100 Hz-100 MHz amp -- can tantalums be used to block DC?

[ezalys]

I'm designing a cryogenic amplifier, and need DC blocking capacitors. Unfortunately the only capacitors that can be used at low temperature are tantalums, NP0 or C0G ceramics, and I need to pass 100 Hz into 50 ohms, so ceramics are out. Is there any way to get away with using tantalums as DC blocks?

[ezalys]

Alternatively I could just design the thing to be a DC-100 MHz amplifier, but this amp follows a previous amp and I don't know how to avoid the previous stage's desired Vds and the next stage's desired quiescent gate voltage from fighting with each other.

Cryocompatibility makes things hard. I can use one kind of BJT, one kind of HEMT, the aforementioned capacitors and metal film resistors, but any inductor will do.

[bob91343]

You need to be more specific.  How much room do you have?  How much capacitance do you need?  How much leakage can you tolerate?

[ezalys]

I have 1 inch by 1 inch by 1/4 inch to work in. I need 100 uF. Leakage is not a huge concern. If it looks like ~1 kohm with a 3 volts across it, I'm fine with that.

[jonpaul]

Cryogenic is vague: What is the storage and operating temp min/max eg +85 C to -40C?

Liquid Ln? Liquid H2?

Is the circuit immersed in the cryo fluid?

You can get high value plastic film caps with low leakage, what is the voltage rating needed?

Jon

[bob91343]

I think a reasonable electrolytic capacitor or two would fit in that space.  The biggest worry about electrolytic capacitors is leakage, both electrical and fluid.  Their low temperature performance is less than stellar.  At high temperatures their leakage increases.

[magic]

Alternatively I could just design the thing to be a DC-100 MHz amplifier, but this amp follows a previous amp and I don't know how to avoid the previous stage's desired Vds and the next stage's desired quiescent gate voltage from fighting with each other.

If Vds > Vgs then dunno, maybe 1kΩ || 100nF network in series with the gate and a few mA current sink at the gate? The current sink could be dynamically adjusted by some sort of DC servo...

[TurboTom]

I'm not informed on the effect of cryogenic temperatures on the behavior of tantalum caps but here's an example of tantalum DC block in the SM300 9k~3GHz signal generator by R&S. They may have used some special unobtainium specimen, but to me the caps appear to be pretty much off-the-shelf 2.2µF 16V AVX varieties.

There may be additional ceramics in parallel on the other side of the PCB but IIRC, there wasn't any RF magic going on there, and the lower impedance path would be going without utilization of vias, so I'ld have had expected the chip capacitors to be on the side with the rest of the RF voodoo going on, and rather have the low(er) frequency (tantalum) caps placed on the other side.

So up to 100MHz, using tantalums shouldn't be an issue at all.

[Gerhard_dk4xp]

I have once compared different sorts of capacitors, including
2 types of electrolytics. One of them failed completely even
at freeze spray temperatures.
That does not show that the better one works at 200 Kelvin.

Given that higher temperature is usually more leakage,
leakage is probably less of a problem.

Just looked through my .pdf datasheet collection: there were no electrolytics
specified for less than -55°C (Panasonic & some Nippon Chemical).
Most go only to -40°C.

In fact, -55°C was the minimum I saw specc'ed for any capacitor.
And even if you find one that fits therm/electrical, I may crack mechanically
when you cool it down depending on your board.

Can't you do it differentially?
And if voltage noise plays a role at low frequency, the input cap
needs to be much bigger than required  for f-3dB.

<   http://www.hoffmann-hochfrequenz.de/downloads/experiments_with_decoupling_capacitors.pdf   >

Cheers, Gerhard

[TimFox]

Can you change your amplifier design so that the first stage input is DC-coupled at 50 ohms, followed by the next stage AC-coupled with a reasonable capacitance into a higher impedance input to remove the DC, thus avoiding 100 uF values?

[Marco]

Alternatively I could just design the thing to be a DC-100 MHz amplifier, but this amp follows a previous amp and I don't know how to avoid the previous stage's desired Vds and the next stage's desired quiescent gate voltage from fighting with each other.

What exactly is wrong with emitter/source degeneration with a parallel capacitor in a common emitter/source cascode amplifier? The DC input voltage sets the bias current.

[David Hess]

I have tested solid tantalum capacitors down to liquid nitrogen temperatures but the application was non-critical decoupling.

To cover a frequency of 100 Hz to 100 MHz, I would drop AC coupling if at all possible.  It is not like DC coupled amplifiers are difficult over that range.

[NiHaoMike]

Could the amplifier be encased in an insulated package with a heater to keep it at a somewhat less cold temperature?

[ezalys]

When I say cryogenic I mean one Kelvin — sorry I should have been more specific. Ultimately no capacitors are spec’d down this low but it’s low temperature physics lore what works well. Solid tantalums, and certain ceramics are acceptable. HEMTs and HBTs are acceptable.

I’m certainly not against DC coupling — I just don’t know how to do it for a 50 ohm input impedance! Could Marco and David say a bit more about designing the input to be 50 ohms and DC coupled? I’ve never designed an amp like this. Could you maybe point me to a reference or a schematic or draw me one?

Heaters are unacceptable — the point of having the amp cold is to reduce Johnson noise.

[David Hess]

When I say cryogenic I mean one Kelvin — sorry I should have been more specific. Ultimately no capacitors are spec’d down this low but it’s low temperature physics lore what works well. Solid tantalums, and certain ceramics are acceptable. HEMTs and HBTs are acceptable.

Usually the most practical thing to do is test and qualify parts for cryogenic operation which is what I did.  I knew to avoid capacitors that relied on a liquid dielectric which would just freeze.

I’m certainly not against DC coupling — I just don’t know how to do it for a 50 ohm input impedance! Could Marco and David say a bit more about designing the input to be 50 ohms and DC coupled? I’ve never designed an amp like this. Could you maybe point me to a reference or a schematic or draw me one?

What are the input and output impedances?  Is everything 50 ohms?  What output power level is required?

If you are limited to HEMTs and HBTs, then circuit configurations will be limited.  I only worked down to liquid nitrogen temperatures where common analog ICs still worked.

[Bud]

50Ohm @ 100Hz - why is this limitation ?

[Marco]

I’m certainly not against DC coupling — I just don’t know how to do it for a 50 ohm input impedance! Could Marco and David say a bit more about designing the input to be 50 ohms and DC coupled?

Sorry, I forgot the pre-amp had to be terminated to 50 Ohm AC.

That said, designing a pre-amp with a large DC offset at the output which has to be AC terminated is just rude. They should have known it would cause problems with available capacitors  ... that's saddling other people with your bad design. That amp should have been servo'd to a low offset.

[Gerhard_dk4xp]

When I say cryogenic I mean one Kelvin — sorry I should have been more specific. Ultimately no capacitors are spec’d down this low but it’s low temperature physics lore what works well. Solid tantalums, and certain ceramics are acceptable. HEMTs and HBTs are acceptable.

I’m certainly not against DC coupling — I just don’t know how to do it for a 50 ohm input impedance! Could Marco and David say a bit more about designing the input to be 50 ohms and DC coupled? I’ve never designed an amp like this. Could you maybe point me to a reference or a schematic or draw me one?

Methinks that Infineon wrote that their SiGe transistors had 200 pV/rtHz at cryo temperatures IIRC.
What is Vbe at cryo? What Beta can we expect? That should not be great, also. 

When the second stage is a cascode, that usually does not add noise worth to talk about.
It has a large output impedance, allowing a large load resistor and thusly, a lot of gain.
But when the beta is really small, effective shot current may be not so small.

In short, we can only generate hot air when we do not know the 1st stage. If it makes some
healthy gain, it should determine the noise, at least most of it.
A Cascode would give the first transistor the opportunity to use a very low Vce, thus low
heat dissipation and no headroom required for AC Vds/Vce. And the cascode
would live on the very same current. Output impedance could be healed by a follower,
That could sit in some distance to the very cold.

Where do the 50 Ohms come from? Just because there is a network analyzer on the table?

Cheers, Gerhard

[David Hess]

50Ohm @ 100Hz - why is this limitation ?

It is a problem because passing 100 Hz with a 50 ohm impedance requires a capacitance of at least 32 microfarads which is inconveniently large in an RF application.

[ezalys]

Haha we can’t curse the first stage designers too much since we came up with it! It’s two transistors in a DIP carrier and bunch of wire bonding, all designed to be immersed in 300 mK 3He. The amp just can’t be that complex. It’s two HEMTs cascaded in their ohmic region. They dissipate maybe 30 microwatts total. 1 milliwatt total will boil all the 3He at once.

That said, we can burn more power at 1 kelvin. 50 ohms since that’s the cryogenic coax we have, and these stages are maybe 2 feet apart.

[Marco]

Lets say about 10 microwatt are available to drive 50 ohm then around 50 mv are available as signal. That doesn't leave an awful lot of offset for an active solution. It would have to involve a longtailed pair, but I don't know if Vbe matching will be good enough.

You could consider bringing out the signal from the preamp to a higher temperature regime with a high resistance low frequency connection, then you can servo the supply rails with a nice opamp.

Then you can make the second amplifier DC coupled with a differential pair (with a resistor between the emitters to set the gain I mean, not a discrete opamp). Offset won't matter as long as it doesn't saturate the amplifier, offset here won't cause dissipation in the primary amplifier, unlike with the servo. No capacitors needed except tiny ones to tune the response and for the supply.

[KE5FX]

Tantalums are fine, just put two of them back to back.  They act like forward-biased diodes when reverse-biased beyond their polarizing voltage, and they behave as non-polarized capacitors below that. 

Or at least they do at 25C...

[Gerhard_dk4xp]

You could also exchange some gain for a smaller capacitor.
Gain is cheap and your capacitor choice may be limited.

It does not need to be so extreme like the example.

[Kleinstein]

From the RF side the tantalums can be OK. However I don't really know how good they behave at the low temperature. If concerned with the RF part one could have a MLCC (C0G) in parallel.  Often one could move the capacitor to the ground side, e.g. to provide some offset to the circuit. It can still be tricky with the termination part - you likely want the 50 Ohms termination to be AC only, so the low temperature stage does not have to provide so much DC current.

One may not need perfet termination for the whole frequency band. So for the termination part a much smaller capacitor (e.g. 100 nF range) may be acceptable. It would complicate the transfer function however. The signal offset could be compensate by a DC servo from the ouput side. This may be in the extremes even be at higher temperauter / room temperature, but would need another wire down. If the DC drift to compensate for is small the DC servo can get away with a smaller capacitor, though at a higher voltage.