Should I use CMOS vs. TTL?

Question

Now, I know that this question has been asked so many times that it seems like I'm trolling, but I must point out that I couldn't find the right answer to this question anywhere.

So yeah, CMOS and TTL are different. Logic levels, speed, drive current, power consumption, and so on.

What I'm wondering is what is their difference from a designer's perspective.

For this question, I'll be designing a small computer, say, a Z80 or a 6502 or 8080 or whatever. Let's use HCMOS and LSTTL as our examples.

Does it really make a difference, practically?

The clock rates are low enough that gate delays are not an issue. I'll power this from AC mains so power is not an issue. I'll stick to one logic family or the other so compatibility is not an issue. So what's the incentive to choose one over the other?

Say I want to make this device as reliable as possible. Also, I want to make this device repairable.

So which logic family should I use?

I'm talking about product lifetime. Should I stick to one type of IC because they are more reliable? What's the MBTF of these devices?

Also what if, in a few years, they stop manufacturing some of these ICs? For example, if you've got a Motorola ECL-family computer, if an IC breaks, you're outta luck. You'll have to kludge your own or scrounge for some old rare ICs. I've heard that large FIFOs and PROMs and SRAMs in the 74xx family have also been discontinued. So if something breaks, I want to be able to buy readily-available parts and fix it.

Answer 1

A few arguments:

• with HC and HCT, the fan-out (at lower frequencies) is essentially unlimited, for LS it s IIRC 10 (or was it 20)?
• the HC and HCT families are newer than LS, so my bet would be that they will be around longer than LS
• HC and HCT use much less current than LS (except maybe for a few gates that switch at a very high frequency), which eases the design and layout. HC and HCT also require much less power supply decoupling
• check the logic levels, and choose the best match: HC or HCT. LS gives you only one option.
• last time I checked, there was more choice in HC/HCT than in LS

Let me add two (weak) arguments in favor of LS:

• inputs can be left open (they will be high)
• more robust against ESD

Answer 2

I'll assume you're talking about 'glue logic' and you're not trying to build a CPU out of logic chips.

I think you're better off using popular types of HC CMOS chips since they'll likely still be available in some years. 4000-series CMOS is too slow when operated from low voltage, and LS TTL is becoming rare, has lower fan-in and requires that you use HCT CMOS if you want to mix families because the output swing is not compatible with HC CMOS.

You may not care about power consumption, but most of us do, and if we're not designing in LS TTL, it will disappear sooner than otherwise.

Anyway, if you use 74HC CMOS you can probably find pin-compatible parts in CMOS, LS TTL and more modern logic families such as VHC CMOS. You can constrain your design to allow the use of any of them. For example, a part that might be used is a ' 573 latch. You could allow for a 74LS573, 74HC573, 74VHC573, 74HCT573 and probably others. All will operate from 5V.

You can look at the distributor stock and get an idea of what parts are on their way out. If they've not made the transition to RoHS and/or SMT you know they're dead chips walking, and if the stock at disties other than the ones that specialize in obsolete parts is low or nonexistent, they're probably as good as gone.

Answer 3

My favourite family for 5V glue logic is HCT for one simple reason, compatibility.

You can feed it's inputs quite happily from 5V TTL, 5V CMOS or 3.3V CMOS and it's output has both the drive strength to drive TTL parts and the voltage levels to drive 5V traditional CMOS parts.

Answer 4

There's one important thing that everyone forgets: more modern logic families might have the same logic levels, but they can switch much faster than old logic families. If your board is not designed to cope with such fast switching, it will either completely stop working, or will work erratically, once you start swapping out old chips for newer, faster logical equivalents.

So, when doing a design, you have to specify not only a particular logic function and logic levels, but also a particular switching speed - rise- and fall times, and you should verify the performance of your design using both the slowest- and the fastest-switching devices. You should also identify what combinations of devices are timing critical: say a fast device's output may violate a slow device's input hold time.

Just because a logic line switches at 100kHz doesn't mean that it's a slow signal. If you're using a logic family that slews the output in under 10ns, you need to treat it accordingly, or things won't be reliable at all.

You should also plan for large device packages becoming obsoleted: volumes are in mobile devices, and there everything is tiny. Not using DIPs is a foregone conclusion nowadays. Even the decades-old SO is not very future-proof for some logic devices. You need to design your board so that the SMT devices have sufficient clearances and thermal reliefs for rework. Repairability without a hot air rework station may not be an attainable goal.

You need to choose the packages with an eye for longevity. You may need to plan for adapter boards to be used to adapt future packages to your design. This was easy for DIP, for SMT packages it's hard as for anything smaller than SO you have to reflow-mount the adapter and clearances are a big issue.

The trend seems to be that single/double gate devices are popular, and are multi-source. So sometimes it may be better to use single/double packages instead of the "usual" quadruple/hex packages.

If you want to allow future use of adapter boards with reasonable reliability, there are at least two options that you may wish to think of ahead of time:

1. castellated terminal adapters from a smaller device to a SO footprint,
2. BGA adapters from a smaller CS or BGA device to a larger BGA footprint.