Cybord Analyzes the Authenticity, Quality, and Traceability Of Components With AI And Big Data

We are very fortunate to have Dr. Eyal Weiss for today’s episode. He is the CTO and founder of Cybord, an AI tool for detecting counterfeit hardware and electrical components.

This is a very exciting conversation. We will talk about a better, practical applications of AI in the electronics industry, outside of design.

Watch through the end and make sure to check all the additional links and resources below.

Listen to the Podcast:

Download this episode

Watch the Episode:

Show Highlights:

• Introduction to Dr. Eyal Weiss, CTO and founder of Cybord an AI tool for detecting counterfeit hardware and components
• Dr. Eyal Weiss's company is using AI-based component inspection services to detect counterfeits and ensure component quality.
• The system utilizes images of individual components taken during the placement process, leveraging big data and deep learning technology.
• The software integrates seamlessly with existing pick-and-place machines' internal vision systems, making it a plug-and-play solution.
• AI plays a crucial role in understanding and processing the vast diversity of component types, enabling accurate identification and fingerprinting of individual components.
• Most component failures are due to component-level issues, not board-level failures, making it crucial to focus on component quality.
• The traditional industry approach treats components as bulks, similar to handling grains, without addressing them individually.
• AI and big data technology now enable the ability to address components as individuals, allowing for better traceability and targeted recalls.
• Dr. Eyal Weiss's company uses AI and image analysis to examine individual components, providing detailed information about quality, authenticity, and potential defects.
• By analyzing individual components using AI-based inspection, manufacturers can conduct more targeted and efficient recalls based on specific fingerprint information.
• AI-driven analysis can also identify mismatches in traceability information, preventing wrong recalls and improving recall precision by targeting smaller groups of components.
• The AI system uses image recognition and natural language processing to interpret component markings and extract relevant data for traceability.
• The individual component level traceability is recommended in the latest IPC standard (IPC-1782B) and can provide the highest level of traceability with minimal additional cost.
• The AI system's feedback allows OEMs and CMs to improve their processes and make informed decisions on recalls, ensuring a win-win situation for both parties.
• Dr. Weiss wishes everyone success and a lifetime of no recalls ;)

Links and Resources:

• Read related articles:

  • The Path Forward for Embedded AI

  • Bringing AI to Market the Easy Way

• Connect with Dr. Eyal Weiss on LinkedIn

• Visit Cybord website

• Learn more about the IPC-1782

• Visit Octopart.com, the Electronic Parts Search Engine

Transcript:

Dr. Eyal Weiss:

The philosophy of what we're saying is that imagine if you could have looked at every individual component before you placed them on the board, just look at them, and based on what you see with your eyes you can say what component it is, what type it is, you can see if there's defects or corrosion or something bad on it. You can read the marking, and then you go to the internet and find out what it means for every individual component, and then you feel safe because you placed the component that you really checked, you know what you put there, you know when, where, whatever. Everything you need to know, you know. The problem is that there are hundreds of millions of types of components, so this is why it was perceived as an impossible task. But this is where big data and AI comes in.

Zach Peterson:

Hello, everyone, and welcome to the Altium OnTrack Podcast. I'm your host, Zach Peterson. Today we're talking with Dr. Eyal Weiss, CTO and founder of Cybord.AI. His software company is using AI in inspection and traceability, and I thought it would be very great to talk to him today about this very interesting area in PCB manufacturing and assembly. Dr. Weiss, thank you so much for joining us today.

Dr. Eyal Weiss:

Thank you. Good to be here.

Zach Peterson:

Yes, thank you. So we don't talk to ... I think we don't talk to enough manufacturers, and specifically folks who do software for manufacturing. So maybe if you could introduce yourself and your company to our audience.

Dr. Eyal Weiss:

My name is Dr. Eyal Weiss, and as you said I'm the founder and CTO of Cybord. I came to this interesting field after a very long life project I was working on for 15 years, inventing a new technology from invention, testing, then prototyping, and then finally fielding. A team of 25 scientists and engineers doing the life project, and then we failed in the field because of a capacitor that failed. When we went to check what happened, we found out that there was corrosion on the components because basically their age was faked, and the project was almost failing.

Eventually, we recovered from it, but think about it, the life project always going to trash because of a one-cent capacitor. So when we went to the manufacturers to ask them, "How come you don't check the components before you use them?" They just told us there's no such thing as testing the components, just buy them from a trusted source and that's it. We had to recall thousands of units very, very broadly to try to fix it. It cost millions, the project was delayed by a year, and then I realized it's time to put some new technology into this field, which was completely manual until then. So after the project was successful eventually, I founded Cybord to do this step.

Zach Peterson:

This is interesting, because I have heard of component inspection services specifically looking for counterfeits, and they usually focus on integrated circuits, making sure that there's actually a die inside the package, things like that, or making sure the die is not fake. I've never heard of any kind of inspection service doing the type of things that it sounds like you're doing, except for maybe some visual checks to check to see if there might be some corrosion for example.

Dr. Eyal Weiss:

Well, this isn't exactly the topic of our talk, but basically making components without a die is old-school. Today, counterfeiting is much more sophisticated. What we're doing is we're doing something completely new. Using the power of AI to look at every individual component placed on every individual board using software only, because we're using the image system already on the pick-and-place machines or the AUI machines. So based on the images of every component we can tell you who's the manufacturer, when it was manufactured, on what site, if there's any quality issues like corrosion or mold or cracks or bent leads or something like this, on the fly for every individual component during placement.

This is possible because today big data technology and deep learning technology has made this possible. Until just a few years ago, it was pursued as an impossible task. So yes, we're using images and we are looking for all aspects of the components from counterfeit, quality to information.

Zach Peterson:

So it sounds like you have maybe a plug-and-play software package. Someone who already has a vision system that is doing inspection on the line, they could interface with the equipment that's already there and then work that in to your software, or work your software into that process I guess.

Dr. Eyal Weiss:

Yeah. Basically yes, but it's not just the inspection software, it's basically the assembly hardware. A pick-and-place machine is using its own internal vision system to make sure the components are aligned correctly before they are placed on the boards, so every component its picture is being taken anyway by the pick-and-place machine. So we take advantage of this action taking place anyway, so you don't have to add any process or any inspection machine or hardware, you can just use this hardware that's already inside.

We are working with ASM, with Fuji, with Yamaha, Universal, and so we are working directly with them to get the images from the machine. We have an API for that, so from the operator point of view it's completely transparent, it just happens in the background, it doesn't slow anything down, just inspect all the components during the assembly.

Zach Peterson:

So you mentioned that AI is involved. I can see how AI is involved just in image processing and recognizing some sort of defect that might be obvious in the image. I think someone might ask, because there is all of the AI hype going on, I think someone who is skeptical might ask, "Why couldn't you do this without AI?" Or is it possible to do it without AI, it's just that the level of compute is too much? What makes AI the secret sauce?

Dr. Eyal Weiss:

All right. Well, it's not just AI, it's understanding what you're doing that makes the AI possible. Some people are saying, "All the data is the same." But it's not really accurate. The philosophy of what we're saying is that imagine if you could have looked at every individual component before you placed them on the board, just look at them, and based on what you see with your eyes you can say what component it is, what type it is, you can see if there's defects or corrosion or something bad on it. You can read the marking, and then you go to the internet and find out what it means for every individual component, and then you feel safe because you placed the component that you really checked, you know what you put there, you know when, where, whatever. Everything you need to know, you know. The problem is that there are hundreds of millions of types of components, so this is why it was perceived as an impossible task. But this is where big data and AI comes in.

If we harness the power of this very large diversity, we can use it to our advantage. So this big diversity becomes, instead of a liability, it becomes an advantage and makes it possible. Now, what we do, the trick here, the secret sauce is not just trying to use it as data, but to understand that every component is an individual component, every component is unique. What we find is we are trying to look for evidence of a fingerprint that this machine that packaged it imprinted on it during the component manufacturing. So instead of trying to identify hundreds of millions of types of components, we only identify a few thousand machines that packaged them.

So instead of being an impossible, slow and inaccurate task, it becomes a super fast and super accurate task. That's our secret patent and sauce, how to do this on every aspect of our product, for counterfeit detection, for quality detection, for figuring out what every marking on every individual component means. So yes, without the AI this would not be possible, because you need big data and AI tools to do that.

Zach Peterson:

I see. The AI is really an automator for parsing through and processing the data, and then doing that comparison in order to build that fingerprint down to the individual machine and package that particular component.

Dr. Eyal Weiss:

Yeah. In this case, this is really the enabler of the technology. We see in many applications today that AI is used to organize data, to make sense of data, to understand things about data, but in this case without AI, it's not just the organizer, it just would not be possible to do this understanding or deep understanding focused on the technology of the components themselves, and figuring out if a component is good, who made it and all the information you need to know about it. So it's really the core of the process that we're doing.

Zach Peterson:

Well, I think someone would rightly ask, "Why do I need to know the fingerprint associated with a particular machine that made that component?" Don't the manufacturers already supply some kind of data that allows me to break these components into lots or into batches, so that if I need to recall something from the field or if I need to reject a batch of components I can do that? It sounds like they don't do that just based on the story that you mentioned when we first started talking.

Dr. Eyal Weiss:

Yeah. Actually, what you're addressing is something people perceive like it's someone else's problem, manufacturers know what they're doing, let's trust them, they have a good process. That's okay. But it's a business, it's an industry, and every aspect of production today is being controlled very strongly and deeply, although it's the same concept, it was made for to make production in high quality, but yet you still monitor the quality of the process during the assembly very, very strongly. This is why the quality is so high today, relatively high.

But the components themselves, nobody's looking into them, so it became like an open loop issue. If I'm selling your product and I'm telling you this is the product and nobody can verify it, nobody except for really final testing, functional testing, which was not covering a lot of the issues, so the process is incomplete because there's no verification. No one is really telling you the components are good, until something fails. Every process that goes in open loop is not a good process. What we did is we just started looking at the components.

So first, we look at the quality of the components and we say, "Okay, we are seeing so many components, so far we looked at more than four-billion, so we see so many components, let's see what we can tell about the quality of the components." About corrosion, about counterfeit, how many are there? Because theoretically they are supposed to be good. We didn't just use low level manufacturers, our partners are really good, top of the line manufacturers whose qualities are the first on the list. You see that for example in corrosion and body defects, there's about 100 to 200 defects per million, that's a lot. In counterfeit, it's about 0.4%.

When you buy them from franchised source, through a trusted well established source, if you go to the brokers' industry it's about four to five percent counterfeit. The traceability information errors, so I mean this is supposed to be this date code for example, but in fact it's something else, the error range is between a half percent to the good quality component manufacturers to eight percent in some cases. So the reality, when you shed light on it, is not as good as you think it is. So even if you are using the best tools for traceability, the data is just not verified and not tested. Because of that, you have so many errors and so many mistakes and inaccuracies in the process that when you do have to do a recall, in many cases you do a not effective recall, because you bring something wrong back and not the one that you wanted to do. The information that you think you have is not accurate.

So it's really interesting, but once you have so much data you see what is the reality of the product. So people are saying, "Oh, I have trust-ability, no problem. Or it's okay, I have AOI, so I'm covered, right?" So yes, you're using the best technology available today, almost the best, but the reality is that because the process is so well monitored and the components are not, it means that almost all the falls that people have today comes from the component. It's really time to shine a light on the components and see where we can save money on components, where we can improve the quality based on the components' quality, traceability and so on.

Zach Peterson:

Well, when we had talked earlier and after hearing this from you again, this is really surprising, especially as a PCB designer, because as designers we care so much about the board being reliable, and we expect board level failures. It sounds like we implicitly assume that the components are perfect because they came from a semiconductor manufacturer, and they do their testing and they do their accelerated life testing, and we just assume they know what they're doing and it's all good. What you're saying is that actually component level failures are more common than board level failures.

Dr. Eyal Weiss:

Yeah. This is because board level failures are addressed very well by today's technology, it's not just chance. It wasn't like this 10 years ago. Before there was AOI or before there was so deep technology going into the process, but today when all this technology's in place, yeah. Almost all the failures today are caused by components themselves.

Zach Peterson:

So are manufacturers aware of this now? Are component manufactures doing anything about this? Or are they just now becoming aware of it thanks to the work that people like yourself are doing?

Dr. Eyal Weiss:

I think that we are trying to get awareness to these issues, because this is our business basically, but people feel that there's something they are not controlling. They're calling it black magic, I don't know why this failed, so let's try to see, this time it worked and this time it didn't, what did we do different? I don't know. So they're trying to find something like playing with them, and people are always looking first for faults that they did. So if you are building a product say, "Okay, this is probably my design." Or maybe the manufacturing place or the workmanship. They're trying to look at things that they can control, but because they can not control the components.

Even very, very large and very, very powerful companies don't really control they components that they use, they just buy them. Even when they buy them, they can control maybe lot information and arrangement of data, but they can not really control the components. So everybody are always focused about what they can control, and components was perceived as something that is just out of the control. We're trying to bring tools to focus on the components themselves, because we're saying that every component matters. Everyone, you don't have to treat them as bulks anymore.

Think about it. The entire industry today is addressing component in bulks. So you bought a component that was produced by a specific site at the specific date or a week, basically a week usually, and they are all grouped together. So they are the same lot. Why is it done like that? Because the technology doesn't allow you to do anything better. So even the best kind of traceability out there in the market today is managing the materials by lots.

Zach Peterson:

So as a PCB designer, we're often thinking about the board and the board level failure potential, and I think we just naturally assume that all of the component vendors know what they're doing and they've done their quality checks and they've done their life testing and everything that they're supposed to do to guarantee that they give us a good component, or at least guarantee the distributors are getting a good component. Then it's getting to us in a reasonable amount of time to where the date codes are valid and it's traceable, and still high quality and usable. So we don't assume that there's going to be a component failure unless we overstress the components, but it sounds like you're saying is that most of the failures, even without overstressing, are just component level failures due to qual;ity issues with the manufacturer.

Dr. Eyal Weiss:

It doesn't have to be the manufacturer, but think about it, it's a complete industry. No industry, no matter how good it is, is completely perfect and is fail-safe and there's no human error in the loop somewhere, and there's no all kinds of things that have to be done to deliver. It's an industry, it's a huge industry, so assuming that everything is perfect is just what people do, because they have no way to cope with the components. You have control of everything that you can control, so you can check the process, check the placement, and you check the functionality, and you're trying to put control into every aspect of product that's revealed and you put everything in it.

But whatever you can not control, it's just out of the scope of your thinking, so people assume that everything is nice in the components themselves, but the reality is it is just another part that can fail of course. Because you put so much effort into the process, the components are left unchecked. All the failures now, almost all the failures now focus on this side. Think about the level of how little effort were put in the components. The boards, we know them by name, so they have unique identification for every individual board, serial numbers, right? So a board is unique, and if a good company have a traceability capability they can pull out all the tests done on this board and the qualifications and what was good and what was bad until it was finally released.

But the components are handled in bulks, like it was corn. So the technology to handle components is the same as handling corn basically. You assume that everything in this bulk, think about a big sack of grain, this is how you basically handle components. You say, "They all came from this big grain, so if this big sack of grain was stored somewhere bad, then okay. Now, let's try to recall all the bread that we make from this large sack of grain, because it was, I don't know, contaminated, sick, whatever." So this is basically how we handle components today. We say, "We don't have grains, we don't have sacks, but we have date codes and load codes and things like this that can handle material in bulks."

So we call it bulk, and we say that, "Everything that came from this bulk is the same, and if we have some issue then we will recall the entire batch." But this is just because until today there was no technology to address the components as individuals, like we do for the boards for example, or for anything else except for components. But today with AI and big data technology this just doesn't have to be like that. You can address components as individual and stop treating them as bulks. They are not grains of, I don't know, corn or rice or whatever. There's sophisticated technology that you can address them individually. I'm not talking only about the expensive ICs and memories, which we do handle specifically, I'm talking about every component.

Because as I've told you before, my big project failed because of a capacitor, a one-cent capacitor. So there's no need to handle them as bulks, you can do it individually. So suppose you have a failure in the field, I can show you the picture of the component that failed and I can tell you what was the results of the analysis for corrosion, defects, authenticity, homogeneity, mold, cracks, whatever. Then we can recall only the ones that show this failure or this issue or the siblings or some kind of similarities to make this recall, instead of being done in a bulk method, do it individually. So we can recall an individual component or individual board associated with the component it was placed on. This is really a completely different approach to handling material, and today it's possible because of this big data and AI technology.

Zach Peterson:

It sounds like in the past, I suppose, that if this type of issue came up and a recall was needed, that the recall was I guess too extensive or too conservative, to the point where they're recalling a bunch of things that not only didn't fail, but probably were not in danger of failing, they just happened to contain a component that came from the same bulk batch, as you described, as components that did fail. So I think someone would rightly question how is it that those components got intermixed like that, and I guess at that point it wouldn't make sense to really do that kind of big recall if you could identify the root cause. But it sounds like the manufacturers don't even know how to identify the root cause, they just put the components out there and so you have to do this really detailed inspection in order to get to that level of granularity.

Dr. Eyal Weiss:

Yes, but it's not costing you any additional work or process, because you already look at every component. So it's just adding software to analyze them using very sophisticated and very lean tools. So yeah, it's about time that this will happen. Now, I'm not talking only about faults. Even if you want to do or handle your material in bulks, which is fine, this is how people are doing business today, you still don't know if it's really correct or not. Let me show you an example, we're working with really large companies and last week we had a recall event that we were involved in. So we had the date code, and the company was using their conventional traceability system, which is the top of the line to do a recall based on a date code.

So there was a crystal that failed and they tried to figure out, "Okay, what failed? What do we have to recall? Where's the fault?" It was manufacturers in different sites around the world, so we pulled out the information, we told them, "Okay, you have the same date code manufactured in different locations around the world, but in fact you have two different manufacturing sites for the components with the same date code. So if you want to recall, recall the one that failed, because this is a completely different product, it was assembled in a different site." It was a crystal by the same company within different sites.

Just based on looking at the top marking of the components we can tell them that there are different manufacturing sites. So instead of recalling the entire lot, the entire date code of the two production sites, they were lucky, it was only the small part of the manufacturer that they had to recall. So instead of recalling thousands of boards, it was only a few hundred that they had to do that. So even if using the conventional, not looking for a specific fingerprint or fault, just by getting the correct information for traceability, you can make your recall much more efficient.

I can give you more examples, there was another case, I think it was about two months ago when we detected a mismatch between the traceability information that was recorded, that was written on the reels, and then one that was actually placed on the boards. This happens all the time, hundreds of times a day, so it's not something uncommon. It's very, very common. But there was a question about the quality of this component, and what we saw was that because of this error they started investigating wrong parts. Because what was marked on the component was one date code, what was marked on the reel was another date code. So when they tried to investigate what happened, they recalled the wrong boards.

So it's very, very common. So basically you're trying to recall the wrong component, you investigate it saying, "Hey, something failed here, but I don't see the failure, so it's black magic. I don't know what happened here." So we just showed them, "Hey, this is a wrong date code that you're trying to recall." So this is another aspect. A third aspect, which is really interesting, is even if you manage material by lot, not by date code, when you see so much material you understand that companies mix in the same reels different date codes and different lot codes, and they just don't tell you about it. Why do you care? It's the same component, it works, I'm signing that this is good, so who cares?

But in fact, we see it's very common that manufacturers, good manufacturers, mix in the same reel different date codes and different lot codes. So okay, so if you are assuming, "Okay, who cares?" So if I'm recalling them, I'm recalling them anyway, right? Yes, but there's a bigger opportunity here, because let's say you have a lot code on the marking, which is this big, and inside it you have five different sub-lot codes on the marking of the components themselves, so if you have a failure you can recall only the sub-lot, you don't have to recall the entire lot. Why should you? The group, the sack of grain is the smaller one, you don't have to use this big sack of grain, you can use a smaller sack of grain.

So if you're doing a recall, you can do it more precisely, first without an error because you're actually seeing what you're using, and second you can choose the smallest group to make this recall event less painful.

Zach Peterson:

I see now, there are multiple ways to get around just the standard manufacturer documentation that's provided, whether it's on the component or on the reel, in order to really target what you want to recall, should you need to do a recall. You can pick out other data, you can make sure that it's correctly identified, whatever is going to get you the most accurate result. I think I'm starting to see where the AI really becomes important, because for a human to go through and process all of that stuff manually coming off of the component would take forever, whereas the AI can automatically parse through the data on the backend very quickly. You basically don't have to worry about it as a human.

Dr. Eyal Weiss:

Yeah. It's also because the standardization of marking on components is not there really. Everybody puts whatever he likes, there's no real standard. So figuring out what every individual component marking really means, so what is the code? What this code means, what this code means, what this code means is an impossible task for a human. So what we did was we did something very similar to what everybody knows about, ChatGPT and NLT, trying to develop a kind of AI that uses the same concept as ChatGPT or NLT, figuring out what it means, but specializing on components.

So we get an image of the component, we get the text and we figure out what it means. So what part means the date code, what part means the lot code, what manufacturing site and so on. Because there's so much variation, we're using something very similar to NLT to figure out what it means. So it's not just reading the text or seeing it, but it's figuring out what it means using AI.

Zach Peterson:

Yeah. This is another instance where ChatGPT invades the conversation again.

Dr. Eyal Weiss:

Yeah, I guess it invades all the conversations recently, right?

Zach Peterson:

Yeah, unavoidable. I think one final question that you could speak on very well is regarding the IPC traceability standard. How will your work or will your work affect the IPC traceability standard going forward?

Dr. Eyal Weiss:

Well, the latest version, the 1782B already contains this individual component level traceability. So it's already there. So it's becoming like the highest level of traceability available today, but it's mixing things up. Because today we're level one, two, three and four, level four the highest level of traceability was the ability to ... It means that you have a serial number on every board, and you have an identification on the reel, and you have a system that also tells you what from this reel came to a specific reference designation on a board.

We are bringing something that makes all these three levels basically the same, because with a piece of software, you don't have to add any hardware, so if you have only level two traceability system you can have individual component level traceability. So the idea is that in the new IPC standard, the latest version, there is already strong recommendation, it says that it's highly recommended for quality manufacturers to use individual component level traceability. So it comes there because this is a new technology, it comes as a strong recommendation, but I guess that maybe in the next versions I hope it will become even stronger than that. Because they didn't want to do it more than strong because it's a new technology, but it really is the next level, the next step of traceability level that is now available.

The good thing is that it's actually cheaper than doing traceability level too. So you can have the highest level of traceability with almost zero fuss, zero work, because it's all automated AI software, real time. So even if you have a traceability issue, a failure, during testing you can immediately use this information to shift your production. This is actually what happened to our large customers I was telling you about, once we told them about the issue they immediately cut this manufacturer out and they continued manufacturing with the other manufacturer on site. So there was no slow down, no halts, because the feedback from the traceability sinking into quality, because it's real time, because it's automated, because it's AI.

So it's really very, very powerful. We're now also working with IPC toward adding a verification of the traceability information. Because traceability standard today only describes the methods that you use in your production site you have traceability, but there's no verification of the data of traceability. It tells you that you have do serialization, that you have identification of the reels and so on, but there's no verification or automated verification or any kind of verification of the results of the traceability that you have. So we are working together with IPC now to make this also a part of the standard, so if you practice traceability you can really make sure what is the quality of your traceability information that you've got based on this.

Zach Peterson:

So when that feedback comes back to, I'm going to assume it's the CM, when that feedback comes back to them what does it look like? Do they just get an email, "Hey, these components, these reference designators, you need to check these or you need to reject them?" Is it that kind of information? Is it really that simple? Or is it a bit more complex than that?

Dr. Eyal Weiss:

Actually, we're more addressing the OEM than the CM.

Zach Peterson:

Okay.

Dr. Eyal Weiss:

The CM of course is in the loop and is part of the solution, but we're working with the CMs and with the OEMs in different ways. For the CM, we're allowing them to have visualization of the traceability information that they have, so if we find a mismatch between let's say what is written on the component and what was inserted into the traceability system, so you can not argue with this mismatch because you see the picture, then someone made an error. Now, it could be an error that was made by the component manufacturer, it could be an error during the reception of the material, it could be an error during the handling of the material.

So it's an opportunity for the CM to have a feedback on their processes. So woops, we have a failure in accepting, let's find out what the issue is and fix it from the source. So for the CM it's a good opportunity to improve their handling of the material. Sometimes they say, "Woops, it's not enough, it's not in our hands because the error was on the components' manufacturer." Not as easy to address, but we did have two occasions where the components' manufacturer said, "Woops, sorry. My bad. It was really an error and thank you for mentioning it." So by feeding this information back to the CM and sometimes to the component manufacturers, it allows the improvement of the processes in both the CM and the component manufacturer.

From the quality point of view, yes, you get an alert and this alert will be handled by the quality people of the production line. So in this case for example, the quality people of the EOM told the CM, "Remove this manufacturer from the allowed approved vendor list for this production immediately, and then continue manufacturing." So it fed back immediately through the OEM.

Zach Peterson:

I see, I see. So it really does come top down, and they still get to maintain control and decide what to do, as far as continuing to manufacture with the vendor. They can make the judgment call or they can go with the advice and just immediately remove that vendor.

Dr. Eyal Weiss:

It has to be a win-win. It's not something that you can do like in a force, it has to be a win-win. So in the tension between the OEM and the CM, everybody wants the product to be good and everyone wants to be efficient about manufacturing it. So your process has to be streamlined and inline with what's important for both the CM and for the OEM. So it has to be a win-win, and this is how it's built.

Zach Peterson:

Okay, okay. Well, this is all very interesting, and I think this is one of the better practical applications of AI in the electronics' industry, outside of design. We've heard a lot of hype lately, and so I think your solution is not only very practical, but also will prevent failures, and also of course help limit the scope of recalls to a reasonable level, instead of just hitting everything with a hammer. So thank you so much for presenting this information for us today, and hope to have you on in the future as all of this develops.

Dr. Eyal Weiss:

Thank you very much. It was really a pleasure talking to you and your audience, and I hope you and wish you successful and not to experience recalls in your lifetime.

Zach Peterson:
 

Absolutely. To everyone that's out there watching and listening, thank you so much. Make sure to subscribe to us on YouTube, you'll be able to keep up with all of our podcast episodes and tutorials as they come out. Make sure to check out the show notes, we'll have some great resources that you can learn from, you can check out Dr. Weiss' company, Cybord.AI, and you can get in contact with them to learn more about their solutions. Last but not least, don't stop learning, stay on track and we'll see you next time. Thanks, everybody.