[time-nuts] Bye-Bye Crystals

Magnus Danielson magnus at rubidium.dyndns.org
Tue Mar 14 18:49:24 EDT 2017


On 03/14/2017 08:39 PM, Attila Kinali wrote:
> On Tue, 14 Mar 2017 13:39:02 +0100
> Magnus Danielson <magnus at rubidium.dyndns.org> wrote:
>> Some claims that MEMS will kill crystals. It will surely eat a good
>> market share, but I think there is applications where MEMS is not mature
>> enough compared to crystals.
> MEMS is quite mature, it's just that it is playing a different game.
> While with quartz (and other piezoelectric crystals) we know how
> to design a crystal to frequency, things aren't so simple for MEMS.
> Simply scaling the design doesn't work apparently.
> What they instead do is to use the MEMS oscillator as a reference
> for a PLL locked VCO. As the whole thing is going to be a few mm^2
> of silicon anyways, reserving some µm^2 for the PLL and VCO don't
> cost much. And it gives the ability to "tune" the oscillator
> for the frequency needed after production (the same technique is used
> with "programmable" crystal oscillators). Of course, having a PLL,
> mostly a fractional-N PLL, causes a lot of spurs in the output,
> which can cause problems, depending on the application.

Actually, as I described in a post before, all this actually started 
with crystal oscillators and was motivated by much simplified production 
as a fixed crystal frequency could be manufactured and then 
resynthesized to any of a large set of frequencies much later. When MEMS 
came along, the whole pipe was already cleaned and "a minor 
implementation detail" could be changed.

> The big promise of MEMS oscillators was, that they'd be cheaper (due to
> integration in silicon) and used less power. As far as I am aware,
> neither promise could be upheld. MEMS need a quite different production
> process than normal digital electronics, hence it's usually more economic
> to have the oscillator on a different die than the digital chip. As for
> power consumption, the low power MEMS are about at the same level as the
> low power 32kHz crystal oscillators (and also in the same frequency).
> One place where MEMS are exceedingly good is temperature characteristics.
> Silabs demonstrated an oscillator, which, prior to any compensation,
> exhibited only <5ppm shift over the full temperature range.

For many purposes, I see MEMS as a nice complementary technology.
The market cut and splices up a bit differently between them than it 
used to be, but not very drastic.

> As for the demise of single quartz crystal units, I think that is not
> going to happen any soon. It is rather that the economics shift. Most
> of the single crystals are used as reference oscillators for digital
> and analog/RF chips. Ie most these chips have an internal oscillator
> that uses an external crystal to drive their internall VCO+PLL.
> As the crystal frequency is dictated by the frequencies these chips
> have to generate, there is a kind of standardization going on due to
> the limited number of protocols that need special frequencies. Two very
> common frequencies are 12MHz, for USB, and 25MHz, for Ethernet.
> 16MHz is base for CAN, some Wifi chipsets and USB as well. Then there
> are a couple of frequencies that are related to GSM, UMTS and the various
> other telephone standards. There are maybe a handfull of these frequencies,
> which "everyone" needs (ie are used in many high volume products). These are
> the crystals we will be able around for the forseeable future. There are
> other frequencies that are less used, which you will still get, but need
> to pay more or are made to order. Frequencies for protocols that are
> not used much anymore, or can be easily generated from another frequency
> that is more common, are bound to die out (as has happend with all those
> UART crystals, which are only used in legacy systems or for historical reasons).

In general, the miniatyrization of synthesis is a much greater change to 
the scenery than MEMS. There is plenty of chips out there that allows 
you to synthesize several frequencies. A re-occurring structure is a 
high-frequency CMOS oscillator, possibly multi-phase, which is locked to 
a reference (say 20 MHz or 26 MHz) and then output dividers provide 
variants. Fractional synthesis is also in there for more uneven 
frequency shifts. FPGAs also comes which multiple DLLs and lately PLLs 
on chip that does essentially the same thing, but maybe not as elaborate 
as the dedicated chips. DDS chips is another branch. In general, these 
types of chips have become much much better in terms of jitter, 
frequency range, control etc.

In this context, MEMS does much less to affect the market. The whole 
synthesiz part have made huge difference for a whole range of products 

> For specialized applications, where the crystal is not directly interfaced
> to a chip that provides the oscillator, it is more convenient for the
> designer to just use a complete oscillator than to design his own oscillator
> with all the problems that it involves. Getting such a device reliable to
> work in production volumes is nothing an average engineer without prior
> experience in can just pull off. Heck, I design my stuff to use oscialltors
> instead of crystals, because that's one thing less I have to care about.
> But even with these oscillators, there is only a limited number of frequencies
> that are easy to get. Those are again the standard frequencies from above,
> and a couple of round numbers (like multiples of 10MHz)

There is a number of more or less odd-ball frequences that occur. For 
instance, 25 MHz isn't used as much as 125 MHz these days for Ethernet, 
as it matches the needs of GE. 148,5 MHz is another, in that range there 
is a number of numbers that fit various gigabit-pipes divided by numbers 
like 10, 16, 20, 32, 64, 66 and such. SAW oscillators is another 
approach used there.


More information about the time-nuts mailing list