Hi cctalk, I asked this question over on vcfed, but in the spirit of knowing more (and avoiding silly, costly mistakes) I thought I'd ask again here. I hope this is okay --- I think there must be a lot of community expertise to draw from, and not everyone is in the same forums.
After seeing CuriousMarc's horror film about the killer transistor that zapped the guts of his HP9825T, I've been working for some time on a solid-state DC power supply monitor device that will chop all power to logic if there's ever any excursion above or below critical voltage thresholds on any power supply channel. I've been pretty successful in the development so far and have a gizmo that accomplishes the basic goal, even if it's not going to win any industrial design awards. It's not a crowbar circuit: each voltage channel passes through a substantial driver IC that can switch the power right off. My device can detect and react to anomalies very quickly. But then you browse through old DRAM datasheets and see warnings like these: "Vbb must be applied prior to Vcc and Vdd. Vbb must also be the last power supply switched off." "Forward biasing this supply [that is, Vbb] with respect to Vss will destroy the memory device." And so even though my device is fast, it's possible that when it's slamming the doors shut, there's a split second where -5V is off and +12V is still on, or even the chance of a Vbb-Vss forward-bias "blip", who knows? Of course you can measure whether this is happening, but it's difficult to know how meaningful that will be: maybe a computer that loads the voltage rails differently will have different behaviour, and remember, the case you really care about is when a power supply behaves abnormally! System characterisation is hard... Anyway, my question is: what did hardware designers in the '70s do to satisfy specified power supply requirements for the chips they were using? The conversation so far on vcfed has two remarks: one observing that a lot of folks just didn't worry about it, and one pointing out an anti Vbb>Vss gimmick in the PSU for the Nascom kit computer involving a ladder of protective diodes between each of the rails: (-12V) ---->|---- (-5V) ---->|---- (0V) ---->|---- (+5V) ---->|---- (+12V) I noticed a similar pattern in a few other PSU schematics, but usually only for the negative channels. Were there any other common tricks? How serious was the danger of getting it wrong? How fast could you fry your DRAM if you did? Thanks for any insight, --Tom