I have encountered an unusual situation in which I require an XOR gate that will function reliably when presented with a square wave input with a frequency between 2 and 3 GHz. I know that desktop CPUs have logic gates that can function at these speeds, but I don't know of any IC that will do this. Should I try to build the gate out of transistors?
Also, at these speeds, do I need to worry about using ground planes, mitered bends, and microstrip?
The fastest logic family long has been and still is ECL. While often overlooked in recent times, developments such as PECL and LVPECL (essentially positive supplty ECL and differential PECL) have kept the family at the forefront of logic switching. The previous limitations of multiple supplies and negative voltages have been eliminated, but with backwards compatability available in many cases.
The MC10EP08 / MC100EP08 devices would meet your requirement http://www.onsemi.com/pub_link/Collateral/MC10EP08-D.PDF
Not quite as good but also almost meeting your spec http://www.onsemi.com/pub_link/Collateral/MC10EL07-D.PDF
Available from Digikey (in stock) http://search.digikey.com/scripts/DkSearch/dksus.dll?Detail&name=MC100EP08DTGOS-ND
In PECL mode these will operate from Vcc = 3.3V to 5V and Vee = 0V.
Maximum frequency is rated as > 3 GHz typical with propogation delays of 250 picosecond (!) typical and 300 picosecond max at 25C with cycle to cycle jitter of < 1 ps.
Digikey list a range of ECL gates.
While 3 GHz operation is probably best left to existing gates such as these ones, it is relatively easy to implement extremely high speed gates yourself using discrete parts with ECL type topology. Looking at the equivalent circuits of older ECL gates gives a good start (modern datasheets typically just give overall functional diagrams with no clues as to how the results are achieved). Gates are essentially very familiar long tailed pair type arrangements. Performance per effort and cost is liable to be vastly better than for most other approaches.
An excellent TI tutorial on "Interfacing Between LVPECL, VML, CML, and LVDS Levels" with discussions on impedance matching, transmission lines, reflections, biasing ... , and includes diagrams of how functionality is achieved.
I suggest a change of approach. You don't say why you need such an XOR, but I will propose that if you're asking questions about mitered corners and ground planes then you don't really have what it takes to do this kind of circuit. Don't take offense at that, as I suspect that 99.99% of the people on this site couldn't do that-- including me, and I have done GHz circuits before! So, rather than trying to do a 3 GHz XOR, I suggest that you find a different way of accomplishing what you want in a way that doesn't need such fast speeds.
Just to make myself clear, here's why I suggest changing your approach... Let's say that you could do a 3 GHz XOR, then here's some of the issues & solutions that you'd face:
You wouldn't do this out of individual transistors, too slow. TTL type parts are also way too slow. Instead you would have to think about some high speed logic parts. Back in the day you might use ECL or PECL parts (a different family, like TTL but not). I have no idea what you would use now, or even if the ECL/PECL parts are still around. Of course custom chips will do it too, at a huge cost.
Ground planes, absolutely. Controlled impedance PCB's, yup. Maybe 6 or 8 layer PCB's, depending on other requirements. At least 4 layers, for sure. Mitered bends, might as well. Microstrip/microplane traces, absolutely. And of course you'll have to play very close attention to PCB layout. Remember that 3 GHz is about 0.333 ns.
Once you get it all built, let's say that it doesn't work. Then what? Get out the o-scope! Most hobbyiest o-scopes top out at around 100 MHz. In my office I have a 1 GHz, 4-channel scope that cost US$10K, but the 1 GHz probe costs an extra US$2K. You'll need at least 5 or 6 GHz and 3 scope probes. I haven't priced them in a while, but that will cost at least US$10K, and maybe up to US$30K.
So, to do it you'll have to use parts that are hard to find, do a complex layout on a multilayer PCB, and when it doesn't quite work right (odds are that it won't) you'll have to spend lots of money on an o-scope to help you figure it out. Then repeat the process again, because at 3 GHz you can't rework your PCB to fix the flaws. Ouch!
And lastly, here's a link to some On-Semi ECL XOR Gate: http://www.onsemi.com/PowerSolutions/product.do?id=MC100EL07 It looks like it might, just barely, be able to do 2 GHz. 3 GHz looks like a stretch, but not completely out of the question. They do have an eval board for that chip (wow, I've never seen an eval board for an XOR gate before). If you insist on going down this path, that eval board might be your best option (US$137 at Digikey). But you'll still need an o-scope.
3Ghz? Dude, you have real troubles :-)
Making out of transistors is not an option - You won't go far over 100 Mhz even with fastest transistors. The main problem is trace lengths & EM interference & sloooooww transistors.
Even if you have separate chip with required speed - You will have to worry alot about transmitting signal with up to 10-15Ghz bandwidth (to have at least some visible fronts, you need to be able to transfer multiple of your target digital frequency). Also, at this speed, signal reflections will require impedance matching everywhere (=i.e. you need not only ground plane, but also specific PCB thickness & trace widths + termination)... World of hell.
The only reliable solution is to leave that XOR gate inside custom ASIC having rest of your device. Even at 0.25um you can have 3Ghz XOR easily.
Probably a bit much for you, but the HMC721LC3C from Hittite is good for 14 GHz. Digikey has 10 in stock as of this writing.
There's some design info that might be useful that you could glean from their evaluation PCB, much of which would be applicable to less demanding requirements.
It's really useful to have a fast sampling scope-- you can see the discontinuities introduced by PCB bends, connectors, visa etc. such a beast can be cobbled together from boat-anchor eBay findings for a few K (dollars) but it won't be very portable.
