48V Point of Load regulators, the time is now.

Discussion in 'Motherboards' started by KD5ZXG, Jun 15, 2017.

  1. KD5ZXG

    KD5ZXG n00bie

    Mar 24, 2017
    Vicor, ST, TI, seems like everyone and their 1st cousin are ready and see this coming but us.
    Problem is power wasted by hot [H]ard switching regulators each time they open and close.
    Yeah, you can have lots of phases, and spread the problem out over an area, but it still adds
    up the same wasted power.

    There is built in parasitic capacitance that discharges like an unseen spark at every closure.
    Thats a problem like a built-in bomb, but the capacitance is also necessary (and sometimes
    more intentionally added), cause it protects the switch from blowing when it reopens due to
    current stored both by necessary and parasitic inductance. Dual purpose bomb and pillow...

    Solution is a resonant power converter that provides periodic opportunities to switch at near
    zero current or voltage. The cap protects the switch at opening, the inductor discharges the
    cap before closing. A dance that is sometimes hard to balance, especially if the load is not
    a steady draw. But when done right, wastes almost no heat.

    Don't actually need 48V to do this, 12V rail works fine too. But 48V gives some advantage.
    Capacitance can be made non-linear with higher voltage swings, and that gives more pillow
    time to catch the voltage while its near each rail without adding a lot of extra stored energy.
    And that means it could maybe operate without heatsink over a much wider range of load.

    I could elaborate to the point of schematics, but only if doing so would not be wasting your
    time. Comments, suggestions, dirty jokes???
    Last edited: Jun 15, 2017
  2. KD5ZXG

    KD5ZXG n00bie

    Mar 24, 2017
    25 views and not a single bite... Topic not interesting, or just need a picture?
    Granted VIcor's ZVS solution looks less klunky than others for the moment.
    But all of them, no heatsink! When you can switch at zero, none is needed.
  3. TheHobbyist

    TheHobbyist Hugs Hard Johnnies [H]ard

    Apr 8, 2008
    Beyond my understanding, but interesting. I hope someone chimes in.
  4. Arcygenical

    Arcygenical Will Watercool for Crack

    Jun 10, 2005
    Certainly an interesting concept but it does have a few drawbacks...

    For one, practicality? Outside of extreme current demands I don't really believe the board power delivery is a huge consideration for most people... What's the bucking efficiency down from a higher voltage? Even a shitty buck converter I built with scrap parts measures around 73% when tuned properly. I personally am not comfortable working around 48v in a consumer desktop, I do too much modification to PSUs and Cases.
  5. Gman1979

    Gman1979 Gawd

    Sep 21, 2011
    I feel, after seeing just how craptastic some of the power designs for AM4 boards has been, if this single chip can make up for some of the bs we see being used then bring it please. Judging by past experiences though, the board manufacturers would probably use it as a band aid over even worse voltage regulation and supply on motherboards :(

    You also gain efficiency by using 48 volts compared to 12 volts. It's similar to how a high pressure sodium light uses less electricity when using a 220 volt ballast compared to a 115 volt one, even if this example uses AC instead of DC.
  6. Arcygenical

    Arcygenical Will Watercool for Crack

    Jun 10, 2005
    Right, but does the efficiency gain actually warrant the price of the current package. TI is selling these converters at 52$ a pop, and sadly the 1v load point eval board (which I would really like to pick up) is in the 300$ range.

    There are no appreciable power savings from using 220v versus 110v ballasting for HPS, though... Unless you're using very shoddy wiring, or an improper ballast factor.

    Also, from further reading, and looking at the power dissipation losses which increase proportional to Fsw for these (and all buck type) regulators... And the required increase in coil size needed to minimize Fsw, I'm just wondering if the efficiency gains are remotely worth the component cost.

    I can absolutely see these being used in extreme overclocking scenarios though. A few in parallel could supply hundreds of watts at low losses for LN2 runs.
    Last edited: Jun 19, 2017 at 8:32 PM
  7. Gman1979

    Gman1979 Gawd

    Sep 21, 2011
    On the same 1000 watt ballast, switching the input voltage from 115 to 220 resulted in around a 15% reduction in Kwh for that greenhouse as shown on the electric bill. The difference may not be the same if using a solid state ballast though. The wiring didn't seem to be an issue as the site was commercially growing flowers and young fruit trees for ages and ages without issue. The entire thing was overbuilt to allow further expansion over time. There also seemed to be a slight increase in bulb lifespan when running on 220, supposedly from a difference in the harmonics inside the bulb itself.
    Last edited: Jun 19, 2017 at 8:36 PM
  8. ecmaster76

    ecmaster76 Gawd

    Feb 6, 2007
    Its separate calculations for load and loss power (different R values) but writing subscripts on here sounds like too much work

    Power losses scale linearly with voltage but exponentially with current. Increasing voltage reduces current required for equivalent power
    Doubling voltage decreases current by half
    (0.5*I)^2 = 0.25*I^2

    So you reduce power losses in the system by up to 75%. More than that if you are running air conditioning. ;)
    In example a system that is 80% efficient at 110v would be up to 95% efficient at 220v

    Gman1979's example is very believable if most of the inefficiencies were

    EDIT/TLDR - 110v sucks - we only use it residential installation so people don't fry themselves as easily. Everywhere else we use much higher voltages whenever practical
    Last edited: Jun 20, 2017 at 9:48 AM
  9. KD5ZXG

    KD5ZXG n00bie

    Mar 24, 2017
    120 or 220 AC/DC conversion is a totally different problem than 48VDC/DC.
    You have concerns with power and crest factor. Also crossings locked out by
    undervoltage. Most lighting converter power and crest factors are crap.

    Unfortunately most of the systems I've measured are properly tuned to only
    one voltage and frequency, though most will function well outside that range.
    I don't understand the full reason why they can't properly adjust themselves.
    But they don't let me design the damn things, just test em...


    TI EVM is spendy cause its bleedin edge Gallium Nitride. But ZVS doesn't
    require GAN, and doesn't require 48V. It just requires the willingness to put
    some time doing genuine engineering.

    Best transistor today for ZVS is an insanely fast switch like SIC in parallel
    with an extremely non-linear capacitance. Perhaps a silicon Superjunction
    MOSFET with the gate permanently wired OFF, just abuse as a capacitor.

    On the cheap cheap, you could use the Superjunction as both switch and
    cap. Though superjunction gate capacitance is almost too slow to advantage
    any extra hang-time it's non-linear drain capacitance creates. Sad these are
    better used for capacitor (and body diode) than as switches.

    Infineon had a white paper explaining the capacitance trick, I forget the link.
    Was probably an appnote regarding the C7 series, Huge useful non-linearity.
    When voltage flails resonantly toward the other rail, it tends to stick there
    where its easy to catch, as capacitance baloons at low voltage difference.

    Superjunction non-linear knee is around 20V, so maybe 12V systems won't
    be able to stretch the load range using the non-linear cap trick. One of a few
    logical arguments why 48V and up might be better. But there are other ways
    to stretch the ZVS load range. Inductors can be made non-linear to current.

    Some day, GAN will catch up and surpass Silicon and SIC. Maybe...
    Neither GAN nor SIC have much in the way of useful drain capacitance.
    The race is on right now for less of every capacitance, which is good for
    hard dumb switching. But not an optimal solution to ZVS. Some external
    capacitance will have to be added in almost every case.
    Last edited: Jun 21, 2017 at 12:40 AM