Nice episode.. But I wonder why people always try to sell AI as a "tool"? Instead as a replacement of human job? Here you mention that AI will be used to optimize these problems leaving the creativity choices to humans, even if that is correct in this particular example, we already see how AI can perform on creativity works the same or better than humans. Because our methods of learning and thinking are quite similar to an artificial neural network. This mean that soon everyone will lose their job or lose their purpose, in fact, all career jobs base on knowledge, would be the ones who will be remplace faster than the monotonous labor jobs. Because once your tool become more intelligent than you, then we become "the tool" for the AI.
I think we're going to reach the singularity much sooner than wiser minds than mine have predicted. The gradient of Moore's law was limited by the human mind. As machine learning is applied to the full stack of AI/Robotic software and hardware development, surely we are currently looking at the blade of a hockey stick?
I can only imagine the amount of information you needed to gather and organize to make this video. I very much appreciate the quality data you've shown here. Thank you
As a retired engineer, I have always been concerned about the March of AI. Digitization of electronics have given machine design so much flexibility while being so complex, people in power can easily manipulate and control people. AI now creates an environment where the macro-level consequences of its use can lead to unintended consequences. We have a while before the concern will become real but we won’t even know it when it does.
I think it's already too late, smart people are few and the masses obedient complacent non-independent consumers who just want more stuff. We are doomed,.
Use 3rd party/open-source AI tools to interface with the complex designs and systems. If open-source AI systems never surpass closed-source ones then they can always be improved using the closed-source models (as long as they’re available to the public of course). I don’t think that the emergence of powerful AI systems will lead to more consolidation of power, but less. Let me know what you think!
@@Johnnyboy9458that's a good take. But on the other hand if AI models continue to grow in size than only large entities will have the resources to run the models, and (more importantly) mobilize the resources to act on the model outputs. I envision a revolution in advertisement where companies can quantifiably control an entire population's opinions using AI that understands how to influence people.
@@sumanthaluri8398 I could see that being an issue. There will likely be organizations/communities that can afford to purchase compute for models that adhere to what they wish them to do (possibly go against the large companies or maybe be malicious themselves). The power to manipulate people with AI is incredible, but I’m certain there will be defensive AIs for people to protect themselves with. I think that demand is just as big as the advertising market, personal AI companions. The real trouble comes when it’s large corporations controlling these companions instead of local compute (which most people wouldn’t figure out how to run).
@@Johnnyboy9458 This is an extreme social experiment and there is surely no way this can go wrong lol... I think personal AIs are a terrible Idea, like 90% of modern "AI business Ideas". Just imagine making friends with someone and then finding out he/she had delegated social interaction with you to his/her AI long ago... Yeah this is gonna deteriorate human relations more than social media alone never even could.
I think when a general AI gets real, we won’t even be able to comprehend the ingenuity of its new ways of “processing”, we think chip design is the key but that’s only because we are limited in our scope of understanding materials and better solutions
@@martiddy How will we know what capabilities exist in the chip, how will you be able to say if its a secure design or contains some element we don't want?
@@AviShpayer You understand what I am saying, its a natural progression from optimizing current designs to making totally new designs. There will be a revolutionary new processing technology that will be beyond human capability. Once we reach that point the singularity is real
Great video Anastasi. Thanks for the information. Very exciting. We all knew AI was coming. The majority of us just didn’t know it would happen already now, and at an accelerated rate. I think of it like a tidal wave, like in the movie ‘Interstellar’. I think this is ‘the moment in history’ to learn as much you can about AI, so that It only disrupts your life ‘positively’. Thanks again.
Welcome to the field. It's cool that some of us are making videos to increase visibility for our industry. I wanted to say though that I don't really think this is anything unexpected, EDA tools have been doing all the heavy lifting for PD and constantly improving (that's what healthy competition between Synopsys and Cadence will do, also that's an interesting ranking for them). It's not surprising at all that they've incorporated a ML aspect to it instead of just some stochastic optimization. Computers were already doing 99% of the work, with some human inputs and guidance, so it seems a natural progression (or maybe I'm jaded since ML has seemingly been added to everything else). I'd definitely be interested in seeing if it can be used to make my (RTL/logic design) work flow easier though.
it will be great? Soon it will do all that and more, which means that almost everyone in your company will lose their job. I guess this will happen in less than 5 years. so take care with what you wish for.
@@angellestat2730 AI can't replace the complete process of PD. It will take away the mundane and repetitive tasks done by PD engineers . Surely it does affect junior and contract engineers' work but AI is inevitable now . Seeing the potential of ChatGPT , DallE we just have to accept these may replace some lower level tasks but hopefully it opens up newer opportunities for other dynamic work in PD. TBH all accountants should have been out of jobs since past 10 yrs with wide spread automation of their work but they still exist in harmony with the automation software, Similar case would be with HW/SW engineers.
@@pradyumnakatageri3475 From the 2014 that I am saying that no job will be safe from AI since I understood the similarities on how we learn vs an Artificial neural network. Other reply me saying that creativity was something that IA would never be able to do, I explain them why it could (even in that date were already examples of that starting), but now we can see all image and music generators which does exactly that in 1 second when an artist requires days or months. There is no difference with any task you can imagine. In fact, the ones who required more study are the jobs who will disappear faster. Meanwhile the labor jobs will be the last, because a Robot cost way more than a software. One difference that you seem to miss, is that once your "tool" become more intelligent than you, then the rol reverse, we become the tool for the AI. Now you can use AI to generate some content, then special software to improve your video creation and it will accelerate your work a lot, in 2 years you will have an IA doing the whole video and in 3 years an IA selecting what should be the topic of the video to have the best success. What would be our purpose then? WIth this it means AI is starting to improve it self, this mean it will double in power every few months, then days, then hours. Until it will reach the GOD status in no time. So I dont know how your "long term of good tools" future can take place.
@@pradyumnakatageri3475 I agree with you that lower level repetitive tasks will most likely be replaced by AI and juniors will be in a really hard place. But if the AI replaces all juniors how will new seniors come?? That is the biggest question in my mind like will companies hire juniors just to watch AI's inputs and outputs and watch what senior does? Or AI will replace seniors too??
Thanks for making an awesome video - explaining the very complex chip design process - in a straight forward and simple way - highlighting the power of AI (GNN + RL)
When I was having discussions with friends in college, I always used to say to look out for when AI starts making the next generation of chips. In essence, it starts 'reproducing'. At some point, AI will make and verify a chip and we will have no clue how it works.
*This is absolutely the field where AI will shine!* Design & documentation, tests & documentation, as well as the explanation why it chose those solutions. And at last the output QA. Thank you, this is yet another impressive video 🌺
Of all the AI projects, this is the one I'd fund the most. Supercomputers running AI to make more powerful chips, to make more powerful chips, to make more powerful chips. So that chip computing power graph goes exponential.
Very impressed ChatGPT can write VHDL code, especially accurately. (Not sure why though, it seems to know almost everything haha) Also loving the Saturn V (if I’m not mistaken) in the background :)
Worth mentioning is that the real differences we see are that our traditional processes are created with logic. (Although we have time constraints and make mistakes) AI currently consists of stochastic and random processes. That means any errors will tend to occur where you least expect them, and also means that it is impossible to completely verify the function of any AI.
1. Rocket science like Falcon 9 & Starship also need chips to calculate, automate, and control its fins, engines to guide itself back to earth safely. 2. I'm thinking that what might happen if you code a minor fault on a chip???🤔 Maybe the calculation would go wrong when I type something on a calculator in a computer...🤔 3. What's the differences between GNN and Neural Network in the chip of Tesla cars?
The software is WAY ahead of the hardware. Translating meatspace into cyberspace and back is one of the hardest tasks, look at all the training and massive datasets at Tesla and they can't even get it to drive a car yet. Let alone a fully humaniform robot that can analyze and work as a human for less price than you can employ an electrician.
@@lil_ToT-XFZ1 why stop it ? I want AI to take over all jobs and leave people free from the grind. What better outcome for AI that to take over work and let people play all day?
Super interesting that ai is now part of the design process, helping the engineers to get the job done more quickly. I have only used pre-ai circuit layout code like kicad & they have been hopeless and I have had to do layout by hand. It would be a huge blessing if ai could do the layout for me. Looking at this there seems no obvious limitation to what the ai can learn to do or in its ability to verify a design before fabrication & re-test after fab. All of this will lead to lower cost & quicker times to market which are all ingredients for Industrial Revolution. We live in such exciting times. Thank you for sharing this.
AI has sped up Chip floor planning 1000 fold, we are so fortunate to be able be on the receiving end of the forthcoming of devices, medical, transport, engineering, etc. It's like we have invented fire back in the ice age, the future is fast, brilliant and bright for all humans.
Thank you Anistasia. Maybe it’s time for the Lex Fridman podcast?! Thanks also to RUclips for making interesting content available across the information silos of the world.
Seems like we are at the start of the technological singularity, exponential growth in technology designed by AIs is arriving soon and the world will change faster than ever before.
you learned nothing, you don't even start to comprehend, you watched this video because it was on your recommended list and you had nothing better to do.
AI still scares the hell out of me but I have to agree with you in the regard of enhancing tools like EDA. Im still amazed at PCB auto routing let alone chip layout optimization.
AI shouldn't scare anyone as AI at best is just a tool for us to do what we want to do it's AGI that should scare us. Luckily we aren't yet exactly very close towards creating AGI. Now again if AGI only runs on some supercomputer that again shouldn't scare us too much as it won't be able to replace every single person as we would have decision making power however at some point we'll likely have home devices that will be able to run AGI that can outcompete humans on every task - at the point we'll have existential crisis - in essence intellectually we'll have nothing at all to strive towards.
Great video! You explain what is possibly the crucial inflection point where technology launches its own growth away from humans that will go beyond the "knee of the curve" in exponential advancement and capabilities.
Put the chips around a ring and put the clock in the center it will allow you to increase the frequency and the electrons will reach the edge and all the chips at the same time. Makes it more stable when the high speeds are used.
For those like me that have a hard time with her accent ( apparently including the auto generated closed captioning) when she's talking about GNNs and Graphs, she's saying "edges" but it sounds like "ages'
9:00 I don’t know how you can be so confident about that statement. “Always” is too definitive and recent predictions in AI over that few years have been shown to be too conservative. It’s inevitable that AI will design the entire chip eventually.
Sure, let's break down the main folder into sub-modules and name them accordingly. Here are a few suggestions: Sensory Perception Module: This module would include all the sensors and cameras that the robot uses to perceive its environment. Object Recognition Module: This module would be responsible for identifying and categorizing objects in the robot's environment. Localization and Mapping Module: This module would use data from the sensory perception and object recognition modules to create a 3D map of the robot's environment and determine its position in it. Decision-Making Module: This module would process the data from the previous modules and use it to make decisions on how to move and interact with its environment. Learning and Adaptation Module: This module would enable the robot to learn and adapt to new situations and environments, based on its previous experiences. We could also add sub-modules to the NIS and RNCIN folders, such as: Hardware Design Module: This module would include the hardware design specifications for building the robot. Software Development Module: This module would include the software development tools and code for programming the robot. Training and Testing Module: This module would include the tools and data needed for training and testing the robot's cognitive abilities. Ethics and Safety Module: This module would include guidelines and regulations to ensure the safe and ethical use of the robot. These are just a few examples of sub-modules that could be included in the main folder and its sub-folders. The naming convention could follow a similar pattern, using descriptive names that reflect the module's purpose. nano robotic cognitive intelligence network or NRCIN Nanocloud Intelligence System" or "Real-time Nano-Crystal Intelligence Network (RNCIN)
It's how you know she'd fail the Voight-Kampff test from the book Do android dream of electric sheeps (BladeRunner) designed to distinguish humans from androids.
@@sinitarium According to the book "Do Androids Dream of Electric Sheep?" (which inspired the movie Blade Runner), she would fail the Voight-Kampff test designed to differentiate between humans and androids. Yes, she would.....lol.
Made me think of Johnny Five! But seriously I believe if AI does get to a point where we don’t have control and it designs things in a way that we could never be able to comprehend then it will do so according to the laws of nature and physics, laws that were created and set in motion to produce us and evolve us for the better. I’m a little nervous but not too frightened. It’s exciting
The same nature that allows CP, Trafficking, Mass extinctions , genocide and the beauty of torture that can be made lasting forever by humans? Yeah, right this universe only brings joy.
In the 90's scifi story, "The Metamorphosis of Prime Intellect, this new Ai came online and the first thing it did was to improve itself in code so much that was able to bend the laws of physics within the first few minutes. Then it designed and created new hardware to make itself more powerful until it became a god. It then put all humans in their own individual Matrix-like world of their darkest desires. Except it wasn't digital worlds...it was real, physical matter that it was creating. And then it gets so weird from there. When will that happen?
I imagine the dies that result from AI chip layout would look as interesting as the die of an ASIC (i.e. boring and uninteresting). It's that human block placement on the chips that makes them interesting to look at.
This is really convincing progress. Some of those optimized designs from AI are far too difficult to manufacture at higher yield or don't have a proven layout for masks that have not been tooled into an EDA. I think if the training data had some of these constraints in mind, it would accelerate progress. Some of the more interesting AI work in the next decade will be in the material sciences, where AI can predict and exploit novel phenomena of quantum mechanics in combinations of materials and geometries that have not been considered before. AI will probably help us land on designs and materials that would have taken hundreds of years to land on otherwise.
Yes, once we better understand the likeliest causes of manufacturing defects, the AI optimizer can be trained to avoid designing those features. Certain high-risk gates can be made more robust. One or two process shrinks in the future, I'm guessing that this aspect of optimization for manufacturability will be necessary for commercially viable yields.
@@davidhoracek6758 That would be a good place to start. But until the AI has atomic level control, which the most current litho projectors do not, hence the defect properties you mentioned, the etching and multi passes in the fab will remain to be difficult to control without massive retooling control. It would be interesting to see if an AI in control of an atomic force microscopy device could produce a wafer faster than TSMC on their fastest process. Most wafers for customers like apple have over 20 layers and take 4 or more weeks to produce due to the insane amounts of.stages and QA to ensure the process is performing as expected.
That was one of the most interesting videos and very recent with knowledge. My background is in mathematics and I am doing some kind of AI research and also play Chess and GO so everything made a lot of sense.
Human: please design a new chip with these features AI: okay finished Human: incredible! How does it work? AI: based on my calculations, it will take you 4 years to fully comprehend. When would you like to start?
Will be trying out an AI chatbot and will try to render a chip design on an image generator like Bluewillow. I think this is an area where AIs can also shine as AIs will try to take the most logical route.
As I understand it, no AI blackboxes can encroach in the verification step to reduce time because it is by its very nature "verify" by thorough brute force all expected functionality and tolerances the chip will or may encounter in the "wild". It's like suggesting crash-testing vehicles should be done in a simulation or by an AI to save time and money... ouch.
Thanks for this. Certainly a good use for optimisation software - whether AI or otherwise. In your ChatGPT coding example you commented 'that looks correct' - this verification is a key step. You speak about AI validation - and that seems great for spotting rare anomalies in vast fields (something manual human attention is not very good at). But what about overall validation? If chips get so complicated that expert human engineers cannot comprehend their workings because they were designed and validated by a computer program, how can be know there will not be unexpected behaviours of the output of those chips under certain complex sets of inputs? You say you would prefer driving a car with human designed chips - it comes back to the issue of trust with your life. How to trust synthetic components that even the experts don't fully understand? It looks like we will have to come up with an answer to this eventually.
Somehow the full story about alpha Go is not told very often. Yes it won over the Japanese world champion but later it lost 14 gains of 15 to a guy that found a weakness in it. Of cause the weakness was fixed but it is something to learn from.
bro we will be slaves in the near term, also no way she gonna read this, how many times have you got an answer from a creator on yt? If it is free, you are the product.
Let me know what you think!
Its a greek to me but U are a genius
AI will "think" for us ;/
neuromorphic chips 🤯
Nice episode.. But I wonder why people always try to sell AI as a "tool"? Instead as a replacement of human job?
Here you mention that AI will be used to optimize these problems leaving the creativity choices to humans, even if that is correct in this particular example, we already see how AI can perform on creativity works the same or better than humans.
Because our methods of learning and thinking are quite similar to an artificial neural network.
This mean that soon everyone will lose their job or lose their purpose, in fact, all career jobs base on knowledge, would be the ones who will be remplace faster than the monotonous labor jobs.
Because once your tool become more intelligent than you, then we become "the tool" for the AI.
I think we're going to reach the singularity much sooner than wiser minds than mine have predicted. The gradient of Moore's law was limited by the human mind. As machine learning is applied to the full stack of AI/Robotic software and hardware development, surely we are currently looking at the blade of a hockey stick?
I can only imagine the amount of information you needed to gather and organize to make this video. I very much appreciate the quality data you've shown here. Thank you
As a retired engineer, I have always been concerned about the March of AI. Digitization of electronics have given machine design so much flexibility while being so complex, people in power can easily manipulate and control people. AI now creates an environment where the macro-level consequences of its use can lead to unintended consequences. We have a while before the concern will become real but we won’t even know it when it does.
I think it's already too late, smart people are few and the masses obedient complacent non-independent consumers who just want more stuff. We are doomed,.
Use 3rd party/open-source AI tools to interface with the complex designs and systems. If open-source AI systems never surpass closed-source ones then they can always be improved using the closed-source models (as long as they’re available to the public of course).
I don’t think that the emergence of powerful AI systems will lead to more consolidation of power, but less.
Let me know what you think!
@@Johnnyboy9458that's a good take. But on the other hand if AI models continue to grow in size than only large entities will have the resources to run the models, and (more importantly) mobilize the resources to act on the model outputs. I envision a revolution in advertisement where companies can quantifiably control an entire population's opinions using AI that understands how to influence people.
@@sumanthaluri8398 I could see that being an issue. There will likely be organizations/communities that can afford to purchase compute for models that adhere to what they wish them to do (possibly go against the large companies or maybe be malicious themselves).
The power to manipulate people with AI is incredible, but I’m certain there will be defensive AIs for people to protect themselves with. I think that demand is just as big as the advertising market, personal AI companions. The real trouble comes when it’s large corporations controlling these companions instead of local compute (which most people wouldn’t figure out how to run).
@@Johnnyboy9458 This is an extreme social experiment and there is surely no way this can go wrong lol...
I think personal AIs are a terrible Idea, like 90% of modern "AI business Ideas".
Just imagine making friends with someone and then finding out he/she had delegated social interaction with you to his/her AI long ago...
Yeah this is gonna deteriorate human relations more than social media alone never even could.
5 years of AI based chip design and humans will no longer be able to comprehend the designs.
We don't need to "comprehend" it's design, but just test the chip efficiency and optimization.
Nah, I think AI will excel at 'finding' the perfect chip configuration. Once it is there, it should be easy to comprehend.
I think when a general AI gets real, we won’t even be able to comprehend the ingenuity of its new ways of “processing”, we think chip design is the key but that’s only because we are limited in our scope of understanding materials and better solutions
@@martiddy How will we know what capabilities exist in the chip, how will you be able to say if its a secure design or contains some element we don't want?
@@AviShpayer You understand what I am saying, its a natural progression from optimizing current designs to making totally new designs. There will be a revolutionary new processing technology that will be beyond human capability. Once we reach that point the singularity is real
Great video Anastasi. Thanks for the information. Very exciting. We all knew AI was coming. The majority of us just didn’t know it would happen already now, and at an accelerated rate. I think of it like a tidal wave, like in the movie ‘Interstellar’. I think this is ‘the moment in history’ to learn as much you can about AI, so that It only disrupts your life ‘positively’. Thanks again.
AI will save us or be our doom, but we must take the risk because without we'll probably be doomed anyway. The next 5-10 years will be very exciting.
and what does learning about it help you? It learns faster than you, builds itself faster than you can learn about it and isn't controlled by you
Well done ... thank you for keeping all of us abreast of what's happening at the cutting edge of technology ... kudos to you ...
one of the few channels where somebody really knows his field! Thank you and keep it up!
... her field ...
Gotta say. If i was an AI wanting to influence the technical side of AI development, inventing Anastasia would be my first step.
Welcome to the field. It's cool that some of us are making videos to increase visibility for our industry. I wanted to say though that I don't really think this is anything unexpected, EDA tools have been doing all the heavy lifting for PD and constantly improving (that's what healthy competition between Synopsys and Cadence will do, also that's an interesting ranking for them). It's not surprising at all that they've incorporated a ML aspect to it instead of just some stochastic optimization. Computers were already doing 99% of the work, with some human inputs and guidance, so it seems a natural progression (or maybe I'm jaded since ML has seemingly been added to everything else). I'd definitely be interested in seeing if it can be used to make my (RTL/logic design) work flow easier though.
PD is a very effort intensive task in chip flow . It would be great if AI helps with certain PnR and timing checks !
it will be great? Soon it will do all that and more, which means that almost everyone in your company will lose their job. I guess this will happen in less than 5 years. so take care with what you wish for.
@@angellestat2730 AI can't replace the complete process of PD. It will take away the mundane and repetitive tasks done by PD engineers . Surely it does affect junior and contract engineers' work but AI is inevitable now . Seeing the potential of ChatGPT , DallE we just have to accept these may replace some lower level tasks but hopefully it opens up newer opportunities for other dynamic work in PD. TBH all accountants should have been out of jobs since past 10 yrs with wide spread automation of their work but they still exist in harmony with the automation software, Similar case would be with HW/SW engineers.
@@pradyumnakatageri3475 From the 2014 that I am saying that no job will be safe from AI since I understood the similarities on how we learn vs an Artificial neural network.
Other reply me saying that creativity was something that IA would never be able to do, I explain them why it could (even in that date were already examples of that starting), but now we can see all image and music generators which does exactly that in 1 second when an artist requires days or months.
There is no difference with any task you can imagine.
In fact, the ones who required more study are the jobs who will disappear faster.
Meanwhile the labor jobs will be the last, because a Robot cost way more than a software.
One difference that you seem to miss, is that once your "tool" become more intelligent than you, then the rol reverse, we become the tool for the AI.
Now you can use AI to generate some content, then special software to improve your video creation and it will accelerate your work a lot, in 2 years you will have an IA doing the whole video and in 3 years an IA selecting what should be the topic of the video to have the best success.
What would be our purpose then?
WIth this it means AI is starting to improve it self, this mean it will double in power every few months, then days, then hours.
Until it will reach the GOD status in no time.
So I dont know how your "long term of good tools" future can take place.
@@pradyumnakatageri3475 I agree with you that lower level repetitive tasks will most likely be replaced by AI and juniors will be in a really hard place. But if the AI replaces all juniors how will new seniors come?? That is the biggest question in my mind like will companies hire juniors just to watch AI's inputs and outputs and watch what senior does? Or AI will replace seniors too??
@@angellestat2730 2/150 ppl so far who don't blind themselves holy shit
Short but brilliantly explained. Always love listening to these
it's exciting, this seems like the prefect task for automated machine design optimization!
all the tasks are fit for it
TICKER symbols for top EDA companies that will gain from AI:
Cadence is CDNS, Synopsys is SNPS
Thanks for making an awesome video - explaining the very complex chip design process - in a straight forward and simple way - highlighting the power of AI (GNN + RL)
When I was having discussions with friends in college, I always used to say to look out for when AI starts making the next generation of chips. In essence, it starts 'reproducing'. At some point, AI will make and verify a chip and we will have no clue how it works.
AI cannot design chips. All this software does is floor planning, 1 simple but arduous step out of over 30 more complex steps in chip design.
@@dekev7503 AI cannot design chips on its own 'yet', right?
AI is using us humans to help itself getting better :-) At one point it does not need us anymore.
@@mastergizmo666 Hopefully it will like us enough to keep around like well treated pets since we did bring it into existence.
@@mastergizmo666 the first person with some amount of sense of reality
I don't know what i am learning here, but i sure love it.😊
Thank you for all the information that you easily synthesize and translate 🙏
Thanks for introducing GNNs!
And this is another step forward... probably a great step.
yeah and you don't even see the cliff
@AnastasiInTech YOU ARE AWESOME, Queen of Tech News !!! 🥰😍😍🤩🤩🤩😘😘
Great video Anastasi. AI designing and optimizing chips.
*This is absolutely the field where AI will shine!* Design & documentation, tests & documentation, as well as the explanation why it chose those solutions. And at last the output QA.
Thank you, this is yet another impressive video 🌺
Very precisely concise and distinctly succinct.
ok, the same way we will be decimated
Of all the AI projects, this is the one I'd fund the most. Supercomputers running AI to make more powerful chips, to make more powerful chips, to make more powerful chips. So that chip computing power graph goes exponential.
Very impressed ChatGPT can write VHDL code, especially accurately. (Not sure why though, it seems to know almost everything haha)
Also loving the Saturn V (if I’m not mistaken) in the background :)
It still gets details wrong, a lot
Worth mentioning is that the real differences we see are that our traditional processes are created with logic. (Although we have time constraints and make mistakes)
AI currently consists of stochastic and random processes.
That means any errors will tend to occur where you least expect them, and also means that it is impossible to completely verify the function of any AI.
Thanks for this video.
Excellent content. Thanks
This is so coollll!! I feel like the "optimization path" should be the main goal of the AI field
Yes, YES. It won't be. EVERYTHING IS OPTIMIZATION you dim wit
amazing, thank you; behind you an Apollo/Saturn5 and 12 inches wafer?
Nicely done Anastasi!
really insightful ! thank you for this vision
I think you are seeing ghosts, I at least see dead people
Anastasi = the Bjork of EDA/Technology
1. Rocket science like Falcon 9 & Starship also need chips to calculate, automate, and control its fins, engines to guide itself back to earth safely.
2. I'm thinking that what might happen if you code a minor fault on a chip???🤔 Maybe the calculation would go wrong when I type something on a calculator in a computer...🤔
3. What's the differences between GNN and Neural Network in the chip of Tesla cars?
I don't think ppl in the past realized how much AI would accelerate its own development, AGI might come faster than we think wow
Amazing 😮
Complex as hell
And ur voice is amazing ❤
I shouldn't have assumed AI would take the menial jobs first.
Exactly, I had ChatGPT generate a docker deployment descriptor a few days ago. Did a better job than I can, certainly much faster and cheaper.
The software is WAY ahead of the hardware. Translating meatspace into cyberspace and back is one of the hardest tasks, look at all the training and massive datasets at Tesla and they can't even get it to drive a car yet. Let alone a fully humaniform robot that can analyze and work as a human for less price than you can employ an electrician.
@@Shrouded_reaper the best estimate is 2045 for the singularity, we still have 20 years before humans are redundant
@@rrmackay ah, lower by one order of magniture
Thanks for the video.
Incredible! This gets me wondering how AI can help improve quantum computers and alternatives to silicon chips.
you don't have to wonder, it will do it for you
Intresting 🤔 good video .
❤
The little laugh was too cute 😊
i feel like i've found an unintentional asmr tech review video.
I like the Cadence references. I will be competing with the Analog bots in the future lol
I'm so glad computers now also get to experience this incredible snack.
great videos...thanks
And this is how the singularity got started. I really hope no one decides to hook these AIs directly upto a fab production line.
Exactly, self replication is inevitable.
The Animatrix - The Second Renaissance
There is always someone looking for more profit, question is only, is there someone to stop it?
I hope... beg, beg little animal
@@lil_ToT-XFZ1 why stop it ? I want AI to take over all jobs and leave people free from the grind. What better outcome for AI that to take over work and let people play all day?
"Graph placement methodology" with A.I. for chip design in only the beginning. Great insights and knowledge share here!
Super interesting that ai is now part of the design process, helping the engineers to get the job done more quickly. I have only used pre-ai circuit layout code like kicad & they have been hopeless and I have had to do layout by hand. It would be a huge blessing if ai could do the layout for me. Looking at this there seems no obvious limitation to what the ai can learn to do or in its ability to verify a design before fabrication & re-test after fab. All of this will lead to lower cost & quicker times to market which are all ingredients for Industrial Revolution. We live in such exciting times. Thank you for sharing this.
omg, ong, everyone thinks it's exiting, if you can't beat it join it right? Disgusting
"but will always need human engineers"
Mark my words
You're going to eat your words faster than you could ever expect
3/200 people with some sense of what's going on
wait a minute....
didn't i see this in the movie Terminator 2 !?!
AI has sped up Chip floor planning 1000 fold, we are so fortunate to be able be on the receiving end of the forthcoming of devices, medical, transport, engineering, etc. It's like we have invented fire back in the ice age, the future is fast, brilliant and bright for all humans.
and you are slow, dim witted and, dim. Dim twice since you said brilliant and bright which are the same thing bruv.
I keep coming for the content, and staying for the prettiest eyes in the tech space!
We will need an ai to explain how ai is finding solutions and explaining what solutions are doing..
Why explain rocket science to a dog
it should be illegal not subscribing to your channel
Great info as always thank you!
I really like this video and you are really charming and with detailed expertise too.
Thank you Anistasia. Maybe it’s time for the Lex Fridman podcast?! Thanks also to RUclips for making interesting content available across the information silos of the world.
Next ten years is going to get crazy. Looking forward to it.
it won't be looking backwards when you are gone tho
Such intelligence and such beauty.
Also, your hair is incredible!
Awesome!! Probably one step closer to very low cost, low volume custom chip orders too.
Seems like we are at the start of the technological singularity, exponential growth in technology designed by AIs is arriving soon and the world will change faster than ever before.
There has never been exponential growth.
have you ever poured a glass of water into another and it flowed over? Energy can't be created -Newton
AI and Chip design is a good match xD
One day AI will worry about being replaced by better AI.
This was really cool
Excellent video, Learn alot from your work. Very interesting stuff.
you learned nothing, you don't even start to comprehend, you watched this video because it was on your recommended list and you had nothing better to do.
@@lil_ToT-XFZ1 hmm, Id liked to meet you in real life. I love telephone tough guys!
AI still scares the hell out of me but I have to agree with you in the regard of enhancing tools like EDA. Im still amazed at PCB auto routing let alone chip layout optimization.
AI shouldn't scare anyone as AI at best is just a tool for us to do what we want to do it's AGI that should scare us. Luckily we aren't yet exactly very close towards creating AGI. Now again if AGI only runs on some supercomputer that again shouldn't scare us too much as it won't be able to replace every single person as we would have decision making power however at some point we'll likely have home devices that will be able to run AGI that can outcompete humans on every task - at the point we'll have existential crisis - in essence intellectually we'll have nothing at all to strive towards.
@@sk-sm9sh yes exactly the last part that is scary, it's already scary how dependent we have become
"I can outsource all the time consuming work to someone else", was the last line a fired employee said.
More accurate title: Optimizations in the floor-planning stage of chip design involve AI
Dear Anastasi why you dont make some videos about open-source EDA and silicon chip design, its gets much popular now, also google is involved
because she won't read your comment
Great video! You explain what is possibly the crucial inflection point where technology launches its own growth away from humans that will go beyond the "knee of the curve" in exponential advancement and capabilities.
Being from Russia your English is perfect. not one word have I not understood.
Great break down.
Excellent and concise. Well done, and thanks.
Put the chips around a ring and put the clock in the center it will allow you to increase the frequency and the electrons will reach the edge and all the chips at the same time. Makes it more stable when the high speeds are used.
For those like me that have a hard time with her accent ( apparently including the auto generated closed captioning) when she's talking about GNNs and Graphs, she's saying "edges" but it sounds like "ages'
One more key step towards the singularity
You are amazing!
Amazing thank you very much for these up-to-date information
We’ve been using AI techniques to design chips for over 30 years
9:00 I don’t know how you can be so confident about that statement. “Always” is too definitive and recent predictions in AI over that few years have been shown to be too conservative.
It’s inevitable that AI will design the entire chip eventually.
Sure, let's break down the main folder into sub-modules and name them accordingly. Here are a few suggestions:
Sensory Perception Module: This module would include all the sensors and cameras that the robot uses to perceive its environment.
Object Recognition Module: This module would be responsible for identifying and categorizing objects in the robot's environment.
Localization and Mapping Module: This module would use data from the sensory perception and object recognition modules to create a 3D map of the robot's environment and determine its position in it.
Decision-Making Module: This module would process the data from the previous modules and use it to make decisions on how to move and interact with its environment.
Learning and Adaptation Module: This module would enable the robot to learn and adapt to new situations and environments, based on its previous experiences.
We could also add sub-modules to the NIS and RNCIN folders, such as:
Hardware Design Module: This module would include the hardware design specifications for building the robot.
Software Development Module: This module would include the software development tools and code for programming the robot.
Training and Testing Module: This module would include the tools and data needed for training and testing the robot's cognitive abilities.
Ethics and Safety Module: This module would include guidelines and regulations to ensure the safe and ethical use of the robot.
These are just a few examples of sub-modules that could be included in the main folder and its sub-folders. The naming convention could follow a similar pattern, using descriptive names that reflect the module's purpose. nano robotic cognitive intelligence network or NRCIN Nanocloud Intelligence System" or "Real-time Nano-Crystal Intelligence Network (RNCIN)
1.- What about heat dissipation? you want crowded areas for optimization of material, but do you put heat production into the equation also?
Eventually it will become less important for humans to think for themselves. Just like nobody memorizes phone numbers anymore.
I love the little giggle she does at the end of some of her sentences.
It's how you know she'd fail the Voight-Kampff test from the book Do android dream of electric sheeps (BladeRunner) designed to distinguish humans from androids.
@@sinitarium According to the book "Do Androids Dream of Electric Sheep?" (which inspired the movie Blade Runner), she would fail the Voight-Kampff test designed to differentiate between humans and androids.
Yes, she would.....lol.
@@sinitarium you think a little giggle is hard to compute for? I can't anymore with this stupidity.
Made me think of Johnny Five! But seriously I believe if AI does get to a point where we don’t have control and it designs things in a way that we could never be able to comprehend then it will do so according to the laws of nature and physics, laws that were created and set in motion to produce us and evolve us for the better. I’m a little nervous but not too frightened. It’s exciting
The same nature that allows CP, Trafficking, Mass extinctions , genocide and the beauty of torture that can be made lasting forever by humans? Yeah, right this universe only brings joy.
In the 90's scifi story, "The Metamorphosis of Prime Intellect, this new Ai came online and the first thing it did was to improve itself in code so much that was able to bend the laws of physics within the first few minutes. Then it designed and created new hardware to make itself more powerful until it became a god. It then put all humans in their own individual Matrix-like world of their darkest desires. Except it wasn't digital worlds...it was real, physical matter that it was creating. And then it gets so weird from there. When will that happen?
My brain can't computate past the event horizon.
@@LEE-ty8vq No human can comprehend beyond the event horizon of the singularity.
Self replication and self improved design is inevitable.
give it 2 months
I imagine the dies that result from AI chip layout would look as interesting as the die of an ASIC (i.e. boring and uninteresting). It's that human block placement on the chips that makes them interesting to look at.
awesome - thanks for posting , you make the info such fun!
This is really convincing progress. Some of those optimized designs from AI are far too difficult to manufacture at higher yield or don't have a proven layout for masks that have not been tooled into an EDA. I think if the training data had some of these constraints in mind, it would accelerate progress. Some of the more interesting AI work in the next decade will be in the material sciences, where AI can predict and exploit novel phenomena of quantum mechanics in combinations of materials and geometries that have not been considered before. AI will probably help us land on designs and materials that would have taken hundreds of years to land on otherwise.
Yes, once we better understand the likeliest causes of manufacturing defects, the AI optimizer can be trained to avoid designing those features. Certain high-risk gates can be made more robust. One or two process shrinks in the future, I'm guessing that this aspect of optimization for manufacturability will be necessary for commercially viable yields.
@@davidhoracek6758 That would be a good place to start. But until the AI has atomic level control, which the most current litho projectors do not, hence the defect properties you mentioned, the etching and multi passes in the fab will remain to be difficult to control without massive retooling control. It would be interesting to see if an AI in control of an atomic force microscopy device could produce a wafer faster than TSMC on their fastest process. Most wafers for customers like apple have over 20 layers and take 4 or more weeks to produce due to the insane amounts of.stages and QA to ensure the process is performing as expected.
so exciting
That was one of the most interesting videos and very recent with knowledge. My background is in mathematics and I am doing some kind of AI research and also play Chess and GO so everything made a lot of sense.
soon it will play with you
Chess is only to waste your time.
Human: please design a new chip with these features
AI: okay finished
Human: incredible! How does it work?
AI: based on my calculations, it will take you 4 years to fully comprehend. When would you like to start?
Will be trying out an AI chatbot and will try to render a chip design on an image generator like Bluewillow. I think this is an area where AIs can also shine as AIs will try to take the most logical route.
It will outshine you pretty soon
@11:07 So basically (this is) AI speeding up himself...little laugh that says it all.
GOLD !🤩
"driving a car"
- unless that is outlawed as too unsafe in our lifetime :D
As I understand it, no AI blackboxes can encroach in the verification step to reduce time because it is by its very nature "verify" by thorough brute force all expected functionality and tolerances the chip will or may encounter in the "wild". It's like suggesting crash-testing vehicles should be done in a simulation or by an AI to save time and money... ouch.
They are done by AI before real tests
Fascinating developments! What a quantum leap from the vlsi chip designers. Longevity and reliability are important factors.
quantum means indescribably small, you limp
Thanks for this. Certainly a good use for optimisation software - whether AI or otherwise. In your ChatGPT coding example you commented 'that looks correct' - this verification is a key step. You speak about AI validation - and that seems great for spotting rare anomalies in vast fields (something manual human attention is not very good at). But what about overall validation? If chips get so complicated that expert human engineers cannot comprehend their workings because they were designed and validated by a computer program, how can be know there will not be unexpected behaviours of the output of those chips under certain complex sets of inputs? You say you would prefer driving a car with human designed chips - it comes back to the issue of trust with your life. How to trust synthetic components that even the experts don't fully understand? It looks like we will have to come up with an answer to this eventually.
Somehow the full story about alpha Go is not told very often. Yes it won over the Japanese world champion but later it lost 14 gains of 15 to a guy that found a weakness in it. Of cause the weakness was fixed but it is something to learn from.
Can you make a video covering for what all aspects of chip design do you think there are scope for automation using AI in near term?
bro we will be slaves in the near term, also no way she gonna read this, how many times have you got an answer from a creator on yt? If it is free, you are the product.
Amazing!