Re: Non linear miller capacitance
Hmmm, something seems suspicious here. This behavior is a property of the BJT model and not the simulator per se. All SPICE simulators have used the Gummel-Poon BJT model since before the days of the personal computer. The junction capacitance formulas used by Gummel-Poon are described here. I can only see two ways in which this capacitance would end up being constant with voltage.
1) If a generic NPN or PNP were used (in which CJC, MJC and VJC are not specified), CJC defaults to zero, which would give zero capacitance, independent of voltage.
2) If the model erroneously specified MJC=0, the capacitance would end up being CJC, independent of voltage. This would imply a faulty set of model parameters. If MJC is not specified, it defaults to 0.33, so simply failing to specify MJC would not cause the capacitance to be constant.
PHEONIX said:
Hmmm, something seems suspicious here. This behavior is a property of the BJT model and not the simulator per se. All SPICE simulators have used the Gummel-Poon BJT model since before the days of the personal computer. The junction capacitance formulas used by Gummel-Poon are described here. I can only see two ways in which this capacitance would end up being constant with voltage.
1) If a generic NPN or PNP were used (in which CJC, MJC and VJC are not specified), CJC defaults to zero, which would give zero capacitance, independent of voltage.
2) If the model erroneously specified MJC=0, the capacitance would end up being CJC, independent of voltage. This would imply a faulty set of model parameters. If MJC is not specified, it defaults to 0.33, so simply failing to specify MJC would not cause the capacitance to be constant.
Mistake
Hello Guys
I have made a mistake about the simulators not simulating non linear capacitance effects. I looked at my original source of information and it looks like I interpreted it the wrong way.
A question fo you Andy are Jfet models just accurate as BJT models. I know this sounds like a stupid question
Regards
Arthur
Hello Guys
I have made a mistake about the simulators not simulating non linear capacitance effects. I looked at my original source of information and it looks like I interpreted it the wrong way.
A question fo you Andy are Jfet models just accurate as BJT models. I know this sounds like a stupid question
Regards
Arthur
Re: Mistake
Hi Arthur,
It's not a stupid question at all!
I don't think the answer can be very clear-cut though. I'd think of it in two ways.
1) Which model has the potential to fit the data better, given the best possible job has been done fitting the parameters of the model to the data of the device?
2) What can one expect of typical models, say the ones you'd download from the manufacturer's web site?
Regarding the first question, the best information I've seen on this comes from Scott Wurcer of Analog Devices. He reports that for some BJT IC designs they've done, the design met every single specification requirement on the first pass, based on simulations alone. This strongly suggests that the quality of the bipolar transistor models they are using is extremely good. But to get models this good requires lots of effort, expense, expertise, and knowledge of details of the process parameters. It also requires measurements of transistor data that you don't see on typical datasheets of discrete devices. So the BJT models can be made to be extremely good, but only with much difficulty.
Models of discrete devices that you download are usually created by third parties under subcontract to the semiconductor company. These third parties only have access to data from the datasheet of the device. The nature of the subcontract may be such that they have an incentive to spend as little time as possible to get the model parameters for each device. For these reasons, the device models created by these third parties are not going to be anywhere near as good as the ones used by Scott and his team.
Regarding JFET models, these have a pretty simple structure, and I suspect that even if the maximum effort were put into fitting the parameters to the data of the device, they wouldn't be as good as the best possible BJT models. However, the simplicity of the model makes fitting the parameters to the device data much easier than with a BJT. So it may be that this ease of data fitting could result in better models in the case of a "typical" model downloaded from the device manufacturer.
I have seen very good and very bad examples of each, so I guess the answer is, "It depends".
PHEONIX said:
Hi Arthur,
It's not a stupid question at all!
I don't think the answer can be very clear-cut though. I'd think of it in two ways.
1) Which model has the potential to fit the data better, given the best possible job has been done fitting the parameters of the model to the data of the device?
2) What can one expect of typical models, say the ones you'd download from the manufacturer's web site?
Regarding the first question, the best information I've seen on this comes from Scott Wurcer of Analog Devices. He reports that for some BJT IC designs they've done, the design met every single specification requirement on the first pass, based on simulations alone. This strongly suggests that the quality of the bipolar transistor models they are using is extremely good. But to get models this good requires lots of effort, expense, expertise, and knowledge of details of the process parameters. It also requires measurements of transistor data that you don't see on typical datasheets of discrete devices. So the BJT models can be made to be extremely good, but only with much difficulty.
Models of discrete devices that you download are usually created by third parties under subcontract to the semiconductor company. These third parties only have access to data from the datasheet of the device. The nature of the subcontract may be such that they have an incentive to spend as little time as possible to get the model parameters for each device. For these reasons, the device models created by these third parties are not going to be anywhere near as good as the ones used by Scott and his team.
Regarding JFET models, these have a pretty simple structure, and I suspect that even if the maximum effort were put into fitting the parameters to the data of the device, they wouldn't be as good as the best possible BJT models. However, the simplicity of the model makes fitting the parameters to the device data much easier than with a BJT. So it may be that this ease of data fitting could result in better models in the case of a "typical" model downloaded from the device manufacturer.
I have seen very good and very bad examples of each, so I guess the answer is, "It depends".
Re: Mistake
Unfortunately the underlying mechanisms in JFET models are a 'hack' at best. I'm not up on what's out there in the rest of the world but we have JFET1 and JFET3 neither of which I can get to make curves (even close) like a 2SK170 for instance. The big money is not in JFET op-amps, you use a diff-pair as input devices and things are sort of OK.
PHEONIX said:
Unfortunately the underlying mechanisms in JFET models are a 'hack' at best. I'm not up on what's out there in the rest of the world but we have JFET1 and JFET3 neither of which I can get to make curves (even close) like a 2SK170 for instance. The big money is not in JFET op-amps, you use a diff-pair as input devices and things are sort of OK.
john curl said:
I agree that SPICE is limited by the quality of the models, and that things would be a lot better if there were better models out there. The poor quality of off-shelf manufacturers' models is frustrating, and that even applies to many of the BJT models. This certainly makes blind dependence on SPICE unwise in many cases.
Nevertheless, even with mediocre models, SPICE simulation and SPICE-based experimentation can provide extremely valuable insight. It is never a replacement for building a prototype in the discrete world in which we audio guys mostly live, but its use can help us get to a better prototype on the first shot very often. Moreover, I have found many times that when I had trouble with a protoype or did not understand something, I was able to figure it out by subsequently doing some SPICE simulation/experimentation. Its ability to let you probe and see things that you would not easily be able to do on the breadboard can be a remarkable source of insight.
BTW, John, you do use BJTs in the output stage, right? In fairness, of course, the models of the output transistors might be lacking.
Cheers,
Bob
john curl said:
Even so SPICE can just be a big calculator at times checking bias levels and current ranges over Idss and Vp variations. It automates a lot of tedious work most of which does not lead to any new insight. In fact if one is clever a visual display of what is going when changing a bunch of variables can give you a new idea.
Some circuits are like converging one of those 60's color TV's with 5 variable inductors and 6 pots that all interact. You could spend hours twiddling when someone with the "algorithm" could get it right in a few minutes.
Since my designs are simple, WHY should I model them, when I can just build and measure them? It is a faster, overall approach. I continue to remind you that I have been doing circuit modeling and simulation for the last 42 years. What does it take for anyone to realize that I don't dismiss circuit simulation. Also, to keep SY from attempting to put his foot out, let us do an ACCURATE spice simulation of SY's tube circuits, as they are even simpler. How about it SY?
john curl said:
John, I don't really care one way or the other. But I disagree, I find it faster overall to simulate before building.
You're the one who said, "I'm with Bob Pease on this one, don't need no stinkin' SPICE". I hope no one labors under the misconception that Bob is designing IC's these days.
You can, but I can't. Please don't compare apples to oranges. I design discrete. I have elaborate breadboards, with supplies included, to make solderless prototypes.
The super design that I made for Sound Technology 25 years ago, with 100KHz OPEN LOOP BANDWIDTH, and with about .001% or so distortion at 100KHz was made on an Elite 2 solderless breadboard. I never wired it together, before releasing it to ST to be used as an oscillatior buffer. By the way, it beat the IC's available at the time. That is why I was commissioned to design it.
You make IC's, and you also have good repeatable models, as I do not.
That is the difference.
In DIY, it can be a toss up as to which approach is easiest and best. While computer simulation is often comfortable for the armchair designer, it is prone to some inaccuracy.
Building a prototype can give more accurate results, but it is true that it is difficult to 'tweak'.
What is the problem here?
The super design that I made for Sound Technology 25 years ago, with 100KHz OPEN LOOP BANDWIDTH, and with about .001% or so distortion at 100KHz was made on an Elite 2 solderless breadboard. I never wired it together, before releasing it to ST to be used as an oscillatior buffer. By the way, it beat the IC's available at the time. That is why I was commissioned to design it.
You make IC's, and you also have good repeatable models, as I do not.
That is the difference.
In DIY, it can be a toss up as to which approach is easiest and best. While computer simulation is often comfortable for the armchair designer, it is prone to some inaccuracy.
Building a prototype can give more accurate results, but it is true that it is difficult to 'tweak'.
What is the problem here?
john curl said:
Its fine for everyone to do what they are most comfortable with and what works best for them. Nobody is holding it against you that you rely more on build-measure without simulation than others. By the same live-and-let-live token, you should limit your derision of those who use SPICE and THEN build and measure. SPICE is a wonderful tool, but it is way from perfect, especially in the wrong hands. That goes for models as well.
You build fine amplifiers with little or no dependence on SPICE, partly because of your strong skill, intuition and experience. I successfully built the MOSFET EC amplifier without use of SPICE at all (the SPICE user interface in 1982 on UNIX was terrible text-only). Based on my experience since then, If I had had LTSPICE or the equivalent back then, I'm sure I would have converged on that successful, high-performing design much more quickly. I can't honestly say whether the end result would have performed better or not had I used SPICE.
Every day I have reason to bemoan the poor quality of some manufacturers' SPICE models. It is incredible how far off they can be, and yet have many, many of the SPICE3 parameters. They often don't even get Vbe vs Ic right. A much simpler model created by a kid in one day just out of school would often be better than some of these models that the sub-contractors generate pretty much blindly. So there, I've complained. I can't believe OnSemi pays their sub-contractor for such cr@p models.
But, nevertheless, even simulations using way less-than-perfect modles are extremely useful and provide great insight.
SPICE can be frustrating and imperfect, but we should not throw out the baby with the bathwater. When I designed linear ICs at Bell Labs (including full complementary bipolar in the 1970's) I had to rely a lot on SPICE and we had to do our own models in many cases. Sometimes we would spend a day just trying to get the darned simulation to converge! Things have REALLY improved.
Cheers,
Bob
Bob Cordell said:
It's a tool, and you chose the tool for he job. Just the other day I was working on the 2nd generation of my ec amp when I had an alternative idea that looked good on paper but I wasn't sure I had the reasoning right. During a spare hour at night I quickly drew it in my sim and ran it, saw that I had made an error assuming something (the output stuck to a rail). A few more checks to be sure, and that was it. Learned something, saved me a lot of time, all in less than an hour. I thought that was pretty neat.
I find such 'proof of concept' sims, when you are out on a design that hasn't been written up earlier, very usefull. So, NOW I will wire the *other* idea up on the breadboard!
Jan Didden
Re: Re: BS
If GKlein get's a good laugh becuase of me, it is worth taking some bashing.
I like them jolly people.
Not the boring ones. (Not mentioning the name of PMA here)
Have some fun! Like Hannes express it
Whether you spice or not, use them bjt or them mosfets.
PS.
If Bob saw my belly, he wouldn't have to worry so much about his own.
DS.
Lineup
G.Kleinschmidt said:
Oh, c'mon. That was the funniest thing I've read in ages.
I want some of what Lineup is smoking.
Cheers,Glen
PS
Bob looks like he needs to go on a diet.
If GKlein get's a good laugh becuase of me, it is worth taking some bashing.
I like them jolly people.
Not the boring ones. (Not mentioning the name of PMA here)
Have some fun! Like Hannes express it
Whether you spice or not, use them bjt or them mosfets.
PS.
If Bob saw my belly, he wouldn't have to worry so much about his own.
DS.
Lineup
Re: Re: Re: BS
I managed to take some off, down to about 190 at 5'10", but I hope to take some more off on my upcoming vacation this Friday to the Grand Canyon and some other places where one can get a lot of exercise walking around.
Cheers,
Bob
lineup said:
If GKlein get's a good laugh becuase of me, it is worth taking some bashing.
I like them jolly people.
Not the boring ones. (Not mentioning the name of PMA here)
Have some fun! Like Hannes express it
Whether you spice or not, use them bjt or them mosfets.
PS.
If Bob saw my belly, he wouldn't have to worry so much about his own.
DS.
Lineup
I managed to take some off, down to about 190 at 5'10", but I hope to take some more off on my upcoming vacation this Friday to the Grand Canyon and some other places where one can get a lot of exercise walking around
.Cheers,
Bob
And I have bought myself a pair of cross-country skis.
You know those all plastic modern ones
.. with a fish-skin like pattern underneath.
No need to use ski-wax, no more.
Sweden is a land of cross-country skiing tradition.
I used to be quite fast & good, when i was young
.. on them skis made out of wood (hickory etc.)
This winter I will try to get in better body shape
by doing some exercise on white snow of Swedish Lapland (north, near arctic circle latitude)
http://en.wikipedia.org/wiki/Lapland,_Sweden
I use other measurements (world standard units, international agreement by scientific standards)
Heigth: 170 cm (1,70 meter)
Weight: unknown .. I do not use to weigh my self
probably something like like 85-90 kg
Normal weight: should be like 70-75 kg
regards
You know those all plastic modern ones
.. with a fish-skin like pattern underneath.
No need to use ski-wax, no more.
Sweden is a land of cross-country skiing tradition.
I used to be quite fast & good, when i was young
.. on them skis made out of wood (hickory etc.)
This winter I will try to get in better body shape
by doing some exercise on white snow of Swedish Lapland (north, near arctic circle latitude)
http://en.wikipedia.org/wiki/Lapland,_Sweden
I use other measurements (world standard units, international agreement by scientific standards)
Heigth: 170 cm (1,70 meter)
Weight: unknown .. I do not use to weigh my self
probably something like like 85-90 kg
Normal weight: should be like 70-75 kg
regards