level shifting at 100 MHz

[poorchava]

The question is about digital circuitry, but the problem itself appears to be more of analog nature, so i'm posting it here.

In my design i need 100 MHz 5V clock signal. The problem is that I'm unable to buy a 100 MHz integrated oscillator operating from 5V supply, only the 3.3V one. That means i've got to translate logic levels somehow. The frequency is what poses a problem. Normally I'd do that using a fast comparator or some general purpose bjt (like bc547 or 2n2222), but they don't work at such frequency.

What should i do?

 

[FvM]

Are you sure, you actually need 5V signal swing? TTL IO standard would be pretty satisfied with full 3.3V swing. If the rise time doesn't count, you can also "boost" the level with a LC circuit. You should clarify the load reuirements.

 

[poorchava]

Oscillator which is available to me has swing between 0.1 and 0.9 of supply voltage, which os from 0.33 to 2.97 in case of 3.3V supply. Unfortunately IC which i intend to feed the clock signal into has high logic level from 3.5V upwards.

I don't know if the rise time counts or not, datasheet doesn't say. Only thing it says is minimal pulse width of 5 ns. At 100 MHz (period of 10ns) that means assymetric dyty cycle and very fast rise and fall times.

Oscillator which i intended to use is CFPS-72-100M. No i see that is has rise and fall times of 6ns. How is that possible for 100 MHz oscillation? I mean how should i interpret that information in datasheet? Does it mean that signal starts to fall before is reaches max level while rising?

 

[dick_freebird]

That would depend on your real load, vs the test condition in the datasheet.
But 6nS rise + 6nS fall means you done run out the clock... I would suspect
somebody cut and pasted from another lower frequency oscillator with less
care than it deserved.

I think you might be able to get something like a 74ACT04 to run at 100MHz
as long as you're not at high temp at least. Use the T version for a skewed
(TTL) logic threshold, eat the excess IDD from never getting to input-railed
and call it good enough?

There are probably other devices in the newer, low-voltage-logic families
that might do 3.3 - 5V.

 

[FvM]

I regularly use 74AHCT series for 3.3 to 5V level conversion, preferably 74AHCT1G125 single gate devices. They are specified with a typical propagation delay of 3.4 ns (with 15 pF load) and good symmetry.

To interface a symmetrical clock of 2.7Vpp to 5V CMOS, capacitive level shifting with 2.5V bias would also keep the threshold requirements.

Considering the siad 5 ns pulse width requirement and the standard 40 to 60 % duty cycle specification of crystal oscillators, 100 MHz operation won't be guaranteed by design, apart from level conversion problem.

 

[poorchava]

Considering the siad 5 ns pulse width requirement and the standard 40 to 60 % duty cycle specification of crystal oscillators, 100 MHz operation won't be guaranteed by design, apart from level conversion problem.

The IC is Intersil ISL5314 DDS unit. According to datasheet it can operate at clock frequency of up to 125 MHz from 5V supply.

Interestingly the said CFPS-72-XX series appears to be some kind of industry standard more or less and it's manufactured by numerous companies. The one datasheet wich quoted all rise and fall times as 6ns was from Semtech. The ds from Zarlink says the same, while the ds for the same device from IQD Frequency Products says that for different voltage the slopes are steeper. I suppose I'll need to get to know the exact manufacturer of the oscillator which i'm gonna by.

I think you might be able to get something like a 74ACT04 to run at 100MHz
as long as you're not at high temp at least.

I'm not expecting operation in non-room temperature. The device is intended to be bench-top.

Right now i'm kinda busy as hell (defending my master's thesis tomorrow ), but when i'll have some free time i'm gonna buy the said oscillator, and logic gates and test these solutions. If they fail, I can still use 80 MHz clock, but this will lower my circuit's performance (DDS output sample rate). Then if i somehow get ahold of 100 or 125 MHz 5V oscillator, i'm just gonna replace it.

My main problem is that I live in Poland, which means that in most cases ordering stuff from big distributors like Digikey, Mauser, Future or Farnell is out of question because the shipping fee is prohibitive (like components for $10 and shipping for $30). Technically Farnell has it's dealership in Poland but most of components are in warehouse in UK which means expensive shipping and long waiting. Future also has it's dealership in Poland, but they rarely sell anything in quantities below 100 or 1000 and shipping is also expensive.