ECL logic SPICE models

[Joerg]

Gentlemen,

Does anyone know of a source for SPICE models of ECL gates?
Unfortunately ON Semi only supplies IBIS models :-(

I know, I know, much of this will fall under trade secrets or NDAs and
can't be shared. Just wondering why some companies seem reluctant to
share this stuff. Makes no sense since providing models can only boost
sales. No, not the models some of you might think about ...

Alternatively, is there a recipe about how to boil one out of an IBIS
model file?

 

[Charlie E.]

Gentlemen,

Does anyone know of a source for SPICE models of ECL gates?
Unfortunately ON Semi only supplies IBIS models :-(

I know, I know, much of this will fall under trade secrets or NDAs and
can't be shared. Just wondering why some companies seem reluctant to
share this stuff. Makes no sense since providing models can only boost
sales. No, not the models some of you might think about ...

Alternatively, is there a recipe about how to boil one out of an IBIS
model file?

Hi Jeorge,
Unfortunately, while PSpice has ECL models, they use the PSpice
digital simulation models that were propriietary to PSpice...

Charlie

 

[Joerg]

IBIS models only model the package.

I have the Spice models... I designed much of it... I can't share it
;-)

Dang, I thought so :-(

 

[Joerg]

Hi Jeorge,
Unfortunately, while PSpice has ECL models, they use the PSpice
digital simulation models that were propriietary to PSpice...

I have a really old license of PSpice but AFAIK those weren't very
suitable when you wanted to "mis-use" the stuff for analog tricks.

 

[Joerg]

What is it you're trying to model?

Mostly linear behavior and change of prop delays with chip temperature.
But it's nothing for a project (yet), just kicking the tires. The newer
8GHz stuff looks very enticing but when I saw the prices I almost
dropped my coffee cup.

 

[Fred Bartoli]

John Larkin a écrit :

Heck, the Onsemi GigaComm stuff is only about $35 a gate. Hittite
wants $200.

John

DTL? (Dollar Transfer Logic)

 

[Joerg]

Heck, the Onsemi GigaComm stuff is only about $35 a gate. ...

Only? ...

Problem is, with many of my designs the whole solution including burden
(assembly, additional rigging costs, etc.) has to be lower than that.

... Hittite wants $200.

Made and sold for one defense project and nobody else is buying?

 

[Joerg]

The only way you're going to get a heap of 5 GHz logic for $10 is to
spin a custom IC and make tens of millions of them. $35 for a gate or
a flipflop is OK if you're building a $5000 piece of test equipment.

8-port gigabit ethernet switches are in the $60 retail range now, and
will surely be a lot cheaper soon. That's an amazing heap of fast
logic. A single 10 MBPS ethernet interface used to cost as much as a
nice new car.

Ok, large scale integrated stuff is a different game. But when $30 are
charged for a single inverter I'll do it with a few BFP620 transistors.
ft around 65GHz, SOT23, about 50 cents in qties

 

[Joerg]

I haven't had much luck doing picosecond-speed time-domain stuff with
discrete bipolars. A 65 GHz SiGe transistor has a beta of 10 at 6 GHz,
dropping at higher currents, and turns out to be a pretty pokey switch
in real life. Package parasitics tend to make it an oscillator if you
try to really use the speed potential. The narrowband RF boys can tune
the parasitics out, but wideband is different.

So far bipolars have been good to me, starting with the old BFS17 in the
80's. To this day I always keep a vial full of those on hand. Then
faster and faster ones showed up. Only problem is that most are from
large European companies and their sales channels are often, ahem, let's
say sub-par.

Take a look at some broadband stuff such as CATV amps. Many are done
with discrete BJT, most likely for cost and 2nd-sourcing reasons. But
yes, parasitics can get you and the layout is not at all trivial.

I have had good luck using discrete phemts as switches, but they're
more like $1 or so.

That I still have to try.

If you can tolerate 1-2 ns prop delays and 600 ps edges, some of the
cmos TinyLogic parts are great, and in your price range.

That's sometimes an order of magnitude too slow :-(

 

[Spehro Pefhany]

Only? ...

Problem is, with many of my designs the whole solution including burden
(assembly, additional rigging costs, etc.) has to be lower than that.



Made and sold for one defense project and nobody else is buying?

Cheap compared to high temperature, let alone space stuff.

 

[Joerg]

Cheap compared to high temperature, let alone space stuff.

My last circuit for that market came in well under that. Lock, stock and
barrel.

 

[Joerg]

[...]

So far bipolars have been good to me, starting with the old BFS17 in the
80's. To this day I always keep a vial full of those on hand. Then
faster and faster ones showed up. Only problem is that most are from
large European companies and their sales channels are often, ahem, let's
say sub-par.

Take a look at some broadband stuff such as CATV amps. Many are done
with discrete BJT, most likely for cost and 2nd-sourcing reasons. But
yes, parasitics can get you and the layout is not at all trivial.


That I still have to try.


That's sometimes an order of magnitude too slow :-(

Some of the cheapest gain-bandwidth around is in darlington MMICS. You
can get 4-8 GHz of stable, 20 dB-ish gain for under a buck. They work
fine in pulse apps, too.

8GHz/20dB for under a buck? That would be my kind of device. Have a
particular one in mind?

One issue I found with RF devices was that you can sometimes only obtain
small signal simulator models that aren't very helpful with pulse apps,
plus require software that costs more than your Volkswagen. Even with
big LDMOS parts I had that happen. One company (PolyFet) stuck out and
provided SPICE models. So the big competitor will be left out ... sorry,
NXP.

 

[Joerg]

John Larkin wrote:
[...]

Ok, large scale integrated stuff is a different game. But when $30 are
charged for a single inverter I'll do it with a few BFP620 transistors.
ft around 65GHz, SOT23, about 50 cents in qties
I haven't had much luck doing picosecond-speed time-domain stuff with
discrete bipolars. A 65 GHz SiGe transistor has a beta of 10 at 6 GHz,
dropping at higher currents, and turns out to be a pretty pokey switch
in real life. Package parasitics tend to make it an oscillator if you
try to really use the speed potential. The narrowband RF boys can tune
the parasitics out, but wideband is different.

So far bipolars have been good to me, starting with the old BFS17 in the
80's. To this day I always keep a vial full of those on hand. Then
faster and faster ones showed up. Only problem is that most are from
large European companies and their sales channels are often, ahem, let's
say sub-par.

Take a look at some broadband stuff such as CATV amps. Many are done
with discrete BJT, most likely for cost and 2nd-sourcing reasons. But
yes, parasitics can get you and the layout is not at all trivial.


I have had good luck using discrete phemts as switches, but they're
more like $1 or so.

That I still have to try.


If you can tolerate 1-2 ns prop delays and 600 ps edges, some of the
cmos TinyLogic parts are great, and in your price range.

That's sometimes an order of magnitude too slow :-(
Some of the cheapest gain-bandwidth around is in darlington MMICS. You
can get 4-8 GHz of stable, 20 dB-ish gain for under a buck. They work
fine in pulse apps, too.

8GHz/20dB for under a buck? That would be my kind of device. Have a
particular one in mind?

There are tons, like

http://www.minicircuits.com/products/amplifiers_monolithic.html#lee

http://products.rfmd.com/productselector.jsp?select_family=Amplifiers (BB and General)

http://www.triquint.com/prodserv/more_info/download.aspx?file=/docs/Datasheets/E/ECG003.pdf

Some of the SOT-89 devices are really interesting.

Those are cool. However, the ECG003 is over three bucks although the
datasheet looks great.

I don't simulate this stuff, I test. Most microwave parts are
characterized small-signal, S-params. Spice models are very rare in
this business. I know things about some of these parts that the
manufacturers don't know.

Lately I have simulated more, and then straight to layout. The one
that's in fab right now is a white-knuckle ride because I am using a
device in a weird way. Usually pans out though.

What often amazes me is how close to abs max people (and sometimes app
notes) bias RF devices. When it says 7V they bias them to 5V or 6V on
the collector/drain. Some day a huge pulse finds its way into the input,
almost saturates the device, then it lets go, the inductor shoots up ...
phut.

 

[Joerg]

[...]

Lately I have simulated more, and then straight to layout. The one
that's in fab right now is a white-knuckle ride because I am using a
device in a weird way. Usually pans out though.

What often amazes me is how close to abs max people (and sometimes app
notes) bias RF devices. When it says 7V they bias them to 5V or 6V on
the collector/drain. Some day a huge pulse finds its way into the input,
almost saturates the device, then it lets go, the inductor shoots up ...
phut.

I've run 7-volt-rated MESfets at 18 volts, and 2 volt RF schottkies at
6. I sometimes test parts to destruction and then back off some. The
performance is often worth the small risk.

I've seen that. But if you are a consultant and this goes into some
aircraft you can't design like that. Even with an indemnification clause
that can cause lots of grief should it go wrong in only one case. It's
got to be by the book.

Lots of RF devices seem to have voltage ratings that assume an RF tank
in the output, so that the actual instantaneous drain excursion is 0
to twice Vcc. The RF guys are really, really terrible when it comes to
realtime specs. You're lucky to get any DC curves. More often there's
an app circuit with an input match, an output match, and a gate bias
trimpot.

All they really offer in many cases is a set of S-parameters. Since I
design a lot of pulse circuitry I almost do a rain dance every time
there is just a snippet of SPICE data.

 

[Joerg]

John Larkin wrote:
[...]

One issue I found with RF devices was that you can sometimes only obtain
small signal simulator models that aren't very helpful with pulse apps,
plus require software that costs more than your Volkswagen. Even with
big LDMOS parts I had that happen. One company (PolyFet) stuck out and
provided SPICE models. So the big competitor will be left out ... sorry,
NXP.
I don't simulate this stuff, I test. Most microwave parts are
characterized small-signal, S-params. Spice models are very rare in
this business. I know things about some of these parts that the
manufacturers don't know.

Lately I have simulated more, and then straight to layout. The one
that's in fab right now is a white-knuckle ride because I am using a
device in a weird way. Usually pans out though.

What often amazes me is how close to abs max people (and sometimes app
notes) bias RF devices. When it says 7V they bias them to 5V or 6V on
the collector/drain. Some day a huge pulse finds its way into the input,
almost saturates the device, then it lets go, the inductor shoots up ...
phut.
I've run 7-volt-rated MESfets at 18 volts, and 2 volt RF schottkies at
6. I sometimes test parts to destruction and then back off some. The
performance is often worth the small risk.

I've seen that. But if you are a consultant and this goes into some
aircraft you can't design like that. Even with an indemnification clause
that can cause lots of grief should it go wrong in only one case. It's
got to be by the book.

Lots of RF devices seem to have voltage ratings that assume an RF tank
in the output, so that the actual instantaneous drain excursion is 0
to twice Vcc. The RF guys are really, really terrible when it comes to
realtime specs. You're lucky to get any DC curves. More often there's
an app circuit with an input match, an output match, and a gate bias
trimpot.

All they really offer in many cases is a set of S-parameters. Since I
design a lot of pulse circuitry I almost do a rain dance every time
there is just a snippet of SPICE data.

Somewhere in my bag-o-tricks is an S-Parameter-to-Spice-Model
converter (and vice-versa).

When I find it I'll post.

That would be very kind.

However, except for super large transmitter devices (and sometimes even
then) the S-parameters have been measured under small signal conditions.
Doesn't help much when pulsing them hard, then you are back to what John
said, experiments on the lab bench. Got one of those coming up. I am
going to do the first run with tubes again to see if the concept pans
out, in order not to have stuff blow up in my face. It's really nice to
know that one can still buy the mil version of the 6146 tube, at
affordable prices. So far I haven't blown one though, only tired out a pair.

 

[Joerg]

John Larkin wrote:
John Larkin wrote:
[...]

One issue I found with RF devices was that you can sometimes only obtain
small signal simulator models that aren't very helpful with pulse apps,
plus require software that costs more than your Volkswagen. Even with
big LDMOS parts I had that happen. One company (PolyFet) stuck out and
provided SPICE models. So the big competitor will be left out ... sorry,
NXP.
I don't simulate this stuff, I test. Most microwave parts are
characterized small-signal, S-params. Spice models are very rare in
this business. I know things about some of these parts that the
manufacturers don't know.

Lately I have simulated more, and then straight to layout. The one
that's in fab right now is a white-knuckle ride because I am using a
device in a weird way. Usually pans out though.

What often amazes me is how close to abs max people (and sometimes app
notes) bias RF devices. When it says 7V they bias them to 5V or 6V on
the collector/drain. Some day a huge pulse finds its way into the input,
almost saturates the device, then it lets go, the inductor shoots up ...
phut.
I've run 7-volt-rated MESfets at 18 volts, and 2 volt RF schottkies at
6. I sometimes test parts to destruction and then back off some. The
performance is often worth the small risk.

I've seen that. But if you are a consultant and this goes into some
aircraft you can't design like that. Even with an indemnification clause
that can cause lots of grief should it go wrong in only one case. It's
got to be by the book.


Lots of RF devices seem to have voltage ratings that assume an RF tank
in the output, so that the actual instantaneous drain excursion is 0
to twice Vcc. The RF guys are really, really terrible when it comes to
realtime specs. You're lucky to get any DC curves. More often there's
an app circuit with an input match, an output match, and a gate bias
trimpot.

All they really offer in many cases is a set of S-parameters. Since I
design a lot of pulse circuitry I almost do a rain dance every time
there is just a snippet of SPICE data.

Somewhere in my bag-o-tricks is an S-Parameter-to-Spice-Model
converter (and vice-versa).

When I find it I'll post.

...Jim Thompson

All my S-Parameter to Spice stuff has been updated on the SED page of
my website... except a good article by Microcap... get that from their
website, so I don't step on any copyright toes ;-)

Thanks. I assume you mean this link:

http://www.analog-innovations.com/SED/CreateS-ParameterSUBCKTinPSpice.pdf

Now I have to learn QuickBasic, whatever that is ;-)

 

[Joerg]

On Sat, 12 Sep 2009 10:01:22 -0700, Jim Thompson

John Larkin wrote:
John Larkin wrote:
[...]

One issue I found with RF devices was that you can sometimes only obtain
small signal simulator models that aren't very helpful with pulse apps,
plus require software that costs more than your Volkswagen. Even with
big LDMOS parts I had that happen. One company (PolyFet) stuck out and
provided SPICE models. So the big competitor will be left out ... sorry,
NXP.
I don't simulate this stuff, I test. Most microwave parts are
characterized small-signal, S-params. Spice models are very rare in
this business. I know things about some of these parts that the
manufacturers don't know.

Lately I have simulated more, and then straight to layout. The one
that's in fab right now is a white-knuckle ride because I am using a
device in a weird way. Usually pans out though.

What often amazes me is how close to abs max people (and sometimes app
notes) bias RF devices. When it says 7V they bias them to 5V or 6V on
the collector/drain. Some day a huge pulse finds its way into the input,
almost saturates the device, then it lets go, the inductor shoots up ...
phut.
I've run 7-volt-rated MESfets at 18 volts, and 2 volt RF schottkies at
6. I sometimes test parts to destruction and then back off some. The
performance is often worth the small risk.

I've seen that. But if you are a consultant and this goes into some
aircraft you can't design like that. Even with an indemnification clause
that can cause lots of grief should it go wrong in only one case. It's
got to be by the book.


Lots of RF devices seem to have voltage ratings that assume an RF tank
in the output, so that the actual instantaneous drain excursion is 0
to twice Vcc. The RF guys are really, really terrible when it comes to
realtime specs. You're lucky to get any DC curves. More often there's
an app circuit with an input match, an output match, and a gate bias
trimpot.

All they really offer in many cases is a set of S-parameters. Since I
design a lot of pulse circuitry I almost do a rain dance every time
there is just a snippet of SPICE data.
Somewhere in my bag-o-tricks is an S-Parameter-to-Spice-Model
converter (and vice-versa).

When I find it I'll post.

...Jim Thompson
All my S-Parameter to Spice stuff has been updated on the SED page of
my website... except a good article by Microcap... get that from their
website, so I don't step on any copyright toes ;-)

Thanks. I assume you mean this link:

http://www.analog-innovations.com/SED/CreateS-ParameterSUBCKTinPSpice.pdf

Now I have to learn QuickBasic, whatever that is ;-)

There are four articles there, each beginning (in the listing),
"S-Param...", all of which are elucidating.

Unless I understand something wrong the first three are for extracting
S-parameters from SPICE or PROBE (Orcad's version of SPICE).

But maybe I'll read them again and some more elucidation will come

 

[Joerg]

Jim Thompson wrote:
On Sat, 12 Sep 2009 10:01:22 -0700, Jim Thompson

John Larkin wrote:
John Larkin wrote:
[...]

One issue I found with RF devices was that you can sometimes only obtain
small signal simulator models that aren't very helpful with pulse apps,
plus require software that costs more than your Volkswagen. Even with
big LDMOS parts I had that happen. One company (PolyFet) stuck out and
provided SPICE models. So the big competitor will be left out ... sorry,
NXP.
I don't simulate this stuff, I test. Most microwave parts are
characterized small-signal, S-params. Spice models are very rare in
this business. I know things about some of these parts that the
manufacturers don't know.

Lately I have simulated more, and then straight to layout. The one
that's in fab right now is a white-knuckle ride because I am using a
device in a weird way. Usually pans out though.

What often amazes me is how close to abs max people (and sometimes app
notes) bias RF devices. When it says 7V they bias them to 5V or 6V on
the collector/drain. Some day a huge pulse finds its way into the input,
almost saturates the device, then it lets go, the inductor shoots up ...
phut.
I've run 7-volt-rated MESfets at 18 volts, and 2 volt RF schottkies at
6. I sometimes test parts to destruction and then back off some. The
performance is often worth the small risk.

I've seen that. But if you are a consultant and this goes into some
aircraft you can't design like that. Even with an indemnification clause
that can cause lots of grief should it go wrong in only one case. It's
got to be by the book.


Lots of RF devices seem to have voltage ratings that assume an RF tank
in the output, so that the actual instantaneous drain excursion is 0
to twice Vcc. The RF guys are really, really terrible when it comes to
realtime specs. You're lucky to get any DC curves. More often there's
an app circuit with an input match, an output match, and a gate bias
trimpot.

All they really offer in many cases is a set of S-parameters. Since I
design a lot of pulse circuitry I almost do a rain dance every time
there is just a snippet of SPICE data.
Somewhere in my bag-o-tricks is an S-Parameter-to-Spice-Model
converter (and vice-versa).

When I find it I'll post.

...Jim Thompson
All my S-Parameter to Spice stuff has been updated on the SED page of
my website... except a good article by Microcap... get that from their
website, so I don't step on any copyright toes ;-)

Thanks. I assume you mean this link:

http://www.analog-innovations.com/SED/CreateS-ParameterSUBCKTinPSpice.pdf

Now I have to learn QuickBasic, whatever that is ;-)
There are four articles there, each beginning (in the listing),
"S-Param...", all of which are elucidating.

Unless I understand something wrong the first three are for extracting
S-parameters from SPICE or PROBE (Orcad's version of SPICE).

But maybe I'll read them again and some more elucidation will come

Most math is bilateral ;-)

Ok, true, I was lazy here. Thought about some SW-routine where you slap
the datasheet on a scanner, press magic button, out comes SPICE
sub-circuit ))

 

[Joerg]

[...]

Ok, true, I was lazy here. Thought about some SW-routine where you slap
the datasheet on a scanner, press magic button, out comes SPICE
sub-circuit ))

I can do that for you... for a fee ;-)

I know

Right now I don't need to convert though because I just went over to
manufacturers who furnish SPICE models. But if we really need to and
it's crunch time I might take you up on that, or a client of mine would.

If you and your son-in-law (the computer guy) get together and write a
program that does this you could probably sell it to numerous
manufacturers. Every time a guy like me walks away because they don't
have SPICE models that means serious money is being left on the table.
Once the BOM is done and done, it takes a major earth-shattering event
to change it. Or very painful price cuts. I had sales reps almost in
tears when they realized they shoulda ...

 

[Joerg]

John Larkin wrote:
John Larkin wrote:
[...]

One issue I found with RF devices was that you can sometimes only obtain
small signal simulator models that aren't very helpful with pulse apps,
plus require software that costs more than your Volkswagen. Even with
big LDMOS parts I had that happen. One company (PolyFet) stuck out and
provided SPICE models. So the big competitor will be left out ... sorry,
NXP.
I don't simulate this stuff, I test. Most microwave parts are
characterized small-signal, S-params. Spice models are very rare in
this business. I know things about some of these parts that the
manufacturers don't know.

Lately I have simulated more, and then straight to layout. The one
that's in fab right now is a white-knuckle ride because I am using a
device in a weird way. Usually pans out though.

What often amazes me is how close to abs max people (and sometimes app
notes) bias RF devices. When it says 7V they bias them to 5V or 6V on
the collector/drain. Some day a huge pulse finds its way into the input,
almost saturates the device, then it lets go, the inductor shoots up ...
phut.
I've run 7-volt-rated MESfets at 18 volts, and 2 volt RF schottkies at
6. I sometimes test parts to destruction and then back off some. The
performance is often worth the small risk.

I've seen that. But if you are a consultant and this goes into some
aircraft you can't design like that. Even with an indemnification clause
that can cause lots of grief should it go wrong in only one case. It's
got to be by the book.


Lots of RF devices seem to have voltage ratings that assume an RF tank
in the output, so that the actual instantaneous drain excursion is 0
to twice Vcc. The RF guys are really, really terrible when it comes to
realtime specs. You're lucky to get any DC curves. More often there's
an app circuit with an input match, an output match, and a gate bias
trimpot.

All they really offer in many cases is a set of S-parameters. Since I
design a lot of pulse circuitry I almost do a rain dance every time
there is just a snippet of SPICE data.
Somewhere in my bag-o-tricks is an S-Parameter-to-Spice-Model
converter (and vice-versa).

When I find it I'll post.

That would be very kind.

However, except for super large transmitter devices (and sometimes even
then) the S-parameters have been measured under small signal conditions.
Doesn't help much when pulsing them hard, then you are back to what John
said, experiments on the lab bench.

The RF boys call that "load pull" testing. People make interfaced
load-pull boxes so you can automate exploring the nonlinear space of
large-signal RF amps. Between explosions, anyhow.

Mostly I just press on, eye-ball the rough L and C values, go into the
garage and see if I have enough suitable variable capacitor for a test.
Then it's off to the plumber's place for some copper tubing, to make the
coils. At Home Depot that's now called "plomeria".

google "load pull tuner"

Got one of those coming up. I am

My first job interview, I told the guy that I preferred tubes to
transistors because tubes were harder to blow up. The smug SOB said
"that won't do" and sent me away. My next interview, I said the same
thing, and Melvin laughed and hired me. I designed $200 million worth
of stuff for him.

I've had a similar event with a potential client a long time ago.
Proposed an unorthodox solution that seemed to be "too simple" for them,
not academic enough. They turned me down. A few years later this company
was at the financial end of the rope and went belly up. The sad thing is
that my solution would have potentially saved them, along with about two
dozen jobs :-(