AVR Freaks

There have been sporadic accounts of strange behavior with USART0 when the '1284 is clocked at a high rate. Despite several attempts, I was not able to replicate this problem, but then I ran my experiments at room temperature.

Just wondering if that the two might be related. Let us know what you find out.

Sound along the lines of what I'd suggest. Another thing I ran into is that if you are using the internal RTC clock, the frequency varies with temperature.

Test I would do if you are using the internal clock, is rework one of the boards with an external 8MHZ oscillator and see if the 'temp problem' goes away.

Also, as the guy above mentioned, condensation can play havoc with electronics.

AKA put unit in the fridge, stops working... then after awhile starts working.... take it out... stops working.... then starts working.... I have seem many engineers dazed and confused by this over the years.

You should also set the brownout reset fuse.

Without brownout reset, the CPU can do crazy things on low voltage.

Especially on high frequency (20MHz) the effect was critical.
The highest brownout level should be selected for 20MHz.
And also the longest reset time.

Peter

Some random ideas...

If you turn an ordinary "air" duster can upside down, it becomes a can of freeze spray. :) It probably doesn't get as cold as real official freeze spray, but it's cheaper and can be bought at 8 PM on Sunday at the general store. The effect lasts for several seconds, depending on how warm the part you have sprayed gets in normal operation.

How is the ESD control at the production line? Maybe somebody forgot to plug in his or her wrist strap first thing in the morning, and didn't mention this to you. This is more likely to show up in the winter than in the summer.

One way I have isolated power supply trouble in the past is just to use a plain old battery, either alkaline or lead-acid, wired into the circuit. You *know* a battery is not noisy and is isolated from the line and everything else. It's even adjustable in 1.5 volt increments. :) Use a 1N4001 in series to knock off a volt, and a Schottky to knock off half a volt. (Okay, it's probably not a good idea to float your average AA cell 200 volts above ground, but for most microcontroller-y things it works pretty well.)

I hope this helps!

You could always grab a few other productions boards and toss them outside, cool them down, and then bring them in and test them. Do they have the same failures? If not, then you probably have 2 bad boards. If other boards fail then you need to dig deeper.

This sound a lot like the problems I was having last week when it turned out to be static, see https://www.avrfreaks.net/index.p.... Try using an earth wrist strap while testing and see if this helps.

What the difference between temperature and static? It all good fun!

This sounds like the trouble I was having the other day and it turned out to be static (ESD), try using an earthed wrist strap while testing the boards and see if this makes any difference. Static can send the uC into the unknown and a power cycle can be the only way to recover. If the design has worked previously then it has to be new source of components or something daft like static. See my comments regarding strange behavior with Tiny828 last week.

@Rfrost2 and OP maybe... high impedance circuits with pin change ints.(inputs)are very touchy sensitive and can easily give the impression that they are static sensitive.
Doesn't sound to me like the OP's problem is a static thing.. Mind you, it is Canada and it is winter...

Well, I'll defer to the experts, but a few more thoughts:

If you would:
Confirm that good PCBs remain good, at both room temp and cold temps.
Confrim that bab PCBs remain bad. Run the same test on a bad PCB tomorrow and it will still show up as bad.

I don't think you have yet determined if this is a HW or a SW issue.

If the error rate is truely higher when the PCBs are cold, and you can take a bad PCB and reproducibly tell the difference between running it at room temp and at cold temps, then do what I suggested above. Run the PCB in the cold but locally heat the uC with a power resistor taped/glued to its back. This lets the PCB run in the cold, while the uC is warm. Does the failure rate match that of the PCB when it is cold, or warm?

You still can put a good frequency counter or O'scope on the USART and measure it's exact baud rate when cold, or warm. Still in spec? (Test software generating "UUUU...".

With a good O'scope look at the power supply when it is warm and cold. As Jim already mentioned, the power supply may be getting noisy as a function of temperature, worse when cold, generating more failures when cold. HW problem with the power supply and not with the uC or the remainder of the PCB.

I assume you have a By-Pass cap across EACH V+/Gnd pin pair, and from AVcc to Gnd? One can certainly have erratic operations if this is not the case.

One could also power the PCB with batteries, for a perhaps cleaner supply for testing. (Keep the batteries warm, though, while testing in the cold). Break the Regulator Vout lines and insert your battery supply line (s), unless your regulators don't mind Vout > Vin, (many do mind!).

The fact that the board still runs but the USART data becomes scrambled makes me think about SW issues, also. Is the stack getting overwritten, trampling on a few variables? Do you have memory to increase the various stacks for a quick test, (faster than trying to measure the stack usage under real conditions)?

Do you have ISRs that are taking too long during the USART routines? Are you using the USART HW module or a SW USART?

Are there variables being modified by ISRs that weren't being read/written atomically? Do you need to disable and then reinable interrupts around any special parts of code. (You could change the USART TxData routine to always output "M" instead of real data to see if the "M" shows up on a terminal dump, or if it, also, is being trashed.) (Pointing to bad data getting to the USART driver routine.)

I still think the first step is sorting out HW vs SW as the primary problem, and the temp variability sure points towards a HW factor. If it was a SW factor alone then one would not expect to see the big difference based upon temp.

For testing HW issues it is difficult to take your work bench outdoors! Much easier to use a small mini-fridge/freezer, (such as those used by college students), to have a cold environment inside, on your bench; and much cheaper than a real environmental testing chamber.

You make make it get quite cold if you pull the cover on the internal thermostat and tweak it such that the compressor won't turn off, (just make sure it doesn't overheat the compressor).

I feel your pain.

JC

If you have been making the boards for years, maybe you are using old component stocks that have not been stored correctly I.e. they have absorbed moisture from the air, then when the board went through the soldering process it caused chip damage(popcorn affect etc), If in dought make some more boards with new component stock and see if things improve.

Note: you have independently reinvented the "screaming TTY". :) http://www.catb.org/jargon/html/...