At 11:55 AM 4/24/02 +0300, Sampo Syreeni wrote: >On Tue, 23 Apr 2002, Riad S. Wahby wrote: > >>This may take more voltage than you want to use in your process, but you >>can engineer the base-emitter junction if you've got a friend in process >>engineering.
You can also use common guard structures to isolate the "HV" part of the chip, without dicking with the Delicate Recipes (process) which you Don't Want To Do And Probably Wouldn't Be Allowed To Anyway. Also helps keep digital switching noise out of the source. >Aren't there dedicated avalanche diodes available with low breakdown >voltages, precisely for this reason? I think they're used in applications >where zeners could be, except for higher breakdown current. > >>One other potential problem is long-term reliability, but that's a >>subject for another email. Actually, we're interested... >Shouldn't be a problem, if you limit the breakdown current. If you're >after entropy, you'd likely want to use a constant current source anyway. And constant-current sources are *sooo* tough to make out of transistors :-) My small point is confirming that junction/avalanche RNG sources are very compatible with standard CMOS fabrication. (I've actually probed test structures on production wafers in a stuffy metal room examining this... man, you don't want to have had too much coffee trying to land the probes.. ..looking at analogue measurements with spectral analyzers and sampled data with statistical tools) The junction structures are "louder" than resistors --they produce more entropy per watt. Intel may have had other, valid reasons for using resistive sources in its real RNG.
