24 Jan
2021
24 Jan
'21
7:07 a.m.
On 1/23/21 6:40 PM, David Moylan wrote:
Reminds me a little of the hardware development of the
BBC Micro. The original design used an NS 81LS95 multiplexer for the memory, and for
reasons that nobody could explain, only the NS component worked in their design - the same
component from other vendors failed.
The core reason was due to the bus speed and how the original infrastructure design was
done, but the choice of many components was critical to the stability of the machine.
I'm not implying NS is better - just saying that it's interesting to compare
stability and long term failure rates in the field and how much they can vary.
Far be it for me to disparage Nanny Beeb, but this type of dependency
is indicative of poor design. Part of the design process involves
calculating out all margins and tolerances to arrive at worst-case
figures for things like logic voltage levels, timing, etc.
The datasheet for a component constitutes a sort of contract. The
manufacturer of the component guarantees that the component's
specifications will match those listed in the datasheet...but nothing
more. All manufacturers can and do to change HOW a component meets
those specifications, and subtle variations can and do occur as a result.
Datasheets usually list minimum, typical, and maximum values for
various parameters. From one fab to the next, from one production batch
to the next, and even from one chip to the next, the component's
performance on a given parameter will vary within that range. All
designers know this (or at least they're supposed to!) and they must
perform end-to-end calculations based on the min/max range of those
parameters.
A dependency on a particular manufacturer's implementation of a
standard and common design pattern like the 81LS95 is a bug. I sure
hope they fixed it in a later revision.
-Dave
--
Dave McGuire, AK4HZ
New Kensington, PA