PDF Solutions, Inc. (NASDAQ:PDFS) Q2 2024 Earnings Call Transcript

PDF Solutions, Inc. (NASDAQ:PDFS) Q2 2024 Earnings Call Transcript August 11, 2024

Operator: Good day, everyone, and welcome to the PDF Solutions, Inc. Conference Call to discuss its Financial Results for the Second Quarter Conference Call ending Sunday, June 30, 2024. At this time, all participants are in a listen-only mode. After the speakers’ presentation, there will be a question-and-answer session. [Operator Instructions] As a reminder, this conference is being recorded. If you have not yet received a copy of the corresponding press release, it has been posted to PDF’s website at www.pdf.com. Some of the statements that will be made in the course of this conference are forward-looking, including statements regarding PDF’s future financial results and performance, growth rates and demand for its solutions.

PDF’s actual results could differ materially. You could refer to the section entitled Risk Factors on pages 16 through 36 of PDF’s Annual Report on Form 10-K for the fiscal year ended December 31, 2023 and similar disclosures in subsequent SEC filings. The forward-looking statements and risks stated in this conference call are based on information available to PDF today. PDF assumes no obligation to update them. Now I’d like to introduce John Kibarian, PDF’s President and Chief Executive Officer; and Adnan Raza, PDF’s Chief Financial Officer. Mr. Kibarian, please go ahead, sir.

John Kibarian: Thank you for joining us on today’s call. If you’ve not already seen our earnings press release and management report for the second quarter, please go to the Investors section of our website where each has been posted. Before Adnan discusses the financials in detail, I have some comments to make about our observations for the second quarter and our view for the market for the remainder of the year. Our bookings in the second quarter were lower than the strong Q1. Due to the nature of some of the larger contracts, we expect lumpiness in any given quarter and therefore, find it meaningful to look at a rolling average. Since SaaS bookings started improving in Q4 of last year, we have been building backlog, which will support our future growth.

The bookings in the quarter are mostly with customers that are either starting to deploy new systems like Sapience Manufacturing Hub and MLOps or expanding the usage of our platform. In both cases, we anticipate many of these contracts to lead to expansion business in the future. Notable deals in the quarter include a large contract for Exensio Process Control for an advanced logic fab, a contract for initial deployment of Sapience Manufacturing Hub for a large logic manufacturer who’s doing a significant SAP S/4HANA deployment. Successful completion of its initial phase is expected to result in a follow-on, much larger, more significant, multiyear license for SMH, tying all their manufacturing to ERP system to facilitate new levels of productivity.

That same customer, having already deployed Exensio and advanced packaging is also entering into a contract with us in the quarter to pilot Exensio for wafer fab analytics. We closed our first contract for MLOps, an AI-based product we announced in Q4 of last year. This contract is for a large fabless customer that is beginning their journey to deploy AI for testing of products. We anticipate successful application of AI for this use, will result in their expanding the use of AI for most tests. A number of customers also expanded Exensio cloud usage. While increasing the annual run rate of these contracts, these expansions also set up for larger renewals, some of which we anticipate occurring in the next few quarters. Finally, bookings for Symmetrix connectivity run-time licenses showed modest improvements in Q2 over Q1 as our customers’ equipment shipments increased.

Overall, given the strong backlog and business model where most of our revenue is typically ratably recognized, we continue to deliver strong results in revenue and earnings. We were pleased with the business results in the quarter as it demonstrates the strength of our business model. Turning to DFI. As we stated before, we have two machines at one customer and another machine at a second customer. A third has the rights to send us wafers this year for us to analyze on the eProbe machine in our facility while they build their new fab. The machine will be shipped to them when the fab is ready. For the first two customers, usage in Q2 was very high. What is clear is that the direct scan application of the eProbe has very unique capabilities that we believe are valuable in bringing up logic product yields and eventually control production of those products.

In both accounts, we’ve begun discussions about expanding the number of machines. We anticipate those discussions may take the next couple of quarters to conclude. Now let me turn to discuss our view on the environment and our perspective on the second half of the year. As we talk with our customers about their business, some are experiencing weakness while others are growing. As a result, we believe that for the overall semiconductor market, growth will be unevenly distributed. It won’t be the case that a rising tide lifts all boats. With that said, our engagement with customers remains high driven by fabs developing advanced logic processes such as 2-nanometer; fabless customers deploying advanced test control software, often with AI/ML to augment conventional test methodologies; and companies engaged in digital transformations, attempting to leverage data, whether that is IDMs, fabless foundries and equipment vendors.

Given these trends and strong customer engagement, we continue to expect revenue growth for the second half of the year to be about 20% over the same period a year ago. I want to thank all the PDF employees and contractors for their efforts during the first half of the year. Now I’ll turn the call over to Adnan who will review the financials and provide his perspective on our results.

Adnan Raza: Thank you, John. Good afternoon, everyone, and good to speak with you all again today. We’re pleased to review the financial results of the second quarter and to bring you up to date on the progress of the business. Our Form 10-Q has also been filed with the SEC today. Please note that all of the financial results we discuss in today’s call will be on a non-GAAP basis, and a reconciliation to GAAP financials is provided in the materials on our website. For Q2, our total revenue was $41.7 million, essentially flat versus the same period a year ago and up slightly versus the prior quarter. Analytics revenue was up 3% to $38.1 million this quarter versus $37.1 million for the second quarter of 2023, and represented 91% of total revenues this quarter.

The growth in our analytics revenue came from increased usage and upsized renewals by our Exensio customers as well as an uptick in our Symmetrix run-time licenses. As John said, we are excited about the level of engagement with our customers during the quarter, including Exensio adoption by a leading-edge fab customer, expansion of Exensio deployment by multiple merchant semiconductor customers, extension with a cloud provider for their internal use of Exensio and an additional win on the Sapience Manufacturing Hub with our partner, SAP. We’re also pleased with the engagement activity for our DFI system and eProbe machine, and you will see us investing further to continue to address the market needs. During the second quarter, revenue contribution from Integrated Yield Ramp was $3.5 million, down $0.9 million or 21% compared to the same quarter a year ago, driven by lower gain share from our Asian customers as a result of the low volumes.

We’re pleased with our backlog, which grew in the first half of this year from $229.8 million at the end of December 23 to $243.2 million at the end of this quarter. The trends John and I have been discussing and the level of customer engagement leads us to believe we will grow our backlog in the second half of the year as well. We reported gross margins of 75% for the quarter, up versus both 72% for the last quarter and 74% for the same quarter of the prior year. We are pleased with our gross margin performance for the quarter, which is in line with the long-term target financial model we shared at our Analyst Day and User Conference last year. On the operating expense side, our expenses for the quarter were slightly down versus the prior quarter driven by better utilization of our head count resources, primarily in R&D, while SG&A expense was essentially flat compared to the prior quarter.

For EPS, we reported a profit of $0.18 for the quarter, improving from the $0.15 we reported for the prior quarter. We ended the quarter with cash and cash equivalents of $118 million compared to $123 million for the prior quarter. We generated a small operating cash flow for the quarter. During the quarter, we used cash primarily for investment to support the development of our DFI system to address the market need and the build of additional machines we mentioned earlier. As we look to the rest of the year, we remain committed to our prior guidance for the year, with revenue growth returning to our 20% long-term target for the second half of the year compared to the matching prior year period. With that, let me turn the call over to the operator for Q&A.

Operator: Thank you, Mr. Raza. [Operator Instructions] Our first question comes from the line of Blair Abernethy with Rosenblatt Securities. Your line is open.

Q&A Session

Blair Abernethy: Thanks. Nice quarter guys.

John Kibarian: Thank you, Blair.

Blair Abernethy: Just two questions, I guess, for me. Just one on the DFI and the other one on the MLOps system. So on the DFI, you talked at length about it last quarter. I just want to see what you see as having advanced or changed over the quarter. And what is your manufacturing plan looking like? I see that CapEx is stepping up. So give us some sense of the ramp-up in capacity to be able to deliver equipment on the DFI. And then on the MLOps system, can you give us some sense of how big are these transactions? It seems like from the time of the introduction of the product, the actual sale is a pretty short selling cycle. Just maybe talk a little bit about that.

John Kibarian: Sure. So on the eProbe, what’s developed in the quarter. As I said in my prepared remarks, Blair, usage was very high at both customers. And we have a pretty good customer base in the fabless community, too. And so we started hearing from the fabless community about how they’ve been seeing the results. Two, the machine is very good at understanding the relationship between the design and the manufacturing yields and most inspection tools just understand that you compare one inspection result to another to find problems, just understands where it is on the design. Therefore, the manufacturer can talk to the designers about what specific design capabilities they’re seeing, et cetera. I think that’s probably why they’ve communicated with the customer base about it.

So that’s always good when you hear it from our customers’ customer. And as we’ve had dialogues, I think we’re starting to get some understanding or indication from at least the first two customers about what would be the potential number of machines they need. We don’t think in the short term, we can meet both of their needs. It will probably spread out throughout all of 2025 if they were both to come in what they say. And there’s a lot of ifs there, so let’s not get too far ahead of ourselves. But as you could see from our CapEx build-out, trying to mitigate as much as possible contingencies around being able to respond to demand. The typical lead time for machines like this is quite long. So we’re trying to do our best to pull it in, but you can pull it in a matter of a couple of months.

You can’t go and make this an instant lead time. So I think the lead times to those are not much different than the lead times for other machines of similar complexity, which is nine months to a year kind of like time frame. So we are very excited about where we are right now. We do think, besides the opportunities with these customers, we think it speaks more broadly. We’ve had a lot of incremental requests in memory and other areas around pilots because of this kind of unique capability of the machine to kind of know exactly where it is in the design and apply appropriate stimulus on each part of the design. So that’s kind of my answer on the DFI eProbe. If I move on to the MLOps, again, we’re seeing with customers, as they’re doing more and more advanced packaging, there’s many more test insertions and they want to be able to see models extracted from earlier test results up into later tests to get better performance, better test time, better quality screening, et cetera.

And this has been the first application for this. This is a customer we started working with. It seems very short from announcement, but we started working with this customer before the announcement as we were getting closed. It was kind of an early lead customer. The initial deployment is on a relatively small number of testers. That’s why I said it’s really only a small portion of their total production. And we believe, as you would roll it out across a larger number of testers, as they would apply to more and more products, it could be a relatively meaningful contract. At this stage, it is a modest contract. It’s not super huge. But it is, again, I would say, what would amount to being a couple percent of their test of a very complex product they want to prove this methodology works out for.

And I expect them to roll it out more as they gain success with AI at test.

Blair Abernethy: Okay. Great. Thanks for the color, John. And maybe just Adnan, just following on the DFI comments. What should we expect in CapEx going forward? I know you’ve got lots of cash. But $5.3 million this quarter, typically, you’ve been running $2 million to $3 million a quarter. What should we be looking for there?

Adnan Raza: Yes, a very reasonable question. To be honest, when we were ramping up DFI many quarters ago, we were also spending more than what we have in the quarters before this one. So look, I mean, for the next couple of quarters, at least, we think CapEx probably stays at similar levels. As John said, this is a pretty unique time for us with the engagement with multiple customers on leading edge, and we want to be making sure that we are well positioned to take advantage of that opportunity. So probably similar to where we saw Q2 come in is a fair estimate. I think the thing to keep in mind also is, look, we have been operating cash flow positive for a long, long time, and we intend to stay that way as well, and in between the other uses of cash we’ll do from time-to-time as the opportunities present or going to relate to, for example, share purchases, which we have done, for example, in Q1.

Blair Abernethy: Okay, great. Thanks very much guys.

Operator: Thank you. Please stand by for our next question. Our next question comes from the line of Gus Richard with Northland Capital Markets. Your line is open.

Gus Richard: Yes, thanks for taking my questions. Nice quarter. It sounds like you’ve got an exciting outlook. I was just wondering, on the DFI and your engagement with your two customers, what is the use case? Is it yield ramp? Is it bringing up a new product? Or is it actually in fab in production?

John Kibarian: Yes. So far, because it’s been applied to very advanced nodes, Gus, it’s been used for bringing up nodes and bringing up specific products. So each product uses the process a little differently, the design layouts are different. We used to refer to these things in the industry as systematics or something unique about the design. But now the process windows are so tight. There’s always something specific about every design. So they’ve been, I think across a couple of customers, using it on many different designs as they come into a node. We believe that even though the initial use is there, you tape out many designs in production for a long time. So it’s going to give you a first kind of use level even in a production fab, not just a development fab.

And we seem to hear that from the customer base. And over time, we believe it results in a control application as well just because the marginalities, the process windows are so tight, the need to monitor will remain. So early on, I think what’s been happening over the last year, or I’d say six months, has been really around bringing up products. I think as you transition into 2025, I suspect that it ends up being not just to bring up the products but to control the technology.

Gus Richard: Got it. And then I’ll just stick with DFI. Can you add a little more color to your build plans for next year? And I’m assuming that the CapEx, because you effectively lease these products, the spending on building the tools and the various assemblies and with your sub-cons is actual CapEx?

John Kibarian: Yes. So build plans, we are ramping up to be able to build more. I think somewhere between four and eight year would be a reasonable assumption about what we would do in the short term, i.e., on a four-quarter basis. We believe our suppliers have the ability to do more than that, be able to build more, but I think this is maybe our next waypoint. In terms of — you’re right that the machines have been provided on a subscription basis. So it is CapEx for us. Over time, I think we may adjust the business model with customers at some point and the machine may be purchased and some software subscribed. On top of that, I think there’s still some flexibility on both the customers’ mind and our mind about what’s the optimal way.

It kind of gets back to your first question, Gus, if you’ve got to run it forever, for the life of the node, would they prefer to purchase the equipment and subscribe to software. That may be the case. So our customers are a factor of hundreds to thousands times larger than us. So we’re going to listen to them about what’s the best way to work with them in the business. But I think over time, the capital may come off our balance sheet if they were to purchase them, particularly if the control application is there.

Gus Richard: Got it. And then the last one for me is on the MLOps. Clearly, there’s an expansion of chiplets and multi-die packaging, if you will. And I was just wondering, are those two related? And if so, are you starting to see increased interest into OSATs?

John Kibarian: They are related. The interest is coming from the product companies, the know-how about chiplet matching is really the responsibility of the fabless companies or the product group because they own the test program and they know how to interpret that, how to build the model to interpret that result, effectively. And then know how to, let’s say, reduce test time, downstream or add additional tests or match chiplets better, so you get an overall system performance that’s better. This is one of the motivations for this customer as well as others that we’ve dialogued with. It is also having us go back and look at our DEX network. We’ve made an investment in having our machines at the OSATs connected to their testers, so they can push data from the cloud.

And MLOps really allows them to manage all of that data traffic up and down their manufacturing flow in order to initiate running of models, let’s say, features for upstream tests to be ready downstream, so they can combine that extracted feature from the upstream test with the testing that’s going on real time to make, let’s say a bidding decision or a sub-bidding decision, et cetera. So yes, you’re right that it is very much related to chiplets and complexities on testing as a result of that. The OSATs have an important role to play because you need to integrate with their MES systems. Customers need to be integrated with their SAP system because they need to know where the chips are going and therefore, where the data needs to be sent. So it does bring up the overall system requirements that are needed.

But the buyer and the user of it is really still the fabless community more than the OSAT today. I think it will stay that way, too, Gus.

Gus Richard: Okay, all right. That’s it for me. Thanks so much.

Operator: Thank you. [Operator Instructions] Please stand by for our next question. Our next question comes from the line of Christian Schwab with Craig-Hallum. Your line is open.

Christian Schwab: Thanks for taking my question. So can you give us an update? You’ve been working for some time on a meaningful semiconductor producer who is now going through a tremendous mess, for lack of any other description. Can you kind of give us an update of what the revenue opportunity over a multiyear time frame could be with that customer now that dates in production could be readjusted?

John Kibarian: Yes. So Christian, thank you for the question. We’re always very respectful of our customers’ proprietary information and what’s going on in any given customers. So we don’t comment on specific customers per se. But I can tell you in general, though, right, our technology is used to help customers be more efficient and more effective. So we always are mindful of the economic situation every customer is going through and think about what’s the best way to work with them. But often, our technology is very important for customers to drive transformation. And so we look at these as opportunities often for us and the customer to be more effective in how they use our systems. And often, it builds a larger business with us over the longer term sometimes.

You don’t know if the short term it does do. And then that’s true for every case when we’re in these situations. In general, our technology is very important for being much more efficient in manufacturing. And I think our track record of being instrumental for customers in change management is quite long.

Christian Schwab: So put another way, do you think this opens up an expansion of opportunity where things could happen faster than previously expected then?

John Kibarian: Customers that are trying to move to advanced nodes, I think the complexity of the technology is opening up opportunities. When customers are really looking for that to happen now and often when customers are going through transformations, they’re looking for that to happen now, I believe our systems are increasingly valuable for those customers. So we look for ways to be able to deliver value, mindful of the fact that when customers are challenged economically, we also have to sharpen our pencils and think of how to be flexible as well.

Christian Schwab: Fantastic, great. No other questions. Thank you.

Operator: Thank you. Please stand by for our next question. Our next question comes from the line of William Jellison with D.A. Davidson. Your line is open.

William Jellison: Good afternoon and thanks for taking the question. I wanted to start out by asking, amongst your existing Exensio customer base, what you’re seeing with respect to trends, adopting the next incremental module for them, what are you seeing amongst those folks?

John Kibarian: Yes. That’s a great question. Thank you for the question. In my prepared remarks, I talked about drivers for the business, and two of them really kind of point to the drivers that we see at customers. First of all, what I talked about early was more automation on tests. That MLOps opportunity is one that I described a lot of times. As I said in my prepared remarks, it’s really being a much more sophisticated test and applying — going from applying rules to applying models, usually ML-based models or AI models, to be ready for chiplet production, more efficient in driving quality, et cetera. The second I labeled broadly in my prepared remarks is digital transformation. And we really see basically that’s really driving two aspects of our business.

The SMH, the Sapience Manufacturing Hub, which has really got the Exensio database inside it. It’s the way of connecting our partners at [indiscernible] from the shop floor to the top floor. When customers want to transform their business, they want to be able to act on whatever AI/ML they apply. Our customers that have been deploying ML models, one of the things they recognize is, I need to know from my ERP system where is this wafer going to go, if I need to send the data downstream to an OSAT who’s going to be testing the package test. So that connection to the ERP system is important for the financial team to be able to get real information and more predictability and better understanding on their economics, but it’s also important for the engineering operations team in order to be able to add more automation typically with AI and their production flows.

So that second opportunity is that SMH piece. We see quite a big opportunity for SMH. And then thirdly, I said that a lot of customers were expanding their cloud offering. One of the things that some of the customers have talked to us about as they begin this journey, the first thing they realize, you need to get all your data in one place. You need to be able to have that data aligned up and down the manufacturing supply chain. Because if you want to apply ML models, you need to — most 80% of your time is often wrangling data and getting it put together. And if you do that with scripts and munging in the typical data science way, it’s hard to then put it online. And it becomes very dependent on the engineer or person that built the model.

If you have first orchestrated your data, sometimes people refer to this as a data lake or a data lake warehouse or different words for it, then you have a way of then building the AI and ML. We’ve had the expansion contracts that we talked about in this quarter where customers are really more and more relying on Exensio to provide more functionality there in terms of orchestrating and managing data. The MLOps product takes advantage of that really and makes that much easier. And so it’s like these three things where we see, I’d say, the majority of our customer base really working one, okay, playing out the test; two, okay, I need to orchestrate my engineering operations with my financial operations; and then three, the foundation there, before I can do any of that, I better have all my data for engineering and manufacturing organized in a central location with a common API, so I can build any of these things.

And those are really the three places that we see opportunity with customers.

William Jellison: Great. Thank you. And then as a follow-up, with respect to DFI, is it still the case that on the revenue generation side of the machines that it tends to scale over time. From the moment you ship a machine, the revenue generation starts out very small. And as the activity scales on that machine over time, it increases. Is that still the way you view it?

Adnan Raza: Yes, I think look, I mean, one thing to keep in mind is our DFI engagements aren’t just ever about just the tool itself. So it’s a combination of software, and it’s a combination of the hardware. And even within the software, there’s many pieces of software, obviously. We’ve talked about the Fire software, for example, which is used to inform our tool about the analytics that will be performed based on the design. And then there’s obviously the other analytics software. So depending on the usage of those, some pieces may be accelerated. For example, if the machine is in the early stage, perhaps that one gets accelerated. We have talked about the lease treatment of the machine that can happen with some contracts. However, look, I mean, on a longer-term, our goal always is that the customer’s usage of the whole system grows over time.

And that is why some of the past contracts that they’ve been done have been on a token basis, such that we expect the customer to be utilizing. And hopefully, we are providing them more value over the time and therefore growing that opportunity.

William Jellison: Right. Thank you.

Operator: Thank you. Please standby for our next question. Our next question comes from the line of Andrew Wiener with Samjo Management.

Andrew Wiener: John, you referenced in the beginning of the call the — I guess, two lead DFIs, I guess you referred to them as customers. But I think one of those two is actually, it’s a manufacturing evaluation, and they’re not most currently a DFI customer per se. But given what you described as sort of conversations about potentially deploying multiple additional machines, is it fair to say that that evaluation is going well and that our sort of confidence level that that will convert into a paying DFI customer has increased?

John Kibarian: That would be correct, Andrew. We shipped the machine at the end of last year, came out beginning of this year. Second quarter was a very heavy usage period, as I said in my prepared remarks. And we saw good results there. They did as well. I think that really speaks to just the value the machine can create. We have a lot of hurdles we still have to get over in both customers. So we’re not by any means, able to just sit back here. There’s a lot of work to be done. But we’ve gotten very positive feedback from that customer. And yes, our confidence is increasing.

Andrew Wiener: Okay. And then I just wanted to clarify, when you said a couple of people asked about your capacity and you’ve referenced a comment about not being able to support, and I realize these are conversations, not orders yet, but sort of support what the two lead customers, I’ll call it, evaluation on a customer would need. Is that meaning within sort of the current four decades, that would be too little? And are you contemplating ways to be able to produce more than eight? I’m just sort of trying to — I realize this isn’t a revenue forecast or an order forecast, but I’m just trying to get a sense of sort of what the demand on these couple of customers are talking about and what you’re talking about from a perspective of what type of capacity you might want to have in ’25?

John Kibarian: Yes. So they’ve both given us ranges. And if it was both from the low end of the range, then I think we would be, okay. If they were in the middle of the range, we’re probably a little short. And if on the higher end of the range, we probably have a bigger issue. And we’re also mindful of the fact that we have customers that are asking for evals and potential demo machines, et cetera, that as we see a number of interesting applications in memory and in other logic manufacturers. So we need to have some slack in that capacity. If we just did everything that was good for them, then maybe we don’t have a way to expand our business in 2026. So we have to be somewhat thoughtful about kind of making sure we’ve got ability to expand out in the marketplace.

So even under kind of their modest or the low end of their requirements where I think we’d be pretty, okay. We’ve also got to look a little bit at where we would be in terms of being able to do manufacturing evaluation, et cetera. We’ve kind of also squeezed ourselves internally to support the three customers we’re supporting right now. And so we don’t really have much capacity as we would like necessarily internally for demos with other customers and some other things right now. So we’re a little bit hamstrung. So yes, we are looking to see what we could do to increase that number. We will make that decision as we get through this year to see how things kind of build out with the customer base. One of them could decide they don’t want to do it, right, so we’ve got to keep a lot of contingencies there.

Andrew Wiener: Has there been any conversations with the third customer that you haven’t shipped the tool to yet, but I guess you’re running wafers internally, as to what their demand could look like or would look like? I mean I know in the past, you’ve talked about sort of most customers who are going to be using it in any real volume would likely want at least a second tool just for redundancy purposes?

John Kibarian: We’ve not had very many conversations with them about that yet, Andrew. It is something we need to do as we kind of think about 2025 and beyond. So it will be something we’ll do in the second half of this year. We’ve been pretty busy in the second quarter, just in early third quarter, just kind of understanding the first two.

Andrew Wiener: And then for some of those other applications you’re talking about, whether it’s other logic players or memory, are they actually currently shipping you any sort of wafer so that you can run internally to demonstrate capabilities? Or is it more, right now, sort of technical conversations?

John Kibarian: Yes. We have at least one customer that I know off the top of my head that’s already shipped those wafers and the memory applications. We have others that are interested in doing that. We are mindful of bandwidth, right? So we’re trying to kind of swap them in a way that we don’t take our eye off the lead customers. But yes, we’ve already gotten memory wafers and starting to show them results.

Andrew Wiener: Okay. And then maybe shifting gears over to MLOps. I know that the first area of focus was the test application. And you had a number of pilots. Just maybe give us a little more color on sort of how you see — are you taking on more pilots now or are you waiting for more pilots to convert to like commercial engagements and then going to use that as sort of proof points to go out to other potential customers. And then maybe the third piece of that is, I think you touched on the idea of getting all the data in one place, and then being able to build machine learning or AI-based applications, for which MLOps is the enabler. Is there any consideration to making efforts to put MLOps sort of in the hands of customers and let them sort of try to figure out how to best use it and develop their own use cases?

John Kibarian: Yes, it’s a great question, Andrew. So I’d kind of go, if I can keep them all, the three-part question in my head. So yes, there are other pilots ongoing with customers and parts of the industry. We are also taking a step back and looking and saying, okay, how could we make it better? What can we do that would make things more effective for customers? And there’s really two parts to MLOps. There’s the operations around orchestrating your data so you can build models. And then there’s managing it once you’re out in the field. The biggest challenge I guess — one of the bigger challenges for that is in the test world, because customers use one company to do wafer sort test, a different company to do package test, a third company to do kind of a card-level test.

And many of our fabless customers are effectively becoming kind of system companies, because they’re making entire cards or even systems at this point. So managing that data flow, once you have a model, monitoring the production, the test production results, and so they can put in rules when they want to trigger a model update or change, et cetera, that’s really the problem MLOps solves for them. In both those problems, we think the second one is the stickier one over the long term. And we want to — look, you take a step back and say, okay, how can we enhance that, how do we make that better for customers. So there’s an activity going on there. And then the third question you asked around putting it in the hands of the customer, the whole intention on MLOps is exactly that.

They can use our environment and build their own models. They can use the stuff that we provide kind of default, example models. They can use a different system. They could orchestrate the data at Exensio, pull it out from the APIs, use a different system from anybody to build their own models and then publish that model back through MLOps throughout their manufacturing flow. So they don’t even need to use our learning environment if they don’t want to. And we design with those three levels of flexibility in mind, because the market really has all three of those types of engineers out there. They’re the really early adopters that have built their own flows for doing model building, but they don’t want to manage the day-to-day, 24/7, make sure all the systems are up and you can get a model anywhere you need to, wherever there’s a test showing up.

That’s not something you want to take expensive data scientists and spending your money on. You want software systems to manage that for you. And that’s what MLOps does. There’s customers that are, let’s say, earlier in their journey and they’re happy to use our environment, but they want to build their own models, because they’ve got a know-how about their products. They’ve got experience in model building, et cetera. That’s kind of like using a little bit more of the system. And the third case is, okay, PDF has an already defined pipeline for test time reduction or quality screening, let me just tune that — let me take that default model and tune it to my application. And those three capabilities exist in the product today. I mean that’s something you’ll see us enhance.

If you have attended SEMICON, at our booth, we had some smaller start-up companies that are doing ML talk at our user on our booth, because we are also looking at how we can work with the broader community. We’re not trying to have the corner on modeling at all. We want to help lots of people in the world bring models to production.

Andrew Wiener: And I guess maybe my last question is, have you made any progress on any of the battery pilots? You made [indiscernible] position in that space.

John Kibarian: Yes. So we’ve been working with battery manufacturers and also battery consumers kicked off a pilot or kicking of a pilot this month. I think actually, it’s in another week or so. Similar to what we did with the eProbe, customers sent us a sample material. We showed them in our lab what the image pipeline could do, how much information they could get about their battery cathode and anode, and now we’re installing that software and system at a manufacturer. So both the manufacturer and the consumer of that battery can work together to identify production control. So that pilot will kick off now. It’s an achievement that we’ve gotten from the kind of stage where it’s really just material in our lab, a very small amount of material, as you can imagine, in our lab to a production environment where there’s noise and there’s — it’s not this vibration and you’re running meters per second of cathode and anode films and proving that the software can keep up with it at that level.

So we’re, I think, at the next milestone right now. We’re very excited about it.

Andrew Wiener: Okay. Great. Thank you.

Operator: Thank you. [Operator Instructions]. At this time, there are no more questions. Ladies and gentlemen, this concludes the program. Thank you for joining us today. Have a wonderful day.