Drat. NTE is gone, and they took the last 5-GHz PNP with them.

There must be some property of this universe that prefers N-type devices.

There seem to be a couple. The matter/antimatter asymmetry means that getting p-type behavior in a practical device relies on holes, and holes have much lower mobility than electrons.

The solid state physics notion of an electron is a bit confusing to lots of people, because despite sounding like an isolated particle, it’s just as much a collective phenomenon as a hole, with properties determined by the band structure of the solid.

Cheers

Phil Hobbs

Thanks. It’s not a 50-ohm system, so using those would be hard. I have thousands for personal use, but can’t put them in licensed designs, which is what I’m moaning about.

Cheers

Phil Hobbs

But you could use the Renesa HFA3096, which includes two 5.5GHz PNP parts.

Can you use mmics? There are some really cool, fast, cheap, low-noise things around. Just because the RF boys test everything at 50 ohms doesn't mean we have to use them at 50 ohms.

They even provide the S parameters, isn't that nice. 8 volts CEO is a bit limiting, though.

Someone must make the NTE2403 as far as I know they're just a re-seller, yeah? Who would have thought repackaging 2N3055s and selling them for $3 a pop wasn't a growth industry..

For the wraparound topology, which is the second-best follower I know of,

VDD 0------*---------* | | R | R | R / | |V *-------| BFT92 | |\ |--* \ In 0-->| CPH3910 | |--* | | | | | *---------*----0 Follower output | V (tail current source)

the BJT needs to be a PNP.

It's a nice circuit, because the PNP reduces the output impedance a lot without adding much noise at all--way better than an NPN follower after the FET.

Because of the local feedback, the transistors need to be fairly different in speed to maintain stability. The FET is about a 750-MHz device, so a 5-GHz PNP is great. The alternative would be a 100-MHz PNP, which would be too depressing to contemplate. :(

The very best follower topology I know about is a fancy bootstrapped version of the White cathode follower, where the feedback is applied via the tail source. That's much harder to stabilize, because there are three transistors in the local feedback loop, but on the other hand its gain is 0.9997 at baseband and above 0.995 at 10 MHz. (You can't readily measure those sorts of numbers directly, so I inferred them from its performance as a bootstrap.)

The reason I care about getting such accurate bootstraps is a bit subtle--probably I'd have enough bandwidth improvement with a gain of

Cheers

Phil Hobbs

The Renesas parts are very disappointing in actual use, on account of their huge Ree’, Rbb’, and package inductance, as I found out to my cost long ago.

Otherwise I’d be all over them.

Cheers

Phil Hobbs

Am 05.09.24 um 20:17 schrieb Phil Hobbs:

They helped me make a nice 20-30 ns 1:1000 time stretcher, even in these funny space-proof flatpacks with even more L.

Gerhard

Really terrible noise and log conformity though, which are what I cared about most.

I was making a 100-MHz laser noise canceller.

Cheers

Phil Hobbs

The CPH3910 is a jfet. Might an un-assisted PHEMT be better?

I’ve used them a fair amount in front ends, generally with a SiGe NPN cascode-slash-drain bootstrap. Their transconductance is a few times higher than a CPH3910’s, but not as good as the local-feedback circuit’s.

pHEMTs have very low drain impedance—the late lamented ATF38143 had a voltage gain of ~0.7 as a follower, even with a current sink in the tail.

Their amazing noise floor (~0.3 nV in 1 Hz) makes them well worth patching up, but they do take some patching.

Cheers

Phil Hobbs

What is the Vdd in your circuit above?

I learned about jfet impact ionization from AoE. Keep the drain voltage low!

It seems, my 1st try did not work

Am 05.09.24 um 16:49 schrieb Phil Hobbs:

I've played with Bob Widlar's FET follower replacement:

Gain seems to be 1.000 and noise is quite OK. I wonder how this would work with today's transistors such as BFP840 & friends. And no PNPs. I have not yet built it.

E1 makes a noisy Vcc to see how many nV/rtHz I need there, Q4, Q5 are just parking lots. V3 prevents Spice from crashing.

BUF602 is a fabulous part, but it's crazy noisy by Phil's standards.

Yup. I run them between 2.5 and 3 V_DS.

That gives full transconductance, low noise, low leakage, and low dissipation.

Cheers

Phil Hobbs

Fun. That’s half of the LM11 input stage?

Cheers

Phil Hobbs

If most your designs are similar enough, perhaps you can consilidate it into a single version and persuade Analog Devices to make it on their XFCB3 process, either if you pay NRE and the design belongs to you, sold as a custom part that they can only sell via you, or that they sell themself and somehow give you a royalty. If you had a patent it might help with that negotiation.

The other option I can imagine is that there might be a fab in China capable of makine fast PNPs and who might find some pride in selling a faster PNP than any western company does.

Am 06.09.24 um 14:47 schrieb Chris Jones:

Some years ago I asked Scott Wurcer, RIP (Designer of AD797) in a different context if AD could't make sth. like PNP or mixed MAT-02, 03,

cheers, Gerhard.