Re: more on weird resets

On Sun, 31 Jul 2005, John Sheahan wrote:

> A while back, decoupling ttl inputs via a pullup was a great idea
> as the damage voltage rating on inputs was less than on the collecter supply. 
> I don't think thats  valid  for cmos, as most gates are
> connected to one supply - and its gate oxide that matters.

It works anyway. The idea is unrelated to the TTL input level setting, 
which was actually important to prevent avalanche latchup afair.

> Your suggested solution works as well its a good capacitor connected to an 
> appropriate virtual earth.

Yes but the appropriate virual earth does not exist in the circuit that 
needs to be 'saved'.

> perhaps you could explain more why a connection from VDD to the reset input 
> is flawed?

Because it precludes ICSP as well as deliberate resetting during 
debugging. I find that a reset button is the second most useful thing 
after the power supply on a prototype board ...

> line pumping from spikes? positive spikes won't reset the chip for an active 
> low clr.

Yes they do. The MCLR line is decoupled by its capacitor and stays at 
constant voltage while the VDD line *rises*. In theory the MCLR line is 
protected by a Schmitt circuit and probably by a monostable. In 
practice, the 220pF/100Ohm solution works many times when there are 
spike sources on the board.

> IO line switching ? This bounces either internal VSS up or VD down.
> An external decoupler does not help here in either case.

It helps to make MCLR follow VDD tighter. The zener curtails these 
excursions from going too high.

> RFI? Reducing the coupling to the reset pin is the key here. depending on 
> where your suggested cap is connected to - it will work as well.
> I don't see any obvious cases where it works better. I do assume the wire I 
> suggested is not longer than the RC individual elements.

Often the RFI causes resets by being coupled into VDD or VSS, not MCLR. 
Older PICs (16C54) had a reputation for being extremely sensitive 
against positive spikes on VDD. Newer PICs have an internal filter on 
the MCLR line.

Peter