Stabilizing pHEMTs

another

bit

lead.

NPN

but

bypassing.

But

since

bypass

Thanks, Fred. The Dremelled hand-made proto oscillated at about 12 GHz, which I measured by watching the amplitude go up and down as I moved my hand within an inch or two of the surface. It went from peak to valley in about 1/4 inch, ergo, roughly 12 GHz.

I have a Tek 11802 scope that's fast enough to see that, but my HP 8568B spectrum analyzer tops out at 1.5 GHz. I might get an 8566B one of these times, but haven't needed it for anything else so far. At IBM, I had an HP 70000 series analyzer with tracking generator that went to 18 GHz. I could probably find one of those on eBay, but you have to mix and match the modules to get it to work right, which is frustrating and expensive. On the other hand, the amount of time I've spent on this one would have bought me one easily.

I'll try snubbing the BJT emitter and see if that helps.

Cheers

Phil Hobbs

--
Dr Philip C D Hobbs
Principal Consultant
ElectroOptical Innovations LLC
Optics, Electro-optics, Photonics, Analog Electronics

160 North State Road #203
Briarcliff Manor NY 10510
845-480-2058

hobbs at electrooptical dot net
http://electrooptical.net

Once you get down to picoseconds, it's easier to do that part yourself, rather than trying to explain it to a layout person. A lot of what I do is sort of instinct anyhow, hard to explain. And PCB layout is fun, done in moderation.

--

John Larkin, President       Highland Technology Inc
www.highlandtechnology.com   jlarkin at highlandtechnology dot com   

Precision electronic instrumentation
Picosecond-resolution Digital Delay and Pulse generators
Custom timing and laser controllers
Photonics and fiberoptic TTL data links
VME  analog, thermocouple, LVDT, synchro, tachometer
Multichannel arbitrary waveform generators

I want a small programmable surface-mount electrical equivalent of a finger. A ferrite bead is the closest, so far.

Hey, there might be some seriousness in that. A couple of people make digitally programmable silicon capacitors, so adding some resistors might be feasible.

--

John Larkin, President       Highland Technology Inc
www.highlandtechnology.com   jlarkin at highlandtechnology dot com   

Precision electronic instrumentation
Picosecond-resolution Digital Delay and Pulse generators
Custom timing and laser controllers
Photonics and fiberoptic TTL data links
VME  analog, thermocouple, LVDT, synchro, tachometer
Multichannel arbitrary waveform generators

another

Cin,

bit

a

lead.

NPN

but

bypassing.

But

since

bypass

That's partly why I did it. The other part is that I have some proprietary things that I'm trying to get going, and I like to have my arms around all aspects of a project even if I'm not the only one working on it. A 1-year license for the pro version of Eagle was $250, which seemed like the right answer.

Cheers

Phil Hobbs

--
Dr Philip C D Hobbs
Principal Consultant
ElectroOptical Innovations LLC
Optics, Electro-optics, Photonics, Analog Electronics

160 North State Road #203
Briarcliff Manor NY 10510
845-480-2058

hobbs at electrooptical dot net
http://electrooptical.net

[...]

In the Gerbers I see the one immediately to the east go to C% but the one farther up north seem to go nowhere. If it really doesn't go anywhere there's an easy fix: The Dremel ... whirrrr ... gone :-)

--
Regards, Joerg

http://www.analogconsultants.com/

another

bit

lead.

NPN

but

bypassing.

But

since

bypass

A snubber is a good idea although tough at such a low impedance node. A bead might be easier but requires a cut. I think if Phil cleans up the layout around R2 that would go a long way. That only requires a bypass cap and a snippet of copper tape.

Check out their new 12.4GHz analyzer, came out very few weeks after I bought their 4.4GHz version:

Being skeptical at first I must say I am impressed now. Ok, the software is still a bit hokey but this thing has already paid itself in the few months I had it. Did a lengthy EMC debug with it a couple weeks ago and this saved us half a kilobuck alone in rental fees. Plus the rental wouldn't have allowed me to listen in via SSB demodulation which saved time during hunts.

I don't think they have a TG for that yet but I have their 4.4GHz TG.

But clean up the R2 connection. Without that it's all like roulette.

--
Regards, Joerg

http://www.analogconsultants.com/

another

bit

lead.

NPN

but

bypassing.

But

since

bypass

I've once found it to be more like 100pF+470R series to ground...

It was on a complicated multiphase PWM current source board, which was incomprehensibly unstable at low voltage. I have been driven crazy for two days when my magic finger (tm) finally found that the ICs ramp generator wasn't linear in the low duty cycle region and the added snubbfinger linearized it. Ten minutes of cut and try later the finger was declared to be about

100pF+470R (unless it was 100R+470pF) and all was working as it should.

--
Thanks,
Fred.

[...]

I've done that in product designs where picosecond inaccuracies had to be auto-calibrated out. Use a SD5400 as variable resistors with one of the FETs in the DC path to servo out the TC. For GHz work one probably needs a different chip of that sort, smaller geometries.

They became expensive for a while but unlike crude oil they came back down nicely. Well under $2.

--
Regards, Joerg

http://www.analogconsultants.com/

another

bit

lead.

NPN

but

bypassing.

But

since

bypass

He he, nice!

12GHz is pretty high. Just think that a 1nH via is 75R at that frequency for example.

--
Thanks,
Fred.

And the full wavelength is about one inch. Which makes that long trace from R2 really suspicious ... :-)

Even the little stub above R2 can make things resonate and then oscillate.

--
Regards, Joerg

http://www.analogconsultants.com/

another

Cin,

bit

a

lead.

NPN

but

bypassing.

But

since

bypass

Yup, so it's hard to ground the sources of the phemt with just a couple of vias. I suggest a topside copper pour, with a lot of vias, and maybe extend the topside pour to pick up other nearby grounds, and maybe a cap or RC from drain to source copper.

I like to make the ground plane layer 2, to keep the ground vias short.

--

John Larkin, President       Highland Technology Inc
www.highlandtechnology.com   jlarkin at highlandtechnology dot com   

Precision electronic instrumentation
Picosecond-resolution Digital Delay and Pulse generators
Custom timing and laser controllers
Photonics and fiberoptic TTL data links
VME  analog, thermocouple, LVDT, synchro, tachometer
Multichannel arbitrary waveform generators

another

Cin,

bit

of a

lead.

NPN

the

cap,

a

input

be

but

concern.

bypassing.

But

since

bypass

Q1

as

Those pHEMTs and SiGe:C NPNs are hotter than a two-dollar pistol. I wish they had better 1/f behaviour, though.

Cheers

Phil Hobbs

--
Dr Philip C D Hobbs
Principal Consultant
ElectroOptical Innovations LLC
Optics, Electro-optics, Photonics, Analog Electronics

160 North State Road #203
Briarcliff Manor NY 10510
845-480-2058

hobbs at electrooptical dot net
http://electrooptical.net

lp

a

z,

y y

=20

ISTM the pHEMT source is effectively "not grounded" at 12GHz. And, feedthrus might even couple and feed back (though I'm not saying they are).

1uF ceramic C1 won't be stiff at UHF either, not the ones I scouted anyhow.

An old r.f. rule was that inductive source + inductive load an oscillator makes. My first 1GHz amplifier breadboards much preferred oscillating at 5GHz to amplifying at 1GHz :-).

Phil, even a frequency counter would give you really good insight about if/how layout contributes to the oscillation. If a wire passed through Q1's source via(s) tunes f.osc up, for example...

Fred's finger-probe is a good idea. Fred needs to send Phil his finger.

--=20 Cheers, James Arthur

They actually do, provided they are small. Size is what mostly makes a modern MLCC suitable or not. 0805 is not so cool here, 0402 or smaller would be better. At least 0603. Trace lengths should essentially be zero and often it can't hurt to make the side with the DC voltage on it a small sub-plane.

Haven't tried 1uF myself since I use mostly 0.1uF but this one looks good:

If Phil has more volts on there maybe this one:

I always thought Murphy is causing oscillations :-)

Usually anything getting close to the area tunes f-osc.

For proper maintenance that should come with a box of Cote-du-Rhone, but only the good stuff.

--
Regards, Joerg

http://www.analogconsultants.com/

Oh, I have a pretty well educated finger of my own. Plus another nine if that one fails. ;)

The pHEMT has two source pads, both of which are grounded by a via right inside the pad. So the ground inductance is way below 1 nH. Apart from the bypass for R2, which Joerg pointed out, there aren't any actual traces longer than about 0.08 inch in the signal path.

The symptom of the oscillation is a DC bias shift that varies depending on where I put my fingers, even when they aren't actually touching the board. The bypass caps are 0603 MLC as well, so I'd expect them to look like ~1 nH inductors above about 10 MHz. What did you measure on yours?

I have a 20 GHz scope and an 18 GHz counter, but the oscillation is too small for the counter to see.

Cheers

Phil Hobbs

--
Dr Philip C D Hobbs
Principal Consultant
ElectroOptical Innovations LLC
Optics, Electro-optics, Photonics, Analog Electronics

160 North State Road #203
Briarcliff Manor NY 10510
845-480-2058

hobbs at electrooptical dot net
http://electrooptical.net

[...]

The trace from collector to the input of U1 is a lot longer. Same for the feedback path around U1. Or the path to U3. Some of that is avoidable, some isn't. The big rule in GHz-work is to "break the line of sight".

But you have to fix the R2 issue. I don't see any way to make this work reliably without fixing that. C5 just isn't going to be able to do its job where it is right now.

A DC shift usually suggests a pretty hefty oscillation. A drab-green helicopter circling the building might indicate that it got out :-)

That's where one of these would come in handy:

--
Regards, Joerg

http://www.analogconsultants.com/

o help

.

pec

ind a

2 GHz,

ed my

alley

ency

d

I didn't actually measure, but went by the datasheet impedance-vs-f curves. I might even have the datasheets in storage somewhere.

That said, 1nH for C1 (if it were that high) =3D 75 ohms at 12GHz, plus the feedthru. That's not trivial.

This (75-ish ohms) very roughly jives with Fig. 4 in this document:

I used 100pF bypasses at UHF, to avoid self-resonance problems IIRC.

(If the transistor's hot enough a suitable cap becomes an LC tank at some frequency, which the transistor will duly find and resonate.)

That said, by the AVX report I linked, typical SMD caps are quite a bit better than I thought. 1uF in 0603 may not be any worse ESL-wise than 100nF, or even 1nF or 100pF in 0603.

That's odd. Oscillations usually rail. I burned up a handful of 8GHz transistors as part of learning that!

A spectrum analyzer made measuring and detecting oscillations and tuning sensitivities trivial. The finger calibration took a little longer.

--=20 Cheers, James Arthur

help

c

d a

GHz,

my

ley

ncy=20

Digging up a few datasheets you're right--it's better than I thought. I did that work over 20 years ago--I'm not sure if capacitors have improved, or if I'm just not remembering something. I remember looking at bypass impedance carefully, and choosing smaller-valued caps for the highest frequencies as a result.

Something like 100nF || 1nF || 22pF was handy in certain situations. I think that effectively prevents resonance peaks too, since no two of the L-C tanks will resonate at any given frequency.

: 2227 3247

Yes, but even that gives you cues. E.g., if approaching one side tunes the thing higher, and the other side tunes it lower.

Also, the fact that a finger in the area changes anything at all is noteworthy--that's a very small amount of coupling.

An adjunct to the finger is a grounded piece of shield, placed between this and that to see if / where they're coupling.

Of course the coupling could be inside Q1 itself--I haven't scrutinized the specs--but external effects can be swiftly gauged and assessed with the methods we've been discussing.

Starting with solid grounded copper foil on top, then removing the absolute minimum sure saves an awful lot of mystery and learning in these situations. =20

--=20 Cheers, James Arthur

help

GHz,

Cool. We are just now starting to use "sideways" caps, like 0306 types. Looks like we can expect about half the ESL of regular caps.

--

John Larkin, President       Highland Technology Inc
www.highlandtechnology.com   jlarkin at highlandtechnology dot com   

Precision electronic instrumentation
Picosecond-resolution Digital Delay and Pulse generators
Custom timing and laser controllers
Photonics and fiberoptic TTL data links
VME  analog, thermocouple, LVDT, synchro, tachometer
Multichannel arbitrary waveform generators