Ladies and gentlemen, thank you for joining us. Welcome to GSI Technology's Second Quarter Fiscal 2022 Results Conference Call. Before we start, the company has requested that I share a safe harbor statement. The discussions in this call may include forward-looking statements about future events and GSI Technology's performance, which involve risks and uncertainties that could lead to actual results differing significantly from those expected. These risks and uncertainties are outlined in the company's Form 10-K filed with the Securities and Exchange Commission. I also want to inform you that this conference call is being recorded today, October 28, 2021, as requested by GSI Technology. Hosting this call is Lee-Lean Shu, the company's Chairman, President, and Chief Executive Officer. Joining him are Douglas Schirle, Chief Financial Officer, and Didier Lasserre, Vice President of Sales. I will now turn the call over to Mr. Shu. Please proceed.

Good afternoon, and thank you for joining us to review our fiscal second quarter 2022 financial results. Our revenue increased by 17% year-over-year for the second quarter of fiscal 2022 to $7.8 million compared to $6.7 million in the second quarter of fiscal 2021. At the high end of the guidance provided earlier in the second quarter through the first half of fiscal 2022, revenue is up 25% compared to fiscal 2021, and sales to Nokia, our largest SRAM customer, have stabilized despite ongoing supply chain challenges affecting our industry. We reduced our net loss by 22% year-over-year due to higher revenue and gross profit, alongside only a modest increase in operating expenses on a year-to-date basis. At quarter end, we had nearly $51 million in cash, cash equivalents, and short-term investments to support the launch of new products and sales and marketing efforts to build a pipeline of opportunities. Our legacy SRAM business is profitable and generates cash flow to support the development and impending launch of our radiation-tolerant devices and the Gemini APU solutions. We continue to advance our key initiatives with ongoing APU and Rad-Tolerant opportunities. We are making progress in current customer engagements for both categories and have had several new beta customer engagements for the APU. We remain focused on allocating our capital towards the APU, which we believe is central to the company's long-term value, considering the unique opportunities presented by our technology. We are committed to our strategic investment in APU. Based on our detailed analysis of market data, we estimate that the global Total Addressable Market for APU search applications is approximately $137 billion. Later in the call, Didier will discuss our assumptions and the market represented to provide more detail about our conviction regarding this strategic investment. We were recently selected to be an organizer and judge for the building scale approximate nearest labor search challenge, the first competition of its kind in large scale approximate U.S. labor search on ANS. The event is hosted by NeurIPS as part of its annual conference on neural information processing systems. GSI is one of the members of a panel led by Microsoft that includes AI thought leaders from both industry and academia. Submissions from participating teams will be evaluated against challenged datasets containing at least 1 billion vector vehicles. The winning teams will be recognized at the NeurIPS 2021 conference in December. Recent advances in A&S techniques for search, recommendations, and ranking require support for handling all large datasets. This competition aims to help establish a consensus on the effectiveness of algorithms at this scale. ANS is crucial for search, recommendation, and ranking. A challenge for ANS algorithm designers is to create a data structure that allows rapid retrieval over the KDS level, even as the database size increases. The tradeoff between accuracy, often referred to as recall, and search latency, typically called credit throughput, is significant in evaluating ANS performance. Several implementations of large-scale ANS are now powering enterprise-grade, mission-critical, and scalable search applications. In these scenarios, benchmarks such as costs, preprocessing time, and power consumption are just as essential as the legal versus latency tradeoff. I have previously noted the APU's advantage in power consumption from our relationship with our Rad-Hard to crypto miners who operate their own printed hardware. We all recognize the importance of obtaining power-efficient hardware to reduce energy costs. The issue of power consumption is critical today regarding how much energy we need to generate power for our machines. Some notable statistics include that global data centers consumed an estimated 205 terawatt hours in 2018, accounting for 1% of global electricity use. The energy used by data centers approximately doubles every four years, indicating that they have the fastest-growing carbon footprint of any sector within IT. This problem is likely to escalate as workloads become increasingly data-intensive and AI-focused, necessitating the development of new hardware and chipsets specifically designed for power efficiency as a crucial element in future data center design. Addressing this challenge is one of the strategic opportunities for our technology. GSI can demonstrate in various applications that our technology provides substantial power savings since our typical system demands a significantly smaller footprint. We have lowered the overall total cost of ownership. Two of the evaluation criteria in the building scale approximate VL search challenge competition will compare participants based on power usage and hardware costs. Like the other organizers and panelists, GSI will submit to the competition. Naturally, we aim to achieve a leading outcome for our submission and gain increased visibility among competitors, including Nvidia, Microsoft, Intel, and other prominent AI hardware vendors. We mentioned in our last call that we are partners in OpenSearch 1.0 launched by AWS with our Elasticsearch k-NN plugin. We are still in the initial stages of developing this opportunity. We have a functioning server in our Silicon Valley data center where we conducted an initial customer demo. Currently, the release of the new OpenSearch 2.0 protocol has been delayed, which has postponed our launch timing. We expect to conduct our next series of demos once the new version of OpenSearch is available. Our goal is to open our facility to beta customers in calendar 2022. During this quarter, we successfully raised GSI's profile in the industry through our participation in the ANS competition. We are pursuing opportunities that have emerged following our recent success in prior competitions with the APU and the announcement of the radiation-tolerant NASA project. The GSI team has been working diligently to elevate our visibility, aiming to attract data customers and build a business pipeline. I sincerely appreciate your support as a fellow GSI shareholder. Now I will hand the call over to Didier, who will further discuss our business performance. Please proceed, Didier.

Thank you, Lee-Lean. As mentioned, we have pinpointed the market segments that are most pertinent to the APU and outlined our total available market, or TAM, along with our serviceable available market, or SAM. For APU search applications, we estimate the global TAM to be $137 billion in 2021, with a projected compound annual growth rate of around 20%, reaching $287 billion by 2025. Our SAM stands at $4.5 billion in 2021, growing to $10 billion by 2025. The various search market segments included in our analysis are computer vision, synthetic aperture radar, drug discovery, cybersecurity, and service markets such as NoSQL, Elasticsearch, and OpenSearch, which we plan to support with a SaaS solution. We are actively engaging with customers in these areas. Computer vision encompasses image and object recognition, facial and body recognition, warehouse robotics, automatic target recognition, and dense registration, which helps align a photo with a previous image for mapping purposes. We are currently focused on opportunities in facial recognition, object identification, SAR, and reidentification. Reidentification involves sorting faces or objects not present in the database, enabling recognition and tracking of patterns related to where and when an image is captured. Regarding the supply chain constraints our industry faces, similar to many semiconductor companies, we have a one-year lead time on substrates for chip assembly, and supply remains constrained. One of our substrate suppliers experienced a fire, but they are now operating at 70% capacity, which is an improvement. A subcontractor in Taiwan shut down for four weeks in June due to a COVID outbreak in their factory, but they have returned to full operations. Concerning the recently announced TSMC wafer price increases, we are expecting a 20% rise in wafer costs, which will take effect for all new orders. The overall impact of rising assembly labor costs, increasing substrate prices, higher wafer costs, and expedite charges will lead us to raise prices for all our customers. In the second quarter of fiscal 2022, sales to Nokia amounted to $1.9 million, representing 23.8% of net revenues, down from $3.4 million or 51.7% in the same period last year, and $3.8 million or 42.7% in the previous quarter. Military defense sales constituted 27.4% of second-quarter shipments, compared to 26.9% a year ago and 20.0% in the prior quarter. SigmaQuad sales were at 52.4% of second-quarter shipments, versus 65.4% in the second quarter of fiscal '21 and 63.6% in the prior quarter. I will now pass the call to Doug. Doug, please proceed.

Thank you, Didier. We reported a net loss of $4.6 million or $0.19 per diluted share, and net revenues of $7.8 million in the second quarter of fiscal 2022 compared to a net loss of $5.2 million or $0.22 per diluted share, and net revenues of $6.7 million for the second quarter of fiscal 2021. And the net loss of $4.2 million or $0.17 per diluted share, and net revenues of $8.8 million for the quarter of fiscal '22. Gross margin was 53.6% compared to 46.7% in the prior year period and 54.4% in the preceding first quarter. The changes in gross margin were primarily due to changes in product mix sold in the three periods. Total operating expenses in the second quarter of fiscal 2022 were $8.7 million compared to $8.3 million in the second quarter of fiscal 2021 and $9.1 million in the prior quarter. Research and development expenses were $5.9 million compared to $5.7 million in the prior year period and $6.1 million in the prior quarter. Selling, general and administrative expenses were $2.8 million in the quarter ended September 30, 2021, compared to $2.6 million in the prior year quarter and $3 million in the previous quarter. The second quarter fiscal 2022 operating loss was $4.5 million compared to $5.2 million in the prior year period and $4.4 million in the prior quarter. The second quarter fiscal 2022 net loss included interest income and other expense net of $8,000 and a tax provision of $42,000 compared to interest income and other expense net of $16,000 and a tax provision of $62,000 for the same period a year ago. In the preceding first quarter, net loss included interest and other expense of $20,000 and a tax benefit of $172,000. Total second quarter pretax stock-based compensation expense was $716,000 compared to $653,000 in the comparable period a year ago and $823,000 in the prior quarter. At September 30, 2021, the company had $50.7 million in cash, cash equivalents, and short-term investments and $2.8 million in long-term investments compared to $54 million in cash, cash equivalents, and short-term investments and $5.8 million in long-term investments at March 31, 2021. Working capital was $53.6 million as of September 30, 2021, versus $56 million at March 31, 2021, with no debt. Stockholders' equity as of September 30, 2021, was $69.9 million compared to $75.6 million as of the fiscal year ended March 31, 2021. We still see the supply chain constraints having a modest impact on our ability to fill all of our orders, but there has been some improvement. The supply chain situation remains fluid, and we do not expect significant improvements from these constraints before next year. Given these variables, current expectations for the upcoming third quarter are net revenues in the range of $7.2 million to $8.2 million, with gross margin of approximately 52% to 54%. Operator, at this point, we'll open the call to Q&A.

Our first question comes from Rajvindra Gill with Needham & Company.

This is actually Denis asking a few questions for Raji. So I'll just start off with a question about the price increases. When do you expect that the price increases that you're passing on to your customers that will start to show up in the top line?

So we are changing backlog and all new orders as of December 1.

All new orders, understood. And my other question is about the Gemini-I APU. A few calls back, you mentioned that the Gemini-I APU was expected to enter volume production this quarter. How is that progressing? Are you starting production, or have supply chain issues hindered that? Can you provide an update on the Gemini-I volumes?

Yes. At the component level, we are currently in preproduction mode, which means we are still qualifying the components. This is happening prior to any demand being observed. Right now, our main challenge lies with the software. We need to make a significant amount of software APIs ready for our customers, so software will be the primary obstacle for hardware availability.

So you're saying that once the software is ready, that will kind of get you one step closer to the actual production of the Gemini-I?

Yes. From the hardware point of view, yes, we are ready. We just need to win the application with software.

Our next question comes from the line of Jeff Bernstein with Cowen.

Congratulations on the new patent, you guys were just issued on the SRAM structure optimized for in-memory computing. And from that patent, you can see there's not a large instruction set here and it's going to be important for you guys to provide compiler and software libraries that make the chip easier for people to use. So can you just give us some detail on what the deliverables are there and the timing for those?

Currently, much of our efforts are focused on raising awareness. We have performed well over the past ten months compared to existing solutions in the industry. However, we still have a lot of work ahead to develop the market and build customer trust. As we’ve mentioned previously, we are actively participating in the competition and striving to win awards, all of which contribute to our brand visibility. We hope to showcase our capabilities to both our peers and customers, and this is our primary focus at the moment.

And Jeff, to add to that, the compiler stack is essential for the deliverables because, as Lee-Lean mentioned, we are currently developing many of the APIs and algorithms, but many people want to create their own, and we can't keep up with all the different applications. Therefore, the compiler stack is crucial. We will be releasing it to our beta clients, and those clients have already been identified. We initially aimed to do this by the end of the year, but it seems that it may now occur in early next year for the beta clients. They will test it thoroughly and try to find any issues. Following that, we plan to make it available to the general public sometime in 2022, hopefully in the first half of the year.

Got you. I’m assuming that for certain types of applications that are somewhat universal, multiple customers might be interested, like the synthetic aperture radar analysis or object recognition, that you are developing libraries to simplify things for users. Is that correct? Could you also discuss when each of the vertical markets you mentioned earlier will be delivered?

Correct. So you have both libraries and you also have further up, the stack, the actual APIs or the algorithms. And so as you mentioned that has either been done or is being done for things like the SAR, the synthetic aperture radar. You know that we've written it also for the BIOVIA pipeline pilot platform. It's already been done. We've done it for facial recognition and object detection. We are in the process of doing it for the dense registration, and we've done some work on reidentification as well. So there has been a lot of work that's been done already.

Got you. So is it fair to say that by the middle of next year, some of these key verticals will basically be covered and very accessible to customers who want to start using them?

If they want to use our algorithms, correct. Yes. And like I said, some of the entities we deal with, getting external software is a problem. And so the answer is, for folks where it's not a problem, those should be available for folks that want to write their own, that's where we're relying on the compiler stack.

Okay. Got you. And then just I want to understand a little bit more about the environment that you're facing out there in terms of sales and understand that you guys are in the model execution, not training part of the market. But what we see in the training market, AWS just did an Elasticsearch instance based on the Intel Habana chips. So they're kind of open to outsiders. They have their own Trainium custom chips. It looks like Microsoft has Graphcore, Groq, and Nimbix as training chip partners that they're working with, where some of the others like Baidu, Alibaba, and Google are only doing their own. Talk about the not invented here? Or who is open to looking at outside chips among those big players out there that could be important partners?

Sure. So you bring up a good point. Almost all of the large big data guys have internal chip development happening, and they've always had that. But with that said, certainly, if you bring a solution that's going to accelerate a search, or it's going to lower the power, or it's going to do something that's important to them, they're open to it. I mean, we haven't seen anybody that basically said, no, we're designing our own. As you said, a lot of the chips these guys are doing are for very specific functions that are important to them. A lot of it is centered around the training, as you mentioned. We haven't seen that level of activity for specific search, which is what we're focused on.

Got you. But your feeling is that delivering these kinds of capabilities to some of these large players, they will be open to looking at these and possibly offering instances on our networks?

Correct. I mean, and we've seen that already play out with the OpenSearch. Again, we've demoed on 1.0, and we're waiting for 2.0, but they've already announced us as a partner on that. So yes, they're open.

Got you. Okay. And then I just wanted to ask on the non-Nokia customers, 48% of revenue, that's the highest since before the pandemic. Any particular customers that are growing outside of military? Or what's sort of the status of non-Nokia customers?

Yes. In reviewing the numbers, Nokia experienced a decline that was greater than our overall revenue decline quarter-over-quarter. However, we saw growth in other areas. There wasn't just one specific area of strength; notably, the military sector showed some positive results. We fulfilled the final part of our Rad-Tolerant order, having shipped half in the June quarter and the other half in the September quarter. Additionally, we noticed a rebound among some of our customers in the automotive equipment and test equipment sectors. The automotive industry faced challenges earlier this year due to chip shortages, which led to a slowdown in the first half, but conditions appear to be improving. As a result, equipment companies have returned to us this past quarter.

Got you. That's great. And then just a request. Jim Ramel, the largest shareholder, asked you guys to get an IP valuation, which makes sense to me. I think we're now your second biggest non-index holder. And just kind of back of the envelope, it looks like you're building that you own at Elko Drive is worth over $10 million potentially. And I think it would be worthwhile to have that appraised and have shareholders understand that there is some additional asset here that's significant beyond the cash that you guys have.

That's true. We believe that the building is obviously worth a lot more than what's on the books. We've been in it for over 10 years now, of course.

Our next question comes from the line of Brett Reiss with Janney Montgomery Scott.

Have any crypto miners, in fact, approached you in possibly using the Gemini system?

We have considered crypto mining and currently, there are specialized chips designed for that purpose. Our APU is not specifically intended for crypto mining. While we've explored the idea, we don't believe our technology is well-suited for it. However, it could potentially serve general purposes, but it is not practical for generating revenue through crypto mining.

Right. Right. Now for the ongoing customer engagements to vest into material revenues for the company, is the additional engineering that must be done in the control of our company? Or is it additional engineering that must be done by the potential customers' engineering teams to make the Gemini system a commercially viable product for us?

So it depends, right? The answer is yes to both, right? I mean, we certainly need to continue to develop more algorithms, more libraries, more APIs. But then we also need to make that compiler stack available so that they can do it themselves. Because, again, there are way too many potential applications going forward for us to keep up and there are some people who just don't want to outsource that. So they'll do it themselves. And supplying them that compiler stack will allow them not to have to basically program at the register level. I mean, they'll be able to do it on a much higher language. And so the answer is, yes, we will need to do some of it and that will be done internally, and then we need to enable our customers to do their own as well.

Our next question comes from the line of George Gaspar, a Private Investor.

Just ongoing with the last question. Can you give us a very specific example of the testing on a specific entry into the market that you're running against competition? Can you give us something that would identify with your capacity being much superior to the others out there?

We are currently working with the customer and are preparing to conduct a proof of concept specifically for the synthetic aperture radar. We reached this point by performing benchmarking for them on a 5 kilometer by 5 kilometer image, requiring a resolution of 0.5 meters in 1 millisecond. We completed the benchmarking on CPUs, GPUs, and our APU, the Gemini, and presented all the test data and results to them. Ultimately, we were selected, and we are now defining the proof of concept with them. This application is a clear example of how they evaluated data from the three different chip technologies and opted for GSI.

Okay. Interesting. Continuing on the defense side, there seems to be an ongoing need for space work to identify and interpret higher and faster speeds. Is there any possibility of expanding your activities in the defense military sector given your capacity compared to others?

Yes, definitely. Many of the applications we've discussed, such as SAR, object detection, and automatic target recognition, are military-focused. We have also worked on projects like signal classification in that area. There are numerous sectors and applications we've explored for the defense market. As you know, we partnered with Space Micro and received a NASA SBIR grant for what NASA refers to as an inference processing unit, which is essentially a ruggedized board that utilizes our technology for space use. Some time ago, we conducted initial SEL testing on the APU, and the results were excellent. We plan to follow up on that testing soon, hopefully in the first half of 2022, and conduct SEU and SEFI testing to complete everything necessary for deploying the Gemini in space. We are definitely committed to continuing our efforts in that direction.

That's very interesting. I got one question on the issuance of stock in the last quarter. How many shares have been issued? And was that for issuance to employees basically for the increase in shares? Can you elaborate on this aspect?

I don't have the actual number here in front of me, George, but it's somewhere between 100,000 and 200,000 shares.

Between 100,000 to 200,000 shares?

That must be somewhere about 100,000 to 200,000 shares.

I see. So those...

And then since the beginning of the fiscal year, you have some ESPP purchases and option exercises. In fact, just this past week or so, we had two directors exercise some options. And I believe there are Form 4s out there for each of those.

Okay. All right. And at this point, being that the stock has been lower for some time here, has there been any purchasing of shares in the marketplace to recover shares that are outstanding?

We haven't done any repurchases for over a year now, if that's what you're asking.

Our next question comes from John Fichthorn with Dialectic Capital.

I guess as a quick follow-up to the last question, I've got to ask with $50 million of cash and a $10 million building and a bright outlook, maybe why you wouldn't be doing share buybacks right now since it was asked?

Well, we've had a long history of buying back shares. We went public in March of 2007 and netted $30 million in the offering. And we've already repurchased over $61 million worth of our stock. So we've already taken, in terms of dollars, 2x off the market based on what the IPO was. And we have cash in the bank, but we understand that the APU is a very exciting product. There is a very large market that we're addressing with that. But there's still work to do. And we just feel that we want to make sure that we're going to be successful. And yes, the stock is cheap. We all believe that. But we think that at this point, given we've already done in terms of stock repurchases, it makes sense to keep the cash at this point and make sure that we're successful with our development efforts.

It makes sense, and it leads me to my second question about your go-to-market strategy. You've mentioned a proof of concept and the competitions you're entering. You've also pointed out some potential customers or even acquirers who may be in-house competitors. I don't fully understand the current go-to-market strategy, especially since it seems to have been pushed to Q1 or Q2 of next year. If you're able to share any details or clarity on this, I would appreciate it.

We have several potential applications, with the overarching theme being search. This is a broad term, but various applications fall under it. Our focus has been on identifying where we can provide the most value in different markets and which customers find our solutions most beneficial. We've been analyzing market size and segments to demonstrate our value to customers. The benefits we offer vary, including lower power consumption, faster query responses, and cost savings. Additionally, we are in the early stages of discussing how our solutions can provide mobility. In the past, certain functions required large CPU or GPU server setups, but now our solution can compactly fit into a couple of racks, making it transportable in planes, vans, submarines, and similar environments. It's important for customers to realize the potential mobility that can come with our solutions.

Maybe I should elaborate a bit on your exciting applications and the significant total addressable market. When do you anticipate being further along in development and seeing initial revenue? When can we expect a customer to sign a contract indicating they want to pay for your solution? What specific application or applications do you think you are closest to achieving this? Are you currently at the proof of concept stage, or have you progressed further?

I would say it's in the government and military sectors. So areas like the SAR application, some of the object detections that would probably be where we're furthest down the line. I mean, obviously, you know that we're going to be working this OpenSearch 2.0, and that's going to kick in sometime in 2022, but it will be instances. So it will take a little longer for that revenue. It's a service. It's a SaaS model. So that will take a little longer to kick in. Some of it will kick in, in 2022. The question is how much is not, and we don't know yet, obviously.

But before the OpenSearch 2.0, will you have revenue from one of these defense-type customers this year, even if it's just NRE?

It’s possible. The timing is difficult to predict with the military sector. We should have had a board or two purchased by now, and it might happen by the end of the year or possibly early next year. Predicting timing with these clients is not always straightforward. However, we are close to securing at least a couple of boards. To clarify, we are not discussing volume at this stage. This is still in the proof of concept phase where they will buy a board, test it, and confirm that it meets their needs before proceeding further. So, it will likely be a few units at a time.

And so in terms of being like production revenue, like real recurring either SaaS revenue or something else, is that really 2023?

Certainly not earlier than the second half of next year. I mean, yes, I don't see any way second half of next year would be the earliest. But yes, it could fall into 2023. Again, hard to really predict exactly when that's going to happen.

I completely understand, but can you tell me how soon it could happen if everything goes smoothly? It seems like the second half of next year is the earliest. To clarify, in the Defense Department sector, the competition is different because you're not up against other products made within the department. However, in many commercial sectors, you are. So, could you explain whether your strategy is mainly about winning contracts and gaining recognition, especially considering companies like Google, Microsoft, and Baidu may still push their own solutions? Or do you have a sales team actively reaching out? I'd like to understand your specific commercial go-to-market strategy.

No, we're in discussion with these parties. As I mentioned earlier, they are all involved in chip designs. I can't recall if it was Jeff who pointed this out, but we haven't noticed anything specifically related to the search aspect. Generally, it's more about functionality. The companies we're in conversation with include OpenSearch, which is part of AWS. As you mentioned, AWS or Amazon is developing their own chips. So, why have they selected us as a partner? Clearly, they aren't pursuing anything internal that competes with our offerings. While we do face competition from internal designs, particularly among chip manufacturers, we haven't observed significant efforts from these companies in the search domain regarding what we bring to the table.

The large-scale database search is still a very new market. AWS OpenSearch has only recently launched. We are seeing a new level of competition in bidding and search, which has never been encountered in the industry before. This is just the beginning. Our Gemini APU is currently the best available hardware for this type of work, and we are optimistic about our performance in this competition. We believe this will highlight our value in this sector.

I am concerned that this is very new, especially since you're about to reach version 2.0 and we might see revenue increase in late 2023. I'm unsure about your backup plan if we find ourselves in this situation at the end of 2023, discussing a potential timeline pushed to the second half of 2024. It's still early, and people are evaluating. What options do you have if we reach the end of 2024 and look ahead to 2025? Is this the only path, or are there alternatives?

No, no. We are more than that. We are looking at search as one area of focus. Additionally, we discuss our defense and government projects, particularly the NASA initiative aimed at developing a device suitable for space. This development allows the prime contractor to integrate the device into satellites, applicable to numerous companies and applications. So it's more than just search. We offer object recognition, sensor anomaly detection, and sensor registration, along with various computer vision capabilities. We are hopeful, although we are uncertain about when production revenue will materialize; nonetheless, we anticipate it will come soon and we believe that.

At this time, we have reached the end of the question-and-answer session.

Thank you all for joining us. We look forward to speaking with you again when we report our third quarter fiscal 2022 results. Thank you.

This concludes today's conference. You may disconnect your lines at this time. Thank you, and have a great day.