SMPS: Safe Operating Temps?

Discussion in 'General Electronics Chat' started by apqo1, Nov 22, 2013.

  1. apqo1

    Thread Starter Well-Known Member

    Oct 5, 2008
    50
    3
    I'm working on a DC-DC buck circuit. I have it working well, with ringing and overshoot on the switched node well controlled, stable output, good transient response, etc. My problem is temperature.

    The converter IC is a 6A-rated part with internal high-side FET, driving a 13A-rated external low-side FET and 11A-rated inductor. Max design load current is 4.8A. Max design ambient temp is 50°C, but no testing has yet been done above 25°C.

    Minimum board size is a major design goal.

    All of the parts have -40 to +125°C temperature ratings, except for a few X5R and NP0 capacitors. All caps are MLCCs.

    I'm seeing case surface temps ~70°C (measured with a thermocouple, confirmed with an IR gun) on the converter IC, FET and inductor.

    I experimented with a heatsink on the back side of the board (exposed copper ground plane, connected to top side with numerous thermal vias), but it only brought temps down ~10°C. Hardly worth the expense in BoM cost and manufacturing complication.

    For manufacturing, I plan to use a DFN package with slightly better thermal performance for the converter IC (proto uses an 8-SOIC/EP), and specify 2-oz copper (proto has 1-oz copper), but I don't expect huge improvements from these changes.

    So, my questions are these:
    - What are safe and sensible operating temperatures for the converter IC, FET and inductor?
    - How much worse are the temps likely to get at the design 50°C ambient?
    - Am I chasing the last degree of improvement for no good reason?
    - Generally speaking, what sort of MTBF penalty is paid as temperature rises?

    Thanks.
     
    Last edited: Nov 25, 2013
  2. crutschow

    Expert

    Mar 14, 2008
    13,009
    3,233
    It's the junction temperature that's the crucial value so you need to account for the amount of power dissipation and the junction-case thermal resistance to determine how much higher the junction temperature is over the case temperature.

    I would generally want to keep the junctions temperatures below 100°C for good reliability

    The temperatures you measured will generally go up directly with ambient temperature, so a 50°C ambient will cause the case and junction temperatures to go up 25°C over those you measured at 25°C.

    I believe the failure rate varies exponentially with temperature so a 10°C change in temperature can have a significant effect on the MTBF.
     
    apqo1 likes this.
  3. bountyhunter

    Well-Known Member

    Sep 7, 2009
    2,498
    507
    Safe maximum is about 110 - 120C JUNCTION temperature...... not case surface, but JUNCTION. That requires info on the package thermal resistance of each part. The junction is always a lot hotter than the surface of the package. Obviously, it's better not to run continuously at the maximum temp.



    It will be roughly linear: all temps at 50C ambient should be about 25C higher than what is measured at 25C.

    I don't know, it depends how hot the junction temps are.


    The rule of thumb we were forced to accept was that an IC's life is reduced by 1/2 for each 10C rise in temp. That was much disputed, and no convincing proof of it was ever seen. Life is reduced at higher temps, but it's not a simple formula.

    We found that when ICs are WELL MADE (no contamination/built in defects) they have very long lives at any op temp up to maybe 120C. At least, they live longer than the product itself (like hundreds of thousands of hours). Parts that die typically have defects in the manufacturing process. We used to do 100 hour burn-ins at 125C to intentionally kill of the defective parts (called infant mortality) and the parts that remained were very long lasting.
     
    apqo1 likes this.
  4. THE_RB

    AAC Fanatic!

    Feb 11, 2008
    5,435
    1,305
    What efficiency, inductor value and operating frequency are you using?

    Most beginners use too high a frequency and with switching losses that will lower efficiency and cause things to run hot.

    Or maybe they use too small an inductor value, causing it to run close to saturation which again reduces efficiency.
     
  5. apqo1

    Thread Starter Well-Known Member

    Oct 5, 2008
    50
    3
    Crutschow and bountyhunter, thank you both. Excellent information. I did some searching since posting my questions and found this, the first few pages of which seem to confirm what you're both saying about temps and their effect on device life.

    So, assuming I have accurate case temp measurements, is it possible to work backwards to estimate junction temp?

    The datasheet for the converter IC lists θjc = 15°C/W and θja = 75°C/W. Let's say ambient temp = 25°C and case temp = 70°C. Is Tj estimable?

    Note that the datasheet specifies that θjc/θja are determined on a JEDEC 51-7 standard 4-layer test board. Since my board is smaller and 2-layer, I presume that I can't calculate Pd from my temp measurement. Also, the datasheet gives no guidance on calculating Pd for a given operating condition.

    Thanks again.
     
    Last edited: Nov 25, 2013
  6. apqo1

    Thread Starter Well-Known Member

    Oct 5, 2008
    50
    3
    THE_RB,

    Efficiency is estimated at ~89%, inductor value is 3μH (as suggested by the datasheet) and Fsw is fixed at 600kHz.

    Inductor ripple current has been measured (see attached image). Inductor saturation rating is 6A.

    Thanks for jumping in!
     
    Last edited: Nov 25, 2013
  7. bountyhunter

    Well-Known Member

    Sep 7, 2009
    2,498
    507
    If there is a defined "case" surface and you can measure it and you can accurately determine the average power dissipation, yes. That is not easy to do.


    That is always a problem with surface mount components. We used to provide an assortment of PCB layout patterns with θja values we determined by using the body diode VBE.
     
  8. apqo1

    Thread Starter Well-Known Member

    Oct 5, 2008
    50
    3
    Sadly, the case surface they specify is the exposed pad, which of course is soldered to the board. I did use a 0.050" via in the pad that is soldered through, so I can get a fairly decent measurement. That said, I have no way to quantify the effect of my board on that measurement, and as I mentioned, it's vastly different than the JEDEC board used by the mfg.

    As for an accurate Pd value, I don't know. If it were a TI part, their WEBENCH tool could get us close, but it's not.

    I've read about that technique of using the known temp characteristics of a diode in the die to measure temp, but I wouldn't have a clue how its done in practical terms!
     
  9. #12

    Expert

    Nov 30, 2010
    16,284
    6,797
    Ten Characters.
     
    apqo1 likes this.
  10. crutschow

    Expert

    Mar 14, 2008
    13,009
    3,233
    For many devices the power can be estimated by the current draw from the spec sheet times the power supply voltage.
     
  11. RamaD

    Active Member

    Dec 4, 2009
    254
    33
    But θca is not known, and estimating θca is tough, as there is 1 Oz of copper acting as heatsink. The connection to the copper island and the layout would determine θca.
     
  12. #12

    Expert

    Nov 30, 2010
    16,284
    6,797
    For a transistor in free air, θCA is known
    Given: θJC = 15 C/W
    Given: θJA = 75 C/W
    75 - 15 = 60 C/W θCA

    I have illustrated the method of modeling a thermal resistance circuit. That method provides a starting point, specifically, the worst case limit. Therefore, I believe I contributed a useful piece of information.

    For instance, if the true wattage is known, the true θCA can be calculated.
     
  13. ronv

    AAC Fanatic!

    Nov 12, 2008
    3,290
    1,255
    If you are shooting the plastic case with the IR gun you are closer to junction temperature than case temperature. Case temperature is the tab.
    I'm not sure why the converter should be warm if it is just driving a FET.:confused:
     
  14. apqo1

    Thread Starter Well-Known Member

    Oct 5, 2008
    50
    3
    #12:

    Thank you.

    That model gets me a lot closer to understanding what's going on that I was before. Even if it ignores differences in the boards, it looks like a pretty good, if inexact, estimate.

    -----

    ronv:

    Thanks for chipping in.

    It was my understanding that the plastic case material used to encapsulate ICs is a poor thermal conductor, and that most heat is conducted away from the die by the legs, tab or exposed pad.

    In this case, the IC is an 8-SOIC/EP and the manufacturer specifies the exposed pad as the place to measure case temp.

    This converter has the high-side FET built in, resulting in higher Pd than you would see in a comparable controller. It also drives an external low-side FET.
     
    Last edited: Nov 25, 2013
    #12 likes this.
  15. #12

    Expert

    Nov 30, 2010
    16,284
    6,797
    Fairly often on this site, a seemingly brilliant customer just needs to be pointed back to a basic concept to get him on the right track.
     
  16. apqo1

    Thread Starter Well-Known Member

    Oct 5, 2008
    50
    3
    Indeed, though I'm seldom mistaken for brilliant, seemingly or otherwise!
     
  17. THE_RB

    AAC Fanatic!

    Feb 11, 2008
    5,435
    1,305
    89% is low for a modern SMPS buck with synchronous rectification! I would expect 93-96% at mid currents.

    Also datasheets for new SMPS ICs like to push tiny inductor values (and very high freq) to be trendy and modern and gain some size benefits but at the expense of efficiency.

    The good news is your inductor current waveform looks to be not saturating, but of course at 600kHz your switching losses will be high as they are directly proportional to freq.

    If you can, 'scope the switching waveform and find the nS of each slope, / and \. That will give you a rough idea of switching losses as percentage of the total losses. The other way is to measure the Rds ON of the two FETs, and the DC resistance of your inductor, and calc the static ON and OFF period losses and subtract from the total efficiency, the remainder of course will be the switching losses.

    My suggestion is to question the datasheet inductor size suggestion and see if you can go up a bit in inductance, and then increase the timing cap to reduce operating frequency. Push the frequerncy as low as you can without the inductor saturating (you will easily see that on the 'scope).

    And changing from 1oz to 2oz copper will make a very significant difference in how cool the SMD switching devices run, especially if you have >= area to the recommended cooling pad area.
     
    apqo1 likes this.
  18. bountyhunter

    Well-Known Member

    Sep 7, 2009
    2,498
    507
    It's a switcher with internal drivers. A high peak current (sometimes a couple of Amps) is needed to snap the gate of the FET on and off quickly. That current pulse flows through the on resistance of the internal driver at the switching speed, that's what generates the heat.
     
  19. kubeek

    AAC Fanatic!

    Sep 20, 2005
    4,670
    804
    The thing is that there will be quite a large difference between what you measure with a thermocouple on the case and what you measure with an IR based thermometer or camera.
    Just last week I was doing this, and the thermocouple showed 55°C in the hottest spot and infra camera showed 63 in the same spot, on a ~230pin package. It seems the IR light can pass through the case material better than the conducted heat, thus fooling such measurements.
     
    Last edited: Nov 24, 2013
  20. #12

    Expert

    Nov 30, 2010
    16,284
    6,797
    Have infrared cameras improved so much that emissivity of the surface is not a problem now? Is the case material transparent to infrared? Should I start my own thread on this?
     
Loading...