Can GPT-4 Revolutionize PCB Design? Exploring the Possibility of Automating Footprints and Schematics Creation from Datasheets

Thread Starter

BusyBeaverWIthPCB

Joined Mar 17, 2023
1
Hello everyone,
I wanted to share an idea that I have been thinking about regarding the use of GPT-4 in automating the creation of electronic components footprints and schematics for PCB design applications. As many of you know, creating footprints and schematics can be a time-consuming process, especially for complex designs with many components.
The idea is to use GPT-4, a state-of-the-art language model, to parse component datasheets and generate footprints and schematics automatically. The process would involve inputting the datasheet for a given component and having GPT-4 analyze it to generate the necessary footprints and schematics.
This could potentially save a lot of time and effort for PCB designers, as well as improve the accuracy of the designs. However, the question remains: is this idea feasible? Can GPT-4 be trained to accurately interpret and generate the necessary footprints and schematics from component datasheets?
I would love to hear your thoughts on this idea and whether or not it could be implemented in practice. Do you think GPT-4 has the potential to revolutionize the way we design PCBs? Let's discuss!
 

Ya’akov

Joined Jan 27, 2019
9,267
Welcome to AAC.

The primary difficulty with your proposal is that, at this time, GPT “hallucinates and lies”. This, according to its own developers. And, though they are working to eliminate this problem they’ve only reduced it.

Something as precise as PCB design can’t afford the ambiguity of the model’s coupling to reality. The use of language models to generate code, for example, has proven dodgy. It can help but only with vigilance from the programmer.

Constraining the function of the model to some subset which is human-vetted might be useful. Or perhaps using an adversarial learning approach could reduce or eliminate errors. I am not an expert in ML, but ny understanding of the current SotA of ML algorithms like GPT is that they can‘t be relied upon to stick to the facts, so to speak, and so anything that requires critical adherence to precise parameters is outside their application space.
 

nsaspook

Joined Aug 27, 2009
13,583
You also need to be careful about using these language models for commercial uses because there are currently questions about copyright and ownership of results that lack human authorship and/or are derived from other copyright sources.
https://arstechnica.com/tech-policy/2023/03/us-issues-guidance-on-copyrighting-ai-assisted-artwork/
Authors risk losing copyright if AI content is not disclosed, US guidance says
Guidance comes after the Copyright Office decided that an author could not copyright individual AI images used to illustrate a comic book, because each image was generated by Midjourney—not a human artist. In making its decision, the Copyright Office committed to upholding the longstanding legal definition that authors of creative works must be human to register works. Because of this, officials confirmed that AI technologies can never be considered authors.

This wasn’t the only case influencing new guidance, but it was the most recent. Wrestling with the comic book's complex authorship questions helped prompt the Copyright Office to launch an agency-wide initiative to continue exploring a wider range of copyright issues arising as the AI models that are used to generate text, art, audio, and video continue evolving.

The guidance offers some specifics on what isn’t copyright eligible when it comes to AI works generated solely by prompts—with no modifications made—which the Copyright Office likens to giving “instructions to a commissioned artist.” These works lack human authorship and, therefore, won’t be registered.
 

nsaspook

Joined Aug 27, 2009
13,583
can we verify the circuit accuracy using chat gpt?
Seems unlikely with the current LLM technology. They are systems optimized to manipulate language/images/sounds using the rules that humans use to manipulate symbols. They don't understand what those symbols mean as an abstract description of some reality the humans used to create the original language/image/sound in the beginning.
1698191054736.png
 
Last edited:

eetech00

Joined Jun 8, 2013
4,056
Hello everyone,
I wanted to share an idea that I have been thinking about regarding the use of GPT-4 in automating the creation of electronic components footprints and schematics for PCB design applications. As many of you know, creating footprints and schematics can be a time-consuming process, especially for complex designs with many components.
The idea is to use GPT-4, a state-of-the-art language model, to parse component datasheets and generate footprints and schematics automatically. The process would involve inputting the datasheet for a given component and having GPT-4 analyze it to generate the necessary footprints and schematics.
This could potentially save a lot of time and effort for PCB designers, as well as improve the accuracy of the designs. However, the question remains: is this idea feasible? Can GPT-4 be trained to accurately interpret and generate the necessary footprints and schematics from component datasheets?
I would love to hear your thoughts on this idea and whether or not it could be implemented in practice. Do you think GPT-4 has the potential to revolutionize the way we design PCBs? Let's discuss!
I've used chatGPT (along with millions of other humans). Fascinating tool...;)
I thinks its worth exploring. I mean, after all, electronics is basically physics and math. Schematics could be built based on a library of functional blocks. PCB design is basically follows a specific set of design rules and constraints. And there are already existing footprint generators as well as simulation model generators. I don't think this is something that is expected to perfect right at the start, and so results would be checked by a human, at least in the beginning anyway. So yes, I say go for it!
 

nsaspook

Joined Aug 27, 2009
13,583
LLM's don't actually understand anything, but rather operate as complex probability machines. Circuit requirements, design, layout and build have major elements of human creativity involved. Sure, a LLM can be a useful tool but that's all it is, with the distinct disadvantage of the being untrustworthy to the point of just making stuff up in very convincing ways.
 
Top