Two Oscillators or Frequency Divider?

[Dubbie]

I am designing a board at the moment with two ICs. One needs a 100Mhz clock, the other, an MCU ~8Mhz-25Mhz.

Is it a daft idea to divide the 100Mhz clock down for the MCU? The clocks don't really need to be in sync for any reason.
The chips will be about 40mm apart on the board.
I plan to use the Silicon Labs 511BBA100M000AAG -  OSC, 100MHZ, ±50PPM, 3.3V, SMD for the 100Mhz clock.

If anyone can give me any issues to consider either way, I'd be grateful.

Thanks!

R

[CJay]

Why not use an Si5351 and program the NVRAM with startup frequencies?

[Dubbie]

Good idea CJay.

I'll look into that.

Thanks!

[Benta]

Is it a daft idea to divide the 100Mhz clock down for the MCU? The clocks don't really need to be in sync for any reason.

It's actually a good idea. It's cheap, simple, and you avoid possible beat frequencies from unsynchronised oscillators.

[Dubbie]

Should I just use a SN74AHC74 (dual D flipflop) for a divide by 4 or is there a better way to do it?

[Z80]

The programmable clock generator CJay mentioned is the modern way to do clocks but needs programming.  The flip flop divider is the tried and tested old school way and either will do the job.  100MHz is right at the limit for that chip but within spec so you should be ok.

[CJay]

The programmable clock generator CJay mentioned is the modern way to do clocks but needs programming.  The flip flop divider is the tried and tested old school way and either will do the job.  100MHz is right at the limit for that chip but within spec so you should be ok.

They do but you can request custom parts from SiLabs that have a start frequency set in factory only NVRAM, not sure if there's a minimum order number but it's meant to be no added cost, might even be able to order a sample.

It gives the flexibility to reconfigure the part 'on the fly' so if your application suddenly needs an 11.0592MHz clock or some other heritage value, it's as simple as reconfiguring a bit of code instead of having to rework the hardware.

[Dubbie]

Thanks for the thoughts guys. The siLabs thing looks perfect. Only two issues is 3x the jitter and the fact you have to program it. For my application that should be ok. I doubt they will want to do a custom run for an order of 5 parts

[CJay]

You may be surprised, I seem to remember it's not that high a number and they're cheap enough that it could be possible to buy the minimum number and still be on the right side of the numbers.

[Benta]

Low volume pricing (Mouser EU):

Si5351: 2 Euro
74AC74: 40 cents

I know what I'd choose. Just sayin'

[Dubbie]

I decided to add an Ethernet PHY. It also needs 25Mhz clock. I think I’ll go with the divider. Simpler and cheaper. Good to know about those clock chips though.

[ebclr]

Active crystal Oscillators modules are so cheap, that isn't worth making anything else than buying 2 oscillators for the frequencies you want

https://www.aliexpress.com/item/100MHZ-100-000MHZ-100-00MHZ-100-000-5X7MM-active-crystal-oscillator-New-Original/32796709255.html?spm=2114.search0104.3.98.fXDao3&ws_ab_test=searchweb0_0,searchweb201602_4_10152_10065_10151_10068_10344_10342_10343_10340_10341_5560011_10307_10137_10060_10155_10154_10056_10055_10054_5470020_10059_10532_100031_10099_5460020_10338_10339_10103_10102_10052_10053_10142_10107_10050_10051_10173_5380020_10326_10084_10083_5370020_10080_10082_10081_10110_10111_10112_10113_10114_143_5570011_10312_10313_10314_10078_10079_10073,searchweb201603_19,ppcSwitch_2&btsid=0da4b0ff-c2f9-4724-a9b2-973e4dc17bf9&algo_expid=5f22a2a3-53f9-4431-b05d-adbb1dab7aa6-13&algo_pvid=5f22a2a3-53f9-4431-b05d-adbb1dab7aa6


https://www.aliexpress.com/item/10pcs-SMD-5070-active-crystal-oscillator-OSC-25MHZ-25M-7050-5-7-3-3V-25PPM/32775872130.html?spm=2114.search0104.3.2.zu2SxQ&ws_ab_test=searchweb0_0,searchweb201602_4_10152_10065_10151_10068_10344_10342_10343_10340_10341_5560011_10307_10137_10060_10155_10154_10056_10055_10054_5470020_10059_10532_100031_10099_5460020_10338_10339_10103_10102_10052_10053_10142_10107_10050_10051_10173_5380020_10326_10084_10083_5370020_10080_10082_10081_10110_10111_10112_10113_10114_143_5570011_10312_10313_10314_10078_10079_10073,searchweb201603_19,ppcSwitch_2&btsid=dbf8eb97-c766-4695-84e1-3f8809c0363a&algo_expid=0d416ab9-e0ef-40cc-afa2-10cc3f536749-0&algo_pvid=0d416ab9-e0ef-40cc-afa2-10cc3f536749

[CJay]

I guess it depends, if the design will be fixed then the 74AC74 plus one oscillator would be a solution, if there's a possibility it may need to be altered in the future or reconfigured after production then the Si device is ideal.

Horses for courses though.

[Leo Bodnar]

My case for separate crystal use:

Si5351 consumes 30mA and needs programming.  I don't know the intended volume but anything above few hundred units is a pain when you actually get to do it.

Divider seems like a logical choice but separate crystal will actually have lower EMI footprint and power consumption as well

Separate crystal also allows independent debugging and layout reuse

MCU crystals cost less than 74AC74

MCU with a crystal can be put to really low power sleep mode

Leo

[Dubbie]

but separate crystal will actually have lower EMI footprint and power consumption as well

This has actually been bothering me... those clock leads running about the board don't seem ideal from an EMI perspective. Especially as I have analogue elsewhere on the board as well.
I think you might have changed my mind.