Fun with spherical cows

[Phil Hobbs]

So I have this gig coming in to build charge amps for a French ion
accelerator lab.

The specs are for 1 kHz - 60 MHz, ideally less than 0.3 nV/sqrt(Hz)
noise, when hung off a detector using 250-mm diameter plates, spaced by
5 mm, connected with a ~80 mm long, teflon insulated cable.

Fun.

SPICE says that it can be done stably, with realistic strays, using
three Mini-Circuits pHEMTs in parallel and a BFU520A NPN cascode.

I have some test boards on order, courtesy of Simon, so in a couple of
weeks we'll see if it can actually be made to work.

With things like this, the first goal is to keep them from oscillating
someplace up in the gigahertz, and the second is to ake them do what you
want.

Parallelling devices with 12-GHz fmax is a good way to make them
oscillate. The trick in this instance seems to be source degeneration
using Murata's magical BLV03VK600SNLD ferrite bead.

Unlike the vast majority of beads, they're specified by the impedance at
**5 GHz** instead of 100 MHz--these ones are 60 ohm, but you can get 220
ohm ones too (BLV03VK221SNLG).

I spent a bit of time using similar tricks to do a lab amp similar to
our LA22 product(*), but with 200 MHz bandwidth instead of 20, and 0.3
nV/sqrt(Hz) noise instead of 1.1 nV. The spherical cows think it can do
all that with 1.8 ns edges and no overshoot. We'll see!

<Loading Image...>
<Loading Image...>

The schematic is a bit busy, as it has to have a lot of strays put in to
get anything vaguely meaningful, and I had to scrunch it a bit
(connecting blocks using flags rather than wires in some cases) to make
it fit in the window. (The actual product schematic will probably be
fairly different, but we'll see.)

I have no idea how accurate the pHEMT model is, so I need to build some
test boards and find out. Fortunately we can get them monstrous cheap
from JLCPCB these days. (**)

Cheers

Phil Hobbs

(*) More details at <https://hobbs-eo.com/products/la-22-lab-amplifier>.

(**) JLCPCB raised a bunch of Series B money in late 2022, so maybe all
those cheap boards are being subsidized by VC money. Enjoy it while it
lasts!

[john larkin]

On Wed, 28 Feb 2024 17:11:52 -0500, Phil Hobbs

Post by Phil Hobbs
So I have this gig coming in to build charge amps for a French ion
accelerator lab.
The specs are for 1 kHz - 60 MHz, ideally less than 0.3 nV/sqrt(Hz)
noise, when hung off a detector using 250-mm diameter plates, spaced by
5 mm, connected with a ~80 mm long, teflon insulated cable.

Yikes. What's the capacitance? You might almost start with a
transformer!

I did a wire chamber amp array for CERN. That's a whole nother story.

Post by Phil Hobbs
Fun.
SPICE says that it can be done stably, with realistic strays, using
three Mini-Circuits pHEMTs in parallel and a BFU520A NPN cascode.
I have some test boards on order, courtesy of Simon, so in a couple of
weeks we'll see if it can actually be made to work.
With things like this, the first goal is to keep them from oscillating
someplace up in the gigahertz, and the second is to ake them do what you
want.

I have a 50 MHz triggered LC oscillator based on a SAV541. It likes to
oscillate at all sorts of frequencies, like 6 GHz (according to our 6
GHz scope.)

Mini has some new, basically repackaged, versions, which may have less
wirebond parasitics.

In my experience, the sources should be hard grounded. The magic beads
in the gate seem to help. But the layout really dominates.

Post by Phil Hobbs
Parallelling devices with 12-GHz fmax is a good way to make them
oscillate. The trick in this instance seems to be source degeneration
using Murata's magical BLV03VK600SNLD ferrite bead.
Unlike the vast majority of beads, they're specified by the impedance at
**5 GHz** instead of 100 MHz--these ones are 60 ohm, but you can get 220
ohm ones too (BLV03VK221SNLG).
I spent a bit of time using similar tricks to do a lab amp similar to
our LA22 product(*), but with 200 MHz bandwidth instead of 20, and 0.3
nV/sqrt(Hz) noise instead of 1.1 nV. The spherical cows think it can do
all that with 1.8 ns edges and no overshoot. We'll see!
<https://electrooptical.net/www/sed/500xLabAmpTransient.png>
<https://electrooptical.net/www/sed/500xLabAmp0.28nVto200MHz1.8nsTr.png>
The schematic is a bit busy, as it has to have a lot of strays put in to
get anything vaguely meaningful, and I had to scrunch it a bit
(connecting blocks using flags rather than wires in some cases) to make
it fit in the window. (The actual product schematic will probably be
fairly different, but we'll see.)
I have no idea how accurate the pHEMT model is, so I need to build some
test boards and find out. Fortunately we can get them monstrous cheap
from JLCPCB these days. (**)
Cheers
Phil Hobbs
(*) More details at <https://hobbs-eo.com/products/la-22-lab-amplifier>.
(**) JLCPCB raised a bunch of Series B money in late 2022, so maybe all
those cheap boards are being subsidized by VC money. Enjoy it while it
lasts!

[Phil Hobbs]

Post by john larkin
On Wed, 28 Feb 2024 17:11:52 -0500, Phil Hobbs

Post by Phil Hobbs
So I have this gig coming in to build charge amps for a French ion
accelerator lab.
The specs are for 1 kHz - 60 MHz, ideally less than 0.3 nV/sqrt(Hz)
noise, when hung off a detector using 250-mm diameter plates, spaced by
5 mm, connected with a ~80 mm long, teflon insulated cable.

Yikes. What's the capacitance? You might almost start with a
transformer!

I thought about it, but it's hard to get the ferrite losses low enough
to achieve those noise levels. The noise temperature (baseband to 200
MHz) is less than 30 K.

Post by john larkin
I did a wire chamber amp array for CERN. That's a whole nother story.

I'm listening intently. Say on!

Post by john larkin

Post by Phil Hobbs
Fun.
SPICE says that it can be done stably, with realistic strays, using
three Mini-Circuits pHEMTs in parallel and a BFU520A NPN cascode.
I have some test boards on order, courtesy of Simon, so in a couple of
weeks we'll see if it can actually be made to work.
With things like this, the first goal is to keep them from oscillating
someplace up in the gigahertz, and the second is to ake them do what you
want.

I have a 50 MHz triggered LC oscillator based on a SAV541. It likes to
oscillate at all sorts of frequencies, like 6 GHz (according to our 6
GHz scope.)

Yeah, something that resonates at 50 MHz probably has a lot of higher
resonances too, not to mention all those PCB traces.

Post by john larkin
Mini has some new, basically repackaged, versions, which may have less
wirebond parasitics.
In my experience, the sources should be hard grounded. The magic beads
in the gate seem to help. But the layout really dominates.

Yes, it does. However, it's quite possible to use 12-GHz f_max pHEMTs
in all sorts of circuits, including bootstraps. That low flatband noise
makes it worth all sorts of pain to get there.

Post by john larkin

Post by Phil Hobbs
Parallelling devices with 12-GHz fmax is a good way to make them
oscillate. The trick in this instance seems to be source degeneration
using Murata's magical BLV03VK600SNLD ferrite bead.
Unlike the vast majority of beads, they're specified by the impedance at
**5 GHz** instead of 100 MHz--these ones are 60 ohm, but you can get 220
ohm ones too (BLV03VK221SNLG).
I spent a bit of time using similar tricks to do a lab amp similar to
our LA22 product(*), but with 200 MHz bandwidth instead of 20, and 0.3
nV/sqrt(Hz) noise instead of 1.1 nV. The spherical cows think it can do
all that with 1.8 ns edges and no overshoot. We'll see!
<https://electrooptical.net/www/sed/500xLabAmpTransient.png>
<https://electrooptical.net/www/sed/500xLabAmp0.28nVto200MHz1.8nsTr.png>
The schematic is a bit busy, as it has to have a lot of strays put in to
get anything vaguely meaningful, and I had to scrunch it a bit
(connecting blocks using flags rather than wires in some cases) to make
it fit in the window. (The actual product schematic will probably be
fairly different, but we'll see.)
I have no idea how accurate the pHEMT model is, so I need to build some
test boards and find out. Fortunately we can get them monstrous cheap
from JLCPCB these days. (**)
(*) More details at <https://hobbs-eo.com/products/la-22-lab-amplifier>.
(**) JLCPCB raised a bunch of Series B money in late 2022, so maybe all
those cheap boards are being subsidized by VC money. Enjoy it while it
lasts!

Cheers

Phil Hobbs

[John Larkin]

On Wed, 28 Feb 2024 18:47:15 -0500, Phil Hobbs

Post by Phil Hobbs

Post by john larkin
On Wed, 28 Feb 2024 17:11:52 -0500, Phil Hobbs

Post by Phil Hobbs
So I have this gig coming in to build charge amps for a French ion
accelerator lab.
The specs are for 1 kHz - 60 MHz, ideally less than 0.3 nV/sqrt(Hz)
noise, when hung off a detector using 250-mm diameter plates, spaced by
5 mm, connected with a ~80 mm long, teflon insulated cable.

Yikes. What's the capacitance? You might almost start with a
transformer!

I thought about it, but it's hard to get the ferrite losses low enough
to achieve those noise levels. The noise temperature (baseband to 200
MHz) is less than 30 K.

Post by john larkin
I did a wire chamber amp array for CERN. That's a whole nother story.

I'm listening intently. Say on!

Image a bunch of planes, I think it was four, each two sheets of
aluminized mylar with a few hundred parallel wires between. The planes
are oriented at different angles. Some particle of scientific interest
blasts through the planes and ionizes some exotic gas and we want to
see of there's a coherent track and what direction it's in. I recall a
big mag field and a final scintillator. There are many millions of
events per second, all tangled, and people want enough data to get
Nobel prizes.

There's a high voltage between the sheets and the tiny wires so a
particle has a lot of avalanche gain, so we get a big, fast signal. I
used an ECL gate in linear mode as the first amps.

The real problem was the torrent of confusing data. My idea was
"progressive enrichment" to reduce the data rate to something
managible. There were two layers of brute-force ECL logic to decode if
the hits might qualify as a track. Then an FPGA, and finally the data
was logged to disk for analysis and prizes.

I did this for some people at UCLA, who rented a site at Cern, a
proton-proton collision thing. I got to meet Lisa Schlein, who you
have probably heard on NPR.

https://pubs.aip.org/physicstoday/online/17771/Obituary-of-Peter-Schlein

[Phil Hobbs]

Post by John Larkin
On Wed, 28 Feb 2024 18:47:15 -0500, Phil Hobbs

Post by Phil Hobbs

Post by john larkin
On Wed, 28 Feb 2024 17:11:52 -0500, Phil Hobbs

Post by Phil Hobbs
So I have this gig coming in to build charge amps for a French ion
accelerator lab.
The specs are for 1 kHz - 60 MHz, ideally less than 0.3 nV/sqrt(Hz)
noise, when hung off a detector using 250-mm diameter plates, spaced by
5 mm, connected with a ~80 mm long, teflon insulated cable.

Yikes. What's the capacitance? You might almost start with a
transformer!

I thought about it, but it's hard to get the ferrite losses low enough
to achieve those noise levels. The noise temperature (baseband to 200
MHz) is less than 30 K.

Post by john larkin
I did a wire chamber amp array for CERN. That's a whole nother story.

I'm listening intently. Say on!

Image a bunch of planes, I think it was four, each two sheets of
aluminized mylar with a few hundred parallel wires between. The planes
are oriented at different angles. Some particle of scientific interest
blasts through the planes and ionizes some exotic gas and we want to
see of there's a coherent track and what direction it's in. I recall a
big mag field and a final scintillator. There are many millions of
events per second, all tangled, and people want enough data to get
Nobel prizes.
There's a high voltage between the sheets and the tiny wires so a
particle has a lot of avalanche gain, so we get a big, fast signal. I
used an ECL gate in linear mode as the first amps.
The real problem was the torrent of confusing data. My idea was
"progressive enrichment" to reduce the data rate to something
managible. There were two layers of brute-force ECL logic to decode if
the hits might qualify as a track. Then an FPGA, and finally the data
was logged to disk for analysis and prizes.
I did this for some people at UCLA, who rented a site at Cern, a
proton-proton collision thing. I got to meet Lisa Schlein, who you
have probably heard on NPR.
https://pubs.aip.org/physicstoday/online/17771/Obituary-of-Peter-Schlein

Interesting. How did the first two layers work?

Cheers

Phil Hobbs

[John Larkin]

On Fri, 1 Mar 2024 13:04:49 -0000 (UTC), Phil Hobbs

Post by Phil Hobbs

Post by John Larkin
On Wed, 28 Feb 2024 18:47:15 -0500, Phil Hobbs

Post by Phil Hobbs

Post by john larkin
On Wed, 28 Feb 2024 17:11:52 -0500, Phil Hobbs

Post by Phil Hobbs
So I have this gig coming in to build charge amps for a French ion
accelerator lab.
The specs are for 1 kHz - 60 MHz, ideally less than 0.3 nV/sqrt(Hz)
noise, when hung off a detector using 250-mm diameter plates, spaced by
5 mm, connected with a ~80 mm long, teflon insulated cable.

Yikes. What's the capacitance? You might almost start with a
transformer!

I thought about it, but it's hard to get the ferrite losses low enough
to achieve those noise levels. The noise temperature (baseband to 200
MHz) is less than 30 K.

Post by john larkin
I did a wire chamber amp array for CERN. That's a whole nother story.

I'm listening intently. Say on!

Image a bunch of planes, I think it was four, each two sheets of
aluminized mylar with a few hundred parallel wires between. The planes
are oriented at different angles. Some particle of scientific interest
blasts through the planes and ionizes some exotic gas and we want to
see of there's a coherent track and what direction it's in. I recall a
big mag field and a final scintillator. There are many millions of
events per second, all tangled, and people want enough data to get
Nobel prizes.
There's a high voltage between the sheets and the tiny wires so a
particle has a lot of avalanche gain, so we get a big, fast signal. I
used an ECL gate in linear mode as the first amps.
The real problem was the torrent of confusing data. My idea was
"progressive enrichment" to reduce the data rate to something
managible. There were two layers of brute-force ECL logic to decode if
the hits might qualify as a track. Then an FPGA, and finally the data
was logged to disk for analysis and prizes.
I did this for some people at UCLA, who rented a site at Cern, a
proton-proton collision thing. I got to meet Lisa Schlein, who you
have probably heard on NPR.
https://pubs.aip.org/physicstoday/online/17771/Obituary-of-Peter-Schlein

Interesting. How did the first two layers work?
Cheers
Phil Hobbs

Well, it's been over 30 years. I recall that the first layer was to OR
pulses in a number of patches of each of two planes, and AND all the
possible patch pairs and see if there was a candidate track. Then do
that for a second pair of planes. Then AND again to see if four
patches had a simultaneous hit that suggests a track. If so, pass all
the wire hit pulses into the FPGAs. It would have been cool to
time-stamp every hit on every wire, but there wasn't the budget for
that, and it would have been a lot of data.

Nowadays it could be most all FPGA, since they are faster now.

Some of these collider sites save a petabyte of data per day. Big
money.

CERN wants to build an even bigger ring now. I don't think it's worth
it, as putting more bootprints on the moon isn't worth it.

[Phil Hobbs]

Post by John Larkin
On Fri, 1 Mar 2024 13:04:49 -0000 (UTC), Phil Hobbs

Post by Phil Hobbs

Post by John Larkin
On Wed, 28 Feb 2024 18:47:15 -0500, Phil Hobbs

Post by Phil Hobbs

Post by john larkin
On Wed, 28 Feb 2024 17:11:52 -0500, Phil Hobbs

Post by Phil Hobbs
So I have this gig coming in to build charge amps for a French ion
accelerator lab.
The specs are for 1 kHz - 60 MHz, ideally less than 0.3 nV/sqrt(Hz)
noise, when hung off a detector using 250-mm diameter plates, spaced by
5 mm, connected with a ~80 mm long, teflon insulated cable.

Yikes. What's the capacitance? You might almost start with a
transformer!

I thought about it, but it's hard to get the ferrite losses low enough
to achieve those noise levels. The noise temperature (baseband to 200
MHz) is less than 30 K.

Post by john larkin
I did a wire chamber amp array for CERN. That's a whole nother story.

I'm listening intently. Say on!

Image a bunch of planes, I think it was four, each two sheets of
aluminized mylar with a few hundred parallel wires between. The planes
are oriented at different angles. Some particle of scientific interest
blasts through the planes and ionizes some exotic gas and we want to
see of there's a coherent track and what direction it's in. I recall a
big mag field and a final scintillator. There are many millions of
events per second, all tangled, and people want enough data to get
Nobel prizes.
There's a high voltage between the sheets and the tiny wires so a
particle has a lot of avalanche gain, so we get a big, fast signal. I
used an ECL gate in linear mode as the first amps.
The real problem was the torrent of confusing data. My idea was
"progressive enrichment" to reduce the data rate to something
managible. There were two layers of brute-force ECL logic to decode if
the hits might qualify as a track. Then an FPGA, and finally the data
was logged to disk for analysis and prizes.
I did this for some people at UCLA, who rented a site at Cern, a
proton-proton collision thing. I got to meet Lisa Schlein, who you
have probably heard on NPR.
https://pubs.aip.org/physicstoday/online/17771/Obituary-of-Peter-Schlein

Interesting. How did the first two layers work?
Cheers
Phil Hobbs

Well, it's been over 30 years. I recall that the first layer was to OR
pulses in a number of patches of each of two planes, and AND all the
possible patch pairs and see if there was a candidate track. Then do
that for a second pair of planes. Then AND again to see if four
patches had a simultaneous hit that suggests a track. If so, pass all
the wire hit pulses into the FPGAs. It would have been cool to
time-stamp every hit on every wire, but there wasn't the budget for
that, and it would have been a lot of data.
Nowadays it could be most all FPGA, since they are faster now.
Some of these collider sites save a petabyte of data per day. Big
money.

Must have been a crap ton of ECL. (That’s ‘merde tonne’ for the SI crowd.)

Cheers

Phil Hobbs

[john larkin]

On Fri, 1 Mar 2024 22:52:57 -0000 (UTC), Phil Hobbs

Post by John Larkin
On Fri, 1 Mar 2024 13:04:49 -0000 (UTC), Phil Hobbs

Post by Phil Hobbs

Post by John Larkin
On Wed, 28 Feb 2024 18:47:15 -0500, Phil Hobbs

Post by Phil Hobbs

Post by john larkin
On Wed, 28 Feb 2024 17:11:52 -0500, Phil Hobbs

Post by Phil Hobbs
So I have this gig coming in to build charge amps for a French ion
accelerator lab.
The specs are for 1 kHz - 60 MHz, ideally less than 0.3 nV/sqrt(Hz)
noise, when hung off a detector using 250-mm diameter plates, spaced by
5 mm, connected with a ~80 mm long, teflon insulated cable.

Yikes. What's the capacitance? You might almost start with a
transformer!

I thought about it, but it's hard to get the ferrite losses low enough
to achieve those noise levels. The noise temperature (baseband to 200
MHz) is less than 30 K.

Post by john larkin
I did a wire chamber amp array for CERN. That's a whole nother story.

I'm listening intently. Say on!

Image a bunch of planes, I think it was four, each two sheets of
aluminized mylar with a few hundred parallel wires between. The planes
are oriented at different angles. Some particle of scientific interest
blasts through the planes and ionizes some exotic gas and we want to
see of there's a coherent track and what direction it's in. I recall a
big mag field and a final scintillator. There are many millions of
events per second, all tangled, and people want enough data to get
Nobel prizes.
There's a high voltage between the sheets and the tiny wires so a
particle has a lot of avalanche gain, so we get a big, fast signal. I
used an ECL gate in linear mode as the first amps.
The real problem was the torrent of confusing data. My idea was
"progressive enrichment" to reduce the data rate to something
managible. There were two layers of brute-force ECL logic to decode if
the hits might qualify as a track. Then an FPGA, and finally the data
was logged to disk for analysis and prizes.
I did this for some people at UCLA, who rented a site at Cern, a
proton-proton collision thing. I got to meet Lisa Schlein, who you
have probably heard on NPR.
https://pubs.aip.org/physicstoday/online/17771/Obituary-of-Peter-Schlein

Interesting. How did the first two layers work?
Cheers
Phil Hobbs

Well, it's been over 30 years. I recall that the first layer was to OR
pulses in a number of patches of each of two planes, and AND all the
possible patch pairs and see if there was a candidate track. Then do
that for a second pair of planes. Then AND again to see if four
patches had a simultaneous hit that suggests a track. If so, pass all
the wire hit pulses into the FPGAs. It would have been cool to
time-stamp every hit on every wire, but there wasn't the budget for
that, and it would have been a lot of data.
Nowadays it could be most all FPGA, since they are faster now.
Some of these collider sites save a petabyte of data per day. Big
money.

Must have been a crap ton of ECL. (That’s ‘merde tonne’ for the SI crowd.)
Cheers
Phil Hobbs

I never got to see the installation. The budget didn't include a trip
for me to Switzerland.

Switzerland is kind of boring anyhow.

[Bill Sloman]

<snip>

Post by John Larkin
CERN wants to build an even bigger ring now. I don't think it's worth
it, as putting more bootprints on the moon isn't worth it.

Not according to John Larkin, whose exquisite judgement also tells us
that climate change isn't happening (and wouldn't matter if it did) and
that Donald Trump has common sense - which is to say, vulgar greed.

[Cursitor Doom]

On Wed, 28 Feb 2024 17:11:52 -0500, Phil Hobbs

Post by Phil Hobbs
So I have this gig coming in to build charge amps for a French ion
accelerator lab.
The specs are for 1 kHz - 60 MHz, ideally less than 0.3 nV/sqrt(Hz)
noise, when hung off a detector using 250-mm diameter plates, spaced by
5 mm, connected with a ~80 mm long, teflon insulated cable.
Fun.
SPICE says that it can be done stably, with realistic strays, using
three Mini-Circuits pHEMTs in parallel and a BFU520A NPN cascode.
I have some test boards on order, courtesy of Simon, so in a couple of
weeks we'll see if it can actually be made to work.
With things like this, the first goal is to keep them from oscillating
someplace up in the gigahertz, and the second is to ake them do what you
want.
Parallelling devices with 12-GHz fmax is a good way to make them
oscillate. The trick in this instance seems to be source degeneration
using Murata's magical BLV03VK600SNLD ferrite bead.
Unlike the vast majority of beads, they're specified by the impedance at
**5 GHz** instead of 100 MHz--these ones are 60 ohm, but you can get 220
ohm ones too (BLV03VK221SNLG).
I spent a bit of time using similar tricks to do a lab amp similar to
our LA22 product(*), but with 200 MHz bandwidth instead of 20, and 0.3
nV/sqrt(Hz) noise instead of 1.1 nV. The spherical cows think it can do
all that with 1.8 ns edges and no overshoot. We'll see!
<https://electrooptical.net/www/sed/500xLabAmpTransient.png>
<https://electrooptical.net/www/sed/500xLabAmp0.28nVto200MHz1.8nsTr.png>
The schematic is a bit busy, as it has to have a lot of strays put in to
get anything vaguely meaningful, and I had to scrunch it a bit
(connecting blocks using flags rather than wires in some cases) to make
it fit in the window. (The actual product schematic will probably be
fairly different, but we'll see.)
I have no idea how accurate the pHEMT model is, so I need to build some
test boards and find out. Fortunately we can get them monstrous cheap
from JLCPCB these days. (**)
Cheers
Phil Hobbs
(*) More details at <https://hobbs-eo.com/products/la-22-lab-amplifier>.
(**) JLCPCB raised a bunch of Series B money in late 2022, so maybe all
those cheap boards are being subsidized by VC money. Enjoy it while it
lasts!

I'm still using ferric chloride but it's getting harder to find these
days. There ought to be a better way using lasers to cut the traces by
now. You wouldn't need that much power, would you? I'd have thought
maybe 4 or 5 Watts would do it for your typical FR4.

[john larkin]

Post by john larkin
On Wed, 28 Feb 2024 17:11:52 -0500, Phil Hobbs

Post by Phil Hobbs
So I have this gig coming in to build charge amps for a French ion
accelerator lab.
The specs are for 1 kHz - 60 MHz, ideally less than 0.3 nV/sqrt(Hz)
noise, when hung off a detector using 250-mm diameter plates, spaced by
5 mm, connected with a ~80 mm long, teflon insulated cable.
Fun.
SPICE says that it can be done stably, with realistic strays, using
three Mini-Circuits pHEMTs in parallel and a BFU520A NPN cascode.
I have some test boards on order, courtesy of Simon, so in a couple of
weeks we'll see if it can actually be made to work.
With things like this, the first goal is to keep them from oscillating
someplace up in the gigahertz, and the second is to ake them do what you
want.
Parallelling devices with 12-GHz fmax is a good way to make them
oscillate. The trick in this instance seems to be source degeneration
using Murata's magical BLV03VK600SNLD ferrite bead.
Unlike the vast majority of beads, they're specified by the impedance at
**5 GHz** instead of 100 MHz--these ones are 60 ohm, but you can get 220
ohm ones too (BLV03VK221SNLG).
I spent a bit of time using similar tricks to do a lab amp similar to
our LA22 product(*), but with 200 MHz bandwidth instead of 20, and 0.3
nV/sqrt(Hz) noise instead of 1.1 nV. The spherical cows think it can do
all that with 1.8 ns edges and no overshoot. We'll see!
<https://electrooptical.net/www/sed/500xLabAmpTransient.png>
<https://electrooptical.net/www/sed/500xLabAmp0.28nVto200MHz1.8nsTr.png>
The schematic is a bit busy, as it has to have a lot of strays put in to
get anything vaguely meaningful, and I had to scrunch it a bit
(connecting blocks using flags rather than wires in some cases) to make
it fit in the window. (The actual product schematic will probably be
fairly different, but we'll see.)
I have no idea how accurate the pHEMT model is, so I need to build some
test boards and find out. Fortunately we can get them monstrous cheap
from JLCPCB these days. (**)
Cheers
Phil Hobbs
(*) More details at <https://hobbs-eo.com/products/la-22-lab-amplifier>.
(**) JLCPCB raised a bunch of Series B money in late 2022, so maybe all
those cheap boards are being subsidized by VC money. Enjoy it while it
lasts!

I'm still using ferric chloride but it's getting harder to find these
days. There ought to be a better way using lasers to cut the traces by
now. You wouldn't need that much power, would you? I'd have thought
maybe 4 or 5 Watts would do it for your typical FR4.

Why not buy boards? Quick-turn 2 and 4-layer boards are cheap
nowadays. You get solder mask, silk, planes, and vias!