With the typical behavioural (that is, typical distortionless) op-amp model it is to be expected that the noise-gain test circuit has to fail in simulation, because the internal circuit is not at all modelled (for any op-amp model I know of), only part of its idealized behaviour. This is usually clearly stated in the opamp model file, or it is just plain obvoius from the model itself.PMA said:
- Klaus
KSTR said:
Specialized filter design software (Linear's FilterCAD (free), TI's FilterPro (free), etc...). They trend to be product centric, but they still do an excellent job. Such software is not only doing the calculations, but is also helping in choosing the best filter topoplogy that would cover your requirements.
Unless you are designing a new (or modified) filter topology, the sensitivities are already pre-calculated. Otherwise, indeed, the Spice .step directive should be your preferred poison.
EKV model update
Hi all,
A while back I was working on model parameter extraction for the EKV model for use with vertical power MOSFETs. I'm providing an update on that here, and posting models for four device types.
Here's what happened with the model parameter fitting:
1) The official documentation for the EKV model is way out of date. The first Visual Basic for Applications (VBA) implementation of parameter fitting that used these equations was off by 10 percent or so from simulator values using the same parameters.
2) An implementation based on some Verilog-A code at the Silvaco web site was found and translated into VBA. It had the same problem.
3) A third implementation was created after finding some VHDL source code buried in the EKV web site. The VHDL implementation was converted to VBA. That implementation was much better, giving data very close to the simulator values for the small devices used for regression testing on the EKV web site. However, when large devices were used, differences of up to 8 percent or so between simulator values and predicted values based on the VHDL source code started to show up. These errors were sometimes very small, and accuracy of the models was very parameter-dependent. This indicates a discrepancy between the VHDL source code and what is actually used by the simulators. LTspice, MicroCap and TopSPICE all agreed with each other, but the translated VHDL code differed from the SPICE simulations.
4) After emailing the EKV developers roughly 10 times or so, I finally got a response. They sent me the "C reference implementation", which was an abstract implementation, not what they actually provided to the simulator vendors. I converted this to VBA and got the same inaccurate results as with the VHDL implementation. I emailed them a couple more times, asking for the SPICE3 implementation that they said in their previous email that they could provide. They did not respond.
I'm not going to do anything more with the EKV model. The developers really know their device physics, but they appear to be completely clueless in the area of software configuration management and software development in general. Their approach of providing only obsolete information on their web site is completely counterproductive to getting more popular acceptance of their model. Screw them.
Anyway, I have models of four devices that I'm going to post here. The parameters were adjusted by trial and error to give results that matched the simulator values pretty well. Each device has an Excel file associated with it. The size of this file makes it so I can only do one per post.
If you load the Excel file, you may get a warning about macros. The VBA code for the EKV model equations is considered a macro by antivirus software. It's fine to disable macros, as the data for comparing datasheet and simulated values is already computed.
The devices are as follows:
IRFP244: Like the IRFP240, but is a 250V device suitable for high-power amplifiers.
FQA9P25: Somewhat like the IRFP9240, but is 250V and does not have the problem with variations of transconductance in the audio frequency range.
IRF614: Like the IRF610, but 250V instead of 200V.
FQP2P25: Somewhat like the IRF9610, but is 250V instead of 200V, and does not have the transconductance variation problem like the P-channel IRF devices.
These models are for LTspice only.
Here is the IRFP244.
Hi all,
A while back I was working on model parameter extraction for the EKV model for use with vertical power MOSFETs. I'm providing an update on that here, and posting models for four device types.
Here's what happened with the model parameter fitting:
1) The official documentation for the EKV model is way out of date. The first Visual Basic for Applications (VBA) implementation of parameter fitting that used these equations was off by 10 percent or so from simulator values using the same parameters.
2) An implementation based on some Verilog-A code at the Silvaco web site was found and translated into VBA. It had the same problem.
3) A third implementation was created after finding some VHDL source code buried in the EKV web site. The VHDL implementation was converted to VBA. That implementation was much better, giving data very close to the simulator values for the small devices used for regression testing on the EKV web site. However, when large devices were used, differences of up to 8 percent or so between simulator values and predicted values based on the VHDL source code started to show up. These errors were sometimes very small, and accuracy of the models was very parameter-dependent. This indicates a discrepancy between the VHDL source code and what is actually used by the simulators. LTspice, MicroCap and TopSPICE all agreed with each other, but the translated VHDL code differed from the SPICE simulations.
4) After emailing the EKV developers roughly 10 times or so, I finally got a response. They sent me the "C reference implementation", which was an abstract implementation, not what they actually provided to the simulator vendors. I converted this to VBA and got the same inaccurate results as with the VHDL implementation. I emailed them a couple more times, asking for the SPICE3 implementation that they said in their previous email that they could provide. They did not respond.
I'm not going to do anything more with the EKV model. The developers really know their device physics, but they appear to be completely clueless in the area of software configuration management and software development in general. Their approach of providing only obsolete information on their web site is completely counterproductive to getting more popular acceptance of their model. Screw them.
Anyway, I have models of four devices that I'm going to post here. The parameters were adjusted by trial and error to give results that matched the simulator values pretty well. Each device has an Excel file associated with it. The size of this file makes it so I can only do one per post.
If you load the Excel file, you may get a warning about macros. The VBA code for the EKV model equations is considered a macro by antivirus software. It's fine to disable macros, as the data for comparing datasheet and simulated values is already computed.
The devices are as follows:
IRFP244: Like the IRFP240, but is a 250V device suitable for high-power amplifiers.
FQA9P25: Somewhat like the IRFP9240, but is 250V and does not have the problem with variations of transconductance in the audio frequency range.
IRF614: Like the IRF610, but 250V instead of 200V.
FQP2P25: Somewhat like the IRF9610, but is 250V instead of 200V, and does not have the transconductance variation problem like the P-channel IRF devices.
These models are for LTspice only.
Here is the IRFP244.
Re: EKV model update
Hi Andy,
First, thank you for the good work on the models.
Second, for more info about the EKV implementation, you might contact Bill Steele from Micro-Cap. (the guy who also helped us to implement your "tanh expression" for Cgd, you know). Normally, he is very responsive, even to those who didn't pay for MC.
Cheers, Edmond.
PS: Here's his email address: support@spectrum-soft.com
andy_c said:
Hi Andy,
First, thank you for the good work on the models.
Second, for more info about the EKV implementation, you might contact Bill Steele from Micro-Cap. (the guy who also helped us to implement your "tanh expression" for Cgd, you know). Normally, he is very responsive, even to those who didn't pay for MC.
Cheers, Edmond.
PS: Here's his email address: support@spectrum-soft.com
Re: Re: EKV model update
Thanks Edmond. I've discussed this with Mike Engelhardt (the LTspice developer) also. According to him, he had to sign an NDA agreement to incorporate EKV into LTspice. So he wasn't able to provide any details because of that. Looks like EKV is not truly open-source software as BSIM is. I wish I had known that when I started.
Edmond Stuart said:
Thanks Edmond. I've discussed this with Mike Engelhardt (the LTspice developer) also. According to him, he had to sign an NDA agreement to incorporate EKV into LTspice. So he wasn't able to provide any details because of that. Looks like EKV is not truly open-source software as BSIM is. I wish I had known that when I started.
As I have already said, in trying to get reasonable models starting from, e.g., OnSemi datasheets, I found quite impossible to get consistent SGP models parametes from the various informations therein.
E.g. I found that the IS, NF values one can get from VBE(on) versus IC (for IC<1), are not appropriate when trying to get the other parameters (BF, IKF, ISE, NE) from hFE versus IC.
Now ! the doubt ... are the OnSemi datasheets reliable ?
I looked at the MOSPEC datasheets (http://www.ortodoxism.ro/datasheets/mospec/2SA1302.pdf ) for the 2SA1302 that should be the same as the OnSemi MJL1302A: they provide different information.
Am I wrong ?
Thanks
E.g. I found that the IS, NF values one can get from VBE(on) versus IC (for IC<1), are not appropriate when trying to get the other parameters (BF, IKF, ISE, NE) from hFE versus IC.
Now ! the doubt ... are the OnSemi datasheets reliable ?
I looked at the MOSPEC datasheets (http://www.ortodoxism.ro/datasheets/mospec/2SA1302.pdf ) for the 2SA1302 that should be the same as the OnSemi MJL1302A: they provide different information.
Am I wrong ?
Thanks
Re: Re: Re: Re: Filters
As long as you are playing with that buggy piece of bloatware from Orcad-Candence (list price $9,995.00), which apparently even lacks the tools for designing filters, I can barely see a good reason why you are whining about the price (starting from $0.00) of Micro-Cap.
syn08 said:
As long as you are playing with that buggy piece of bloatware from Orcad-Candence (list price $9,995.00), which apparently even lacks the tools for designing filters, I can barely see a good reason why you are whining about the price (starting from $0.00) of Micro-Cap.