Page 1 of 2 12 LastLast
Results 1 to 10 of 12

Thread: The Integrated Voltage Regulators

  1. #1

    The Integrated Voltage Regulators

    This thread is to talk about and confirm IVR technology.

    IBM's POWER8 iVRMs is using Switched Capacitors Regulators, correct?
    https://i.imgur.com/wpdLy7B.png

    Intel's current FIVR is using Low Drop Out(LDO) Linear Regulators, correct?
    https://i.imgur.com/rGaRxaM.png

    AMD's upcoming IVR is using Buck Converters, correct?
    https://i.imgur.com/43BaOn5.png

  2. #2

  3. #3
    Quote Originally Posted by Seronx View Post
    This thread is to talk about and confirm IVR technology.

    IBM's POWER8 iVRMs is using Switched Capacitors Regulators, correct?
    https://i.imgur.com/wpdLy7B.png

    Intel's current FIVR is using Low Drop Out(LDO) Linear Regulators, correct?
    https://i.imgur.com/rGaRxaM.png

    AMD's upcoming IVR is using Buck Converters, correct?
    https://i.imgur.com/43BaOn5.png
    Wow, didn't realize that there were plans to integrate voltage regulators on the silicon. Makes sense for handheld and practical if Memory is HBM (power efficient) and 3D attached to Logic using the same supply. So I guess this was predictable.... Makes FD-SOI more attractive doesn't it?

  4. #4
    i guess this is required for the near threshold voltage regulation. which intel and amd want to adopt. it also improves reaction times for voltage clock scaling in there processors.

    also how can intel have a higher efficiency with those linear power regulators doesn't it just convert all the unused energy to heat? efficiency = Vout/Vin.
    or does the leakage current and capacity charge get lowered enough to make it more power efficient?

  5. #5
    Quote Originally Posted by jamie965 View Post
    didnt intel is going to abandon ivr?
    http://wccftech.com/intel-abandon-in...oarchitecture/
    The IVR is cool when it comes to low cost mobile stuff since it simplifies the power delivery circuitry to the processor die significantly. Ideally if the input voltage to the processor could be something like 5V, there's a lot of the motherboard that could be done away with.

    But I can see it being a problem for the high power stuff. Run a power calculation for an LDO-style regulator:

    Vin = 1.25V (Motherboard supplied)
    Vout = 1V (Internally regulated down - just a guess figure)
    TDP = 100W (just to pick a number out of the air)
    =>Iin = 100A
    Power(LDO)=(1.25V-1V)*100V = 25W

    So using that extremely crude calculation you've an extra 25W dissipated on the processor die which will raise the junction temperature accordingly. It's no wonder the Haswell TDPs actually went up relative to Ivy Bridge. I also wouldn't wonder why the temperature issues for the high-end desktop chips are as prolific as they are at higher motherboard-set voltages.

    Quote Originally Posted by Seronx View Post
    AMD's upcoming IVR is using Buck Converters, correct?
    https://i.imgur.com/43BaOn5.png
    Doubtful I'd say. A buck converter needs discrete inductors/chokes or whatever you want to call them. They can't be integrated on-die so they'd need to be either external to the package or on the package itself (both would be external to the die either way). I couldn't see them sticking it on the package tbh.

  6. #6
    Quote Originally Posted by Zoot View Post
    Doubtful I'd say. A buck converter needs discrete inductors/chokes or whatever you want to call them. They can't be integrated on-die so they'd need to be either external to the package or on the package itself (both would be external to the die either way). I couldn't see them sticking it on the package tbh.
    actually i think it is possible AMD already uses inductors on die for there resonant clock mesh. the inductors for GHz clocks where in the size of .75nH to 1.25nH at 100um^2. so i guess they can possibly do it at those 1GHz clocks. while normal converters work at kHz speeds.

    source at page 14 warning it is a pdf

    more info can be found from "piledriver isscc 2012" search.

  7. #7
    AMD slide looks more subtle than plain, almost steady buck conversion
    Android is not an OS.

  8. #8
    Quote Originally Posted by Zoot View Post
    The IVR is cool when it comes to low cost mobile stuff since it simplifies the power delivery circuitry to the processor die significantly. Ideally if the input voltage to the processor could be something like 5V, there's a lot of the motherboard that could be done away with.

    But I can see it being a problem for the high power stuff. Run a power calculation for an LDO-style regulator:

    Vin = 1.25V (Motherboard supplied)
    Vout = 1V (Internally regulated down - just a guess figure)
    TDP = 100W (just to pick a number out of the air)
    =>Iin = 100A
    Power(LDO)=(1.25V-1V)*100V = 25W

    So using that extremely crude calculation you've an extra 25W dissipated on the processor die which will raise the junction temperature accordingly. It's no wonder the Haswell TDPs actually went up relative to Ivy Bridge. I also wouldn't wonder why the temperature issues for the high-end desktop chips are as prolific as they are at higher motherboard-set voltages.
    https://www.google.com/patents/US20140070876
    Here is the particular patent for Intel's FIVR.
    Quote Originally Posted by Zoot View Post
    Doubtful I'd say. A buck converter needs discrete inductors/chokes or whatever you want to call them. They can't be integrated on-die so they'd need to be either external to the package or on the package itself (both would be external to the die either way). I couldn't see them sticking it on the package tbh.
    https://www.google.com/patents/US20130257525
    Here is the patent particular to AMD's IVR.

    --
    Intel is Low Drop Out and NMOS, so it kind of points to Linearity. (If you ignore it states Linear Voltage Regulator...)
    AMD is using a Buck Regulator, so it points to Buck Converters.

  9. #9
    Quote Originally Posted by Seronx View Post
    https://www.google.com/patents/US20140070876
    Here is the particular patent for Intel's FIVR.https://www.google.com/patents/US20130257525
    Here is the patent particular to AMD's IVR.

    --
    Intel is Low Drop Out and NMOS, so it kind of points to Linearity. (If you ignore it states Linear Voltage Regulator...)
    AMD is using a Buck Regulator, so it points to Buck Converters.
    Thanks for the patent links.

    Looking at Intels, their setup is a whole bank of your standard LDO that uses a pmos as the main pass device across the whole die. It's the same setup you'll find in most off the shelf LDOs you can find out there.

  10. #10
    Checking this image its both PMOS and NMOS;
    https://patentimages.storage.googlea...313-D00006.png

    MP = PMOS
    MN = NMOS

    The fullset for Linear Regulation;
    https://i.imgur.com/6f84s2l.png
    https://i.imgur.com/VKUDgyv.png
    https://i.imgur.com/rGaRxaM.png

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts
  •  
WordPress Appliance - Powered by TurnKey Linux