Although spice is a useful tool to all involved in electronics, I have often found (as widely reported by others also) that the spice models available to us are often wildly inaccurate, sometimes by an order of magnitude, when compared with real life measured results. I breadboarded a basic common emitter amplifier stage a few years ago and compared measured results such as
amplification, bandwidth and distortion with simulated results of the same parameters for just half a dozen or so common BJT's I had to hand. The correlation was tenuous in most cases and scarily wide of the mark in others!
To get to the point, I am curious as to if anyone on here has done a more comprehensive and scientific study of this (or just good old practical experience) and can recommend some models for some BJT's/MOSFETS/FET's that can be trusted to simulate somewhere near reality??
It's all very well having thousands of models available for high Ft or low Cob or high gain linearity super modern devices but if they are not accurate then what's the point? Especially if in practice the difference between lets say a real 2N3055 and a simulated one is greater than the actual difference between a real 2N3055 and a real MJL3281
And no I haven't made that test..... Those part numbers are just random but I do seem to vaguely recall a real TIP41C in a CE amp measuring about 30 KHz -3dB and a simulation giving about 10 MHz -3dB so this is the sort of discrepancy we can be up against!
amplification, bandwidth and distortion with simulated results of the same parameters for just half a dozen or so common BJT's I had to hand. The correlation was tenuous in most cases and scarily wide of the mark in others!
To get to the point, I am curious as to if anyone on here has done a more comprehensive and scientific study of this (or just good old practical experience) and can recommend some models for some BJT's/MOSFETS/FET's that can be trusted to simulate somewhere near reality??
It's all very well having thousands of models available for high Ft or low Cob or high gain linearity super modern devices but if they are not accurate then what's the point? Especially if in practice the difference between lets say a real 2N3055 and a simulated one is greater than the actual difference between a real 2N3055 and a real MJL3281
And no I haven't made that test..... Those part numbers are just random but I do seem to vaguely recall a real TIP41C in a CE amp measuring about 30 KHz -3dB and a simulation giving about 10 MHz -3dB so this is the sort of discrepancy we can be up against!
Check Bob Cordell site. He has a couple of models for BJT's that are the most accurate models, possible to find.
Also, you have to know, how your spice software is treating this model.
For example, LTSPICE is switching from Gummel and Poon model to the Ebers-Moll model when certain parameters are not specified.
Also, you have to know, how your spice software is treating this model.
For example, LTSPICE is switching from Gummel and Poon model to the Ebers-Moll model when certain parameters are not specified.
Andy_c has done some improved models that are accurate.
Most known are the MJL3281A MJL1302A:
Improved SPICE Models for MJL3281A and MJL1302A - Section 1
Most known are the MJL3281A MJL1302A:
Improved SPICE Models for MJL3281A and MJL1302A - Section 1
Normally purchased product are pretty good such as that from MicroCap. When simulating and testing results are exactly the same. but as Eugene mentioned you have to know what you are simulating.
Stuff available that are for free are exactly that, free and you cannot blame anyone for the accuracy or applicability.
I have downloaded many manufacturers spice files and have not found any inaccuracies their models as these are based on actual measurement and not hearsay.
Maybe it is the quality of the software and calculations, making use of some parameters and ignoring others. I cannot see why manufacturers of components will provide false models, as their livelihood depends on capable engineers designing those components into products that is sold and pays salaries. I cannot imagine any reason for providing lies.
I have experienced some differences in results between free spice simulators with limited capabilities and professional purchased products.
Stuff available that are for free are exactly that, free and you cannot blame anyone for the accuracy or applicability.
I have downloaded many manufacturers spice files and have not found any inaccuracies their models as these are based on actual measurement and not hearsay.
Maybe it is the quality of the software and calculations, making use of some parameters and ignoring others. I cannot see why manufacturers of components will provide false models, as their livelihood depends on capable engineers designing those components into products that is sold and pays salaries. I cannot imagine any reason for providing lies.
I have experienced some differences in results between free spice simulators with limited capabilities and professional purchased products.
Well this could certainly explain some of the results I've had with LT spice
I'll certainly check out those models on Bob Cordells site! Thanks!
These are about the only improved models I was aware of and already have.
This is surely a prime example of how the original models from manufacturers are not always trustworthy....
I have most of the available simulation programs but must admit that when I had rogue results with a particular model on LT spice I didn't do the same test using another programme... Surely a model should generally be interpreted the same by various spice engines??? Or maybe not according to Eugene
The results with the TIP41C were using a manufacturers model!
Basically I'm thinking that a few known accurate models for "general purpose" small signal transistors, plus a few commonly used drivers and a few outputs, would be much more useful than a vast range of possibly totally erroneous ones...
One problem is that BJTs vary widely within a given part number. That's one reason why Harold Black's clever circuit technique is so valuable
I'd love to have a cookbook procedure (follow step 1, follow step 2...) for measuring various BJT and FET parameters and creating a model; some of the model parameters are a bit mystifying- can any of the smart people help me?
I'd love to have a cookbook procedure (follow step 1, follow step 2...) for measuring various BJT and FET parameters and creating a model; some of the model parameters are a bit mystifying- can any of the smart people help me?
MicroCap has some models, that are hard to find. And I like their user interface. But overall quality of the models, supplied with MicroCap is pretty much similar to the others. Never seen anything interesting within PSPICE library. In any event, I will never use model as a black box - this makes simulation useless. As an example, Linear technology provides encrypted models for their components. It you look at performance of their LT1128, it shows unrealistic performance.
One problem is that BJTs vary widely within a given part number. That's one reason why Harold Black's clever circuit technique is so valuable
I'd love to have a cookbook procedure (follow step 1, follow step 2...) for measuring various BJT and FET parameters and creating a model; some of the model parameters are a bit mystifying- can any of the smart people help me?
That was one important doodle whilst on a ferry journey
Yes of course there are wide variations as you point out but it would be nice to at least have some accurate models for a "typical" example of a particular device... Rather than be guessing if it is an order of magnitude out or not!
Haven't got any further with that Genrad Digibridge as yet...
It's all very well having thousands of models available for high Ft or low Cob or high gain linearity super modern devices but if they are not accurate then what's the point? Especially if in practice the difference between lets say a real 2N3055 and a simulated one is greater than the actual difference between a real 2N3055 and a real MJL3281
!
And I can tell you that quite a few years ago I found gross errors in the MJL3281 and 1302 spice models ( can't just remember which of the pair had the glaring fault at the moment ). The manufacturer and model provider not interested in fixing the errors!
Spice is meant to be a guide to your circuits results.
Given even transistors with the same model number vary then you cant just use one model to be correct.
Use 2 transistors in series then the number of possibilities varys as X squared.
I rarely use spice but tend to use my skills as a design engineer and a calculator to work out a circuit.
Given even transistors with the same model number vary then you cant just use one model to be correct.
Use 2 transistors in series then the number of possibilities varys as X squared.
I rarely use spice but tend to use my skills as a design engineer and a calculator to work out a circuit.
For many years I resisted spice altogether, then I started to use it but often was first to mock the sometimes dubious results.
Today I regard it as a VERY useful tool in the designers arsenal so long as you always remember that "spice speak with forked tongue".... often. Sometimes it's astonishingly accurate but those cases are usually not discrete amplifiers
The point is that if we have models for discrete devices that we can trust then the results we get can be taken with less shovels of salt!
Guys, firstly one must know from experience what to expect the answer to be. That is engineering in its simplest form, you actually have a simulator in your head that does many calculations and correct things through experience. Once running a spice and the numbers spewed out is vastly different from what you expect then scrutinize the models.
There is one more area and that is the minimum, typical and maximum spec of a component, which one do you use. Now Microcap can use all three, but if you have measured one typical to the batch of transistors that you are using, then obviously insert what you know is correct.
Lastly it is worth performing Monte Carlo that takes all variants into consideration and provides you with typical results over a broad range of parameters anywhere between minimum and typical as well as tolerances of resistors and capacitors, etc.
No simulator that I have seen take lumped constants into consideration, these are the stray elements of your layout, so there remain grey areas. Furthermore, I have not seen spice that can step one component through temperature range, it by default steps all which is not the case in real world scenario.
Spice is a great tool that will derive many operating parameters from a design and working correctly in spice means that there is a good possibility that it may be practically realized. To draw specifications from spice modeling such as noise and distortion from the modeling is virtually impossible, in real life their numbers can be several orders of magnitude short of what the simulator predicts.
Use spice as a check on your designs, but use good engineering principles to derive your design. Many element consist of more than one pole and quadratic and more sophisticated forms must be considered to solve for X. This spice models do not do it, it is really expected that you have a bit of engineering savvy and can do basic maths to define a starting point.
There is one more area and that is the minimum, typical and maximum spec of a component, which one do you use. Now Microcap can use all three, but if you have measured one typical to the batch of transistors that you are using, then obviously insert what you know is correct.
Lastly it is worth performing Monte Carlo that takes all variants into consideration and provides you with typical results over a broad range of parameters anywhere between minimum and typical as well as tolerances of resistors and capacitors, etc.
No simulator that I have seen take lumped constants into consideration, these are the stray elements of your layout, so there remain grey areas. Furthermore, I have not seen spice that can step one component through temperature range, it by default steps all which is not the case in real world scenario.
Spice is a great tool that will derive many operating parameters from a design and working correctly in spice means that there is a good possibility that it may be practically realized. To draw specifications from spice modeling such as noise and distortion from the modeling is virtually impossible, in real life their numbers can be several orders of magnitude short of what the simulator predicts.
Use spice as a check on your designs, but use good engineering principles to derive your design. Many element consist of more than one pole and quadratic and more sophisticated forms must be considered to solve for X. This spice models do not do it, it is really expected that you have a bit of engineering savvy and can do basic maths to define a starting point.
Last edited:
Well said! Agreed
The thread was however not meant to be philosophising on the nature and limitations of spice itself but more a plea for known accurate (ok, as accurate as spice can ever be and YMMV) models in order that the results can be taken more at face value than if the models may be dodgy!
Certainly yes if the circuitry being designed is high frequency and high impedance then you will get nowhere without adding some realistic stray capacitances. Same for stray inductance if low impedance..
Looking for Good Models
I would recommend buying a copy of Ian Getreu's book on bipolar transistor modeling and measurements.
see
Ian Getreu's Storefront - Lulu.com
to order a copy...
I just ordered a copy. It's a great book, and it came in under 1 week.
I would recommend buying a copy of Ian Getreu's book on bipolar transistor modeling and measurements.
see
Ian Getreu's Storefront - Lulu.com
to order a copy...
I just ordered a copy. It's a great book, and it came in under 1 week.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.