Repair methods (was Cromemco 3101/Beehive B150 score)

Tony Duell ard at p850ug1.demon.co.uk
Sun Jun 4 16:04:23 CDT 2006


> 
> On 6/4/06, Tony Duell <ard at p850ug1.demon.co.uk> wrote:
> > 'Digital circuits are built from Analogue parts' (one of Vonada's lawas
> > IIRC).
> >
> > I haev never understood how you can understand digitial electronics
> > properly and not understand analogue electronics. I certaimly couldn't
> > understnad digital stuff until I understood things like transmision
> > lines, termination, etc.
> 
> Well... as someone who considers his own knowledge of analog circuits
> to be inferior to his own knowledge of digital circuits, let me say
> this - while I _know_ that at the lowest level, all digital circuits
> are really just collections of analog parts, in practice, as long as
> you aren't working with a design at the edge of the envelope, you can
> ignore 90% of analog theory and still have an understanding that works
> on the circuit in front of you.

The problem comes when you end up with a circuit that works 90% _of the 
time_ becuase you've got a reflection somewhere that you shouldn't, or 
coupling between gates (either due to stray capacitance between traces, 
or becasue of inadequate power rail decoupling, or...) that you don't 
want. 

One thing that's _very_ important to rememebr is that the frequency you 
need to consider when designing is not determined by the clock frequency 
you're using, but by the sharpness of the edges of the logic signals 
(that is, the rise/fall times). Most modern chips (things like CPLDs, 
etc) have quite fast rise/fall times. You may only have a 1kHz master 
clock on your board (or whatever), you still need to do the design 
assuming there are multi-hundered-MHz signals around.

> OTOH, I've seen mysterious problems take a long time to solve when
> *completely* overlooking the analog aspects of a TTL circuit - one in
> particular was undershoot and ringing with a 74S409 DRAM controller
> hooked up to a field of 32 4164s.  The eventual solution was to insert
> 33 Ohm resistors in-line on the CAS and RAS lines to dampen the

Yes, series termination. Very common in such circuits. And that's another 
thing I recoemnd. Read -- and understand -- as many complete circuit 
diagrams as yoy can. Not the cut-down, ideallised ones you find in some 
so-called text books. But real ones, from actual working products. You'll 
soon see where termination is considered essential.

No, I am not suggesting copying complete commercial designs and claiming 
them as yor own. But you can pick up tricks from such circuits. The list 
of references in my Ph.D. thesis includes several to PDP11 printsets and 
technical manauls. Not because I copied the PDP11 (I wasn't even building 
a processor), but because some of the tricks pulled in the PDP11/45 to 
save a few ns were, when translated into F TTL and ECL, were very useful. 

Alas most students never get to see a complete, working circuit diagram 
of anything. Once I met a chap who'd just graduated in engineering, and 
had specialised in RF work. I put the scheamtic of a discrete-transistor 
AM radio (nothing special, a superhet of course) and asked him to explain 
what each of the components was for. He was totally lost. I don't think 
he even spotted the IF transformers. 

> undershoot to where the ringing didn't cause a problem - it was still
> there, but at an amplitude that didn't have a negative effect.  I was
> amused to see the identical solution 2 years later in a Commodore
> "slap-on-the-front" RAM expansion for the Amiga 1000 - the card was
> literally the connector, one capacitor per DRAM, the DRAM, and a
> couple of 33 Ohm (38 Ohm?) resistor packs.

I am trying to think of a DRAM board using 4116ws, 4164s or 41256s  that 
doesn't sereis-terminate the RAS/ and CAS/ lines and probably the address 
lines and WE/ line as well. 33 Ohm or 68 Ohm resistors are common for 
this. I rememebr soldering in dozens of the darn things when I upgraded a 
part-filled Unibus RAM board.


> So there are certainly many cases where it _is_ important to have a
> good grounding (pun intended) in analog theory (transmission lines,
> parasitic capacitance, hidden inductance, RF effects, etc), but for a

I learnt years ago that the most difficult part of a digital circuit to 
understand is the wire (OK, PCB traces :-)) linking the gates together.

-tony




More information about the cctalk mailing list