rsense
    tomorrow, Also I will try to get the scales and such nice info into the
    shots.

    Also I will try exchanging the sepic capacitor to a large 1uF ceramic
    capacitor that I have in stock.

    Thank you so very much for your help. I will sleep much better now that I
    see a possible solution to my problems ;-)

    Best regards
    Henrik
     
  12. Dear Sir

    Thank you for your valued reply, I will try bumping up the sense resistor
    tomorrow and post a few more and better screen shots.

    Best regards
    Henrik
     
  13. Joerg

    Joerg Guest

    BTW it makes it easier for others to follow and contribute ideas if you
    write below the quoted text.

    Stack a few of those for the test. 1uF may be a tad too small for a 1A
    switcher (tho it'll work). Nowadays you can get pretty good 10uF or at
    least 4.7uF caps but even there you may have to parallel in order to
    ease and share the current load. They have become quite cheap.

    With switchers it is really important to carefully study the datasheets
    of even the most mundane parts. We take those caps for granted in bypass
    situations but here they'll have to slosh around a lot of energy. I
    leaned that the hard way. Everything was fine for an hour or so. Then a
    very loud bang, pieces flying around, and what was left from the
    ceramic material of a capacitor had turned into green bubbly glass.
    Quite pretty actually but it could have been a bad situation if I hadn't
    worn eye protection glasses. Best to wear those when you work on
    switchers and similar gear that can go "exotherm".

    Since you are in Denmark this would be the perfect time for an ice-cold
    shotglass of Aquavit. Good stuff :)

    [...]
     
  14. Yes, mea culpa. I actually know this, it was just plain sloppiness on my
    part.
    I stacked a few of my 1uF 50V on top of each other.

    A chilling history indeed, I once experienced something similar. A colleague
    of mine got a small part of his earlope shot away by and exploding tantalum
    drop-type capacitor. He bled quite a lot and was really shocked.
    Aquavit is good stuff indeed! An I soon might need some of it to calm myself
    down ;-) It is really really hot in Denmark these days and this switcher
    design of mine, does not exactly cool anything down. The bugger keeps
    getting hot beyond my wildest fears.

    I tried experimeting with the Rsense and I have found out that anything
    above 20mR in value makes my output not stable when input is below 24V.

    I have tried various types of mosfet that I have stocked, but no
    considerable change in the gate waveform came from this.

    I tried changing the inductors to 33uH, but still stuff gets really hot when
    I draw 1A.

    I have made a few new screenshots, where the scaling information is present,
    perhaps this could be of some help in the fine assistance I have already
    received.

    The shots are made from a running system with 20mR Rsense and 33uH inductors
    of the CDRH127 series.

    www.impc.dk/files/drain.jpg
    www.impc.dk/files/source.jpg
    www.impc.dk/files/gate.jpg
    www.impc.dk/files/fb_pin.jpg

    Thanking you in advance!

    Best regards
    Henrik
     
  15. Joerg

    Joerg Guest

    Haven't seen it this bad but one guy had a shirt full of holes after a
    dozen tantalums exploded. I do not use tantalums in designs, except in
    low current timing circuits. And even there only once in a blue moon.

    Look at the bright side: Once you've mastered a few designs like this
    you'll become quite the expert in SMPS design. There certainly aren't
    many of those engineers left in the industry. Many younger engineers
    just take an app note or sample design from a datasheet and when it
    doesn't work they give up. Others are so afraid of SMPS that they always
    buy "modules in a can".

    The high art later will be to build switchers without special PWM chips,
    just from logic chips and discrete parts. Can be real fun.

    20mohm still allows >8 amps. That is way too much for the inductors you
    chose.

    Instability for higher values: The switcher then goes into what's called
    continuous conduction mode or CCM. This can lead to right half plane
    zero instability, often called RHP-zero. If you really have to live with
    such small inductors you'll have to read up on the topic and change the
    feedback compensation. Read under slope compensation, page 10 of the
    LM3478 datasheet.

    Bottomline your 3.6A inductors are rather marginal for the power level
    you need. If you want to avoid CCM altogether you need much bigger ones.
    But you cannot use 20mohms with the higher value CDRH127 inductors,
    they'll saturate.

    This usually happens when the FET has really high Cgd. The LM3478 is not
    a very strong driver so you can't use FETs with Rdson in the range of a
    few milliohms, the are too large.

    Those inductors cannot take 8A or current peaks.


    This you need to crank up a lot more. It's hard to see the slope at
    Isense. Either ignore the spikes or connect to Rsense via a coax to get
    rid of them. Best to connect it where that 1000pF muffling capacitor on
    the sense line is.


    Looks like you are getting a lot of noise into that node. Do you have a
    full ground plane? The LM3478 needs that.
     
  16. Well, I try and keep myself focused on the silver lining, even when burning
    my fingers on various inductors ;-) The app-notes and sample designs are
    fine for ideas, but the true art of our trade is in understading, so only
    one way to go. Into the quagmire. I NEVER give up. being the youngest of 4
    boys taught me that. I get beaten sometimes, but I never give up, neither on
    this matter ;-) Aquavit to the rescue, though. As Mr. Sean Connery so
    brilliantly points out in "the Rock" : "Loosers always whine about [giving
    their] best, winners go home and **** the prom queen". Well, I am no Nicolas
    Cage, but I certainly did marry the prom queen, so no caving in.
    Well that sound like a job for the Picaso's of our esteemed trade ;-)
    I hope that I am not way off in my current understanding of the matters, as
    described in the following:

    I rechecked all my calculations once again and found out that 33uH should be
    sufficient with my chosen frequency. I then glued in a couple of TDH series
    inductors from Chilisin I had laying around. Lo and behold, they do not get
    hot at all! (of course, a little, but nowhere near "hot" like I am used to).
    However, the downside is that it it still only possible for me to get 24V
    out using no more than 20mR as Rsense. As soon as I crank up this to 50mR
    the output falls and varies with the input voltage. According to my
    calculations I have duty limits of respectively 0.44 and 0.62 for my input
    voltage limits, which is close to or above the 0.5 "risklimit" mentioned in
    the datasheet regarding feedback compensation.

    So if I understand correctly, I need bigger current capaple inductors, and i
    order to make my Rsense work, I need to determine a sufficient resistor to
    put into my current feedback.

    Maybe even if I get this feedback issue working, my CDRH127 serie would be
    alright afterall for the 1 amp output. I calculated my Iavg maximum to 3,1A
    which is still below the limit for these inductors.

    I can see that bumping the frequency reduces the ripple current, but I guess
    there might be some problems with e.g just bumping the frequency to 500kHz?
    I use 400KHz in the current setup.
    I will try a produce better screenshots. Hopefully my next shots will be of
    a better running system :)

    Thank you for your valued input!

    Best regards
    Henrik
     
  17. Joerg

    Joerg Guest

    Henrik [7182] wrote:

    [...]
    The most memorable comment at the beginning of a design review when this
    slide came up: "What the hell does that circuit do?"

    I haven't run your numbers but you may be right in that you really do
    need less than 50mOhms. That was just a suggestion to see whether it
    will run the CDRH127 without them becoming hot. Since the LM3478 is a
    current mode switcher and Vsense is 165mV your 20mOhms means that the
    inductor must withstand 8.25A plus some margin. No matter how you toss
    and turn that you'll need a 10A inductor there for peace of mind.

    Still I think something isn't quite right here. Just key the numbers
    into the boost model below, for example 15V in and 24V/1A out. The
    inductor shouldn't go beyond 2A.

    http://schmidt-walter.eit.h-da.de/smps_e/aww_smps_e.html

    AVG doesn't matter. What matters is Ipeak. The instant the current
    exceeds the limit for core saturation the inductance collapses. Still
    puzzled why you need this much when your lowest input voltage is 15V.

    400kHz should be ok here. Your FET switching losses are quite high,
    probably too big a FET. With higher frequency that'll become worse.

    Oh, they were ok this time, just crank up the scope gain for the Vsense
    node.
     
  18. legg

    legg Guest

    The smallest of the TDH series (open form) bobbin inductors has double
    the surface area and two or three times the magnetics x-section of the
    low profile closed parts originally used. Neither parts give much core
    material loss information. Losses with accumulate from both features,
    with core loss having a high positive temp coefficient above 60-80degC
    (depending on material type and grade).

    You should find that the duty cycle is more highly dependant on load
    variations.

    Scoping with single traces at close to 50% duty makes visual
    correlation a problem. Try to trigger on the same portion of the
    operating period or wave edge, regardless of the display; if neccesary
    by taking and uncalibrated signal from the pwm oscillator to the
    direct trigger input.
    The switch carries the current of both inductors.


    The apparent lengthy (250nSec?) negative-going excursion in the
    current signal (at the source)during turn-off may be an artifact of
    the scope digitization. You need to look at it with a shorter
    time-base and a shorter probe ground clip.

    EL
     
  19. JosephKK

    JosephKK Guest

    Mmmm. Aquavit. Once in a while i get some Aalborg Jubilamus variety
    aperitif, sugary but very tasty.
     
  20. Joerg

    Joerg Guest

    True, they don't have a SEPIC sim. But the 2nd inductor doesn't carry
    that much. I really don't see where 8A peaks should be necessary. I had
    to swing a few designs from boost to SEPIC because clients revised the
    specs in mid flight. That has not caused any serious changes in peak load.

    Or with a real scope. SCNR :)

    I am not all that happy with those Tek lunch-box size scopes. Often I
    advise clients to go on EBay and bid on a Tek 2465. When it arrives it's
    like as if someone had turned the lights on.

    The other downside that hits in cases like SMPS designs is that many
    Teks only have 2.5K memory. Not enough. My scope has 25K so I can
    trigger, digitize at full bore and then blow up individual regions to
    look at a spike, zoom back out to see whether my RHP-zero situation is
    kosher, and so on.
     
Ask a Question
Want to reply to this thread or ask your own question?
You'll need to choose a username for the site, which only take a couple of moments (here). After that, you can post your question and our members will help you out.
Electronics Point Logo
Continue to site
Quote of the day

-