How to make a super fast sampling head?

[Joerg]

I've done it twice - once for real, with an 800MHz clock and GaAs
counters for the coarse intervals and a built in time to voltage
converter to work out when the trigger arrived within the the 1.25nsec
clock period. My 800MHz clock was poor and we had about 60psec of
jitter on the sampling edges.

A few years later I redesigned the delay-generating part of the
circuit around a 500MHz clock (which would have been crystal
controlled with less than a picosecond of jitter) and ECLinPS counters
for the digital part of the delay, and I was planning on using the
MC100E195 for the fine delays - it offered 2nsec of delay range and
20psec resolution. The MC100EP195 looks even nicer.

The delays through the MC100E195 are temperature dependent, and the
design did depend on recalibrating these delays against the crystal
controlled clock every few minutes, by using the system to set up a
pulse-width modulated waveform and digitising the DC level to find
what each delay really was. We should have been able to run through
all 128 discrete delays within a millisecond or so.

That's where an electronically controlled servoed delay line comes in,
to do the calibration automatically.

 

[John Larkin]

Another part you might want to look at are those blazingly fast
transistors from your companies over there, Infineon and NXP. For
example the BFP620. Ok, it's a bipolar transistor but it sports a
whopping 65GHz ft and is very cheap. Just don't push it much past 2V.

I've played with the 45 GHz SiGe bipolars, and always found them to be
slow as time-domain amps or switches. And they love to oscillate in
wideband circuits.

PHEMTS on the other hand are phenomenal. Today I was testing a cheap
NEC part, NE3509M04. Gate capacitance measures under 0.8 pF, drain is
about 0.35, Rds-on is about 6 ohms at zero gate voltage, and it
switches on/off fast and clean with ecl-type gate swings. It has no
equivalent of bipolar saturation delay; pull the gate to -0.4, and it
just turns off.

The weird thing is that Rds-on has a negative TC. I've never seen a
fet do that before.

(Used my trusty old green Boonton analog c-meter. The bottom meter
range is 0 to 1 pF, and it has provision for offsetting strays, 2/4
terminal measurements, and dc bias injection.)

John

 

[Jan Panteltje]

Skyworks SMS7621-079, 0.25 pF schottky. I have a reel of them, 23
cents each.

John

Great, can afford 4 of those I think.

 

[Jan Panteltje]

Neat! That's a great deal.

Jan, in Europe you can get this brand via BFI Optilas.

Found them in .nl, their wesite sucks though.

In case you want
to look these up:
http://www.skyworksinc.com/products_display_item.asp?did=798

Got the pdf, needs some reading and getting used to very low inductances
like .3 nH

 

[Joerg]

Great, can afford 4 of those I think.

Hey, even a shotglass of Beerenburger with Friese Nagelkaas ist over
four Euros these days ...

 

[Joerg]

I've played with the 45 GHz SiGe bipolars, and always found them to be
slow as time-domain amps or switches. And they love to oscillate in
wideband circuits.

I haven't tried the BFP620 yet, running other tests right now. So far my
projects don't require more than the BFS17 or maybe the BFR93. But even
6GHz ft is probably considered a slowpoke in your world.

PHEMTS on the other hand are phenomenal. Today I was testing a cheap
NEC part, NE3509M04. Gate capacitance measures under 0.8 pF, drain is
about 0.35, Rds-on is about 6 ohms at zero gate voltage, and it
switches on/off fast and clean with ecl-type gate swings. It has no
equivalent of bipolar saturation delay; pull the gate to -0.4, and it
just turns off.

Thanks! Entered into my Dear Santa list.

The weird thing is that Rds-on has a negative TC. I've never seen a
fet do that before.

(Used my trusty old green Boonton analog c-meter. The bottom meter
range is 0 to 1 pF, and it has provision for offsetting strays, 2/4
terminal measurements, and dc bias injection.)

I am using my dipmeter for small Cs. Sometimes the analyzer but that
thing makes such a racket with its airflow.

 

[blackhead]

J Larkin was saying his super sampling head costs 750$.
I was thinking how can you make one yourself, what else is needed?

It seems complete and utter madness doing this stuff discretely in
this day and age. You'll have more problems getting the signal into a
fast ADC, without distortion.

 

[Joerg]

It seems complete and utter madness doing this stuff discretely in
this day and age. You'll have more problems getting the signal into a
fast ADC, without distortion.

So, how do you build a 50psec sampler non-discrete?

 

[John Larkin]

It seems complete and utter madness doing this stuff discretely in
this day and age.

The only thing I can say to counter your statement is that we do
picosecond electronics discretely in this day and age, and charge lots
of money for it. At the low to moderate quantity that specialized
electronics is sold, and the cost/yield of custom GaAs and other
exotic materials, discrete design is perfectly reasonable.

Surface-mount parts are so small and fast they have pretty much
destroyed the hybrid business; the Tek 20-50 GHz sampling heads were
hybrids. The "wall in the sky" for surface-mount design is at about
100 picoseconds, where things start to get tough.

You'll have more problems getting the signal into a
fast ADC, without distortion.

The point of equivalent-time sampling is that the adc can be
arbitrarily slow; the front-end acts as a picosecond-gate
sample-and-hold.


Really, all you need is the Digikey catalog and ebay and some
persistance to build good high-speed signal processing stuff.

John

 

[John Larkin]

Thanks! Entered into my Dear Santa list.

In stock at Digikey.

John

 

[John Larkin]

So, how do you build a 50psec sampler non-discrete?

The fastest samplers, well above 250 GHz by now, are custom GaAs
parts. They use a nonlinear transmission line (shock line) to sharpen
the edge from an external SRD, and integrate the sampling diodes at
the end. Megabucks of engineering.

Agoston Agoston, of Hyperlabs and formerly Tek, did a 20 GHz sampler
on FR-4 with surface mount parts, at least 10 years ago. It looks like
any ordinary pc board.

Tek and HP sort of called a truce at about 70 GHz, probably because
signals this fast berely go through coax or connectors. PSPL sells a
100 GHz shockline sampler, and I think maybe LeCroy has built a scope
around it.

John

 

[Spehro Pefhany]

In stock at Digikey.

John

Your recommendation seems to have cleaned them out.

Best regards,
Spehro Pefhany

 

[Joerg]

In stock at Digikey.

Nope, you guys must have cleaned them out.

 

[Joerg]

The fastest samplers, well above 250 GHz by now, are custom GaAs
parts. They use a nonlinear transmission line (shock line) to sharpen
the edge from an external SRD, and integrate the sampling diodes at
the end. Megabucks of engineering.

Agoston Agoston, of Hyperlabs and formerly Tek, did a 20 GHz sampler
on FR-4 with surface mount parts, at least 10 years ago. It looks like
any ordinary pc board.

Tek and HP sort of called a truce at about 70 GHz, probably because
signals this fast berely go through coax or connectors. PSPL sells a
100 GHz shockline sampler, and I think maybe LeCroy has built a scope
around it.

Ok, full custom it can be done, of course. But that's mega-bucks just in
NRE before you even have anything on the bench.

 

[Joerg]

The only thing I can say to counter your statement is that we do
picosecond electronics discretely in this day and age, and charge lots
of money for it. At the low to moderate quantity that specialized
electronics is sold, and the cost/yield of custom GaAs and other
exotic materials, discrete design is perfectly reasonable.

Surface-mount parts are so small and fast they have pretty much
destroyed the hybrid business; the Tek 20-50 GHz sampling heads were
hybrids. The "wall in the sky" for surface-mount design is at about
100 picoseconds, where things start to get tough.

AFAIK even stuff like this is done on circuit board (but not FR-4):
http://www.ahsystems.com/catalog/data/pdf/PAM-1840.pdf

I used to love hybrids, designed some of them. Active laser trim is a
fantastic design tool, you can do things you wouldn't even dream about
in regular design or chip design. But yeah, in the early 90's the bottom
fell out of that business :-(

The point of equivalent-time sampling is that the adc can be
arbitrarily slow; the front-end acts as a picosecond-gate
sample-and-hold.


Really, all you need is the Digikey catalog and ebay and some
persistance to build good high-speed signal processing stuff.

Plus one minor detail: Potential customers with enough budget.

 

[[email protected]]

AFAIK even stuff like this is done on circuit board (but not FR-4):http://www.ahsystems.com/catalog/data/pdf/PAM-1840.pdf

I used to love hybrids, designed some of them. Active laser trim is a
fantastic design tool, you can do things you wouldn't even dream about
in regular design or chip design. But yeah, in the early 90's the bottom
fell out of that business :-(






Plus one minor detail: Potential customers with enough budget.

Tell me about it - though my 500psec sampling head was stroboscopic
electron microscope, which sold for about $100,000 dollars before we
added our timing and data-collecting electronics.

 

[Robert]

I've played with the 45 GHz SiGe bipolars, and always found them to be
slow as time-domain amps or switches. And they love to oscillate in
wideband circuits.

PHEMTS on the other hand are phenomenal. Today I was testing a cheap
NEC part, NE3509M04. Gate capacitance measures under 0.8 pF, drain is
about 0.35, Rds-on is about 6 ohms at zero gate voltage, and it
switches on/off fast and clean with ecl-type gate swings. It has no
equivalent of bipolar saturation delay; pull the gate to -0.4, and it
just turns off.

The weird thing is that Rds-on has a negative TC. I've never seen a
fet do that before.

(Used my trusty old green Boonton analog c-meter. The bottom meter
range is 0 to 1 pF, and it has provision for offsetting strays, 2/4
terminal measurements, and dc bias injection.)

John

TriQuint Semiconductor has a Prototype Chip Option program where they sell
you a place on a shared Mask Set for a lot of their PHEMT & HBT processes.
You get a small amount of chips for a smaller price with the shared Mask
running at regular intervals. The details, and prices, are listed here.

http://www.triquint.com/prodserv/foundry/proto_sched_pco.cfm#sched3

There's also a larger shared Mask program, called the Prototype Development
Quickturn, PDQ, listed on another page.

I don't see their new combined HBT and PHEMT process, TQBiHEMT, listed on
the prototype schedule but it's possible the Web page hasn't caught up yet.


Robert H.

 

[John Larkin]

TriQuint Semiconductor has a Prototype Chip Option program where they sell
you a place on a shared Mask Set for a lot of their PHEMT & HBT processes.
You get a small amount of chips for a smaller price with the shared Mask
running at regular intervals. The details, and prices, are listed here.

http://www.triquint.com/prodserv/foundry/proto_sched_pco.cfm#sched3

There's also a larger shared Mask program, called the Prototype Development
Quickturn, PDQ, listed on another page.

I don't see their new combined HBT and PHEMT process, TQBiHEMT, listed on
the prototype schedule but it's possible the Web page hasn't caught up yet.


Robert H.

Cool. What we want is...


4 3
| |
+---+
|
d
+--g
| s
| |
+---+-----+----5
| |
| |
d d
1----g-----g
s s
| |
| |
+-----+
| |
| |
2 6


which looks a lot like an old GigaBit Logic depletion-load logic gate.
It works pretty well with the NEC parts, but integrated it would
really scream.

John

 

[Robert]

Cool. What we want is...


4 3
| |
+---+
|
d
+--g
| s
| |
+---+-----+----5
| |
| |
d d
1----g-----g
s s
| |
| |
+-----+
| |
| |
2 6


which looks a lot like an old GigaBit Logic depletion-load logic gate.
It works pretty well with the NEC parts, but integrated it would
really scream.

John

It may be worthwhile to you at those prices (bottom of the page). You'd know
better than me. But unless you're talking about larger FETs the part will
probably be dominated by the bond pads. You might be able to find someone
that can bond you up some die using less than standard size bond pads to get
more on an order.

There are free tools around to lay it out. TriQuint used to supply free
layout Libraries for IC Editor's ICED layout tools (Layout, DRC, LVS). I saw
from their web site recently that they've gone to open source and have a
free download.

TriQuint would be able to tell you about what tools their Library supports.
I used to layout customer designs in ICED and for your circuit all you'd
need to do would be to streach their standard FETs to the size you'd need.

http://www.iceditors.com/


Robert H.

 

[Robert]

It may be worthwhile to you at those prices (bottom of the page). You'd
know better than me. But unless you're talking about larger FETs the part
will probably be dominated by the bond pads. You might be able to find
someone that can bond you up some die using less than standard size bond
pads to get more on an order.

There are free tools around to lay it out. TriQuint used to supply free
layout Libraries for IC Editor's ICED layout tools (Layout, DRC, LVS). I
saw from their web site recently that they've gone to open source and have
a free download.

TriQuint would be able to tell you about what tools their Library
supports. I used to layout customer designs in ICED and for your circuit
all you'd need to do would be to streach their standard FETs to the size
you'd need.

http://www.iceditors.com/


Robert H.

I should mention that they used to supply all their device Models for PSpice
as well. You'd have to model the strays yourself but they had some guidance.
Or used to.

Robert H.