I'm interfacing to a microprocessor system, and purposefully not using programmable logic. Feel free to call me a masochist. There's a lot of address decoding etc. happening and I may be able to cut costs and save board space if I use some diode logic. I'd probably also be saving in propagation delay — some of the gates are three levels of cascaded 74HC32s or 74HC08s involving 8 to 16 inputs.
The functions I'm considering replacing sit between HC (possibly some AC) family ICs, all running at 5V. They always sit between an HC IC and another HC IC (usually a '138), so signal restoration shouldn't be an issue. If necessary, I can change those to HCT to give more leeway. They're going to be running at clock periods of 500ns to 250ns at most (that's 2–4 MHz).
Can I use wired gates with relative impunity?
Do you have a suggestion on the type of diode to use (ideally available as through-hole and surface-mount), or shall I go with my all-time favourite, the 1N4148?
Best Answer
The main problem with diode logic is slow rise time due to the (relatively) weak pull-up in combination with the ever-present stray capacity in your circuit (assuning a wired AND circuit). The use of a buffer transistor might get you into slow switching due to saturation. I don't think you will get into (addiotional) trouble from the less-than-perfect switching characteristics of your diode.
If you want to cut PCB size consider using some of the very small one-gate-per-package chips. OTOH, if you realy are a masochist, go for all the trouble you can find! But that does not match with asking here...
So impunity: NO. Check your timing margins, if you have a few us to spare I guess brew-it-yourself logic might work. Below 100ns I would not try it.