EMS@C-LEVEL
As Forbes, Entrepreneur, Fast Company and SCOOP writer, Philip Stoten, continues to talk to EMS (Electronic Manufacturing Services) executives he learns more about their individual and collective experiences and their expectations for their own businesses and for the entire electronic manufacturing industry.
EMS@C-LEVEL
The EMS (Eric Miscoll Show) - Is Inspection Leading the AI Revolution in Manufacturing?
The application of Artificial Intelligence (AI) is reshaping the manufacturing landscape, particularly within the inspection sector. In the latest episode of The EMS (Eric Miscoll Show), we dive into how inspection processes are not only evolving but becoming a linchpin in the AI revolution. Join Eric and Philip Stoten as they welcome industry experts Oshri Cohen, CEO of Cybord, and Joel Scutchfield, General Manager of SMT and Semicon Business Operations for Koh Young America. They share invaluable insights into the realms of AI, quality control, and the future of manufacturing.
Listeners will gain an understanding of how AI significantly enhances reliability and accuracy in inspections, moving beyond traditional methods to create a more efficient production environment. We discuss the critical role of data collection in shaping effective AI systems and how manufacturers can leverage these technologies to reduce defects and improve traceability. As customers increasingly seek automated solutions, the episode explores their expectations for immediate action vs. alert systems when it comes to quality assurance.
The important conversation touches on collaboration within the manufacturing ecosystem, emphasizing that sharing data across platforms can maximize the efficacy of AI solutions. With predictions on the future trajectory of AI in manufacturing, this episode is packed with thought-provoking insights and practical applications for industry leaders seeking to navigate the future of manufacturing.
Don’t forget to subscribe, share your thoughts, and leave a review! Join us in exploring how AI can revolutionize quality control and elevate operational excellence in the manufacturing space.
Like every episode of EMS@C-Level, this one was sponsored by global inspection leader Koh Young (https://www.kohyoung.com).
You can see video versions of all of the EMS@C-Level pods on our YouTube playlist.
Hello and welcome to this first episode of the Eric Miskell Show webinar series for 2025. Glad you could all take a time to join us today. Our topic today poses the question is inspection leading the AI revolution in manufacturing? Poses the question is inspection leading the AI revolution in manufacturing? Ai has been all the buzz for the last few years within the industry and we are fortunate to have two industry experts joining us today to present their solutions and to answer some questions and have a discussion as to what this whole movement is doing and what it is generating the benefits for the industry. Before I begin, let me, as always, welcome my lovely and attractive co-host, phil Stoughton. He is joining us from Australia again today. Good to see you, phil. Thank you and just a few housekeeping issues for the audience. As we begin, everybody will remain muted during the course of the webinar. If you do wish to pose some questions, we encourage you to do so using the question tab at the bottom of the screen and will be rebroadcast next week on EMS Now, and we encourage you to share it with all your friends and family and hopefully they get some benefit out of it as well.
Eric Miscoll:So let's begin Again. The issue is inspection, leading the AI revolution in manufacturing, joining us today. We're pleased to have Mr Oshri Cohen he's the CEO of Cyborg, and Joel Scutchfield, who's the general manager of SMT and Semicon Business Operations for CoYoung Americas. Gentlemen, thank you for joining and participating in this today. I wanted to begin by allowing each of you just to set some context, to kind of give the audience a brief overview of how AI is currently being used in your solutions. I'm struck by the fact that your solutions are deployed on different parts of the line, so to speak. So why don't we begin? And, oshu, why don't I begin with you, since yours is really at the very front end?
Oshri Cohen:Yeah, okay, thank you, eric. So, yeah, well, you know, I first just to say a word about myself. So I come from the end customer position. I've been many years with the Mellanox and NVIDIA, so I got the heat of the problems that we are today taking care of both Koyang and ourselves taking care. I got the heat from all over. I was responsible for supply chain and the troubles that I got, whether from the production line or later on from the customer field, were really bad problems, severe problems that we had to manage, and this is exactly what we are doing.
Oshri Cohen:So in Cyborg, what we do, actually, we utilize the power of AI in order to get images from the line. We're getting from two different positions in the line. The first one is the pick and place position, where we can see 100% of the component from the bottom side, and later on we get from the AOI position. We are getting there all the images from the top side specifications or product integrity or even product cybersecurity we are able to enforce and inspect, and this is what we do with AI.
Oshri Cohen:The reason why we need AI is, I mean, you know, for me it's obvious, but for the audience who hear us, actually you know the classic industry looking to have a golden unit and rely on a golden unit every time. They compare something in terms of visual inspection. Well, in electronics it doesn't work, because the variety is just almost infinite. So, if you don't use AI, you won't be able to cover enough and, of course, you won't be able to get good enough, reliable enough results. Therefore, this is why you need, you must use AI, and this is exactly what we are doing.
Eric Miscoll:Excellent Joe. Why don't you give us an overview on Koyang?
Joel Scutchfield:Sure, absolutely. I think most would agree. We are the global leader in inspection technology for PCB slash SMT assembly, offering both solder paste inspection technology along with automated optical for both pre and post reflow, and now kind of migrating toward the back end with through hole technology and dispense process for conformal coding, underfill, et cetera. And even on another front, which I know is not the topic today, into the Semicon world, which is really starting to generate nice momentum here, in particular in the US with the CHIPS Act et cetera. We have been from day one a measurement-based technology in terms of the use of the specific technology and approach that we are taking with our systems, and what I mean by that is, again, measurement-based, objective, parametric information. No gray, it's absolute. I like to say we have the right kind of data, being just that objective, parametric, measurement-based, being just that objective, parametric, measurement-based 3D. We get it the right way by employing the right types of subsystems in our machines to acquire accurately, and we get the right amount statistically relevant, a data set that you can make decisions with and make good decisions with. So we have that very robust, very accurate data set. Now that's great on multiple levels, but it particularly comes in handy when we are trying to create AI engines to, basically, you know, at the end of the day if I can simplify this it's really it's using this data, it's using this, this, this ideally massive data set, or whatever we can get our hands on right Another topic I'm sure we'll talk about as the as the webinar goes on but using this, this superior data set, you know to to really think like a human right, and we, and we do that, you know, by using a series of algorithms or models, or what we call engines, you know, which can be trained by a process expert and, you know, at the end of the day, you know achieve a specific expected result. So that's really what we're trying to do, and you know, and we do that in a couple of different modes.
Joel Scutchfield:There's this application of convolutions, which is basically the plus minus on the factors based on the data values themselves, and then activation functions, which really comes back to for my generation, basic language programming going back many years, where it's if-then and and-or, type statements and initiatives that drive activity right, that create the change, the generative change, right, and so, in the simplest terms, that's really what we're looking at. So it's really natural that we use this data set and we can use this data set and I think in a way that many others are maybe a little bit of a disadvantage because they don't have that data set and the technology, the robust measurement based technology, to drive it. So we've broken this down really into five core areas, starting with really easy and fast product development, or, I guess, from a manufacturing floor perspective, easy changeover, auto-programming right, I need to generate the next program quickly, I don't have hours Go, go go. All right, we're going to do that. We're going to do that with the least amount of human intervention possible, and that's where things get started. The second piece to that is now process optimization. So now I'm using the results that I'm finding at post-print SPI and post-mount AOI to potentially make adjustments to those systems to ensure that bad product is not being built, based on the data that we're collecting and the information that we're gathering.
Joel Scutchfield:The third piece of that is obviously assurance.
Joel Scutchfield:Now that you know the and I guess we could really start with this assurance that the product that's being built is absolutely 100 percent perfect. Right, we want no escapes. We want to make sure that we are validating the results of everything we've done prior to that last piece of inspection. And then we get into things like using AI to improve the inspection capabilities of the various tools. Right Now, I can use AI to create better inspection when it comes to a through-hole technology, or maybe things like flux inspection or, in the case of our DPI systems, bubble detection and so on and so forth.
Joel Scutchfield:And then, finally, predictive maintenance, where again we can look at things like OEE and anomalies that are kind of outside of the normal trend line, if you will, to be able to say, hey, something looks a little amiss here. Right, something might be getting ready to fail, but again, it's all based on this data. So you know, we have specific tools and we'll talk about those, I think, certainly today as the webinar goes on. But that's really our approach and that's our focus, and I think it will continue to be as we go forward until we've really reached the point where we've exhausted all of this. You know we can, and then we expand from there. But, in a nutshell, that's really our approach and how we're going about using this data, using these proprietary engines, to create an ecosystem based on everything I talked about. Based on everything I talked about, all of those five steps reduce human intervention and make sure that quality is there and our customers are as profitable as they can be.
Philip Stoten:Yeah, I think that's really interesting, joel, and it's fascinating that we've got you on as the kind of leading inspection equipment provider and then we've got Oshri on as a hardware-free or hardware-agnostic, software-only company, because it's a very different thing. But all of these AI tools are the same they only work based on the data you put into them. We've all played around with different chat, chat, gpts and anthropics and all these different tools, and if you don't give them the right prompt and you don't give them the right data, what you can get out to be can be pretty messy. How do you see that from your point of view, ashri? When you're dealing with lots of different placement machines, for example, how do you make sure that the synthesized data that you use is the right data?
Oshri Cohen:Well, ai like AI and, exactly as Joe has said, it is being fed by data. And data is gold if you use it cautiously. So, every time we collect data from the lines, we make sure that, first, the data is validated, because using unvalidated data will make more damage than benefit, and this is definitely not what we want our customers to experience. Second, we won't be able to get enough values from the data that we are collecting. So verification is rule number one. Every time we get the data, we must verify. This is one.
Oshri Cohen:And then, since we collect data from different stages and, by the way, we are doing something very different from what Coyang are doing Coyang are super concentrated in process, while we are doing something very different from what Koyang are doing. Koyang are super concentrated in process, while we are only concentrated on materials. So we are looking on each and every component and making sure that we are recognizing the material itself. We know that Koyang are doing fantastic job and Fuji and ASM are doing great job on their side on the process, and we know that the material itself, the raw material, the component itself, are actually overlooked. And this is what we are focusing on.
Oshri Cohen:And again, validating the data, making sure before we get this data into our models. Making sure before we get this data into our models, we make sure that this data is relevant is good enough to use. I'll give you an example. If you would like to build a model in AI that validates authenticity right, then you must make sure that the data set you build your model based on must be validated as original and franchise material right. Otherwise, you mistakenly can teach the model to look for counterfeit materials instead of original materials. So this is one simple example to make everyone understand what I'm saying when I'm indicating the validation part of the data.
Eric Miscoll:You know, it strikes me, as I hear you speaking, oshriya and Joel, to riff off of what you were saying too. You know, using terms like assurance and validation and what have you, which is really at the heart of what we're trying to do and improve the process and the outcome there. And do and improve the process and the outcome there, but speak to as this has evolved, you know, within the industry. You know it gets to standards too, right? What is being done? How is this being? It's not the Wild West, right? This is the electronics industry. So we're going to put in place certain standards, we're going to try to regulate this. We're going to try to ensure that the processes are replicable and that we can do this. We're going to try to ensure that the processes are replicable and that we can do this. Speak to that and kind of what's in place and kind of what may still be needed within the industry in this regard.
Oshri Cohen:Well, standards are a must. I give you a bad example and then a good one. Let's look on the marking. What standard the component manufacturer are following in terms of marking their products? The answer is nothing. No such standard. Every manufacturer do whatever he wants, whatever is more convenient to him to do this. And that's making the traceability task even much more complicated. And you know, even the same manufacturer can decide to change his own marking on the same type of component every few months because they decided to change the method of marking their products. And again, there is no standard.
Oshri Cohen:This one is totally neglected and this part making it very hard to analyze. From the other side, assembly assembly is good, in a good shape, because we have the. You know, the bible of assembly is the ipc. Right, we all look over the ipc a610, no matter what class, one, two, that doesn't matter anyway, this one is very clear. So when we, when we fine-une our tools, when we fine-tune our algorithms, we definitely lie on these kind of specs and these kind of standards. I'll give you an example Aerospace standards. I'll give you an example aerospace. So if you don't look on the AS9100, then you're probably not going to help this industry Because again, this is the kind of Bible or the very basics of this industry. And again, the tools must rely and must follow these standards. So definitely, even if we use AI, the guidelines will always come from specific standards and I can give you more example for the automotive or whatever industry.
Joel Scutchfield:Yeah, I'll add to that if I can. Eric. From our perspective and I'll speak to both of those, because Australia is absolutely right there's really nothing for part markings, polarity markings, those types of things which we address with our smart review tool. We really look at this in two ways, so there's the static element and then there's the adaptive element, right? So the static element is I would equate that more to, as Asha said, the IPC standards right? Let's take KAP, for example, our auto-programming tool, koyaan Auto-Programming. With that we can define things like package types and sizes and again measurement-based information.
Joel Scutchfield:It's pretty well-defined, right. The components within a certain range are going to be of a certain size and thickness, et cetera, et cetera. Now, that's not to say there can't be variation because of the component manufacturer's process, but we can develop AI engines, and it is, you know, this is another, I guess, misnomer, maybe that maybe we can clarify here. You know, within our auto-programming tool, it's a tool that consists of multiple engines, it's not a single engine, right and those engines are designed to do different things, right. So, again I mentioned, identify the package type, that's one. Identify the outline of the component, that's two. And on and on, and on. So it's this series of engines or algorithms, algorithmic models that we create, that make up the tool. But we can do this, really, we can do this off-site. We don't necessarily need customer interaction, we don't necessarily need data from the customer necessarily to do that. It's out there, it's in the specification, it's in the guideline, it's in the guideline, right, it's in the spec, if you will. We move to smart review and now we're looking at basically post-reflow solder joints. Again, polarity part marks Could be doing part marking polarity pre-reflow as well. Ideally we should, but just in the context of this discussion. Now we're using more of an adaptive approach because the tolerances can shift based on, again, ipc class 1, 2, 3. But even within that, the customer themselves typically has a twist on their specification and their tolerance settings for good, bad and so on and so forth. So when we're trying to help them classify judgments and use this information to determine, ideally, at the end of the day, adjust tolerances on the inspection tool itself so we're not creating calls that shouldn't be called it gets a little more complicated, right. So, and again that's the adaptive piece, and you know, again it's the training element. Is there as well, right? So we have to gather data, test and then potentially train Right, and that's much more easy, easily accomplished in a, in a controlled setting, right, where we have the ability to do all of that ourselves. But at the end of the day, we have to give that ability to to the, to right To the user, and potentially create tools. Maybe it's through an online interface that we can convert that data into data that we can use to train that engine right, and it can be done real time. So it's probably more of a little bit of a futuristic approach, but yeah, I guess.
Joel Scutchfield:Coming back to your original question, eric, the lack of standards to some degree, you know, do kind of make it still a bit of a wild west, even within our electronics world. So we'll see. We'll see where that goes. I know Ainemi was doing some work to try to start that process. It's really an interesting conversation. I'm not sure. I was asked this question about a year ago at an SMT chapter event where I did a presentation and you know, and I think everybody's thinking that yeah, there's a way to create standards around this. It's so I don't know, I'm not sure.
Philip Stoten:Yeah, I think it's really hard to create standards in any application of AI.
Joel Scutchfield:For how to create an engine. I'm not sure you know. That's more on the proprietary side. Yes, certainly for part markings and things like that. But yeah, it's a little bit all over the map, as Ashri said.
Philip Stoten:And there is an IPC standard for inspection, isn't there, Ashri? There are IPC standards that talk to what volume of components are inspected and those kind of things.
Joel Scutchfield:Absolutely yeah. The starting point is there, the defaults are there.
Philip Stoten:Yeah, I think that's fundamental.
Joel Scutchfield:If everybody followed the defaults, we'd be in great shape, right yeah?
Philip Stoten:absolutely. I just wanted to kind of put the cart back before the horse and go back to the initial premise. And what I'm curious? Because often with these technologies, we talk about them amongst ourselves and we're very excited about it as technologists, as journalists, as people that study the industry. What I'm really interested to hear from you guys is what the customers are asking for. Quite often, they're asking for a particular outcome or for a particular action from their inspection system. Are they asking for you to do that with AI? Are they saying just fix this problem. However, you can fix this problem. If AI is the solution, great. What are customers driving you to deliver to them? You're both nodding your head. So, Oshri, maybe you can take this one first.
Oshri Cohen:I can tell you, I can testify, what my customers are asking us. They are asking us to provide them confidence because they feel like they are doing a lot of efforts and getting still not enough confidence when it comes to materials they're using. You know they are at the end of the day, they are buying from franchise sources, they're doing, you know, inspection, all this COC stuff, everything Like it looks like they do everything in a correct manner and still they have huge amount of issues during their production and at the end of the day, the end product still contains bad materials. And the question is why? What? Why it happened, why did we do wrong and why? How can we improve? So they ask us to build the confidence, whether we do this with ai. You know ai is just.
Oshri Cohen:You know people get convinced as the times go on that maybe, maybe ai is the right solution because if you look 10 years ago, five years ago, everywhere where you saw a problem in this industry, the classic way to resolve it was to bring more and more and more people. That was the solution and I know Joel agree with me because I see his head and because I know that we did the same. We did the same. We had an issue in the process, in the assembly process. We ran to the line with a million people there and everyone had to look on a different component and all this stuff and you know what, at the end of the day we still got a bad product with some defects in it and usually related to materials, although sometimes it was even related to process.
Oshri Cohen:Now, these days, the situation is different because people start to realize that maybe AI can replace significant part of what people are doing. And it's not that people don't want to do good. They're doing great, they're doing the utmost effort they put there and still the results are not good enough. So AI is kind of the new magic that maybe will allow the customers to have a clean product without bad materials. And in case of, you know companies like Koyang.
Oshri Cohen:So you know, instead of placing 10 people for two, three days to program the machine, they can run the auto adjustment or the AI program, the auto-adjustment or the AI program, and then within I don't know one hour, less than one hour, they get the perfect program for the AOR. And then you know lines like, or tasks like today you know, getting a new board, what we call new product introduction, or MPI, into a line. That's a huge burden, huge burden in terms of setup. And if you come with the capability, real capability of AI to the line and you say, don't worry if you replace, even to an unknown board, it will take us an hour and we are able to start monitoring your product, that's just fantastic. It's a whole new era that this AI brings and I think this is the change that this industry finally gets, which it didn't get it for the last I would say 20 years, but you know I would be cautious the last 10 years.
Philip Stoten:Are you selling confidence, Joe?
Joel Scutchfield:And I think a lot of that comes along with the technology right, With the advances in processing power and just the ability to use AI in total over the last four or five years has really ratcheted this thing at a high rate. You know, Usher is correct. I mean, they really don't care how the problem gets solved, they just want the problem solved. But if you can use AI to do it and in the process again speed things up, ensure quality, reduce the amount of human intervention needed to accomplish the task, they're all for it. So I see kind of a mix of you know, again, they don't really care how the problem gets solved, but they do want to try to understand in many cases what it is that we are doing. You know, what is our commitment, what is our level of commitment to AI? You know, I saw an advertisement from IBM recently in one of the trade journals and it went something to the effect I think it was for one of the Watson tools, if you will, but before you use AI to help you get where you're going, you need to trust what it's doing.
Joel Scutchfield:And I think people are now starting to ask those questions where again, initially it was like oh, it's just magic, right? So, no, it's not. It's everything we've discussed up to this point where there's a lot of effort being put in to making sure we get the right kind of data and have the people, the brainpower to create these algorithms, at least initially, and then have the generative piece kind of take over as we go. But, yeah, it's, you know, again, I think there's this concept that we can still achieve this lights out phenomena, right, and that maybe at some point we will. And certainly I think this particular world, again, if we can get the right kind of data, has the ability to, you know, to to move closer and closer to that over time.
Philip Stoten:So we'll see, yeah, we'll see, We'll see, yeah, it's it's confidence and trust in the AI as well, isn't it? And it's how much you allow the AI to do. When you look at customers' expectations, do they see AI as a co-pilot that's going to say, hey, you've got a problem here, it needs fixing. Or do they see it as the operator to take the action? Are they looking for action? I'm thinking, with Cyborg, for example, of the RTI product you've been talking about recently, where it's actually removing foley components as it finds them at line speed. I'm thinking about rejecting boards and then maybe looking at them and finding root cause later. Are we in that phase where we start with AI telling us where maybe we should be doing better before we're at the stage where AI says, yeah, you know, I've got this, we'll just take care of this and we'll pass the boards that need passing and we'll fix those that can be fixed. Either of you can take that one.
Joel Scutchfield:Well, that's a classic example of what we went through with developing our KPL tool, our process optimizer for both the printer and the mounter right the pick and place machines, where it was an iterative process. It's taken years right through that development horizon, a lot of eyeballs, confirming, you know, what the AI was actually doing in terms of the result right, is it making the right decisions? Is it correct in what it's telling the printer to do in terms of adjustment to speed, pressure and snap off, release or clean cycle on the on the printer itself, to now basically saying, okay, hey, we, we, we know that this was designed to be a, basically a hands-off, a lights out. You know a tool, if you will, for those, for those variables, for those adjustment variables. Same with the, the mounter.
Joel Scutchfield:So, yeah, I think we do. We go through that. We go through that phase where you know there's a uh, there's a little bit of uh, kind of wait and see, um, but as we get further into it and as we prove that we can, we can be successful in these incremental uh, uh slices that we're that we're that we're creating. I think we gain more confidence from the customer base in terms of allowing us to do more with it and, trusting that we're doing the right things, that we are going to give them, ultimately, exactly what they're looking for.
Philip Stoten:Yeah, moving from a co-pilot model to something much more autonomous. Yeah, moving from a co-pilot model to something much more autonomous.
Oshri Cohen:Oshri, yeah, I would say that I see three levels where AI integrates into the industry and that's actually. Customer will be happy to get it in all three levels. It depends which customer we're talking, but usually they will be happy to get it in all three levels. The first one is actually taking actual actions, immediate actions, like our RTI, the one that just you know. Look at a component from the bottom side. If there is any defect in a component, just instruct the machine, throw it away to the dump. That's one example.
Oshri Cohen:Yeah, customers, like particularly EMSs, are expecting us to be able to help them with taking these immediate actions. The second layer, or the second level, is providing the alerts. So not always you can take immediate action and not always the customer would like you to take the action instead of them. They would like to be able to see some alerts and get a decision by themselves and again, they are willing to get a reliable alerting system that they can rely on and they get a better decision based on these alerts. This is second layer. The third one is actually a much higher one and that's exactly as Joel started in his first sentence. We are living in a world of data. Data is the new gold of our era and since this is the case, then customers are expecting us to collect all this information and run some smart AI algorithms to slice and dice this data so eventually they're able to understand what can they do better than what they're doing today, and they expect us to do this for them. Of course, they can sit, you know, and analyze the data manually.
Eric Miscoll:Let me build off of what you both were just saying there, because I think of it in terms of you know, obviously the initial thing is to improve quality and all that. But another consideration I think then would be or a benefit is also the traceability aspect of what you're doing and what your systems have. Kind of could you and now we just have Joel, so now maybe it's just Joel Can you speak to how critical the traceability element is to the solutions that you provide? And, just out of curiosity, how often are you called upon to actually provide that traceability data in the course of operations?
Joel Scutchfield:Well, in our case, I mean, that's probably more of an Australian initiative because there really are component level right across the board. That's their expertise. So he actually might be able to answer that a little bit more astutely or broadly. I should say, in our case it's really on the customer to. You know, we give them the opportunity to maintain all of this data, this data, this, this traceability, again right down to the component, through through KSmart. It's it's all there, okay, it's all there, it's all very accurate, and it's it's it's real time, it's it's for every, every board they inspect. But now they have to really archive that data.
Joel Scutchfield:We will help them, you know, we can help them do that, do that, but really, if they're doing things properly by basically taking the results data we're providing, saving the images they need to save, saving the text files they need to save to be able to retrieve that quickly, then that's basically our part. We've given them what they need. Now they have to manage it and, believe me, that's basically our part. Right, we've given them what they need.
Joel Scutchfield:Now they have to manage it and, believe me, that's not easy and we certainly have been called upon to help them structure our data in a way that it can be managed much more easily. It's not so expansive, right. We have to condense that. We can't ask a customer to, you know, put in you know, 40 terabytes a day to just you know, or whatever, pick a number, right. So that certainly is something that we get requested ongoing to continue to help with the management of that. But again, once those results are generated, then it's really up to them, from our perspective, to archive that and to file it in a way that they can get back to it quickly. And again, we certainly help where we can.
Eric Miscoll:So, Sri, what's your sense of that?
Oshri Cohen:I think that from our end, the approach is a bit different because, you know, we are not the one to generate the data. We are the one to take the data and work on the data and process it for the customer. So, as Joe would say, okay, it's on the customer. He's definitely right. By the way, if a customer doesn't get a decision we would like to have traceability and we are going to invest in traceability they won't be able to have traceability, that's it. No matter what Koyang will provide them or even what we will provide them. It's related to the integrity level of your product. Do you know what's in your product? If you don't know, then there's a big issue going to come up. Okay, so what we do actually? We translate, we recognize the data, we get all the images, recognize the data, translate it for the data. We get all the images, recognize the data, translate it for the customer and then we digest the information, we compare, even with his current traceability information that was collected all over the the other stages of the production, and then we provide them the analysis. Again, it's not an immediate action that they're taking, but that's a kind of alerting system, but more like an analysis, a tool that will keep their back and verify and fix the problems that got into their traceability system.
Oshri Cohen:Big companies, big OEMs, will always strive to have a perfect traceability system, because they know what happens when you don't have it. They start to suffer severely from recalls. You know, instead of recalling 100 boards, 1,000 boards, they start to recall 100,000 boards. And all those boards are sitting out there in data centers in different places, which are you really don't want to get there and stop these machines to work and replace everything up front. Very bad situation. So this is why customers these days, the big customers, are willing to have this traceability and quality enough. It's not good enough. I mean, it's good, it's essential, it's very basic, but it's no longer good enough. Quality you need to provide them traceability on top of the quality. That would be the complementary part that will complete the picture and provide them the insurance level they are expecting to have.
Philip Stoten:That's the real data ad, isn't it? Or the real value ad actually being able to provide that and if a customer has a recall, for whatever purpose, you being able to identify exactly where you know where a particular issue, where a particular component has been used, where a particular batch number has been used, that kind of thing.
Joel Scutchfield:so and on our site. That's where the case mark comes in. As I said, right, we we we do do structure it in a way that it can be archived easily. It can be gotten back to, referred to very quickly. Again, we realize the importance of that. As Ashley said, anybody that's doing automotive, medical, anything with any liability absolutely has to have traceability. But it's also very helpful just within the context of operating the manufacturing floor efficiently, and especially if we're now going to tie our systems at some point into the SAP or the ERP system, right, where supply chain is part of this mix and AI is going to be playing its role in that particular sector, which it is already as well, right? So I think we're going to talk more about this as time goes on, but ultimately, the data's there, it's typically there in a very structured and easy to access fashion.
Philip Stoten:We just have to manage it, yeah, and it's an ecosystemic play, isn't it? Exactly, exactly.
Eric Miscoll:Hey, listen, in the interest of time here I want to kind of start looking at at wrapping this up. I also wanted to to uh acknowledge both of you gentlemen will be participating or your companies will be at apex here uh, this night in a couple weeks. Everybody where the industry gathers in anaheim, so people have an opportunity to come and at this place on earth earth. Eric, Thank you very much.
Philip Stoten:That's right.
Eric Miscoll:For Ko Young and for Cyborg. Encourage you both to anybody watching who wants to know more about this and meet these people. Please plan on doing so there. It'll be three fun-filled days in Anaheim, but listen as we look forward though this is. You know, I'm struck by the fact that you know. The speed of change within this industry is amazing, and we see what's happening with AI, you know, in the industry. What do you think the inspection industry within the EMS industry is going to look like in a year or two, with the way that the trajectory and how things are going? How do you see things changing? How do you see things changing? How do you see things improving and getting better?
Joel Scutchfield:From our perspective, you know we have a fairly clear path, which I laid out a little bit earlier in our conversation here, with those five pillars that we're really driving to maximize, first and foremost. You know the next piece to this is, as we've talked about many times is to be able for all of us to play nice together in the sandbox right, and that means you know we have to move beyond. You know the screen printer manufacturers that we've developed. You know specific tools with the mounter manufacturers that we've developed, specific tools with the mounter manufacturers that we developed specific tools with. I think we were all hoping that CFX will evolve to the point where it's going to allow more and more of that, the data sharing up and down the line. Again, the more data the better, as long as it's good data right. So if we can have more access to more information, more data that we can then contextualize to be able to utilize for those five pillars that I referred to and now other things, expanding that ecosystem. We really didn't talk about the backend, but it is there and before we really get into the nuts and bolts of AI, we really the conversation needs to start withilot and then eventually, you know, be able to achieve the same types of realizations and results that we have on the front end. So you know it's going to continue to evolve.
Joel Scutchfield:I think, again, you're going to see. You know, I think the first place we're going to see it is just the less need for human intervention. You know, at the machine, interacting with the machines themselves and even interacting with the data analysis, having to again have a person go mine for that information, or even to the degree that they still have to today, which is much less than it was years ago. So, yeah, the evolution is going to continue. I don't think there's any doubt about that going to continue. I don't think there's any doubt about that. And again, I think it's going to be less people directly interfacing. Let's redeploy them to do things that we need people to be able to do, that the automation cannot.
Oshri Cohen:I think that if we look on this industry today, we see huge efforts being invested in fixing the problems after they find them. Moving forward five years from today, you will see most of the efforts being invested in preventing the problem. Efforts being invested in preventing the problem and ai I think that's the um, that's the big news. Ai bring to this industry the capability, the, finally the ability to prevent the problems before these are even occurs. And and you know, it's not just preventing by just dropping the bad components when you see bad components, you will have the ability to digest a huge amount of data that this industry generates every minute, every day, and get very smart decisions, a very smart conclusion based on this information. By doing this, this industry will definitely, for sure, will become better, will become better, and not become better because they will bring more people, not become better because they get more machines or they get more pieces of software. No, it will become better because the approach will change, since the capability is different now. It's much more developed, really Okay. So that's exactly where I think this industry is going to In order to achieve this for sure.
Oshri Cohen:I 100% agree with Joe CFX or all these type of communication between the different brands, different machine has to happen. If it will not happen, we're just losing here a huge advantage. You know, if the, if the pick and play sees something wrong and it it is incapable, not capable today to inform about this issue to the AOI, that's unbelievable. I mean, we are missing here a huge part of data that we already saw. We saw it in one place, why shouldn't we prevent it in a different place? Well, that's where this industry goes to for sure.
Philip Stoten:I think it's an exciting world, Oshri. I love the idea of moving from a reactive process to something much more proactive and much more preventative. The other thing I would throw into the mix because I know Eric's just about to close is the idea of agentic AI and the idea that within an agentic system, within an agentic system, there would be the ability to interrogate the cyborg AI, interrogate the co-young AI, interrogate the various different AIs and make sure that data is synthesized and provided how it's needed, where it's needed, in pretty much natural language models. I think that's part of the future that we're going to see in this industry, Eric.
Eric Miscoll:No, that was very good and I'm struck by, as you guys are talking too, just the requirement for collaboration across all of these companies, and we see that within the industry constantly. I know you, gentlemen, are involved in some of that yourselves, with all these partners. We all have our solutions, but in order to realize the potential, we really need to optimize the whole there, right, all of that across and that sharing of the data and the information and getting to exactly what Oshri is talking about there, and I think that paints a picture for a brighter future. Thank you, I like that so good. Well, again, gentlemen, thank you, this has been excellent. I want to encourage anybody listening again to please come to Apex, come and meet these gentlemen in person. They're much taller in person and personable, and I think you would benefit from the interaction and I think it'll be valuable time spent. So, thank you all for all of this. Appreciate all of your insights today. Thank you Very welcome. Enjoyed today. Thank you Very welcome, enjoyed it very much.