So Chat GPT can spit out Spice models for anything you ask it.
Like I asked it to generate a circuit for Phono Preamp that uses two transistors and a 18 volt power rail.
It can also output .cir files.
But try as I might I cant get the files to open in Tinia TI, Protesu, Or LT Spice.
The hint seems to be in the first line which starts with the words Version 4.0
Do you know which spice simulator creates .cir or .asc files with version 4.0 in the header.
Here is the .cir file it created.
If you want the spice model files it created which I cant open I can share that too.
I think it would be a great asset if we could get AI to output the modells. We then open those in LT Spice and do the final tweaks and checks. You can also prompt the AI to output in a specific format. Like create a LT Spice XVII Model for xyz. And it will change its syntax.
Version 4
SHEET 1 880 680
WIRE -448 80 -704 80
WIRE -320 80 -448 80
WIRE -192 80 -320 80
WIRE -48 80 -192 80
WIRE -704 144 -704 80
WIRE -192 144 -192 80
WIRE -704 224 -704 144
WIRE -48 224 -48 80
WIRE -48 224 -704 224
WIRE -192 288 -192 224
WIRE -48 288 -48 224
WIRE -192 400 -192 368
WIRE -704 496 -704 400
WIRE -704 496 -736 496
WIRE -192 560 -192 480
WIRE -48 560 -48 528
WIRE -48 560 -192 560
WIRE -192 640 -192 592
FLAG -704 496 0
FLAG -704 144 0
FLAG -192 144 0
FLAG -192 80 0
FLAG -192 400 0
FLAG -48 80 0
FLAG -48 224 0
FLAG -48 288 0
FLAG -192 560 0
SYMBOL voltage -704 208 R0
WINDOW 3 24 56 Invisible 0
WINDOW 123 0 0 Left 0
WINDOW 39 0 0 Left 0
SYMATTR InstName VCC
SYMATTR Value 18V
SYMBOL cap -320 144 R0
SYMATTR InstName C1
SYMATTR Value 10nF
SYMBOL res -448 64 R90
WINDOW 0 0 56 VBottom 2
WINDOW 3 32 56 VTop 2
SYMATTR InstName R1
SYMATTR Value 47k
SYMBOL cap -192 256 R0
SYMATTR InstName C2
SYMATTR Value 330pF
SYMBOL res -320 256 R0
SYMATTR InstName R2
SYMATTR Value 47k
SYMBOL cap -192 416 R0
SYMATTR InstName C3
SYMATTR Value 330pF
SYMBOL res -320 416 R0
SYMATTR InstName R3
SYMATTR Value 47k
SYMBOL npn -48 192 R0
SYMATTR InstName Q1
SYMATTR Value NPN
SYMBOL npn -48 352 R0
SYMATTR InstName Q2
SYMATTR Value NPN
SYMBOL npn -48 512 R0
SYMATTR InstName Q3
SYMATTR Value NPN
SYMBOL res -48 288 R0
SYMATTR InstName R4
SYMATTR Value 1k
SYMBOL res -192 352 R0
SYMATTR InstName R5
SYMATTR Value 47k
SYMBOL res -192 512 R0
SYMATTR InstName R6
SYMATTR Value 47k
SYMBOL res -48 432 R0
SYMATTR InstName R7
SYMATTR Value 1k
SYMBOL cap -48 432 R0
SYMATTR InstName C4
SYMATTR Value 1uF
SYMBOL cap -48 592 R0
SYMATTR InstName C5
SYMATTR Value 1uF
SYMBOL cap -192 624 R0
SYMATTR InstName C6
SYMATTR Value 100uF
SYMBOL voltage -704 560 R0
WINDOW 3 24 56 Invisible 0
WINDOW 123 0 0 Left 0
WINDOW 39 0 0 Left 0
SYMATTR InstName Vin
SYMATTR Value AC 1V
TEXT -680 664 Left 2 !.ac dec 10 10 10k
Like I asked it to generate a circuit for Phono Preamp that uses two transistors and a 18 volt power rail.
It can also output .cir files.
But try as I might I cant get the files to open in Tinia TI, Protesu, Or LT Spice.
The hint seems to be in the first line which starts with the words Version 4.0
Do you know which spice simulator creates .cir or .asc files with version 4.0 in the header.
Here is the .cir file it created.
If you want the spice model files it created which I cant open I can share that too.
I think it would be a great asset if we could get AI to output the modells. We then open those in LT Spice and do the final tweaks and checks. You can also prompt the AI to output in a specific format. Like create a LT Spice XVII Model for xyz. And it will change its syntax.
Version 4
SHEET 1 880 680
WIRE -448 80 -704 80
WIRE -320 80 -448 80
WIRE -192 80 -320 80
WIRE -48 80 -192 80
WIRE -704 144 -704 80
WIRE -192 144 -192 80
WIRE -704 224 -704 144
WIRE -48 224 -48 80
WIRE -48 224 -704 224
WIRE -192 288 -192 224
WIRE -48 288 -48 224
WIRE -192 400 -192 368
WIRE -704 496 -704 400
WIRE -704 496 -736 496
WIRE -192 560 -192 480
WIRE -48 560 -48 528
WIRE -48 560 -192 560
WIRE -192 640 -192 592
FLAG -704 496 0
FLAG -704 144 0
FLAG -192 144 0
FLAG -192 80 0
FLAG -192 400 0
FLAG -48 80 0
FLAG -48 224 0
FLAG -48 288 0
FLAG -192 560 0
SYMBOL voltage -704 208 R0
WINDOW 3 24 56 Invisible 0
WINDOW 123 0 0 Left 0
WINDOW 39 0 0 Left 0
SYMATTR InstName VCC
SYMATTR Value 18V
SYMBOL cap -320 144 R0
SYMATTR InstName C1
SYMATTR Value 10nF
SYMBOL res -448 64 R90
WINDOW 0 0 56 VBottom 2
WINDOW 3 32 56 VTop 2
SYMATTR InstName R1
SYMATTR Value 47k
SYMBOL cap -192 256 R0
SYMATTR InstName C2
SYMATTR Value 330pF
SYMBOL res -320 256 R0
SYMATTR InstName R2
SYMATTR Value 47k
SYMBOL cap -192 416 R0
SYMATTR InstName C3
SYMATTR Value 330pF
SYMBOL res -320 416 R0
SYMATTR InstName R3
SYMATTR Value 47k
SYMBOL npn -48 192 R0
SYMATTR InstName Q1
SYMATTR Value NPN
SYMBOL npn -48 352 R0
SYMATTR InstName Q2
SYMATTR Value NPN
SYMBOL npn -48 512 R0
SYMATTR InstName Q3
SYMATTR Value NPN
SYMBOL res -48 288 R0
SYMATTR InstName R4
SYMATTR Value 1k
SYMBOL res -192 352 R0
SYMATTR InstName R5
SYMATTR Value 47k
SYMBOL res -192 512 R0
SYMATTR InstName R6
SYMATTR Value 47k
SYMBOL res -48 432 R0
SYMATTR InstName R7
SYMATTR Value 1k
SYMBOL cap -48 432 R0
SYMATTR InstName C4
SYMATTR Value 1uF
SYMBOL cap -48 592 R0
SYMATTR InstName C5
SYMATTR Value 1uF
SYMBOL cap -192 624 R0
SYMATTR InstName C6
SYMATTR Value 100uF
SYMBOL voltage -704 560 R0
WINDOW 3 24 56 Invisible 0
WINDOW 123 0 0 Left 0
WINDOW 39 0 0 Left 0
SYMATTR InstName Vin
SYMATTR Value AC 1V
TEXT -680 664 Left 2 !.ac dec 10 10 10k
Member
Joined 2009
Paid Member
Jan i get the same results.
Like I posted you can also ask chat gpt to generate spice modells which you then save as a .asc file.
It does that too. If you want I can share the .asc files it generated.
Its possible that its an issue of version mismatch. I was kinda hoping somebody with an old version of LTSpice would be able to open it.
Chat GPT is getting better and better at this by the day. When I first tried this back in Jan of 2023 it would not even attempt it.
But AI is improving by the day.
I have now started to use the other AI platforms also. Like right now Im on Bing.
Bing handles images also. And it seems to do a lot of things better than chat gpt.
Its only a matter of time before this starts to work.
Meanwhile Im confused.
Which s/w should I start with. All my friends here use either Proteus or 5Spice or Micro Cap.
Nobody uses LT Spcie. But I see LTSpice is popular here so I loaded that. Proteus I have used for my automotive circuits.
I spent 20 years mastering IC engine simulators. And designing engines. Not sure why I have this mental bug where I spend decades mastering something and then loose interest in it. And then find something new where my knowledge level is zero. and try to master that.
It takes me about 5 years to get up to speed and then another 10 years to master it.
Its very frustrating esp as you age. Things seem to get harder and harder. I see my friends they stick to one subject their whole lives and they never get bored. Or have any desire to learn something totally new. I envy them.
Today I need to figure out how to do a frequency sweep in LTSpice. In proteus its easy peasy.
I think Ill stick with LT Spice and from there slowly move to Micro Cap.
Like I posted you can also ask chat gpt to generate spice modells which you then save as a .asc file.
It does that too. If you want I can share the .asc files it generated.
Its possible that its an issue of version mismatch. I was kinda hoping somebody with an old version of LTSpice would be able to open it.
Chat GPT is getting better and better at this by the day. When I first tried this back in Jan of 2023 it would not even attempt it.
But AI is improving by the day.
I have now started to use the other AI platforms also. Like right now Im on Bing.
Bing handles images also. And it seems to do a lot of things better than chat gpt.
Its only a matter of time before this starts to work.
Meanwhile Im confused.
Which s/w should I start with. All my friends here use either Proteus or 5Spice or Micro Cap.
Nobody uses LT Spcie. But I see LTSpice is popular here so I loaded that. Proteus I have used for my automotive circuits.
I spent 20 years mastering IC engine simulators. And designing engines. Not sure why I have this mental bug where I spend decades mastering something and then loose interest in it. And then find something new where my knowledge level is zero. and try to master that.
It takes me about 5 years to get up to speed and then another 10 years to master it.
Its very frustrating esp as you age. Things seem to get harder and harder. I see my friends they stick to one subject their whole lives and they never get bored. Or have any desire to learn something totally new. I envy them.
Today I need to figure out how to do a frequency sweep in LTSpice. In proteus its easy peasy.
I think Ill stick with LT Spice and from there slowly move to Micro Cap.
I use LTspice for sims, everybody here seems to do that, it is free and has a huge support base.
I also have Proteus but only use that for schematic drawing and PCB design.
Forget about ChatGPT in this context. If it puts together a text from text fragments it finds on the 'net, it is still text.
With asking it to assemble a spice model or circuit from pieces of models and circuits it finds on the 'net, that cannot but created nonsense.
Even if it 'looked' OK, how can you be sure it is a working circuit?? Or that the model is kosher? Forget that.
And unless your current CAD misses something you absolutely need, don't switch to another system, you'll lose at least a year in learning. Unnecessary.
As you say, at a certain age, you don't want to waste months or years. I'm also in that group!
BTW, What is 'IC engine simulators'?
Jan
I also have Proteus but only use that for schematic drawing and PCB design.
Forget about ChatGPT in this context. If it puts together a text from text fragments it finds on the 'net, it is still text.
With asking it to assemble a spice model or circuit from pieces of models and circuits it finds on the 'net, that cannot but created nonsense.
Even if it 'looked' OK, how can you be sure it is a working circuit?? Or that the model is kosher? Forget that.
And unless your current CAD misses something you absolutely need, don't switch to another system, you'll lose at least a year in learning. Unnecessary.
As you say, at a certain age, you don't want to waste months or years. I'm also in that group!
BTW, What is 'IC engine simulators'?
Jan
In LT Spice, you just run the AC analysis instead of transient, and the frequency start-stop is part of that command. AC voltages have separate values for transient and AC stimulus analysis.
I would like a PSpice to LT Spice converter because reading old PSpice files now requires a "free" install of Cordell software that is no small PITA. I suppose I could write an awk script as I have done long ago, but I don't need it that bad.
I would like a PSpice to LT Spice converter because reading old PSpice files now requires a "free" install of Cordell software that is no small PITA. I suppose I could write an awk script as I have done long ago, but I don't need it that bad.
Because basically, the way it works is it looks at all the data it's been trained on to find things similar to the question it's been asked, and works out from that data what is most likely to be given in response. For a lot of plain English questions this ends up working well. For more technical things, it yields plausible-looking nonsense. It doesn't really understand what it's saying, even if it looks like it does sometimes.; it sees that people asking for Spice models are given responses that contain certain things like "WIRE -448 80 -704 80", so it cobbles together something containing similar text without knowing if it makes sense.Why does it generate garbage ?
What Cordell software? You mean his models?In LT Spice, you just run the AC analysis instead of transient, and the frequency start-stop is part of that command. AC voltages have separate values for transient and AC stimulus analysis.
I would like a PSpice to LT Spice converter because reading old PSpice files now requires a "free" install of Cordell software that is no small PITA. I suppose I could write an awk script as I have done long ago, but I don't need it that bad.
Jan
Exactly! Plus it does a lot of cut-and-paste which also often looks plausible with text but not with spice models and circuits.Because basically, the way it works is it looks at all the data it's been trained on to find things similar to the question it's been asked, and works out from that data what is most likely to be given in response. For a lot of plain English questions this ends up working well. For more technical things, it yields plausible-looking nonsense. It doesn't really understand what it's saying, even if it looks like it does sometimes.; it sees that people asking for Spice models are given responses that contain certain things like "WIRE -448 80 -704 80", so it cobbles together something containing similar text without knowing if it makes sense.
Jan
Chat gpt can't be trusted! I also palyed with C programing language with chat gpt and found that chat speak nonsense, you need to tell him 5 times the same question and every time before question you need to tell him that he is wrong, after few - ten times sometimes you get correct answer! He apologizes a lot and when you criticize him for making things up and lying you get very interesting answers, like a living being manipulating you! Try that!
Several of the design firms where I worked, had an entire group of people whose full time job was to measure real-life semiconductor devices and create SPICE models to match. "The Modeling Group" was typically 4-7 people and they would spend hours and hours making tradeoffs: we can increase modeling accuracy of (phenomenon X), but that decreases modeling accuracy for (phenomenon Y). What's the best compromise? Hey designers, what do YOU think, which is preferable?
I doubt that BS-generator software like ChatGPT could do as good a job. ChatGPT is optimized to fool people into thinking it understands a topic, by spewing plausible sounding & domain relevant nonsense. Which isn't what you want (at least, isn't what I want) from SPICE .MODEL statements.
Now if what you actually want "AI software" to do, is to automatically generate complete & correct & optimized circuit designs ("SPICE netlists" rather than SPICE .MODELS), you'll have to wait a while. The state of the art in 2023 isn't that good yet.
I doubt that BS-generator software like ChatGPT could do as good a job. ChatGPT is optimized to fool people into thinking it understands a topic, by spewing plausible sounding & domain relevant nonsense. Which isn't what you want (at least, isn't what I want) from SPICE .MODEL statements.
Now if what you actually want "AI software" to do, is to automatically generate complete & correct & optimized circuit designs ("SPICE netlists" rather than SPICE .MODELS), you'll have to wait a while. The state of the art in 2023 isn't that good yet.
Sorry, not Cordell, I was thinking Corel, but its Cadence. They make you create an account to download the "free version", and hound you with marketing, so I just dropped it.What Cordell software? You mean his models?
Jan
ChatGPT is quite bad for anything that requires quantitative accuracy. It can do languages where one thing follows another, this even includes computer code to an extent. However, for something where you need logic that goes beyond things that seem to fit together, it often fails badly. It's not even particularly good at arithmetic.
Please read this article
https://www.newyorker.com/tech/annals-of-technology/chatgpt-is-a-blurry-jpeg-of-the-web
Then consider whether this technology is the right one to solve your problem.
https://www.newyorker.com/tech/annals-of-technology/chatgpt-is-a-blurry-jpeg-of-the-web
Then consider whether this technology is the right one to solve your problem.
I doubt that BS-generator software like ChatGPT could do as good a job
ChatGPT seems to be kinda OK at most things. The more specialized and detailed, the more it spews well-written garbage. I suspect with something as specific as circuit design, the model would have to be trained on a curated set of data rather than the internet-at-large. We all know loads of hot garbage are posted daily and ChatGPT has all of that hot garbage in it's reference library (at least 2021 and prior).ChatGPT is quite bad for anything that requires quantitative accuracy. It can do languages where one thing follows another, this even includes computer code to an extent. However, for something where you need logic that goes beyond things that seem to fit together, it often fails badly. It's not even particularly good at arithmetic.
Member
Joined 2009
Paid Member
- Home
- Design & Build
- Software Tools
- Getting ChatGPT generated spice models to open in LT Spice.