GSI Technology, Inc. (GSIT) Earnings Call Transcript & Summary

Good afternoon. I'm Lee-Lean Shu, President, Chief Executive Officer and Chairman of the Board of GSI Technology, Inc. I will act as Chairman of the Annual Meeting of Stockholders and would now like to call the meeting to order. This meeting is being held pursuant to the notice of annual meeting mailed to all of our company's stockholders. Today's virtual-only annual meeting is a live audio webcast. Thank you very much to those who are participating in our virtual meeting today. Douglas Schirle, the company's Chief Financial Officer, has been appointed to act as Secretary of this meeting to record the minutes. Brooke Webb of Computershare has been appointed Inspector of Election for the meeting and is responsible for determining the exact number of shares present at the meeting. If you need a copy of the annual report and proxy statement, the links are provided on the meeting webcast page. I would now like to introduce the members of the Board of Directors and the officers of the company, who are attending today's annual meeting. I have Robert Yau, VP of Engineering and the Board member; Didier Lasserre, VP of Sales and the Investor Relationship; Douglas Schirle, our CFO, and also have a Board member, Jack Bradley, Art Whipple, and Kim Le together with me. And I believe Elizabeth Cholawsky also in the line with us. I would also like to introduce Cassie Hartogs, representing BDO USA, LLP the company's independent public firm. I will ask Mr. Schirle, Secretary of the meeting for his report. Mr. Schirle?

Mr. Chairman, the list of the holders of common stock of the company, close of business on July 7, 2021, which is the record date of this meeting, is available for examination to any stockholder present, any proxy holder representing a stockholder on the meeting webcast page. We also have a copy of the notice of this meeting, together with the declaration as to the mailing of a copy of the notice of each stockholder record at the close of business on July 7, 2021. Notice of meeting and the declaration of mailing are available for inspection by any stockholder present or by any proxy holder representing a stockholder on the meeting webcast. Only holders of common stock on the record date are entitled to vote at this meeting. I'm advised by the Inspector of Election that the holders of a total of 19,562,409 shares of the company's common stock are represented at this meeting in person or proxy. There were 24,166,000 shares of common stock of the company outstanding at the close of business on July 7, 2021, the record date. More than a majority of the outstanding common stock is represented here today in person or by proxy. Quorum is therefore present, and the meeting is authorized to transact business.

Thank you, Doug. At this time, we will consider the items of business on the agenda. Today's meeting agenda is posted on the meeting webcast page. We will strictly follow the meeting agenda in conducting this annual meeting. After formal business portion of the meeting has been concluded, we will adjourn and have a short presentation on the company's business and operation followed by a question-and-answer period. I call your attention to the rules of conduct provided for this meeting. These are available to each stockholder in the file section of the meeting webcast page. We ask that all stockholders review and abide by these rules of the annual meeting, including the question-and-answer portion of the annual meeting. Matters to be considered at the meeting today, which are further described in the proxy statement dated July 19, 2021, that was mailed to all stockholders of record with the notice of the meeting. First, the election of 8 directors to hold office for the ensuing year and until their respective successors are duly elected and qualified. The Board of Directors consists of 8 directors and the Board of Directors has nominated those persons set forth in the proxy statement for this meeting. Except for Barbara Nelson, all nominees currently serve on the Board of Directors. The Board's nominees are Jack A. Bradley, Elizabeth Cholawsky, Haydn Hsieh, Kim Le, Ruey L. Lu, Barbara Nelson, Lee-Lean Shu and Robert Yau. As it is present in the proxy statement, stockholders wishing to make nominations of directors for election at annual meeting must comply with the advanced notice requirements set forth in the company's bylaws. Since no notice of additional nominations was received by the deadline, the nominations are closed, 8 nominees receiving the highest number of votes will be elected. The other 3 matters to be considered at the meeting are, the proposal to ratify the appointment of BDO USA, LLP as the company's independent registered public accounting firm for the fiscal year ending March 31, 2022. Third proposal to approve an advisory non-binding resolution regarding the fiscal 2021 compensation of the executive officers, then in the summary compensation table included in the proxy statement for the annual meeting and a proposal to approve the amendment and the restatement of our 2016 Equity Incentive Plan. Each of these proposals is described in detail in the proxy statement. Approval of each of these proposals requires affirmative vote of a majority of the share present in person or by proxy and voting on the proposal. The proposal to approve the fiscal 2021 compensation of the company's named executive officer is a non-binding advisory for as described in the proxy statement. Are there any questions, comments that anyone would like to make on either of these proposals? It's 2:07 Pacific Time and the polls are now open. Let me remind you that if you have already sent in the proxy, there is no need for you to vote. If you have not voted or wish to change your vote, you may do so now by clicking on the link provided online. Votes cast during the meeting will not be reflected in the preliminary voting results announced during this meeting, but will be reflected in the final voting results that we will report on the Form 8-K filed with the Securities and Exchange Commission within 4 business days of this meeting. [Voting]

It's 2:08 Pacific Time and the online voting will now be closed. We will now hear the report of the secretary of the meeting regarding the results of the election. Mr. Schirle?

Mr. Chairman, based on the preliminary review of the votes cast, the Inspector of Election has indicated that each of the nominees for the Board of Directors received an affirmative vote of more than a majority of the shares voted, and thus each has been elected to serve until the next annual meeting with stockholders when their respective successors are duly elected and qualified. Proposal to ratify the appointment of BDO USA, LLP as the company's independent registered public accounting firm for the year ended March 31, 2022, received an affirmative vote of more than a majority of the shares represented at this meeting. Therefore, the proposal has been ratified. Advisory resolution to approve the fiscal 2021 compensation of executive officers has received the affirmative vote of more than a majority of the shares represented at this meeting, and therefore, the proposal has been approved. Proposal to approve the amendment and restated of our 2016 Equity Incentive Plan has received the affirmative vote of more than a majority of the shares represented at this meeting. Therefore, the proposal has been approved. Final voting results will be reported on a Form 8-K, which we will file with the Securities and Exchange Commission within 4 business days of this meeting.

I would like to thank all of you for your interest and attendance at this meeting. There are no other matters than have properly come before this meeting for consideration. This concludes the formal business of the meeting and the formal part of this meeting is now adjourned. We will now proceed with a brief presentation followed by a question-and-answer session. Before responding to any questions or comments, just a remainder it's possible that today's meeting, including some of our comments and some of our responses to your questions may include forward-looking statements. They are based on certain assumptions and are subject to a number of risks and uncertainties. The risks, uncertainties and assumptions that could affect these forward-looking statements, including risks that are included in the company's SEC reports, including our Form 10-K for the fiscal year ending March 31, 2021, and Form 10-Q for the first quarter of fiscal '22. To quantify all forward-looking statements by these cautionary statements, and except as required by law, we assume no responsibility for updating any forward-looking statements. We will start the accompanying presentation. Next page, please. Okay, as you can see, we -- the company are putting -- run out the main development is on the APU. And this page, we show you our potential customers and our collaborators, we are working with, including, for example, NASA, AWS, Space Micro and the Microsoft Research, Baidu, MAFAT, many, many companies working with us. Okay. Next page, please. Let me just rehash how the GPU -- APU. APU is short for the Associated Processor Unit. We have the uniqueness of a 2 million DRAM processor in the chip, and we compute by contact in the memory. They are total 1 peta bit operations per second, we compute. One peta means 10 to the 15, the millions of millions. It's probably the highest number of operations in all the processors today in the market, okay? And we have 220 tera bit per second internal I/O then coverage is also at the highest bandwidth you can find in the market. For the application, we are looking at when we compare the APU to the NVIDIA V100 for the signal processing application, we are up to a 6x power performance sector for single chip and up to 50x for the server. We will show the example of how we do this. The road map, right now, we have a 28-nanometer Gemini I currently in the market we are promoting. And we are coming out 16-nanometer version, the GEMINI II for the later 2022. And also, we are planning for the 5-nanometer Gemini III, 2 years from that, okay? And for this generation, now we believe we can achieve other magnitude improvement, the G2 to the G1 and the G3 to the G2. Next page, please. The next page. Now we just, okay, review what made it so good on our APU. The typical -- how computer working is you have the ALU, the central processing unit, and the working with the memory. The ALU when needed to compute, they will give an instruction to fetch the data to the ALU, okay? So that's the famous from lumen architecture. We have a fast HQ and the store. ALU apply by the Moore's Law double the performance every year pretty much, but the memory assets is really limited by the memory bus bandwidth. So right now, we have this so-called a memory war, okay? But no matter how fast your ALU can perform, I mean you have a data assets that bottleneck. Now, next page, please. How the APU working is we can compute -- do the computation directly on the memory already. Basically, we issue the command to open the war line at the same time. So the pulling operation is performed on the bit line directly. So by doing this, we eliminated the memory bus bottleneck. Next page, so you can see how we organize the internal architecture. Basically, we found we call MMB, Main Memory Block, which is the computation memory block, okay? We formed an array like a 2k by 24. That is one section. And we connect all these sections together, stitching together we form a bank, and we have closed couple of the memory IO Memory shared by also the memory block, okay? So by looking at this, and it's a pure like memory architecture is very tightly organized. And each bank, we have 32-K bit processor in it. And each column, we have a 16 bank and 50-K bit and for the whole chip, and you have a whole column together that we have 2 million bit processor in it. Now there's a chip architecture. You can see how this one does like this. If you just look at the image of this chip, it is almost like a memory chip, okay? But the way they operate is we do the culling operation, the bit line operation basically, they are a vector -- certain vectors, we saw one section of the bit line and another vector -- another section of the bit line put the results into another section of the bit line, so each bit line has performed just like a bit processor, and we have 2 million of these. Okay, next page please. Now I start talking about the application, okay? First thing I want to mention, we have got the MAFAT Challenge that's a Defense Force research department. We got the first place. Next page, please. Okay, the MAFAT Challenge basically is we use radar to detect the animal and the human. This started on July 15, 2020, and end October 15. They had over 1,000 contestants and over 4,000 submissions, and we are the #1. We have a 91% accuracy to recognize between the human and animal. Because of this, we're working with the MAFAT on the follow-on project. And actually, we developed many applications based on this effort. Okay. One of the great use for the APU is on the computer vision. And let me show you the use case and the benchmark. As you can see on this page, they are all computer vision application, which is currently developed by the APU application. And also, we are still working on many others. The first thing I want to talk about is SAR Back projection algorithm. SAR is the short for the Synthetic Aperture Radar, basically, you send a radar pulse from the high altitude, let's say, a satellite or airplane and then you send it to the ground and you have a reflective signal back to the satellite. And by capture this, you capture sending the signal and the reflect signal, you can get the image of the -- what you are trying to target. The next page. So we do the benchmark with APU against NVIDIA GPU. We -- for the calculation, we are much better, much faster than the APU versus GPU. We can see the APU Leda-S and the APU Leda-E. There's 2 versions of our APU Leda against the NVIDIA GPU Leda and also NVIDIA GPU V100 card and also NVIDIA GPU T4 card. There are 2 different type cards. As you can see our speed power, we are much faster than both and the power is much lower than low. So compared to the NVIDIA V100, we are almost 10x faster in terms of speed power, okay, against the NVIDIA T4 GPU we are 12.7x better. Just you're looking on the single card point of view. And the table under this is showing the performance on the server, because we have a much lower power, so we can put a lot more cost on the server. So if your server base, if you look at the speed, power and the form factor, then we are compared to the GPU-based server the V100 base server, we are 59x better. And compared to T4 base server, we are 29x better. So now we are working with the customer, the server-based application on this. Okay. Next one, I'd like to talk about is a dense registration. Basically, data application is you can have -- we form the image compared to your -- the fixed image and then you can register which one it belongs to, okay? So like the picture showed in the slide. That's an okay angle picture, and then we can -- you can -- by comparing to the library, you can find out which picture is this. So this is a very iterated calculation. So many, many -- you need to repeat many, many convolution calculation to do this, and APU is a great to perform this, okay? So basically, it's in-memory calculation. Next page, please. So again, when we compare the APU Leda card against the NVIDIA V100 and then NVIDIA T4, you can find now compared to NVIDIA V100, we run about the similar speed, but we are much, much lower power. And then compared to T4, we are both higher performance and lower power. So from the card point of view, speed and power, we are 2.7x better compared to V100 and 3.5x better compared to T4. So compared to NVIDIA, if you look at the form factor, and the speed and the power, our server can be 18.5x better and Leda onetime better compared to a T4. Okay. The next one I'd like to introduce is our cybersecurity. Okay, next page, please. It's great for APU to work on the SHA. SHA is short for the Security Hash Algorithm. This is the basic application is the cybersecurity, password cracking in the block chain. So we're also working now with a customer on this. And next page. And the other major application for the APU is on the fast vector search. The run there's one major platform, the Elasticsearch we're working on and OpenSearch, which is a derivative from the Elasticsearch. And right now, we are working with our -- one of our collaborator on this. So what it looks like is if a customer has a database in the cloud, let's say, AWS Cloud, which is the host that they operate on the OpenSearch and the user can send the queries to the AWS OpenSearch cloud. And it was like the user extending their query to the AWS OpenSearch. But inside OpenSearch, there is a GSI Elasticsearch popping and you will send a query to GSI server, and we have a GSI server in the cloud direct connect to the AWS Cloud and we do the computation to search for the query and then send back the result back to the AWS Cloud and back to the customer. So while we are doing this because of the OpenSearch or the Elasticsearch, they are not very good in terms of speed and power on the search performance and also they are very limited on the feature on the search. For example, they couldn't do very well. If you want to search the image and the text together, they cannot do very well. But with APU, we have a great feature. We can solve all this issue. So basically, what it does is the customer has -- next page, please. Customers have a 14-point database, which is unstructured data, and we can organize it into the binary vector and the performance search on the binary vector. And of course, search is using the KNN, that's the popular search algorithm. And we can do the batch search and also in the server, we can put multiple client in the server with the multiple data sets. Basically we embed result of top K basically, the closest result back to the customer. Okay? So that's pretty much all the application we are offering. And the final page is, -- next page, please. Yes, next page, yes. Just a quick introducing -- introduction of how we are offering from a hardware point of view. Basically, we -- right now, we can offer all the way from the chip to the PCIe card and even to the server, okay? So we are working with the customer with all these solutions, the processor, the application base even on the Software-as-a-Service platform. Okay. So, thank you. This concludes my presentation. Any questions? Okay. So that conclude our question-and-answer session. Thank you very much for joining us today. We are grateful for your interest and the continued support of GSI Technology.

This concludes the meeting. You may now disconnect.