lists at ozindfw.net
Sun Aug 15 12:25:11 UTC 2010
On 8/14/2010 2:56 PM, J. Forster wrote:
>> On 8/14/2010 10:08 AM, J. Forster wrote:
>>> FWIW, IMO any engineer who uses undocumented or uncontrolled parameters
>>> instructions in a production design is a fool.
>>> If you are that silly, you must fully specify the selection criteria.
>> This is, easily said, a wonderful goal, and absolute fantasy. It's
>> optimistic at best to expect someone to anticipate all contingencies.
>> It's certainly good practice to specific critical parameters, but it's
>> rarely makes economic sense to specify every possible detail.
> OK, important uncontrolled parameters.
> For example, I'd consider things like hFE; VCEsat; VCBO, fT and others
> important, but not the package capacitance in a low frequency transistor.
> There are clearly unimportant parameters and essentially irrelevant ones.
> That's where experience and good judgement comes in.
> If your circuit is not stable with a high fT part, that needs to be tested
> or the design fixed.
Right, but the definition of "high ft" varies over time. Few reasonable
designers as recently as 10 years ago would have anticipated the ft of
today's CMOS processes. It's also likely they wouldn't have expected
their designs to have lasted 10 years, and the vast majority haven't.
Oddly enough, it seems like most of the really long life designs are the
lower volume ones. These are usally the same ones that won't justify
extensive up front analysis and cost unless they are DoD or aerospace
>> As to relying upon unspecified parameters, most datasheets are woefully
>> incomplete. If you are going to use any significant number parts, it's
>> unlikely that you'll be able to get everything specified, much less get
>> compliance commitments for each parameter. Few vendors are willing to
>> do the testing required to guarantee a substantial number of parameters,
>> and the simple reason is no one is willing to pay for it.
> If your design is that critical, you may have to do incoming
> inspectrion/selection or send the parts to a company that does.
But the portion of the discussion that is the root of this branch was
precisely about designers being surprised by dramatic and unanticipated
changes in component performance. Few rational companies are going to
test for parameters in that category.
I think we are almost making the same point here. Certainly we agree.
My point is that there is an economic tradeoff. There are a number of
parameters that are critical to any circuit's operation that it's
reasonable to decide are not likely to vary outside critical
parameters. It makes no economic sense to test these.
mailto:oz at ozindfw.net
Southlake, TX 76092 (Near DFW Airport)
More information about the time-nuts