Anyone make PCBs with Othermill?

Lol. Bit of a "hand-wavy" explanation there :)

Was this before you added the FB on the BFP640 base?

I know when to quit. I learned enough from that breadboard to know that the switcher that I had in mind would take too much engineering to develop, and might never work well enough. The hysteretic switcher idea was interesting enough to try. The requirement was for amps of precise constant current with wide adjustment range and high output impedance from DC to about 1 GHz, in face of big fast load changes. I wouldn't try five different toplogies. The final design is a cascode linear regulator with heatsinking.

I like to build and solder and test stuff now and then, to get away from computer screens and telephones and people. The switcher breadboard was extreme from a complexity standpoint; most of my breadboards are just to test a part or a simpler circuit, something I can do in an hour or two. If we want to test a complex circuit, we lay out a real multilayer board, with ground and power planes, vias, solder mask, all that. While we're waiting for that to arrive, we can always work on other parts of a project. It just takes a little planning for that to be efficient. I'd be seriously unhappy if any of my people had to iterate 10 times on a circuit.

You're just not going to be able to make a medium-complex fast circuits with modern BGA and chip-scale parts on a milling machine. Maybe the technology that you are developing, RF or something, is suited to milling.

One way is to attach a variable resister (grounded via a large cap) and adjust it until the AC output is halved from the n/c output. Take the resistor out of circuit and measure the resistance; that's the resistive component of your output impedance.

I use this technique in LTSpice for measuring the impedance at a node.

Why do all Gerber viewers have such nasty colour schemes?

I do mostly one-offs for feasibility demos, so hand wiring makes a lot of sense. It's also therapeutic, like coding.

I could see combline filters being done that way, especially if the design software generated the NC file directly. Have to have a nice sharp mill, though--copper is nasty gooey stuff, especially if you're doing it dry.

Cheers

Phil Hobbs

That's a low-res screen capture from PADS. We have a color convention for the various layers.

Yeah, like working with wood or metal. Satisfying.

I can see RF, antennas and microstrip filters and MMIC testing, as being good places to use a PCB mill. I couldn't Dremel a serious filter.

Oh, it seems I misinterpreted, and talked about the output impedance.

Is it valid interpretation to say that the "drain impedance" is the impedance that appears in parallel with the drain resistor (assuming that's what it is) to form the output impedance?

I've used stacked abrasive discs spaced out with nylon washers, like a dado, to cut parallel traces for mounting SC-70s. Not too flexible, but did work.

Cheers

Phil Hobbs

I use the Bellin snap-apart adapter boards, except for picosecond stuff.

Me too. This was my first ATF38143 / BFP640 cascode amplifier, for an amplifier that was supposed to be able to measure 1 nA in a 100 MHz bandwidth at the shot noise limit, for a biochip application. The customer was the research arm of a large Far Eastern electronics firm. (Very very large.)

I told them it was probably impossible, since that's only 31 electrons per inverse bandwidth (one-sided). The shot noise of that is 5.5 electrons.

I got within 6 dB of the shot noise, though, which I thought was pretty good for a built-up circuit.

The first proto oscillated at around 12 GHz. I didn't have a microwave spectrum analyzer at the time, so I measured the frequency by an interesting manual wavemeter method: if I held my hand over the board at different heights, the oscillation amplitude varied periodically, with a period of a bit less than half an inch.

The actual board was fine, after a bit of Dremelling and adding a bunch of parallelled bypass caps. With 40-50 GHz transistors and lots of gain in a small space, half a nanohenry here and there can ruin your whole day.

Cheers

Phil Hobbs

The Dremelled proto never really worked right, but was good enough to show that the high frequency gain was adequate. The actual circuit board was much better, partly because of that nice 5-ohm ferrite bead, which is more like 50 ohms out in the gigahertz where I needed it.

The drain impedance of JFETs is a lot lower than your average BJT, but pHEMTs are really putrid that way--for a 20-mA pHEMT, the drain impedance can be 160 ohms. (That's an actual measured value for an ATF38143--the generally similar SKY65050 is a few times better, but its

The ATF38143 is so droopy that you can't even use it as a source follower, even with a really stiff tail current source.

Cheers

Phil Hobbs

...because you're going to make more dumb mistakes on the final product.

Phil, what does "droopy" mean?

Also, what is "drain impedance" and how do you measure it? I looked through the ATF38143 datasheet and can't find anything that comes close:

google wasn't much help either. A lot of books mention it but don't say how to measure it. There seems to be several different meanings, but they are not clear. I did find one reference in

that states on page 13, "The approximate drain impedance is obtained from Z(ohms) = VOS^2 / 2Po, where Po is the amp1ifier output power in watts."

This sounds a bit like matching a transmitter power output stage to the antenna, but I don't think that's what you meant since you measured it.

Can you help?

Thanks

It's simpler than that--drain impedance is just the reciprocal of the drain admittance, i.e. 1/(dI_d/dV_ds).

In BJTs, collector impedance is Early voltage divided by collector current. Normally that's a big number.

It needs to be high in order for a follower or current source to have good performance.

SiGe:C transistors such as the BFP640 have effectively infinite collector resistance, so cascoding pHEMTs with SiGe:C BJTs gives you the best of all worlds: noise temperatures below 30 K and high, linear gain.

Cheers

Phil Hobbs

Yes.

See also: plate resistance.

Phil, do you have any hints at "plate curves" for these beasts? Have triodes actually returned, in the sense of being a (at least somewhat) constant voltage plate/drain characteristic?

Tim

Thanks. That is a big help.

You have mentioned this a few times. For example, in the thread "Re: Very Low Power Preamp", you state:

"Besides the 1/f noise, the main down side of pHEMTs is that they have pathetically low drain impedances, down in the hundreds of ohms. That makes them useless for followers, for instance, and seriously limits the voltage gain you can get out of them. Cascoding them helps a lot."

Why does low drain impedance kill the follower performance?

From your previous post:

What do you mean by "droopy"? What does it do to the signal?

Phil, here is a related question.

In the thread "Re: emitter follower gain", you posted the following:

The post is

I searched through the thread you referenced and could not find the circuit. I searched google groups for all the LTspice circuits you posted, and searched your web site, but could not find it.

Do you have a copy of the circuit and if so, could you post it again?

Thanks

Don't havee it right handy, but it's near the end of the "electrometer front end for DMM" thread from May '14.

Cheers

Phil Hobbs

Because it looks like a resistor from source to drain, which forms a voltage divider with the transconductance and whatever is providing the source current. That makes the gain quite a bit less than 1. It also varies with V_DS, so the follower is intrinsically nonlinear, again even if you have a stiff bias current.

They're very fast, though, and very quiet in the flatband (the ATF38143's voltage noise is about 0.3 nV/sqrt(Hz), and its gate current is a bit under a nanoamp, so its current noise is pretty low, like 5-10 fA per root hertz. The other nice thing about them is that they're surprisingly stable. It isn't that easy to make a pHEMT oscillate, IME, whereas the SiGe:C transistors sing like little gigahertz birdies at the slightest provocation. (They're three times as fast, of course, and have higher gain.) I've successfully used this technique in hand wired protos like this one

but you have to keep the collector current down below a milliamp.

The pHEMT is just above the ferrite bead on the wiper of the bias pot, and the SiGe:C BJT is immediately to its left. All the 1/8-W resistors and leaded caps and stuff are in the bias circuitry.

See above.

Cheers

Phil Hobbs

Haven't measured, but they sure make crappy followers.

Cheers

Phil Hobbs