| ??? 03/05/07 17:59 Read: times |
#134308 - suggestion vs. proof Responding to: ???'s previous message |
Erik Malund said:
Erik Malund said:
What's been observed in a few failures doesn't prove anything except that these failures can occur. what else do you want 'proven' The fact that adding a supervisor seems to help may be encouraging, but it is by no means a proven solution to the problem, since the problem itself is not clearly defined. if you do something that makes the problem go away is that not a 'proven' solution? Yes, if you have a headache it can, in some cases, be fixed with aspirin or surgery and you can, with some justification, state that the aspirin is not the 'proven' solution, but surgery is; however, should that keep you from taking aspirin? If you have TWO problems, as may very well be the case here, and you appear to solve one of them, though I have my doubts as to which, as well as how you know you've solved it, it probably warrants closer examination. Lynn indicated that the reset function is dependent on the oscillator because it isn't simply an asynchronous reset. It's no secret that "something" has to happen in the flash-based MCU to cause that content-corruption. This "something" requires the clock. Kai tells us that the built-in oscillator on at least SOME of the 805x variants will operate considerably below the Vcc range in which we can expect reliable operation from any of our logic, including, presumably, a supervisor. One other thing that has piqued my interest is that Maxim freely published information indicating that their DS89C4x0's, in which I'm quite interested, have a flash corruption problem despite the fact they took the trouble to build in their 1232 supervisor. I'm not sure what that means, but it does indicate there may be more than one problem involved. BTW, I frequently get severe headaches. I sometimes take aspirin, or, more often, another over-the-counteranalgesic. At other times, while not resorting to surgery, I do take a very costly medication (zolmitriptan) intended specifically to address these severe headaches. Whether the over-the-counter analgesic does the trick or whether I resort to the very costly prescription medication, is really a guess. I have to guess what the cause is. If I think it's a sinus problem, I rely on over-the-counter preparations, including a decongestant and a pain-reliever. Since the effectiveness of zolmitriptan is considerably enhanced by early administration, I really can't take the OTC preparations first and wait for the results. It's a guess. I believe that's where we are with respect to this reset/runaway-MCU problem. Insofar as it's possible, don't you think it's worthwhile investigating it? I'm not asking YOU, Erik, to do it. I'm just asking you to stop asserting what hasn't been proven and published with a basis either in a large statistical sample under carefully controlled test conditions, or a long statistical sampling of a smaller number of instances. It's too bad none of the chip makers are willing to publish what they know about this. They probably have a large-enough set of statistics, or, at least, SHOULD have, to make it more understandable. If I tell you that is not safe to drive faster that 38 MPH on route 4711, do I have to tell you why in order to make you not kill yourself? If the posted speed limit is 55, perhaps you should. I've never denied that these events occur. I do have doubts that these all-too-popular solutions touted by the manufacturers' people are the true "fix." For one thing, I believe that there's a larger risk that the FLASH can be corrupted during power-down than during power-up. Which is EXACTLY why a supervisor is required and a RC reset is not worth $#!^ Do any of the popular chip makers indicate that a supervisor is required? Clearly, it can also be clobbered by a runaway MCU as well as by defective code. When the power drops, the Vpp charge pump is presumably fully operational. During power-up, all bets are off.
I'm thinking that the negative-going reset on the MAX1232 can serve to gate-off/stop the external oscillator that drives the MCU. sure it can, but WHY when you have ZERO, NADA, NONE problems when using a proper supervisor. That hasn't yet been shown to be the case. It appears that there are at least two possibly separate and distinct problems that have been lumped together. I'll have to wait for MAX1232 samples to arrive, as Kai has told me that the DS1232's that I have in house are probably rubbish. An analysis I had to make due to some reset problems led to the following results: ALL supervisor worked with a 100nf decoupling capacitor soldered directly across the chip (actually added on top of the chip) only the MAX and possibly the LT worked satisfactory with a 1" trace to the decoupling cap. Thus I am confident that a lot of 'negative publicity' regarding supervisors have been due to adding the decoupling cap nilly-willy ("where it fit on the board", "I am using protoboards", "I put them in a socket and wire wrap to them"), ...) Well, I routinely build up circuits on wire-wrap boards, not the ones with the long strips of power and ground connections, but ones with solid power and ground planes. These allow me ot put a component anywhere I like, and to situate the supply bypass components at the power connections, which ARE the pins. That ensures the Vcc and GND path lengths are minimized. All of this implies that, for those using the really cheap MCU's, the reset and oscillator circuitry costs more than the MCU. I don't know how well that will sell. Well if the purpose of your design is to use a "really cheap MCU" you will take other shortcuts as well. One thing that should not be forgotten is that in substantial volume basically all MCUs are "relly cheap" and so will supervisors be. I is not my job to disclose what you can buy a full-fledged MCU for when you buy 1,000,000 pieces, but it is not much. Since concern for chip price should be ZERO in low volume designs, the curve of HW cost ves importance is basically a straight horizontal line. If you make ten thingies and your development cost is, say, $5000, why should you be concerned about saving $2 in hardware, especially since that 'savings' is likely to make your development cost rise Erik The real question is whether to use an 805x core at all, in view of the fact there are other MCU's costing just as little that aren't plagued by this set of issues. If you use a $1 MCU with a $0.20 crystal and a couple of cap's and an RC reset, or use a $1 MCU and a $.90 oscillator that you can gate on/off, and a $1 supervisor of some sort that can actually do that in a way that works, well the boss may not like the decision, since his component cost is increased by $1000000 or more in the million-piece example you raise. The development cost is neglegible in such a case, regardless of the cost, since it's divided by 1000000. RE |
