Looking for a FET to switch @ 20MHz

[Everton]

I am looking for a FET that will switch at a frequency of 20Mhz or higher for a function gen.  The problem I am having is that the parametric searches on newark, digikey etc.  don't seem to have the switching speed/delay as a selection.  How does one then look for a component barring looking at individual datasheets at rabdom?

Thanks

[codeboy2k]

I've never used this FET, but I read a thesis once where the author created a 75MHz resonant boost converter.  He used the ST PD57060.

So it works up to 75MHz at least, should work at 20.

[Zero999]

I think more information is necessary.

Are you talking about switching as in a multeplexer, a linear amplifier for the output stage or a square wave?

I've never used this FET, but I read a thesis once where the author created a 75MHz resonant boost converter.  He used the ST PD57060.

So it works up to 75MHz at least, should work at 20.

You'll probably find special techniques are used such as zero voltage switching which aren't applicable to this application.

[T3sl4co1l]

Although selection tools are often inadequate, this is one case where they are not (at least... among the distributors with excellent databases, like Digikey).  If you're having trouble selecting a part, it's probably due to your insufficient knowledge on the subject.  And, you'll probably have just as much trouble designing and building the circuit.  So, perhaps we should teach about how to use (and I mean, really *use*, in switching and RF applications) transistors first?

Tim

[codeboy2k]

I've never used this FET, but I read a thesis once where the author created a 75MHz resonant boost converter.  He used the ST PD57060.

So it works up to 75MHz at least, should work at 20.

You'll probably find special techniques are used such as zero voltage switching which aren't applicable to this application.

Yep. I just re-read the thesis.  Tuned impedances on the layout, and the load network and switch capacitance is also tuned so that they achieve ZVS at the drain.

I hadn't looked at that datasheet before now either. That particular part is an RF power transistor good up to 1 GHz. Probably overkill for the OP. 

I agree with T3sl4co1l, digikey does have a great database, I can usually find what I need using their parametric search.

[Alex Eisenhut]

I think more information is necessary.

Are you talking about switching as in a multeplexer, a linear amplifier for the output stage or a square wave?

Sounded like a mixer to me, or a job for diode switches.

[Everton]

Wow.  thanks for all the replies guys.

The application is a simple DIY function that I am building for fun/self education.  I have an AD9811 DDS chip controlled by a TIVA micro which generates a 1V peak to peak sin wave. 

I am generating a variable duty cycle square wave with a TLV3501 comparator from this, but want to be able to vary the amplitude.  I figure that I can control the output voltage of a linear regulator using PWM from the micro and have this variable voltage drive a set of complimentary FETs in an inverter or push-pull configuration which would be my variable amplitude output.  (see attachment).

I figured that a FET would not have aproblem with this switching speed but hadn't thought it would require much drive.  But now thatyou point it out, it makes sense.  Ic= C * dv/dt which at 20 Mhz would be large.

If you're having trouble selecting a part, it's probably due to your insufficient knowledge on the subject.  And, you'll probably have just as much trouble designing and building the circuit.  So, perhaps we should teach about how to use (and I mean, really *use*, in switching and RF applications) transistors first?

I would be happy to hear any suggestions and learn if you are willing to teach me.
I originally was going to use a linear amplifier to do this, but the high frequency components of the square wave means this will not work and so I figured a switching solution with power supply control would be the ticket.  I guess things are never quite as easy as you make them out to be 

[T3sl4co1l]

So you've got a signal source of some sort, and need to amplify and buffer it for 50 ohm output?

That's analog wideband amplifier territory, all the way.  No switching about it!

AD9811 doesn't exist... did you mean 9911??

First thing we need to know is, what is the output signal like?  AD9911 specifies open drain outputs, from a differential pair (http://www.analog.com/media/en/technical-documentation/data-sheets/AD9911.pdf Fig. 4).  In typical application, they draw 0-10mA (and 10-0mA, respectively), from a voltage very near AVDD.  And each one is about 3pF, and that's about all you need to know.

What's characteristic about an "open collector" or "open drain" output?  The source is essentially a current, and that's about it.  It's a sinking current, that'll only sink so far (the compliance), and only tolerate so-and-so voltage (probably, there are internal diodes, not shown, which will shunt the IOUT_P/N pins to AVDD if they go much above, hence the AVDD+0.5V limitation).

Nothing is a true current source/sink, so there is some impedance in parallel with it.  There will be a resistance corresponding to the devices used.  In MOSFETs (the technology used in this part -- it seems to be a CMOS chip), this effect is called channel length modulation.  The analogous term for BJTs is Early effect, and for tubes, it's just called plate resistance.  However, this resistance will be far, far higher (probably in the 10s of kohms to megaohms range!) than the other impedances connected to this node, so we can safely ignore this resistance.  The other impedance is due to capacitance, 3pF per pin.  This has no effect at DC (of course), but becomes dominant at higher frequencies.  The biggest challenge to getting high bandwidth from this device will be avoiding the effect of this capacitance.  Fortunately, at 20MHz, it's not much (2.6kohms).  This chip (the 9911, anyway) is capable of far more than this, which suggests a maximum limit on the resistance we can load it with: at 500MSa/s, it can output a maximum of 250MHz, where that capacitance will have a reactance of 208 ohms.

We also have the electrical limitation of simply not having much voltage to play with.  A +/-0.5V range is 1Vp-p, for 10mAp-p.  1V/0.01A = 100 ohms.

So we should consider loading the IOUT_P/N terminals with no more than 100 ohms for linear operation, and if we can push that even lower, we will extend bandwidth: the pin capacitance acts in parallel with the load resistance, forming a lowpass network.  The cutoff frequency of 100 ohms with 3pF is 530MHz -- pretty generous for a mere 20MHz generator, but it will have a subtle delay and phase shifting effect on the harmonics of square waves.  Subtlety counts, when you're doing test equipment!

By the way, the fundamental problem of making nice amplifiers is this: sure, you can apply feedback up the wazoo, and achieve comically exaggerated impedances -- like the "virtual ground" (ideally, zero ohm impedance) of the negative feedback summing node in an inverting amplifier -- BUT, this only applies at frequencies where the feedback is so comically large.  (Voltage feedback) op-amps only have maximum gain at pretty well DC, so this is useless to us in wideband circuitry!  The reality is, there will necessarily always be some voltage generated, and current flowing, around these circuit nodes; our job is not to attempt to maximize or minimize one or the other, but to optimize them both.

So ultimately, we need to consider the system as a whole: where are we coming from, and where are we going to?

In this case, we are starting with a resistance of 100 ohms or less, with a current of 10mA and under 1V swing.  We need to amplify this to the desired output (you didn't mention -- 5V, 15V, 1000V?), with enough current capability (or really, power capability in the abstract) to drive the desired load (50 ohms, most likely).

Now... do understand that all this fuss about crappy voltage-mode op-amps and comically-adjusted impedances only applies when you're pushing the bandwidth of the part.  Op-amps are available today with GBW in the GHz.  Not that they're necessarily easy to use -- they're pretty hot shit, and won't behave themselves in a poor layout -- but, given that limitation, they do allow an unprecedented reach for pushing crappy textbook circuits out into what used to be real-man's bandwidth territory.

So at this point, you can kind of decide whether you want to use op-amps or discretes.

You can probably put together a very nice circuit using a "fully differential" type op-amp to convert the DAC current into a general purpose voltage signal, then add gain and buffering with stock parts to get the line level signal.

Or you can forge through the discrete design, which even on a simplified level, is more complex, but perhaps more rewarding, even if it's more of historical than practical value anymore.

(The main advantage to discrete, these days, is economic: you can blow $30 easily on the IC solution, while the discrete amp can be built with a handful of cents-per-transistor plus a few power devices.  The IC approach will probably have passable noise performance, and not nearly the full bandwidth of the DDS.  But it'll be quick and dirty and easy to design.  The discrete design can be fine tuned until the cows come home, but can be designed for far lower noise than the DDS is even capable of, and if you want bandwidth from DC to light... you're welcome to try.)

Tim

[dannyf]

I am looking for a FET that will switch

The answer depends on what you meant by "switch".

[KM4FER]

T3sl4co1l,  What a great explanation of an Engineer's thinking.
Thanks for that.

earl...

[c4757p]

I am looking for a FET that will switch

The answer depends on what you meant by "switch".

Considering the amount of detail in preceding posts, was this reply meant to add anything other than a demonstration of your superior knowledge?

[Everton]

I'm not sure why I typed AD9811.  It's actually an AD9850, but the specs of the output DAC are the same as the AD9811 except for the capacitnace which is 8pF.  Even still this doesn't make the chip the limiting factor in the design by a long shot. 

My goal is to design a wideband amplifier that will support 20Mhz sine wave as well as square wave.  I'd like to control the output amplitude from 0 to 20Vp-p and have the ability to apply a DC offset of +/- 10V (ie. The entire dynamic range of the amplifier). 

I would like to do it with discrete components to learn more.  Slapping in an opamp with GHz of GBW becomes a layout challenge more than anything and just seems like cheating anyway.

I am familiar with standard amplifier topologies, but have never had to do any wideband design and don't really know where  to start with it.  Obviously getting transistors with low capacitances (be it FET or BJT) is key, as is the topologies used (ie. differential and cascode) but I am sure there is a lot more to it than that. 

I appreciate any help that is offered

[dannyf]

My goal is to design a wideband amplifier that will support 20Mhz sine wave as well as square wave.  I'd like to control the output amplitude from 0 to 20Vp-p

You could handle that easily with variable gain amplifiers ("VGA") in the analog domain, or similarly PGA. AD603 is one popular one (because it is cheap). A dual gate mosfet (BF998 for example) can do that as well.

You may also think about current feedback amplifiers for speed.

[cosmicray]

I am looking for a FET that will switch at a frequency of 20Mhz or higher for a function gen.

There is an ARRL article discussing using an IGBT at 20 MHz. That might give you some information, and/or head you off in an interesting direction. It is retrievable as a PDF ...
http://www.arrl.org/files/file/QEX_Next_Issue/2014/May-Jun_2014/Horowitz.pdf

[T3sl4co1l]

Ok, so linear amplifiers.  Current input.

Understand that, for a clean 20MHz output, you will want at least a Bessel (minimum delay / linear phase) shaped frequency response, and for crisp square waves, probably you'll want much more -3dB bandwidth than this.

Consider the waveforms in the recent "what's your pulse look like?" thread: fully 100% of the entry-level generators are doing the exact same thing -- some DDS generator-in-one chip, with an output filter and buffer to clean it up in the specified bandwidth.  The cleanup quality varies, but they all share in common the poor step response corresponding to cutting bandwidth as sharp as possible.  In other words, the 20MHz "square" is nearly indistinguishable from the sine.  This is cheap, but ugly.

So, you can attempt to do at least as well, and maybe actually achieve even better.  Downside being, to get that crisp square wave, you need analog bandwidth well beyond 20MHz, like 50 or 100MHz.  But with a 125MHz maximum clock rate, you'd probably want to cut it below 62.5MHz.  So you'll have to decide where in that 20-60MHz range you want to cut it, and how sharp at that.  Filtering is a whole discussion in itself...

For sake of discussion, suppose we do the worst case, then: 100MHz solid.

Consider the basic amplifier types: common emitter, base and collector.

CE is good for general purpose amplification: the input impedance is modest (the base equivalent impedance is usually in the 100s to 1k's ohm range), and the output is a constant current (which can be used for modest voltage gain).  The input DC voltage is referenced to a Vbe drop above the emitter, which might itself be some amount above "GND" (using an emitter degeneration resistor), and the collector has to be above the emitter.  The main downside is Ccb, which acts as negative feedback -- Miller effect, making any useful amount of voltage gain *and* bandwidth difficult to achieve.  Miller effect only moderately affects the output side of Ccb (Cout ~= Ccb * (1 + 1/Vgain)), but strongly affects the input side (Cin ~= Cbe + Ccb * (1 + Vgain)).  Which means you need a much lower source impedance than you expected, for a given bandwidth; which means you'll end up with less total gain than you may've been hoping for.

CB is great for wideband use, but has the downside that the input impedance is very low (r_e ~= 25mV / Ic, a few ohms under typical conditions!).  Current gain is small (indeed, slightly less than 1: alpha = 1 - 1/hFE).  Input DC voltage is referenced below base, and output (collector) voltage is referenced above base (actually, it can pull slightly below the base voltage, because Vce(sat) < Vbe, but that's not much help).  The main advantage is the isolation of capacitances: Ccb works only to load the output, and Cce is very small so there is little feedback to worry about.

CC, more commonly known as the emitter follower, is complementary to both: it has high input impedance, low output impedance, low voltage gain (slightly less than 1, for a complementary reason as the CB amplifier having current gain slightly less than 1), and very high current gain.  Output DC voltage is referenced slightly below input, which has to be below collector voltage (again, input can actually be slightly above collector voltage, but this isn't usually very easy to achieve, or very useful, so we don't worry about it).

Some popular combinations of these exist:
CE into CB: an emitter (grounded base) driven by a collector (or other high impedance or current source signal) is called a cascode.  This has the low feedback of the common-base amplifier, with the modest input impedance of the common emitter (unhindered by Miller effect!) and high gain.  The main downside is the voltage drop: output voltage won't pull [much] below common-base voltage, which in turn has to be something above the input voltage, which is above GND.

CC into CB: two emitters connected together form what's called a differential amplifier.  If they're both the same polarity (NPN and NPN, or PNP and PNP), the total DC bias has to be drawn from the emitters  (double the bias current of a single transistor), and although we've started with the concept as one base grounded, the other as the input; what really matters is the base-to-base voltage.  Note also, Vbe's cancel (if matched), so we also get low offset voltage.  The circuit is symmetrical, so we can take the output from either collector: but understand that, if you draw voltage gain from the same side you're driving, you still get Miller effect.  If you take it from the opposite side (whose base is bypassed or grounded), you eliminate Miller effect, for the same reason it's gone from the cascode.

You can also make a differential amplifier using complements (PNP into NPN), but the Vbe's add rather than subtract, and rather than having an interesting current-steering effect (the collector currents move in opposite directions), the operation is strictly on-on.  Handy when it's needed, but not usually useful.

Complementary circuits aren't always obvious.  You can make further variations on these by inverting one or the other transistor, adjusting for DC bias (the voltages are flipped, and the currents add or subtract instead), and seeing what happens.  One example is the folded cascode: such an example is here,



Observe the 4N35 collector driving the top PNP emitter (with 51 ohm "pullup" resistor).  This is called a "folded cascode".  One advantage to this topology is the low input voltage: the 51 ohm resistor has a small voltage drop (it can be as low as 10s of mV, though it's about 400mV in this example), which places the input very near the respective supply (in this case, +12V), and allows the common-base transistor to produce a nearly rail-to-rail voltage swing.

Feedback is central and inseparable to all of this.  Note that the feedback loop is short: it only spans a few transistors.  This keeps phase shift low, and allows the loops to be stable at much higher frequencies.  ICs can afford to be vulgar in their available bandwidth, because they can use very small transistors capable of 10s of GHz on chip; you can get discrete transistors that fast, but you can't build a circuit that small, not with packages and pins and printed copper in the way.

Tim

[codeboy2k]

There is an ARRL article discussing using an IGBT at 20 MHz. That might give you some information, and/or head you off in an interesting direction. It is retrievable as a PDF ...
http://www.arrl.org/files/file/QEX_Next_Issue/2014/May-Jun_2014/Horowitz.pdf

A great read!  Thanks for posting the link.

From that article, there's a flat washer for a heat sink! awesome! That got me thinking you could probably stack alternating large and small flat washers for more area. I'm going to make a trip to Home Depot in the future

[Everton]

Thanks for that reply.  I'm familiar (to a certain extent) with all the configurations except for the folded cascode. I don't have the time to look at it in detail now, but will do so as soon as I am back home. 

So for 100Mhz bandwidth, would I be best off using a differential or cascode configuration at the first stage?  Also,what is the best way to get variable gain from the amplifier using a micro?  I figured I could PWM a pin and low pass filter it to give me an analog out that could be used to control the gain, but am I better of building a fixed gain amp and building a variable attenuator in front of the input, or is there a better way?

This sounds like its going to be the beginning of a new project that will last for a while!

Thanks

[T3sl4co1l]

One way is to modulate the current through a diff pair, which varies r_e (the internal emitter resistance) of the transistors in the pair; but this is only possible if r_e is dominant, which it probably won't be, because you'll be needing much R_E (external emitter resistance; emitter degeneration) to reduce gain and distortion and increase bandwidth.  This is a handy trick to keep in mind for variable RF amplifiers (where you don't mind the distortion or poor bandwidth -- the harmonics are filtered away, and the capacitance is resonated with inductors), audio synthesis, analog computing and other applications, though.

If you don't need continuously variable gain, you can set the base-emitter circuit for as much transconductance as you can reasonably deal with (within limits of impedances and distortion), and change the collector load resistance, which can be done with diode gates, transistors or relays.  Even better than PWMing, you can design a power-of-2 series of these for direct binary control.

Mind that the capacitance of each open switch plus its attached resistor adds an RC pole-zero pair to the response, which will do weird things near cutoff.  Each one might not be much (a few pF plus a resistor in the 100 ohm to kohm range), but together they add up, so don't put too many on a single stage, either.

In situations like these, it can help to actually short out the "inactive" segments; for instance, back in the 1A1 series input plugins for Tektronix 500 series scopes, they used a rotary switch to select from an array of resistor dividers, for the input attenuator.  All precision wirewound resistors, paralleled with compensation capacitors as needed.  Only one attenuator 'top' was connected to the input at any given time (keeps input resistance and capacitance constant; each one had a trimmer for input capacitance to ground, as well), and only one 'middle' was connected to the input buffer stage at any given time.  All unused attenuators were grounded, top and middle, with a switch contact having an inverse pattern: the input and output wafers connected "1-of-N", while the shorting wafers grounded "N-1-of-N".

Cheap scopes would skip the grounding step, with the result that, stray capacitance to the nearby attenuators, at the input and output ends, would introduce subtle time constants, resulting in "hook" -- anomalous rise or dip in the step response, due to the unmatched pole-zero pairs resulting from the stray coupling.

By the way, a good old fashioned attenuator divider or string would be perfectly acceptable as well -- you have the advantage that no nodes must be high impedance, so you can arrange to place it anywhere you like: at the DAC output, an intermediate stage, just before the output buffer, etc.  It's probably beneficial to choose more than a few, to spread it out.  Do the binary 1,2,4 at the DAC, add the 8,16,32 on the next stage, etc.

You get to use low impedance dividers, so compensation isn't an issue (or much of one).  You can select from taps; you can add or remove resistors in parallel and series; you can add or remove whole stages; etc.  You can use old school relay, diode or transistor switches, or use proper analog switch ICs, or integrated programmable attenuator chips!

I will add this note, however: avoid using large attenuation ratios early in the signal chain -- that increases SNR.  You can minimize noise at all settings by placing the attenuation as late as possible -- usually on the output itself.  An attenuator there will probably be relay switched (for reasons of power handling and linearity).

Tim