blast from past

Oops, good point. The TO39 version of the PZT2907 is actually the

2N2905.

Somebody used to make a cool push-on TO5 heatsink that was made of folded sheet metal, springy, looked like Lady Liberty's hairdo.

John

Multiples would complicate life, and we'd still have to get rid of the same heat the same way... big high-C copper pours.

Oh, we were wondering how small an output-drive resistor we could get away with. Say, 47 ohms from the gaafet output to the SMA/B connector. Worst-case dissipation will be maybe 0.6 watts, and that only if the customer programs +5 DC out and then shorts the output. Conclusion:

0603. With reasonable pads and traces, an 0603 has a hot-spot temp rise of 140 K/W, about the same as a 1206, and needs a couple of watts to actually desolder itself.

John

I remember those. But last time I only found slotted aluminum parts. Not enough springiness. Maybe some of the materials in the Lady Liberty versions were not kosher anymore. Copper beryllium comes to mind.

Sometimes you can get away with smaller SOT223 copper heatsink areas when you spec that layer >2oz.

--
Regards, Joerg

http://www.analogconsultants.com/

"gmail" domain blocked because of excessive spam.
Use another domain or send PM.

Oh man, you guys are really pushing it. Never saw any FR-4 discoloration around it? Or noticed a whiff of that high-amperage smell?

--
Regards, Joerg

http://www.analogconsultants.com/

"gmail" domain blocked because of excessive spam.
Use another domain or send PM.

Why not pick the best SMT part (several have been mentioned) and then instead of a big copper pour, put a fat trace to a through-hole that you can solder a "heat sink" to. That would stick up into your airflow. You might even find something machine-placable like a testpoint or a small inductor (to nothing).

--
Ben Jackson AD7GD

http://www.ben.com/

JOHN 2907 IS A TO18 WAY TO SMALL TO DISSAPATE 1 WATT

Right, it should be a 2905. That's the TO-39 equivalent of the PZT2907.

John

We were thinking about a flat faston tab or two. Some go up and then turn 90 degrees, which would add area.

Somebody should make some flat punched copper things with a couple of pins/tabs.

____________________________ | | | | | | | | | | | | | | | | | | | | ____ ______________ ______ | | | | | | | | | | | | | | | | -- --

John

They do. Ok, it's not copper but still conducts some heat: Modular RF dividers. Off the cuff I don't have a source but this was discussed here a while ago. Placing a dummy RCA jack can also work as those have three large tabs and a sizeable metal body sticking up like a tower. If you take a straight-up version and drill out its guts you might even get a wee draft :-)

Can't you void the heat sink area for a SOT223 from all other planes and traces or is this one of those super-dense boards? That should cut the capacitance way down.

[...]

--
Regards, Joerg

http://www.analogconsultants.com/

"gmail" domain blocked because of excessive spam.
Use another domain or send PM.

I couldn't find the full CENU60 datasheet, either, but I found a MPSU60 datasheet. I'll post it in the schematics group.

The series inductor may be the best solution. If you know the bandwidth of the voltage swing signal then you can pick an appropriate one. Hopefully it won't be too big.

Bob

--
== NOTE: I automatically delete all Google Group posts due to uncontrolled 
SPAM ==

I'm still wondering if there couldn't be some way to electrically isolate the SOT-223 tab from the heatsink, then drive the heatsink as a guard / shield, bootstrapping its capacitance.

(Yes, of course that could be a mess broadband--it all depends.)

Or maybe splitting the dissipation up in a cascode, where the 'lower' element is fast, and the 'rail' element takes most of the heat.

Or an upside-down NE-251-dual-gate GaAsFET sort of thing, if one could do such a thing. (Too bad they don't exist).

Or some such.

Cheers, James Arthur

Tricky. It would need a fast amp of some sort, driving a plane between the transistor's copper pour and the board ground plane. And by the time I do that, I may as well, hmmmm....

replace the current source with a series R-L whose top end is bootstrapped against my output node. A fixed dc offset in the bootstrap path defines the static current. Then I can use a fast opamp, or an NPN transistor, or a n-channel gaasfet, as the top device.

That's worth looking into!

John

Some looney suggested making our own pseudo-transistor: a round or square, 1 square inch roughly, PCB with a TO-223 transistor mounted in the center, then three wires/pins coming down to pin sockets on the main board. Lots of surface copper, up in the air stream, low C, small net pcb footprint. You could even put some other parts on it.

You could panelize these as breakaways and make a lifetime supply in an afternoon.

John

It's not clear what you are trying to accomplish.

You are not going to get far using descretes.

For 8pF:

Xc = 1 / (2 * pi * 8e-12 * 1e9) = 19.89 ohms reactance Not much of a current source.

Wideband drivers are available:

So it can be done. You could review your goals and find a different way to drive whatever you are trying to drive.

Or break out the wallet and find a commercial IC that does what you need to do.

It's worse that that: the 1 GHz is a square wave.

I think I mentioned that we're putting a couple of ferrite beads in series with the collector. That makes the very-fast-edge impedance good. The very slow impedance is of course excellent, since the transistor is in a closed loop, current sense resistor and opamp. It's the middle that's tough. The generic problem is how to make a super-wideband current source; bias tees have much the same issues, how to make a DC-daylight hi-Z inductor.

This circuit is supposed to make a logic level that has programmable Vhigh and Vlow, up to 5 volts swing, 50 ohm output impedance, fast (ballpark 150 ps) edges, and works from DC to 1 GHz. It's a "pin driver" of sorts.

Dearly departed Gigabit Logic used to make a GaAs IC that did just this; I still have a sample somewhere. It was about $200 a pop in the early 90's. I don't know of any other (available) IC that can do anything like this. The new SRS clock generator uses a Maxim laser driver chip, surrounded by lots of stuff, to make logic outputs up to

2 GHz, but the p-p swing is low and it does have a good deal of personality over frequency.

What does the Hittite chip cost? We've used their HMC465LP5, an 18 GHz distributed amp, as an optical modulator driver, but it's about $185 and a brutal power hog; and we'd probably need two per channel to get enough gain; about $3K for the amps alone! DC coupling would be a real nuisance, too.

John

BFQ149? The package is rated at 1W, so you might need two - particulary granting your problem with loading the output with capacitance to ground - and it looks like one of those Philips/NXP transistors where you start production with a handful of free samples against the day that NXP makes another batch. Mouser carries it and will supply small qunatities, but don't have any stock at the moment.

-- Bill Sloman, Nijmegen

GGHHHAAA!

I think the cone inductor works pretty good. Available in various frequency ranges.

[...]

Time for Plan B?

Why do you need a 5V swing? Most devices that can use 150pS edges have much lower logic swings.

The HMC-AUH232, DC to 45GHz, is $164.45 ea, qty 10-99. Oddly enough, the HMC-AUH312, 0.5GHz to 65GHz, is only $129.25, 10-99.

The HMC465LP5, DC to 20 GHz, is now $226.26 ea, qty 10-99. You got it at a good price for only $185.

You can see a complete list of Hittite amplifiers at:

Minicircuits has a selection also. Cheaper but the frequency response is a roller coaster:

I think most of these amplifiers are Darlington. The DC parameter only means you supply the external DC blocking capacitors for whatever frequency range you are interested in. That means finding a wideband capacitor, which may cost a bundle.

I don't think they are suited for true DC coupled applications. The thermal drift would be too great for most instrumentation needs. Also the logic zero output voltage would be too high.

Yup, Piconics and the Coilcraft clones. Expensive and very hard to apply, but they still don't solve the problems caused by collector capacitance on the other side. Oh, did I mention that we have about 1 square inch per channel?

But it basically works. I just need to get the current-source capacitance down, and keep the transistor from frying; so I can't surface-mount the PNP.

Other spoilsports have made that same comment. I just think 5 volts of swing would be a good selling point.

Oops, the AUH thing appears to be a bare die. And I need the "DC" thing. Not Hittite's definition of DC, which has burned me in the past, but *real* DC.

I said *real* DC, pardner. One can do the compound thing, split paths and recombine at the output, at a high price in size/cost/power/hassle. There the Hittites or TriQuint distributed amps look interesting.

I've been playing with variations on this circuit for a decade or so. Given my swing requirements, it may never be integrated, so it's still a classic circuit design problem. The really fierce Agilent and Anritsu pulse generators use custom GaAs, but seldom swing more than 2 volts.

John