| ??? 08/31/06 14:34 Read: times |
#123439 - the conclusion Responding to: ???'s previous message |
The problem was a combination of not-so-perfect design, and bad luck.
The regulator (78M05) output was actually 4.89V. There was a defunct filter in way, which was not needed in this configuration, and the designer decided to replace it with an 1Ohm resistor. That produced around 50mV of drop, so the mcu (uPSD3212) plus supervisor (ADM707) saw 4.84V. The supervisor's typical trigger voltage on powerup is 4.65V (there is a hysteresis of nominally 40mV so on powerdown it is less), the processor is happy with 4.5V. So far it seems still quite OK. Powerup, on ADM707 4.84V, but reset still active. What-a-#$&^... Wiggling around with the scope, I touched the 27pF capacitor on XTAL1... startup. Ehhhm... Bad crystal? Bad cap? Replacing ADM707 with an older one from a previous batch (the "old"'s timestamp 0501, the "new"'s is 0503). Immediate startup. Arghhhh... Bad ADM707? --- It turned out, that Erik was right. The crystal sits between the mcu and the supervisor. There is a decoupling cap next to the mcu. VCC trace from mcu to 707 is short, GND everywhere. However, this configuration resulted in around 200mV p/p noise when measured across the 707's VCC&GND terminals; most of it at 66MHz (3rd harmonics of the 22MHz oscillator). Nothing like half of that anywhere else on the board, ONLY on the 707... 4.84-(around)100mV is slightly below 4.75 which is the max. trigger level for ADM707, it can make sense. If even from time to time the amplitode of the noise went higher, it was enough to keep the supervisor in reset for another 200ms or so. If it once released the reset, the hysteresis was enough to keep it in released state. A 220pF cap on the terminals reduced the noise to around a half... Touching the crystal cap probably had loaded the oscillator a bit more and had decreased the harmonics or such. Kicking out also the non-functional 1Ohm resistor, just to be sure. --- This was not my HW design as I don't do HW (except hobby). But I don't accuse the designer (who already left our company anyway). The supervisor is obviously a bit too fast, and that batch probably had the trigger level set slightly higher (according to datasheet it's max.4.75 - so, at the end of the day, one shouldn't use it together with 78xx, isn't it?)... I know, I know, stay on the safe side... Honestly, did this never happen to any of you? Jan Waclawek |
