[sci.electronics] Reliability of Milspec vs Commercial ICs

mark@mips.COM (Mark G. Johnson) (06/11/88)

In  <5770009@hpscdc.HP.COM>, rkarlqu@hpscdc.HP.COM (Rick Karlquist)  says

  > By the way, are JANTX [mil-spec] parts really more reliable than
  > commercial parts (on the average) or is it merely that more is
  > known about the reliability?  I have heard that some JANTX parts
  > are actually less reliable.  

Mil-spec DRAMs are indeed less reliable than commercial DRAMs.

These parts are built in exactly the same wafer fab line with
the same masks... the only difference is the testing.  Batches of
wafers come out of the fab, yy% of them go to the military test
facility, and 100-yy% of them go to the commercial test facility.

The milspec tests are much more severe.  Temperature extremes
and voltage ranges are larger.  And, of course, every milspec
DRAM receives an extremely thorough burn-in.

The nub of the problem is this: the temperature and voltage extremes
used in milspec test & burnin put tremendous stress on the DRAM
storage cell (a MOS capacitor).  Much, much moreso than commercial
testing.  As you'd expect, the storage cell is made as tiny
as possible (to get a small die, lots of em on the wafer, ==> low
cost).  This increases the cell's fragility.  Cell stress is large
enough that the test procedures for milspec DRAMs actually drive the
parts into the "wearout" phase even before they leave the factory:

  FAILURE RATE
	|
	|.                                                    .
	|  .                                                .
	|    .                                            .
	|      .                                        .
	|        .                                    .
	|          .                                .
	|            .                            .
	|              ...........//.............
	|
	+-------------------------//--------------------------------> TIME
   infant mortality----> <------normal life-----> <----wearout---->


So the manufacturer has to make a decision: will s/he build a DRAM with
a large, rugged memory cell capable of withstanding milspec tests,
but economically inferior in the commercial marketplace because of
high die cost?  Or instead, will s/he build a teeny memory cell and
hope that some fraction of chips will survive military testing?

Thus far, given the relative sizes of the commercial and military
markets, most manufacturers have chosen the latter approach.

Postscript......
	Note that as cell sizes have gotten smaller (64K-> 256K-> 1024K)
	mfr's have learned how to make better (higher field strength)
	capacitors.... giving the same capacitance in smaller area.
	So the big-mother cells used in 64K DRAMs of 1982 are about the
	same reliability as the puny cells in today's 1024K parts.

-- 
 -- Mark Johnson	
 	MIPS Computer Systems, 930 E. Arques, Sunnyvale, CA 94086
	{decvax,ucbvax,ihnp4}!decwrl!mips!mark	(408) 991-0208