OL3.NOV–RM
Which comes first, known-good die or multichip modules?
Work underway at Micron Technology and other semiconductor makers, as
well as Sematech, should lead within a year to the availability of fully
tested unpackaged semiconductor chips at prices comparable to commercial
packaged ICs. This will be an enabling technology for mass-produced
multichip modules. Multichip modules have existed for a long time in the
form of military-avionics hybrids. These have been fantastically
expensive by commercial standards, but are used anyway where their
characteristics, such as weight, space, and ruggedness, are absolutely
necessary. One reason they are so costly is the need to allow for rework
when one die is bad in a package that may contain 100 or more. The die
yield problem is elementary. If the die are 99% good, then a 100-chip
module has a probability of 0.99
die are only 90% good, the module yield drops to a sickening 0.000026,
even if there is never a problem with wire bonds or other package parts.
Various methods for improving yields for military ICs are in use or under
development (see Electronic Products, October, p. 41 for an overview).
Micron is currently offering tested military parts, but its test procedure
involves soft wire bonding into temporary packages. This is a
labor-intensive procedure much too expensive for commercial parts, but
Micron engineers think they have an answer. They have devised a carrier,
like a ceramic DIP package, in which die can be mechanically mounted and
held by lid pressure. The contacts are not bonded, but make contact by
pressure. The carrier can be used with normal bonding pads or bumps of
gold or solder. Pressure is maintained by the lid of the package. Once in
the carrier, the chip goes into the existing handling, testing, and
burn-in apparatus. There it can go through whatever procedures are needed
to ensure spec-sheet quality and weed out infant mortality. The costs at
this stage are the same as for packaged parts. The test fixtures
themselves are die-specific, or at least pad-ring specific, and making
them costs more than programming an automatic wire bonder. Micron expects
the up-front cost to be amortized such that the price will equal packaged
parts for production runs of 50,000. For larger volumes, it should be even
cheaper. Known-good die, at commercial prices, should make multichip
modules a plausible option in many more markets than before. –Rodney
Myrvaagnes
Advertisement
