Connect with us

chip swelling up and getting fried

Discussion in 'Electronic Basics' started by DJ, Jul 14, 2004.

Scroll to continue with content
  1. DJ

    DJ Guest

    Hi all,
    I am stuck with this problem of a BGA chip getting fried within a
    few seconds of powerup.The chip takes in 2 power supplies.1v for core
    and a 3v3 for i/o's.There are no overshoots or undershoots that i can
    see in the oscilloscopes.
    The chip has worked on the same board for sometime but eventually gets
    burned out or swelling in the package is seen.I have done the power
    sequenceing as per the manufactures requirement but still having major
    I have exhausted all the options i can look into......please help me
    start looking for some thing that can lead me to the problem.
    Can someone tell me where the likly problem can be?.
    The CMOS chip works for a few seconds but goes on getting hot till
    the swelling appears and then the chip is dead.It has a plastic
    package.The chip has a PCI interface.will the overshoot on the signals
    damage the chip so badly?

  2. Boris Mohar

    Boris Mohar Guest

    Don't use no name chips ;)
  3. DJ wrote...
    Sounds like SCR latchup, wihch once triggered causes a high-current
    capable turned-on SCR to appear across the supply rails. If you are
    sequencing the power supplies correctly, then you may have an input
    that exceeds one of the supplies and is injecting enough current
    through the chip's static-protection diodes to initiate SCR latchup.

    - Win

    (email: use hill_at_rowland-dot-org for now)
  4. Bob F.

    Bob F. Guest

    Have you bothered to measure the current being drawn by this device? Take a
    look at the data sheet and see what the expected power consumed by the
    device should be then check to see how much power it is consuming in your
    design. Remember, current measurements are made by connecting the DVM in a
    serial connection.

    What is the device? How many bumps does it have? Double check your
    design/layout to ensure that all of the inputs/outputs are correctly pulled
    up or pulled down according to the MFG's guidelines.
  5. clive

    clive Guest

    Is the chip 5 volt tolerant on PCI bus or are you using a 3.3V PCI
    (relatively rare). Was chip definitely working?

  6. John Larkin

    John Larkin Guest

    If it's a ram-based FPGA, it could be a weird configuration file. It's
    possible to program some of these parts to self-destruct.

    Could the PCI bus be pulling the i/o's above 3.3?

  7. CWatters

    CWatters Guest

    Check the power sequencing. Does the chip require 3v3 to be up before the

    What about inputs to the chip? Do they appear before the rails are up?
    Perhaps you are "latching up" an unprotected input pin (perhaps an analog
    I/O pin- they don't always have protection).

    Are the clocks running? Some dynamic devices get a bit hot and bothered if
    they aren't clocked.

    Heatsinks not big enough? With some BGA you need to extract heat through the
    PCB as well as from the top of. They need the correct PCB footprint and
    weight of copper.

    Reset not long enough to allow correct operation? Bit unlikely though.
  8. DaveC

    DaveC Guest

    Recollection of Motorola's "CFBU" (catch fire and burn up :) opcode. That
    was that in the first 6502's, wasn't it?
  9. John Larkin

    John Larkin Guest

    S/360 had a bunch of reserved opcodes...

    My favorite is BKO Branch and kill operator

    The PDP-11 had the (real) LandMine instruction,

    MOV -(PC), -(PC)

    which copied itself into all of memory.

  10. S/360 had a bunch of reserved opcodes...

    I haven't hunted up details, but two favourites of mine (that were
    real instructions) were named:

    EBRS: Emit Burnt Resister Smell (on an early computer). Caused a
    particular resister to overheat.

    HCF: Halt and Catch Fire (caused such a tight loop in microcode that
    part of the ucode ROM would melt, think this one was at Intel).

  11. Ken Smith

    Ken Smith Guest

    The MOT 6800 had a HCF. It didn't litterally catch fire. All of the
    busses had squarewaves on them etc so a scope could be used to check the

    The RCA 1802 had a SEX instruction. It stood for "set index".

    The Z80 had / has several that are sort of "load and ignore value". They
    cause a ram read but nothing happens to the value.
  12. Bill Bertram

    Bill Bertram Guest

    Motorola didn't make the 6502, I think you mean the 6800.

  13. One of the microprocessors I worked with (6800? NatSemi
    PACE? - too many to remember;) had an LSEX instruction
    (Load with Sign EXtended).
    Some current processors have such instructions. For
    example, the PowerPC's DCBT (Data Cache Block Touch) is
    used for data cache prefetching (and streaming) and to
    load the TLB.

    Not an instruction, but one could get the monochrome
    monitor on the original IBM PC to release its magic
    smoke by writing an I/O register.
  14. Gee, this might be as simple as an unrecognized floating input. Or, it
    might also be due to needing more capacitance on the power supply terminals.
    Simple, simple. Try the most obvious options first.


    Sir Charles W. Shults III, K. B. B.
    Xenotech Research
  15. Rich Grise

    Rich Grise Guest

    I believe the 8008 (I know it was one of them where the
    instructions are almost microcode themselves) had a "store
    immediate", kinda the converse of load immediate, i.e., it
    would write the contents of A to [PC+1].
  16. Daniel Rudy

    Daniel Rudy Guest

    And somewhere around the time of 07/14/2004 13:12, the world stopped and
    listened as John Larkin contributed the following to humanity:

    Let's not forget the infamous Pentium F00F bug that would lock up the
    CPU so hard that a reset was need to recover.
  17. Daniel Rudy

    Daniel Rudy Guest

    And somewhere around the time of 07/14/2004 08:43, the world stopped and
    listened as John Larkin contributed the following to humanity:
    That can happen!?

    *goes and triple checks RAM-FPGA config files*
  18. John Larkin

    John Larkin Guest

    I loved the 286 ("brain damaged CPU" to quote Bill Gates) trick to get
    out of protected mode back into real mode. The CPU designers forgot to
    allow an instruction to do this, so somebody patented the idea of
    sending a command to the keyboard controller to reset the CPU. I think
    early versions of Windoze actually used this technique.

  19. Nico Coesel

    Nico Coesel Guest

    Daniel Rudy
    With Xilinx Spartan devices: Definitely YES.
    Hmm, a 0.5 amp fuse will do fine to protect your FPGA, just don't (be
    stupid like me and...) shunt it with a wire when it blows, but replace
    it. Or get a polyswitch / polyfuse.
  20. Daniel Rudy

    Daniel Rudy Guest

    And somewhere around the time of 07/17/2004 13:42, the world stopped and
    listened as Nico Coesel contributed the following to humanity:
    Ok, I'm using the parts from Lattice Semiconductor.
    That's a good idea. My question now is how an errant config can kill
    the device? Loop something around from output to input? Or is there
    some way that it can short VCC to GND? I'm thinking the latter since
    you mentioned the fuse...
Ask a Question
Want to reply to this thread or ask your own question?
You'll need to choose a username for the site, which only take a couple of moments (here). After that, you can post your question and our members will help you out.
Electronics Point Logo
Continue to site
Quote of the day