I want to use a retriggerable monostable multivibrator as a hardware watchdog. This is for an amateur satellite project so I need to use basic rad hard logic chips (as opposed to more sophisticated watchdog chips).

I'd like the timeout to be 5 seconds.

Is there an advantage to using a 74123 part vs a 4538? It appears from the datasheet of one 74123 part that the pulse timeout might be limited to .5 seconds. Is this true for all 74123 parts? Some datasheets don't seem to specify. At least one 4538 seem to be able to sustain a timeout of up to 10 seconds.

Are these timeouts true of all parts of that family? Or is it chip specific?


"Rad hard" - there is no guarantee that a "basic" chip will be more "Radiation hard" suitable that a more modern one, although smaller geometries and greater complexity tend to be more radiation prone. If you want genuine rad hard parts you may need to start with a list of suitable ICs and go from there. Re using a small microcontroller as mentioned below - if the uC has its own watchdog, may it not be as suitable? Maybe not. Is it possible to adequately shield a very small core of key components?

Old iron: The 74123 datasheet here seems to be pretty much a denizen of another age and not at all comfortable at the 10 second range.

Modern alternatives: The MC4538 and variants seems much better suited BUT are probably not the best choice. A number of ICs contain oscillators and digital counters and can provide monostable functionality out to many hours.

A better watchdog: Very suitable in many cases and possibly most cost effective would be a small microcontroller dedicated to the watchdog task. One which itself has a quality watchdog to protect it from falling over would be a good idea. One possible choice, based mainly on size and prices is the PIC 10F series - PIC10F20x datasheet here
These have power on reset and watchdog timer, come in SOT23-6, PDIP-8 and DFN8 packages and cost from under $US0.50 in 1's.

Achievable results:

A microcontroller solution can achieve delays of any sensibly conceivable length and complexity of pattern. As an example of achievable watchdog complexity that would be hard to achive any other way - "A watchdog may usually allow say 10 seconds between triggering events but if it was takem to more than 9 second twice in a row or 6 seconds more than 4 times in a row or more than 3 seconds twenty or more times in the last hour it should force a reset." Not a lot of demand for this degree of ability in most cases but trivially easy for a microcontroller solution.
Programming is required, but in any volume the net cost of implementation is liable to be lower than most alternatives.

The table below from the Onsemi datasheet here provides the useful table below at the bottom of page 12.

The MC14536B datasheet hereat 1 uS to 23 hours recommended "looks good" [tm].
it does rather cheat by using an internal digital divider, but cam]=n almost certainly do what you want. Similarly the MC14541

Note that the 4538 is shown as recommended to 1 second but usable up to 10 seconds. The main limitation seems to be leakage. Datasheet here from OnSemi. They say pulse wdths from 1 uS to 10 seconds. Longer for he excessively enthused.

enter image description here

| improve this answer | |
  • \$\begingroup\$ When I say "rad hard" I do mean actually spending $2,000 for 15 radiation hardened logic chips. A rad hard microcontroller would probably be a $10k+ part and is not an option. We will be flying non-rad hard COTS parts for our microcontroller, but want rad hard for the watchdog and a few supervisory circuits. My prime candidate is st.com/internet/com/TECHNICAL_RESOURCES/TECHNICAL_LITERATURE/… I will check availability for a rad hard 14538, looks like Intersil might have one. \$\endgroup\$ – braddock Mar 6 '12 at 16:13

My inclination for a watchdog circuit would be to have an oscillator drive a counter, and then use a circuit like this to interface it to your processor. Note that the processor should only toggle its 'feed watchdog' output when the 'feeding required' signal is active. Attempting to feed the watchdog at other times may result in an immediate reset. One of my peeves with many watchdog chips is that a processor which gets into a 'random pin wiggle' state (can happen very easily on some parts which can use some of their I/O as an external bus) will keep the watchdog happy. Using a circuit like the one indicated can avoid that problem, since random pin wiggling will reset the processor within about half a watchdog period. Further, if the only places your code reads the watchdog-ready pin are in the one of the sequences:

  1. Disable interrupts; Read watchdog-ready pin; Disable interrupts; Update feed-watchdog output; Reset all registers to startup state.
  2. Disable interrupts; Read watchdog-ready pin; Disable interrupts; Update copy of pin in RAM; Examine all registers for consistently; If registers are inconsistent, force reset, Else copy watchdog-ready pin to output

then it is very unlikely that your system will get into an erroneous state that keeps the watchdog happy (btw, note that I specify disabling interrupts before and after reading the watchdog-ready pin; if interrupt vectors are in RAM, it would otherwise be possible for the system to get into a pattern of reading the pin and having an erroneous interrupt vector jump to the "update pin" routine.

| improve this answer | |

Maybe I can help here. You do not need to use a chip for this function, build it in transistors. Stick to J-Fets, and low gain bi-polar transistors, BSV52, BSR12, they are good. Always give a leakage path base emmitter, 10k or less. Run the transistor at a high percentage of it's maximum current. High current helps. Low gain means you get a decent base current, helps stops charge build up. Jfets just work.

| improve this answer | |

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.