Paralleling two MOSFETS to handle more current.

photomankcphotomankc Posts: 929
So I am working on a soft-switch for my robot project. It involves a PICAXE and a P-Chan MOSFET used to switch the high-side of my main power supply based on a user button press or the RaspberryPi sending a power-down signal. I would like to have plenty of head-room in current capacity and I have two options on hand:

- One is giant-sized SMT MOSFETs that are rated to 25A or something stupid high like that. I don't think I could reasonably even make traces to handle that but it would run most any project I have without a sweat.
- The other is an SOT-23 package that is rated to 3.5A peak.

What I thought would be an ideal match is to parallel two of the smaller ones for 7A combined and that would be 75% headroom over what my peak draw is right now.

My question is.... is it really that simple? Can you just parallel the MOSFETs and drive the two gates the same? I know often times that does not work out as intended as one component is ever so slightly different and ends up taking on all the load, fails and then the next in line repeats it's demise. Reading some posting seems to suggest that this is not the case with MOSFETs but I wanted to ask directly before I start working up a PCB.

Comments

  • 17 Comments sorted by Date Added Votes
  • ElectricAyeElectricAye Posts: 4,561
    edited March 2013 Vote Up0Vote Down
    photomankc wrote: »
    ...
    - One is giant-sized SMT MOSFETs that are rated to 25A or something stupid high like that. ....

    Can you use an IRF3708? They're rated at 62 Amps. They are N-channel, however. Does that make a difference for your application?
    http://www.digikey.com/product-search/en?x=15&y=12&lang=en&site=us&KeyWords=irf3708

    And, instead of traces, could you just wire the heavy current lines?
  • Mike GreenMike Green Posts: 22,756
    edited March 2013 Vote Up0Vote Down
    MOSFETs are different from junction transistors in that they will "share" a load without any special circuitry. As one MOSFET gets hot from running too much current through it, its resistance rises and the current takes the "path of least resistance" and preferentially goes to the other device. That said, I'd go with the larger single device. The on-resistance is likely to be lower and there'd be less heat produced for a given amount of current. You don't have to make traces rated for 25A unless you plan to use that much current.
  • LoopyBytelooseLoopyByteloose Posts: 12,537
    edited March 2013 Vote Up0Vote Down
    I have seen Lithium cell controllers for electric bikes with the main switching having as many as 5 MOSfets in parallel to handle the current. Yes, it is that simple.

    About the only reason you might see the practice disappear is that MOSfet have continued to grow larger and larger in capacity and what once required 5 in parallel can now be done by one.
    Hwang Xian Shen, Puddleby-on-the-Marsh.
    All things considered, I can live and thrive without Microsoft products. LINUX is just fine.
  • photomankcphotomankc Posts: 929
    edited March 2013 Vote Up0Vote Down
    Can you use an IRF3708? They're rated at 62 Amps. They are N-channel, however. Does that make a difference for your application?
    http://www.digikey.com/product-search/en?x=15&y=12&lang=en&site=us&KeyWords=irf3708

    And, instead of traces, could you just wire the heavy current lines?

    I strongly prefer to switch the high side for this application. I don't like to have a circuit (robot that I'm playing with) that is live just waiting for an accidental ground to be found when it's supposed to be off. I don't think there is any other compelling reason beyond that and the fact I prefer it conceptually.

    I can go on and use the bigger part I guess. Might not save a whole ton of real-estate vs the tracing out parallel SOT-23s anyway. No plated via's in Kyle's home PCB lab. 25A is plenty for this application.
  • jmgjmg Posts: 12,094
    edited March 2013 Vote Up0Vote Down
    photomankc wrote: »
    My question is.... is it really that simple? Can you just parallel the MOSFETs and drive the two gates the same?

    Pretty much that simple, but you should also do an inrush and thermal budget, and often that selects the device more than rated currents.
    For Thermal you use i * i * R powers, to check what temperature rise you can expect.
    Quite often designs use FETS rated on paper > 10x current, to get the heat under control.

    Inrush budgets involve asking "Who will be the fuse?" and checking startup/surge/stall current levels, which can be significant on any Motor load.

    There are high side Switch devices out there, that include a FET ( P, or N+Charge Pump) and thermal and current sense elements.
  • User NameUser Name Posts: 1,451
    edited March 2013 Vote Up0Vote Down
    I love Mike's advice. Very sound.

    Sometimes we do funny things, though. I recall an Aussie ham radio operator who wanted a power amplifier for his 6M transmitter (50MHz). He boldly went forth and paralleled a bunch of commodity MOSFETS of the sort one might find at Radio Shack. Worked great. IIRC, he ended up with a power output around 250 W.

    Edit: I found the article here. He used IRF510 MOSFETs.
    Platåberget
  • Mark_TMark_T Posts: 1,684
    edited March 2013 Vote Up0Vote Down
    photomankc wrote: »
    So I am working on a soft-switch for my robot project. It involves a PICAXE and a P-Chan MOSFET used to switch the high-side of my main power supply based on a user button press or the RaspberryPi sending a power-down signal. I would like to have plenty of head-room in current capacity and I have two options on hand:

    - One is giant-sized SMT MOSFETs that are rated to 25A or something stupid high like that. I don't think I could reasonably even make traces to handle that but it would run most any project I have without a sweat.
    - The other is an SOT-23 package that is rated to 3.5A peak.

    What I thought would be an ideal match is to parallel two of the smaller ones for 7A combined and that would be 75% headroom over what my peak draw is right now.

    My question is.... is it really that simple? Can you just parallel the MOSFETs and drive the two gates the same? I know often times that does not work out as intended as one component is ever so slightly different and ends up taking on all the load, fails and then the next in line repeats it's demise. Reading some posting seems to suggest that this is not the case with MOSFETs but I wanted to ask directly before I start working up a PCB.

    You would never run a MOSFET at its rated current continuously unless prepared to spend money on big heatsink and fan or liquid cooling,
    the max current rating is almost invariably the abs max power rating expressed as a current. Peak loads that really are occasional peaks are
    OK. I've seen 25A MOSFETs that can handle 25A if dissipating 100W - not my idea of a sensible way to do things though!

    Always use the Rds(on) to calculate the heat dissipation for your load current and decide that way - you need to get a handle on how much
    heat the package can dissipate with various heat-sinking options and take that into account too. And then decide if paralleling up devices
    is worthwhile (doubling up halves overall power dissipation, reduces package dissipation by 4). Sometimes just getting the right device
    is simpler (A nice 3 milliohm Rds(on) FET can solve a lot of problems, but it still won't handle 25A without a heatsink - typically such devices are
    rated for 170A, but that's 90W of dssipation!)

    Don't go for devices with over-generous voltage ratings (Vds), since they have much higher Rds(on) for the same money. Pick Vds about
    twice your supply for a good safety margin but no more.

    [ edit: take the mighty IRFS3107-7P, rated at 2.1 milliohm and 260A - yet AFAICT the bonding leads to the chip melt at 240A - the datasheet calls
    this "package limited" !! ]

    [edit: to answer the original question - so long as you parallel devices of the same part number there shouldn't be issues]
  • evanhevanh Posts: 5,445
    edited March 2013 Vote Up0Vote Down
    All good advice. Definitely don't operate a component near it's ratings, including the wattage. Always design with headroom.

    As has been noted, 25 amps is not a large current for a single power rated mosfet, so you certainly don't need to go multiple devices.

    If you do want to use multiple smaller mosfets then there is a little trick with the gate control that should be observed. The turn-on threshold voltage is a little different for each mosfet. If the gates are all shorted together, just like the sources and drains are, then the one with the earliest threshold will take the higher initial switching load.

    So to keep the switching currents more evenly distributed it's advised to use a low value series resistor on each gate. Basically the same as what is done on a single mosfet but having a resistor on each individual mosfet gate.
    "Are we alone in the universe?"
    "Yes," said the Oracle.
    "So there's no other life out there?"
    "There is. They're alone too."
  • photomankcphotomankc Posts: 929
    edited March 2013 Vote Up0Vote Down
    Wow thanks for all the good advise here. Really the switch is to be designed as logic-system power switch. I do not envision trying to feed multiple power hungry motors directly through it. just the Raspberry, and any uC / sensors / and MAYBE small servos. I would likely plan to use relays for the really major motor loads in my outdoor bot. This is just the master on switch so the system can protect the batteries. I did think that the issues with heating were generally more pronounced in PWM scenarios where the FET spend more time in the no-mans land between off and on.

    I know that you can not operate at the absolute max current on anything and expect longevity but was not aware that 7 to 10 times could be needed. I thought 2 to 3 would be reasonable. I was thinking that the 25A would be good for 5A to maybe to 8A without extraordinary measures. However, reality is the RaspberryPi has a nominal 500 to 700mA draw and I can't see any extra's I'd add topping double that so I think the 25A I have on hand would handle that with little heating. Then the main power switch-on could trigger relays to power up the seperate H-bridges for the motors.
  • jmgjmg Posts: 12,094
    edited March 2013 Vote Up0Vote Down
    photomankc wrote: »
    I know that you can not operate at the absolute max current on anything and expect longevity but was not aware that 7 to 10 times could be needed. I thought 2 to 3 would be reasonable. I was thinking that the 25A would be good for 5A to maybe to 8A without extraordinary measures.

    You need to also consider thermal budgets. Some numbers :
    If we take a '25A' fet as being a nominal 80 milli-ohms, these are the power points

    8A => 8*8*.08 = 5.12W << serious heatsink needed, and ~640mV of DC drop may be a problem
    3A => 3*3*.08 = 0.72W << Still needs large PCB copper, but might save a heatsink.
    2A => 2*2*.08 = 0.32W << getting down to SMD parts, without excess PCB area


    So you can see why a designer might choose a '25A' FET, but run it at under 2A
  • photomankcphotomankc Posts: 929
    edited March 2013 Vote Up0Vote Down
    Ok, that's eye opening. I see those 3.5A parts are out of the question for this. I'll check out the numbers on the bigger part and see what I get from that.
  • photomankcphotomankc Posts: 929
    edited March 2013 Vote Up0Vote Down
    Ugggh so reading the specs again I may have an issue. The max drain-source voltage is 20V. The nominal Vbatt would be 12V. The gate-source max is +/-8V. If I read that right then I must not pull the gate voltage to lower than 4V or I exceed the rated Vgs. Right now I'm pulling the gate right down to zero on a different breadboard part but it has a far larger Vgs rating. I guess a proper resistor divider can be constructed to make sure that when the gate drive transistor is switched on that the voltage is only dropped to 5V rather than 0V? Or am I lost there? I always thought MOSFETs were conceptually simpler than BJT's but then when you got to the nitty gritty they are more complicated than they seem. The datasheets are a bear.

    This is the large guy I have on hand:
    http://www.fairchildsemi.com/ds/ND/NDP6020P.pdf

    Vds -20V
    Vgs +/-8V
    Ids (cont) -24A
    Ids (pulse) -70A
    Rds @ -4.5V 0.050


    So following your calculations:

    1A ==> 1 * 1 * 0.05 = 0.05W Easy
    2A ==> 2 * 2 * 0.05 = 0.20W Ok
    4A ==> 4 * 4 * 0.05 = 0.80W Harder
    8A ==> 8 * 8 * 0.05 = 3.20W Hard
    12A => 12 * 12 * 0.05 = 7.20W Holy crap!
  • photomankcphotomankc Posts: 929
    edited March 2013 Vote Up0Vote Down
    Well I experimented this weekend. I removed the really horrible P-Channel I sourced locally and used the NDP6020 in it's place. I went on the assumption that I really do need to make sure that Vgs doesn't exceed 8V even if the battery is at 12V. To that end I created a voltage divider at the gate so that when it is switched on the voltage is is only dropped about 6.6V below the source for a 12V input voltage. It should tolerate 6.5-14V pretty well although you start to lose turn-on mojo at low voltages. I created a spreadsheet to calculate the wattage, voltage drop, temperature rise, and derated watts based on the data-sheet info. Seems to suggest I could get this particular part to 4-5 amps with decent use of copper as a heat-sink. More than that and I would just use the through-hole part and mount a good heat-sink. I think 3-5A would be a generous power budget for most RaspberryPi hobby projects so I'm going to give the SMT part a whirl and make a home-made board to test it out with. Now I need some type of high power load testing device so that I can dial up the current demand to see if reality matches the calculation.

    I'm working up the Schematic now.
  • jmg wrote: »
    photomankc wrote: »
    I know that you can not operate at the absolute max current on anything and expect longevity but was not aware that 7 to 10 times could be needed. I thought 2 to 3 would be reasonable. I was thinking that the 25A would be good for 5A to maybe to 8A without extraordinary measures.

    You need to also consider thermal budgets. Some numbers :
    If we take a '25A' fet as being a nominal 80 milli-ohms, these are the power points

    8A => 8*8*.08 = 5.12W << serious heatsink needed, and ~640mV of DC drop may be a problem
    3A => 3*3*.08 = 0.72W << Still needs large PCB copper, but might save a heatsink.
    2A => 2*2*.08 = 0.32W << getting down to SMD parts, without excess PCB area


    So you can see why a designer might choose a '25A' FET, but run it at under 2A
    Sir i am a beginner, could you explain me about the voltage drop behind the power dissipation.
    Sir you said that for 5.12W, there will be 640mv voltage drop. Can you explain me , is any mathematical calculation behind that voltage drop.
  • jmgjmg Posts: 12,094
    Budd wrote: »
    jmg wrote: »
    photomankc wrote: »
    I know that you can not operate at the absolute max current on anything and expect longevity but was not aware that 7 to 10 times could be needed. I thought 2 to 3 would be reasonable. I was thinking that the 25A would be good for 5A to maybe to 8A without extraordinary measures.

    You need to also consider thermal budgets. Some numbers :
    If we take a '25A' fet as being a nominal 80 milli-ohms, these are the power points

    8A => 8*8*.08 = 5.12W << serious heatsink needed, and ~640mV of DC drop may be a problem
    3A => 3*3*.08 = 0.72W << Still needs large PCB copper, but might save a heatsink.
    2A => 2*2*.08 = 0.32W << getting down to SMD parts, without excess PCB area


    So you can see why a designer might choose a '25A' FET, but run it at under 2A
    Sir i am a beginner, could you explain me about the voltage drop behind the power dissipation.
    Sir you said that for 5.12W, there will be 640mv voltage drop. Can you explain me , is any mathematical calculation behind that voltage drop.

    Yes, the "mathematical calculation" is based on Ohm's Law, and is the 8A*8A*.08Ω I wrote above.
    More can be found here
    https://www.evilmadscientist.com/2012/basics-power-dissipation-and-electronic-components/
  • photomankc wrote: »
    ........ Now I need some type of high power load testing device so that I can dial up the current demand to see if reality matches the calculation.

    I'm working up the Schematic now.

    You can use a 2N3055 transistor and a heat sink along with a few additional components to build a simple adjustable constant current circuit that makes a good adjustable load. Got one kicking around somewhere that I have used occasionally for many years.
    In science there is no authority. There is only experiment.
    Life is unpredictable. Eat dessert first.
Sign In or Register to comment.