I’m a developer and when an AI generates code, you can always put it in a sandbox, run it, and see if the thing actually works before investing any time on it. If it does and the code looks good, you have a nice : low labour and low brain cell count vs high productivity ratio. With PCB design, what are you supposed to do? Do a full review of what the AI has generated!? Double-check that the generated MCU pinout is actually valid, that the symbol creation has no bug, that the manufacturer’s recommendations are met, etc. That seems to me like doing the full job anyways. Indeed, the tool spits a full schematics, but then what? Who in his right mind would send that to layout and production without doing a meticulous review … which is the most time consuming and requires most expertise. It’s a convenient first draft or alternatives exploration at most, to me. AI would be terrific in a DRC on steroids that analyse your schematics logically, runs calculations on components selected, compares your design to thousands of others and pinpoint some specific concerns. To me, this company’s vision is misguided.
This is actually a great question As an additional step, we run detailed verification checks with a separate engine (advanced rules-based DRC) after the Netlist has been generated. In fact, some customers are looking to use that engine standalone to verify their manual design. Also, our algorithms that generate the circuits are deterministic. They are mathematical, designed to be precise and accurate, to SOLVE for a circuit (kind of like a simulation).
The title of this video is misleading. It should describe it as a comparison on schematic design not PCBs. This schematic could be used for any kind of implementation, from breadboards to wire wrapping(!) to ASIC or something.
@@Zachariah-Peterson So the goal of Altium Academy is to get clicks by any means rather than to be informative? Altium as a company has really been going downhill the last few years.
@@DaleMitchell1367 Literally nowhere in the title does it say "AI designed my PCB" or "AI did my PCB layout", there is nothing suggesting that AI created the PCB itself. PCB designers also design circuits sometimes, so the video is relevant to PCB people.
@@laserdan The video is about the capabilities of an AI platform that could help perform tasks that are relevant to PCB designers. Are you saying the video was not informative or relevant in that regard?
I have to admit that having an AI version of the design will help management and engineers do a better job. Management will quickly realize that Input requirements matter and Engineers will possibly get another view of available parts they originally may not have known or thought about. Every design should have design reviews, Adding a machine review at the requirements, design, manufacturing and test phase will only help reduce mistakes and oversites as soon as possible.
Yes that's a good idea, although the filtering requirement would depend on the spec one is trying to meet or if there is a specific standard that the design is targeting. Just writing "EMI filter" can be a bit misleading, especially when the design is running on battery power. If it were running from a bench supply or external unregulated power it would be quite different.
I could say a lot regarding this video, but I'll just say this, if the human didn't know about the first integrated component, was he also so amateurish not to know that gyroscopes and accelerometers are commonly integrated? He must have separated them intelligently.
That’s why I think this company’s vision is misguided. Maybe the designer could source those separate components easily, had full knowledge of them from past design, has them stocked in house, allow for easy replacement from other manufacturers, etc. This AI bombards you with “niche” components that locks you with a specific manufacturer, and pricier options without knowing if it necessary given your constraints. It’d be better if it would analyse your schematic and tell you how you could do differently. Or better, understand the general approach you follow and tell you where you could improve.
@@adrienferreira4666 The idea that an algorithmic platform cannot constrain on all the things you said is not a really good assumption. The platform can select components based on your company’s preferences, stock, pricing etc.
@@akintomideadesanmi7598 I never said that it wouldn’t be possible. I said it’s not what they’re doing, so it’s borderline useless. I can’t imagine a sensible company with a running design throwing it all into the trash to let an AI start a new design from scratch with niche components they’ve never experienced. With GitHub copilot, it reads your code and suggests a likely suggestion to what you want to do next using your coding style. Give me the same with schematic capture and I’m onboard … and even then, I would have to do a meticulous schematic review of what the AI suggested because unlike code, you can’t run it to at least see if it works. So the time and effort saved is not clear to me.
Also, can we talk about that 1.8nH inductor on the antenna that's likely going to low pass out all your 2.4GHz signal with even just expected trace capacitance for the C values? The AI has designed a circuit that just won't work
That's fair but it would need to be tested with different caps anyways because the PCB layout can induce parasitics which modify the performance at that frequency.
I believe that chip manufacturers should embrace AI by creating dedicated AI models for their chips. These models should be trained on a wide range of information, including datasheets, user manuals, development boards, reference designs, and more. Additionally, chip manufacturers should provide AI-driven models for schematics and PCBs. A tool like Altium could then import these models directly when a chip is added to the design, allowing AI to assist the designer throughout the schematic and PCB layout process. This assistance could include guidance on optimal design practices and error prevention. For component selection, AI could either automate the process entirely or offer suggestions to assist designers in making the best choices. The core idea is that semiconductor companies should create AI models for their chips, just as they currently provide schematic, simulation, mechanical and PCB models.
CircuitMind can leverage these models to drive innovation by generating proposed schematic and PCB documents. These documents could then be imported into Altium, offering added value based on the models. Alternatively, these models could be directly integrated into Altium to assist designers throughout the design process.
id like to see the pcb design part from the schematics. This is equally interesting - i would say with current status of AI - even more than schematics.
come hai fatto a registrare nel sito circuit mind , email non e' valido e inserire email di lavoro.. quale hai messo ? il mio gmail com non la accetta... e come posso procedere per valutare questo programma?
The A.I. design is only good if 1) time is important over anything, and 2) Price is not an issue, and 3) low production run! The A.I. is at best a sometimes good helper!
Because they saw "AI" and "PCB design" in the title and assumed AI-based routing, meanwhile forgetting that PCB designers also need to make schematics.
Integrated power modules are CRAP on price, and only makes sense where tight specs reliability are important! Yes there a re SOME cost saving in that but nowhere enough to compensate for the RIP-OFF prices!
So it's just reusable blocks that "AI" connects together? If I have to manually create a high level block diagram I could do it in Altium with reusable blocks and voila, same result. Somebody just needs to share a good library of reusable blocks or maybe Altium/chip manufacturer could step up and provide them. The problem is that no working pro will ever use this because every one likes a different resistor designator, uses different mechanical layers, has proven footprints that will never change in production, etc...
Not everyone has that much design IP, why not have a system that can generate some of it? As far as semiconductor manufacturers providing this, there is another platform called CELUS that takes this approach and has gotten some traction with the semiconductor manufacturers.
Nope! It's not reusable blocks. We create a digital twin of components from datasheets i.e. all the information you'd normally read in the datasheet is now in a digital model (in our database) that an algorithm can read and understand. Then we've developed a proprietary algorithms that uses circuit design rules to solve for a circuit. The algo takes your requirements, reads the digital model, performs compatibility checks, assigns pins, and adds all the circuit ancillaries (pull-ups, pull-downs, decoupling, level translators, crystals etc.) all at the same optimizing for cost, size, power, supply chain, etc. All our customers are professional engineers and companies.
@@tomideadesanmi1673 I’m sceptical on the MCU pinout side of things. I’ve worked on +100 pin MCU designs where each pin has exclusive IP capabilities that, when selected, excludes functionalities for other pins and create interlocking problems leading to dead ends. Some functionalities require specific VDD that forces to change the voltage level for a whole bank of IO. Some specific pins can deliver so little current that you have to be careful even for standard GPIO operations. When set into reset, some pins of the MCU become outputs and drive a specific voltage high by default to accommodate for firmware update, so connecting it on any kind of output would burn the chip when it goes into reset. I can go on forever with this. Again, I’m very sceptical that your tool accounts for all of that. But I let you prove me otherwise. If your tool generates a schematic and I’m then forced to dive fully into the component’s datasheet to double-check each and every pin anyways, I don’t see how it saves me any labour, expertise, risk, or brain power. This, and the fact that creating a schematic from scratch with niche components I’ve never experienced or that have nothing to do with manufacturers I already work with or that I have components sourced and tested. What is exactly your target market? Companies with no existing designs or that have enough money to start a new design from scratch on things that already work?
@@adrienferreira4666 You know how you just described a really detailed optimisation problem with lots of rules and exceptions to apply? You know what computers are really good and and humans are really bad at? That exact thing.
ai is good but i trust human AND furiour,FFT,EMI,EMC,Ground loop,KCL,KVL, INTUITION so on, bcuz, when "heat" on component is concerned, ai saying "good" , there is big No No!!
I’m a developer and when an AI generates code, you can always put it in a sandbox, run it, and see if the thing actually works before investing any time on it. If it does and the code looks good, you have a nice : low labour and low brain cell count vs high productivity ratio.
With PCB design, what are you supposed to do? Do a full review of what the AI has generated!? Double-check that the generated MCU pinout is actually valid, that the symbol creation has no bug, that the manufacturer’s recommendations are met, etc.
That seems to me like doing the full job anyways. Indeed, the tool spits a full schematics, but then what? Who in his right mind would send that to layout and production without doing a meticulous review … which is the most time consuming and requires most expertise. It’s a convenient first draft or alternatives exploration at most, to me.
AI would be terrific in a DRC on steroids that analyse your schematics logically, runs calculations on components selected, compares your design to thousands of others and pinpoint some specific concerns.
To me, this company’s vision is misguided.
This is actually a great question
As an additional step, we run detailed verification checks with a separate engine (advanced rules-based DRC) after the Netlist has been generated. In fact, some customers are looking to use that engine standalone to verify their manual design.
Also, our algorithms that generate the circuits are deterministic. They are mathematical, designed to be precise and accurate, to SOLVE for a circuit (kind of like a simulation).
I've used plenty of autorouters. I'm not worried.
This is not about autorouters. But I also am not worried.
@@Zachariah-Peterson Ню-ню!
Every experienced and not so experienced PCB layout designer knows that autorouters are actually forbidden. I did not use them since 2004.
I once developed a simpler autorouter which outperforms commercial ones by quality at nets number
Definitely just joking here. I'd love an AI that is better than a human for PCB layouts.
If AI engineers are so smart, they would first replace the high salaried, useless MBA VPs
but you have to be aware that the AI direction is determined by some of those MBA VPs
The title of this video is misleading. It should describe it as a comparison on schematic design not PCBs. This schematic could be used for any kind of implementation, from breadboards to wire wrapping(!) to ASIC or something.
But we got you to click on it right?
@@Zachariah-Peterson because it is misleading. I was expecting a video on PCB design, not circuit design. Your title is bad and you should feel bad.
@@Zachariah-Peterson So the goal of Altium Academy is to get clicks by any means rather than to be informative?
Altium as a company has really been going downhill the last few years.
@@DaleMitchell1367 Literally nowhere in the title does it say "AI designed my PCB" or "AI did my PCB layout", there is nothing suggesting that AI created the PCB itself. PCB designers also design circuits sometimes, so the video is relevant to PCB people.
@@laserdan The video is about the capabilities of an AI platform that could help perform tasks that are relevant to PCB designers. Are you saying the video was not informative or relevant in that regard?
I've been following Tomide from Circuit mind for a while, and what he's built is quite impressive. Naija for life🎉🎉
Thanks Solomon!
Why do I get the feeling that AI will be coming to the next version of AD?
I have to admit that having an AI version of the design will help management and engineers do a better job. Management will quickly realize that Input requirements matter and Engineers will possibly get another view of available parts they originally may not have known or thought about.
Every design should have design reviews, Adding a machine review at the requirements, design, manufacturing and test phase will only help reduce mistakes and oversites as soon as possible.
whats your opinion bout fluxai?
I believe that the high integrated power module also needs an EMI filter which is not included.
Yes that's a good idea, although the filtering requirement would depend on the spec one is trying to meet or if there is a specific standard that the design is targeting. Just writing "EMI filter" can be a bit misleading, especially when the design is running on battery power. If it were running from a bench supply or external unregulated power it would be quite different.
You would always need someone to supervise, correct, test and improve
Ai failes when thermal relief and manufacrurer capabilities too
Absolutely!
I could say a lot regarding this video, but I'll just say this, if the human didn't know about the first integrated component, was he also so amateurish not to know that gyroscopes and accelerometers are commonly integrated? He must have separated them intelligently.
That’s why I think this company’s vision is misguided.
Maybe the designer could source those separate components easily, had full knowledge of them from past design, has them stocked in house, allow for easy replacement from other manufacturers, etc.
This AI bombards you with “niche” components that locks you with a specific manufacturer, and pricier options without knowing if it necessary given your constraints.
It’d be better if it would analyse your schematic and tell you how you could do differently. Or better, understand the general approach you follow and tell you where you could improve.
@@adrienferreira4666 The idea that an algorithmic platform cannot constrain on all the things you said is not a really good assumption.
The platform can select components based on your company’s preferences, stock, pricing etc.
@@akintomideadesanmi7598 I never said that it wouldn’t be possible. I said it’s not what they’re doing, so it’s borderline useless.
I can’t imagine a sensible company with a running design throwing it all into the trash to let an AI start a new design from scratch with niche components they’ve never experienced.
With GitHub copilot, it reads your code and suggests a likely suggestion to what you want to do next using your coding style. Give me the same with schematic capture and I’m onboard … and even then, I would have to do a meticulous schematic review of what the AI suggested because unlike code, you can’t run it to at least see if it works. So the time and effort saved is not clear to me.
Also, can we talk about that 1.8nH inductor on the antenna that's likely going to low pass out all your 2.4GHz signal with even just expected trace capacitance for the C values? The AI has designed a circuit that just won't work
That's fair but it would need to be tested with different caps anyways because the PCB layout can induce parasitics which modify the performance at that frequency.
@@Zachariah-Peterson nah, calculate it as a basic CLC filter, it will low pass out anything above a GHz with only normal trace capacitance in the pF
I believe that chip manufacturers should embrace AI by creating dedicated AI models for their chips. These models should be trained on a wide range of information, including datasheets, user manuals, development boards, reference designs, and more. Additionally, chip manufacturers should provide AI-driven models for schematics and PCBs.
A tool like Altium could then import these models directly when a chip is added to the design, allowing AI to assist the designer throughout the schematic and PCB layout process. This assistance could include guidance on optimal design practices and error prevention.
For component selection, AI could either automate the process entirely or offer suggestions to assist designers in making the best choices. The core idea is that semiconductor companies should create AI models for their chips, just as they currently provide schematic, simulation, mechanical and PCB models.
CircuitMind can leverage these models to drive innovation by generating proposed schematic and PCB documents. These documents could then be imported into Altium, offering added value based on the models. Alternatively, these models could be directly integrated into Altium to assist designers throughout the design process.
id like to see the pcb design part from the schematics. This is equally interesting - i would say with current status of AI - even more than schematics.
I would like to see PCB routing comparison!
"why did the AI add $30 to the BOM? Must have been ignorance of the original designer"
come hai fatto a registrare nel sito circuit mind , email non e' valido e inserire email di lavoro.. quale hai messo ? il mio gmail com non la accetta... e come posso procedere per valutare questo programma?
The A.I. design is only good if 1) time is important over anything, and 2) Price is not an issue, and 3) low production run!
The A.I. is at best a sometimes good helper!
Why are so many folks in the comments talking about autorouters while this isn't about autorouters AT ALL?
Because they saw "AI" and "PCB design" in the title and assumed AI-based routing, meanwhile forgetting that PCB designers also need to make schematics.
God tool to bootstrap the project then a human can takeover saving time
Integrated power modules are CRAP on price, and only makes sense where tight specs reliability are important! Yes there a re SOME cost saving in that but nowhere enough to compensate for the RIP-OFF prices!
So it's just reusable blocks that "AI" connects together? If I have to manually create a high level block diagram I could do it in Altium with reusable blocks and voila, same result. Somebody just needs to share a good library of reusable blocks or maybe Altium/chip manufacturer could step up and provide them. The problem is that no working pro will ever use this because every one likes a different resistor designator, uses different mechanical layers, has proven footprints that will never change in production, etc...
Not everyone has that much design IP, why not have a system that can generate some of it? As far as semiconductor manufacturers providing this, there is another platform called CELUS that takes this approach and has gotten some traction with the semiconductor manufacturers.
Nope! It's not reusable blocks. We create a digital twin of components from datasheets i.e. all the information you'd normally read in the datasheet is now in a digital model (in our database) that an algorithm can read and understand.
Then we've developed a proprietary algorithms that uses circuit design rules to solve for a circuit. The algo takes your requirements, reads the digital model, performs compatibility checks, assigns pins, and adds all the circuit ancillaries (pull-ups, pull-downs, decoupling, level translators, crystals etc.) all at the same optimizing for cost, size, power, supply chain, etc.
All our customers are professional engineers and companies.
@@tomideadesanmi1673 I’m sceptical on the MCU pinout side of things.
I’ve worked on +100 pin MCU designs where each pin has exclusive IP capabilities that, when selected, excludes functionalities for other pins and create interlocking problems leading to dead ends.
Some functionalities require specific VDD that forces to change the voltage level for a whole bank of IO.
Some specific pins can deliver so little current that you have to be careful even for standard GPIO operations.
When set into reset, some pins of the MCU become outputs and drive a specific voltage high by default to accommodate for firmware update, so connecting it on any kind of output would burn the chip when it goes into reset.
I can go on forever with this.
Again, I’m very sceptical that your tool accounts for all of that. But I let you prove me otherwise.
If your tool generates a schematic and I’m then forced to dive fully into the component’s datasheet to double-check each and every pin anyways, I don’t see how it saves me any labour, expertise, risk, or brain power.
This, and the fact that creating a schematic from scratch with niche components I’ve never experienced or that have nothing to do with manufacturers I already work with or that I have components sourced and tested.
What is exactly your target market? Companies with no existing designs or that have enough money to start a new design from scratch on things that already work?
@@adrienferreira4666 You know how you just described a really detailed optimisation problem with lots of rules and exceptions to apply? You know what computers are really good and and humans are really bad at? That exact thing.
@@zyeborm Exactly
ai is good but i trust human AND furiour,FFT,EMI,EMC,Ground loop,KCL,KVL, INTUITION so on, bcuz,
when "heat" on component is concerned, ai saying "good" , there is big No No!!
It's Adidas shirt with Altium logo or Altium shirt with Adidas logo? :) It's first I assume. The Altium logo is poorly sewed.
How much does it cost to AI-design this circuit ?
@tomideadesanmi1673
The AI will improve fast.
Прикольно, но похоже в качестве человек-разработчика выступал первокурсник. Как по мне более полезен ИИ был бы как раз тря трассировки.
Having this could help out new guys to PCB design
I like it because it can generate an example of a schematic for a component I have never used before. Then I can go modify it to suit my needs.
Software needs 200MB, AI need internet, cookies, datasheet, shared repo, T&C ,cookies, and live login But why!!, i just need to route 40 mm track , 🤪
This AI has nothing to do with routing and layout, this one is about circuit generation
@@Zachariah-Peterson thanks for input, my topic was local generative AI, with datasheet as modeling
@@Zachariah-Peterson i really appriciate your efforts , You are A Great teacher.
ASIC design?
I will never touch an autorouter.
It's a good thing this video is NOT about autorouters
Ai helps , but just reading datasheets positive? Little, negative? More (as of now,dont know future)
Однозначно. Но мы продержимся чуть дольше обычных программеров пока 99 процентов и нас спишут в утиль.
After 5 years from now all the circuit and PCB designers are going to loose their jobs
Sorry bro it's gimmick. No Ai will replace the pcb design. It will take more than 2 decade.
No.
the designer will oversee the AI projects
The human keeps pushing FBs where it shouldn't. 'He doesn't learn from mistakes'.