layout of switch capacitors in VCO

[Mehran Bakhshiani]

I am using switch capacitor bank in my VCO design, with C,2C,4C,8C values. I used unit capacitor architecture, for example putting 4 Capacitors in parallel to make 4C value. Is it degrading the Q factor and should I use a single 4C cap value instead of it?

 

[crutschow]

I don't see any significant difference in the Q factor for either configuration. That would be caused by any difference in the series resistance for the two configurations and I would expect that difference to be quite small.

 

[zorro]

There should not be difference between several Cs in parallel and a single bigger C if switches and Cs are ideal.
But even in that case the result in the whole circuit can be different by other reasons, depending of the topology in the VCO.
For example, if the C is part of a tank and Q factor is dominated by the series resistance of the inductor, as Q=ω₀L/R_s, Q of the tank is less when C is higher (because ω₀ is less). So, the VCO could fail to oscillate.
Regards

Z

 

[Darktrax]

The Q-factor need not be degraded. The ESR of the capacitors end up effectively in parallel, and so the total for four capacitors is 1/4 what it would have been for only one. That is not all the story. The fact of 8 connection points and traces converging to the inductance tends to decrease the Q. The end result for Q is probably not much change.

The frequency of operation is important. Putting capacitors in parallel for a switch-mode PSU, or audio frequency device is a lot different than at HF. The need for 4 capacitors suggests the frequency is low unless you are switching them in to extend the frequency span of the VCO.

A different consideration arises whenever devices with very low resistance are used in parallel, and that is the difficulty of making them share the alternating current equally. Even a tiny inequality in the resistance route to one capacitor would cause the current to split un-equally, such that one capacitor ends up having to suffer the majority of the current.

Fortunately, the ESR for each capacitor is usually enough to ensure that the current sharing is sensible. Make the 4 capacitors of the same size and type, and attempt to make the layout offer near equal routes for each.

 

[Mehran Bakhshiani]

When I do post layout simulate the RC extracted version for the VCO it fails to oscillate for 8C built by 8 parallel C while in C extracted version, it works fine. so there should be something in interconnects of these 8 parallel C caused this problem. any suggestion?

 

[Darktrax]

Since the capacitors are all part of a VCO, and (guessing now!) they are the C part of a LC circuit that determines the oscillation frequency, then having an inductance be connected to a spread of 8 inductances (tracks) to 8 capacitors, then another 8 tracks inductances, back to the other side of the inductor proper, possibly via or with active devices, might well make a network that has problems.

If one side of the capacitors is on the ground plane, then by all means lay them out radially all feeding the inductance. Even if not one side grounded, use the concept of "star point", more often known for preventing current returns sharing common-mode track in ground planes. What I mean is, try and have the inductance, and any tracking involved in getting to its ends, land at a "star point", meaning a wide part of track where all the capacitors can be common. This is avoiding each capacitor having separate routes with currents that might share a path to the inductor.

The inductor, with it's tracking, should see "one capacitor" star points, instead of having many separate bits of track, even if "one capacitor" is actually several.

 

[Inder7]

how to build a capacitor bank with nmos varactors and how to measure their capacitance value?

 

[Darktrax]

Gosh - answering a very old thread here. Please allow that even in the original post, I had to kind of imagine what layout he was working with.

You can use varactors in place of capacitors. The only "extra" bit that involves is to find some way to add in the bias voltage, which alters the capacitance.

Normally we are talking about one varactor. If by some chance you need to have more than one, all in parallel, which was what the original post was about, it may be possible to use just one common bias voltage for all of them.

In the original posting, the core of the question was about ESR and capacitors in parallel. He did not make the context clear. There is a world of difference if we are talking about (say) a switchmode power supply stack compared to (say) a set of several capacitors to decouple a microwave active device.

Introducing the bias voltage is done via a high impedance. This can be simply a high value resistor, or in a more elaborate circuit, an inductive track, possibly quarter-wave long, or a straightforward inductance as RF choke, with the high value resistance at the back of it.

A trick to measure the varactor capacitance is is to put a known inductance across it, and find the resonant frequency where a output develops across it.
Alternatively, put a known inductance in series with a varactor across a signal path, and find the frequency which makes the signal line collapse, as in get short-circuited by the combination. In simulation, this would make S21 of the signal line show a dip. S11 would also show a near total reflection.

The capacitance is then discovered using the well-known formula F0=1/(2*pi()*(sqrt(LC) ) see https://en.wikipedia.org/wiki/LC_circuit.

If necessary, you may have to take into account any stray capacitances, and self-capacitances of the inductance to add to the capacitance supplied by the varactor(s), unless the strays are very small in comparison.

Of course, you can just look at the varactor data sheet, and get the capacitance for a given bias voltage, though what you get when you build it may not quite agree.