On Dec 2, 2016, at 12:59 AM, Tony Duell
<ard.p850ug1 at gmail.com> wrote:
On Fri, Dec 2, 2016 at 2:31 AM, Noel Chiappa <jnc at mercury.lcs.mit.edu> wrote:
So I have this memory of a set of law promulgated
by an engineer at DEC, one
of which was something to the effect that 'all digital circuits are made out
of analog devices'. However, my memory doesn't recall where I saw this, and my
Google-fu is not strong enough to turn it up. Can anyone help?
It's one of Don Vonada's laws. I am pretty sure I first read them in
'Computer Engineering', that book produced by DEC. I think the
original is 'Digital circuits are made from analog parts'
Another of the laws is 'There is no such thing as ground' which
(a) means voltmeters have 2 leads, so you can take whatever
you want as a reference
and
(b) all connections have impedance (the inductance is the important
bit in general) so even if you have 2 points connected to a wire you
call 'ground' they won't necessarily be at the same voltage all the time.
There are probably lots of people who have spoken or written along these lines. I heard
the "everything is analog" from my manager Tony Lauck, the head of the DECnet
architecture and A-D group and a polymath (as were a number of others in that group). He
specifically emphasized the point in connection with metastability issues. But since the
statement is both true and all too often forgotten, I assume it has been said many times
before.
On the bit about grounds, I still have tucked away somewhere a very funny (but also
serious) memory by Bob Steward (lead designer of a number of VAXen) entitled "Do not
cut your grounds". He described how a poorly chosen PCB layout turned a ground plane
into a slot antenna, among other things.
These are the sort of issues that supercomputer designers at CDC or Cray all understood
thoroughly -- you can't build machines like that without this knowledge. But a lot of
logic designers working at slower speeds weren't taught properly and/or didn't
really get it, and they might not get in trouble right away.
A somewhat related area is the many different ways in which a number of chip designers
messed up the random number generators needed in half duplex Ethernet. I remember Intel
as a particular offender, but probably not the only one. AMD got it right in the Lance
partly because of a lot of coaching from the DEC Ethernet designers.
paul