Dear Google Answerers,

I'd like pointers to article etc. that give the following info about
the cost of fixing bugs in electronic design versus when the bug is
accidentally created, and when the bug is detected.

The data should be drawn from interviews and surveys of real world
(i.e. commercial) engineering projects in fields of communications,
aerospace, automotive, computer, and other hightech electronic systems
and the semiconductors used in them.

I'm particularly interested in hardware bugs, and even more
specifically digital hardware bugs. Analog hardware bugs and embedded
software bugs are of secondary importance.

In the following, you can interpret "design stage" as "percent of the
project complete" or qualitatively as "Research, Specification, System
Architecture, Design, Verification, Implementation, Tape-out, Unit
Test, System test" or other phrases that indicate the phases a design
goes through.


1) At what design stage are what percentage of errors are introduced on average?

2) At what design stage are what percentage of errors are detected on average?

3) What is the cost of fixing an error (e.g. loaded cost of staffing,
re-spins, opportunity cost of being late to market) versus design
stage for a "typical" project?

I'm aware of Ronald Collett's reports, but they're beyond my budget.

TIA

built2last-ga

The cost of bugs is mostly time-to-market, so it's hard to pin down.
There are very few cases of digital logic bugs, that get to the field
and result in cost, either thru replacement or other reasons.

Intel divide bug is one. There are plenty of other bugs in existing
Intel P4 and Athlon parts (about 50 each, see their web site) that don't
cause return of parts, even though wrong answers and such are possible.

It's hard to nail down delays in time-to-market to a single bug. Each
chip spin typically fixes multiple bugs. So you could argue it only helps
if you eliminate all bugs and eliminate a spin early.

The clearest cost, is when everything else is ready to release new systems
or chips, and the ship date is pushed out to wait for a new spin of the chip.
This does happen. Typically costs one, maybe two quarters. So you can look
at the impact of 2 quarters of deferred revenue. Some argue that the product
is dated then, so has a shorter lifetime, so you don't make up the money.

So the cost of a bug? It can range from $0 to a lot of dollars.

The worst, is when the bug prevents finding other bugs. If you can't do
some amount of stress testing, because of other key bugs, then that's bad
and delays things too.

You won't find a one true answer, because there isn't one. And a "mean" value
doesn't mean much.