Why recent CPUs voltage reduced ?

Discussion in 'General Electronics Chat' started by Hayder, Mar 8, 2016.

  1. Hayder

    Thread Starter New Member

    Mar 6, 2016
    9
    0
    Hi;
    I would like to ask about CPU voltages…

    - What is main reason to make companies reduce the voltages from 5v to 3,3v and to 1v or less?

    - How can the inside transistors work ?

    - And what is difference between RAM and CPU voltage? Because the RAM voltage is still more than 1V?


    Thanks for your help
     
  2. WBahn

    Moderator

    Mar 31, 2012
    17,715
    4,788
    These sound like homework questions. Are they?
     
  3. Hayder

    Thread Starter New Member

    Mar 6, 2016
    9
    0
    No... just a question in mind after midnight
     
  4. ISB123

    Well-Known Member

    May 21, 2014
    1,239
    527
    CPU is at 1.3V-1.5V,RAM is at same.
     
  5. shteii01

    AAC Fanatic!

    Feb 19, 2010
    3,377
    494
    Miniaturization.
     
  6. GopherT

    AAC Fanatic!

    Nov 23, 2012
    5,984
    3,720
    Less heat. And...

    Because smaller seems to be better in many ways. To get smaller, there are trade-offs. A smaller chip means more chips per silicon wafer. To get all (or more) gates on a smaller wafer, they have to be closer together. To get them closer together means narrow line widths and that causes more leakage and cross-talk so you need higher dielectric materials for deposition on the chip and lower voltage to reduce currents and reduce leakage and so on and so on and so on...

    Good luck!
     
    Hayder and Roderick Young like this.
  7. WBahn

    Moderator

    Mar 31, 2012
    17,715
    4,788
    Lower voltage means less power (a LOT less power since power goes as the square of the voltage). Plus, as transistors get smaller the gate thicknesses get smaller and it takes less voltage to perforate them.
     
    JWHassler and Roderick Young like this.
  8. Hayder

    Thread Starter New Member

    Mar 6, 2016
    9
    0
    That's exactly what I thought all the night.... so its not about power dissipation ?
    but to make more and more transistors in the die
     
  9. GopherT

    AAC Fanatic!

    Nov 23, 2012
    5,984
    3,720
    It is about the things I said in the bottom paragraph of my post but it is also about what I said in the first line of my post - "Less Heat".

    The gates turn on and off so there are switching losses (rise time is not instantaneous). Since each generation of chip has at least as many gates as previous generations, you are going to generate the same amount of heat. That same amount of heat on a smaller chip just doesn't work. Therefore, they drop the voltage and minimize the heat (power) which is proportional to Voltage squared.
     
  10. atferrari

    AAC Fanatic!

    Jan 6, 2004
    2,644
    759
    And then we will have to stop when approaching the Si's 0,7V barrier and look somewhere else, isn't it? In fact, they are looking around since long time.

    Maybe time for optical transistors or it is time to press Josephson into service?
     
  11. kubeek

    AAC Fanatic!

    Sep 20, 2005
    4,669
    804
    Hayder likes this.
  12. Hayder

    Thread Starter New Member

    Mar 6, 2016
    9
    0
    I read that if the voltage reduce the switching transistors is not completely closed, so there is a small current, so the heat may be more if the voltage is reduced !!
     
  13. GopherT

    AAC Fanatic!

    Nov 23, 2012
    5,984
    3,720
    Right. To move to the next generation, IBM is using silicon "contaminated" with germanium. The germanium strains the lattice and allows higher electron mobility.

    ST and others have MOSFETS that can handle much higher temperatures with lower R(on). Currently, they advertise up to 200C but that is mostly limited by the polymer case and bond wires, not the semiconductor. Those can work just fine up to 600C. (45 amp, 1200 v mosfet)
    http://www.st.com/web/en/resource/technical/document/datasheet/DM00053079.pdf

    They are $25 each but will become cheaper with volume (not the price of silicon but lower).

    A last material is GaN, currently used in high brightness LEDs. The issue here is the cost of gallium and whether or not to use GaN as the whole chip or build a GaN on a silicon substrate or sapphire substrate.
     
  14. GopherT

    AAC Fanatic!

    Nov 23, 2012
    5,984
    3,720
    A standard Mosfet, for example, is specified for 10V gate-source voltage. A logic level is specified at 4.5 to 5 V for ate voltage and considered fully on. There are more and more "super Logic" MOSFETS that switch down to 1.1 volts (for power MOSFETS). Even less is possible if currents needs are lower (as in a CPU).
     
  15. Hayder

    Thread Starter New Member

    Mar 6, 2016
    9
    0
  16. dannyf

    Well-Known Member

    Sep 13, 2015
    1,768
    357
    Most power consumption from a modern CPU is from charging and discharging those little capacitor, gazillions of them in a given device - you will notice that current consumption increases with frequency.

    As charges are proportional to the volt squared, reducing voltages allows less charges to be moved or added, and allows the capacitance to go down - thinner and smaller devices. All of that leads to lower power and heat.
     
  17. dl324

    Distinguished Member

    Mar 30, 2015
    3,211
    619
    It would be helpful if you limited the process nodes considered.
    To reduce static and dynamic power dissipation and to decrease electric fields for reliability.

    The move to 3.3V core voltages was done before motherboard voltage was reduced. My recollection is that the processor dropped the voltage internally to 3.3V, but I/O was still 5V. In later process generations, voltages were reduced significantly; they operate at much lower voltages than discrete devices.
    The threshold voltages of the transistors can be more tightly controlled than discrete devices because they're manufactured at the same time and are in a similar environment.
    Are you referring to external RAM or L1/L2/... in the microprocessor? In the microprocessor, there are different sleep states and power management methods where RAM voltage can be lowered in several increments; depending on how quickly you want to resume from sleep and whether state needs to be maintained.

    Before about 1995, power management wasn't a standard feature and power dissipation was proportional to C*V*V*f and static power dissipation wasn't considered a problem (though it was still non-zero). With advanced power management and increased static dissipation, actual power dissipation can't be calculated very accurately.
     
    Hayder likes this.
Loading...