GSI Technology Inc (GSIT) 2020 Q2 法說會逐字稿

Ladies and gentlemen, thank you for standing by.

Welcome to the GSI Technology's Second Quarter Fiscal 2020 Results Conference Call.

(Operator Instructions)

Before we begin today's call, the company has requested that I read the following safe harbor statement.

The matters discussed in this conference call may include forward-looking statements regarding future events and the future performance of GSI Technology that involve risks and uncertainties that could cause actual results to differ materially from those anticipated.

These risks and uncertainties are described in the company's Form 10-K filed with the Securities and Exchange Commission.

Additionally, I've also been asked to advise you that this conference call is being recorded today, October 24, 2019, at the request of GSI Technology.

Hosting the call today is Lee-Lean Shu, the company's Chairman, President and Chief Executive Officer.

With him are Douglas Schirle, Chief Financial Officer; and Didier Lasserre, Vice President of Sales.

I would now like to turn the conference over to Mr. Shu.

Please go ahead, sir.

Good afternoon, everyone, and thank you for joining us to review our second quarter 2020 financial results.

Net revenue for the quarter were $11.7 million, within the range of our guidance provided on the first quarter call of $11.6 million to $12.6 million.

Gross margin in the second quarter was 56%, at the high end of our guidance, 54% to 56%, provided earlier last quarter.

Doug Schirle, our CFO, will provide additional comments on our financial results in his commentary.

This is an exciting time for GSI Technology.

In the second quarter, we successfully advanced our Gemini-1 chip on numerous fronts.

We are on track for upcoming customer training sessions in similarity search applications at our US headquarters.

We have also begun benchmarking our device against industry-standard benchmarks.

We are currently testing our device against the 1 million item database benchmark and will soon be testing against benchmarks with larger databases of 10 million, 100 million and 1 billion items, all of which we expect to complete before calendar year-end.

We are also preparing to run the large database benchmark on multiple CAST systems early this calendar year to demonstrate the scalability of our device.

The third silicon spin of Gemini-1 is expected back from TSMC later in the third fiscal quarter.

In addition, we are bringing out the production cart in preparation for customer demonstrations and are expecting the boards to be ready early next calendar.

Our AI team continues to develop the software libraries and algorithms for our expected release of Gemini-1 in calendar 2020.

We are very fortunate that our core business generates sufficient cash to support our new product road map and the funding of our ongoing R&D budget.

The focus of our R&D spending and the work by our engineering team is the Gemini road map, which include Gemini-1 and the Gemini-2.

All the preliminary work that we have done with our customers and our benchmarking shows that our device will increase data processing speed with very low operating power and a scalable architecture.

The most distinct feature of Gemini is the memory circuit design, combining GSI's core capability In-Memory design -- In-Memory chip design with our AI software expertise.

Gemini has impressive capability, including the program interface that allow multiple levels of custom programming, from assembly code up to TensorFlow Python.

We have made tremendous progress on our path to launch our new products while sustaining solid financial performance in our core business to fund this goal.

There are numerous exciting new market opportunities ahead for the company, and we are making key progress on attaining our goals.

I would like to thank my team for all their hard work and the commitment to our vision.

Now I will hand the call over to Didier, who will discuss our business performance in further detail.

Please go ahead, Didier.

Thank you, Lee-Lean.

Switching gears to our new product category.

We have made significant progress on our radiation-hardened SRAM.

The RadHard device continues to track completion of our internal qualifications by the end of this year.

RadHard sales, however, have been delayed.

The prime contractor awarded the program has made a decision to stay with the legacy design due to the aggressive time schedule of the program.

On the radiation-tolerant family, we are currently tracking 2 programs that we anticipate to close in the next 2 quarters.

Looking at our sales breakdown, in the second quarter of fiscal 2020, sales to Nokia were $5.3 million or 45.2% of revenue compared to $6.0 million or 46.6% of revenue a year ago and $6 million or 45.7% of revenue in the prior quarter.

Military and defense sales were 23.4% of second quarter shipments compared to 16.4% a year ago and 21.0% in the prior quarter.

SigmaQuad continues to be our best-performing product category, with sales of 63.5% of second quarter shipments compared to 66.4% last year and 67.9% in the preceding quarter.

I'd now like to hand the call over to Doug.

Doug, go ahead please.

Thank you, Didier.

For the second quarter of fiscal 2020, we reported a net loss of $1.8 million or $0.08 per diluted share on net revenues of $11.7 million compared to a net loss of $351,000 or $0.02 per diluted share on net revenues of $12.8 million for the second quarter of fiscal 2019; and a net loss of $125,000 or $0.01 per diluted share and net revenues of $13 million in the first quarter of fiscal 2020.

Gross margin was 55.9% compared to 62.6% in the prior year period and 63.3% from the prior quarter.

Total operating expenses for the second quarter of fiscal 2020 were $8.5 million compared to $8.4 million in the second quarter of fiscal 2019 and $8.5 million in the preceding quarter.

Research and development expenses were $5.8 million compared to $5.8 million in the prior year period and up from $5.6 million in the prior quarter.

Selling, general and administrative expenses were $2.7 million in the quarter ended September 30, 2019, compared to $2.7 million in the prior year quarter and down sequentially from $2.9 million in the previous quarter.

Second quarter fiscal 2020 operating loss was $1.9 million compared to $394,000 in the year ago period and $229,000 in the prior quarter.

The second quarter fiscal 2020 net loss included interest and other income of $210,000 and a tax provision of $55,000 compared to $145,000 in interest and other income and a tax provision of $102,000 for the same period a year ago.

In the preceding quarter, net loss included interest and other income of $147,000 and a tax provision of $43,000.

Total second quarter fiscal 2020 pretax stock-based compensation expense was $642,000 compared to $552,000 in the comparable quarter a year ago and $651,000 in the prior quarter.

At September 30, 2019, the company had $68.1 million in cash, cash equivalents and short-term investments and $4.8 million in long-term investments compared to $61.8 million in cash, cash equivalents and short-term investments of $9 million -- sorry, and $9 million in long-term investments at fiscal year ended March 31, 2019.

The upcoming third quarter of fiscal 2020, our current expectations are net revenues in the range of $9.7 million to $10.7 million, with gross margin of approximately 60% to 62%.

Operator, at this point, we'll open the call to Q&A.

(Operator Instructions) It appears there are no questions this time.

I'd like to turn the call back to you for additional remarks.

Pardon me, we do have a question from Ari Shusterman of Needham & Company.

So first I want to start off with the delay.

Can you provide a bit more color on this?

And yes, when -- what would be the time line now for the delivery of the product?

And how can we be sure that no more delays in the future?

I'm assuming you're talking about the radiation-hardened product?

Correct.

Correct.

Yes.

So there's multiple entities involved, and what I mean is the prime contractor who's ultimately responsible for the shipment or the delivery of the assets, made the decision.

So they -- the part of the program that we're in is actually with a subcontractor.

It's actually a -- also a prime contractor but in this case, it's a sub.

The sub was pushing for our solution because obviously, it gave the program a much higher performance.

At the end of the day, the ultimate prime contractor was up against the wall for a schedule.

And what they ended up doing was, any new products -- and it didn't affect just GSI, it affected other companies as well.

All new products that haven't been tested in space yet were not allowed in the system and so they went with an old legacy design.

With that said, what we're -- what we need to do is continue with our qualification, which I mentioned, will be completed at the end of this year.

And then we need to push for some heritage, which is getting into space so that when the next deployment of the program gets put in place, we can be part of it.

As far as the timing, unclear, but it would be sometime in calendar 2020.

Okay.

That makes sense.

When it comes to your cooperation with the Weizmann Institute, can you -- has that been segueing into something larger?

Or how has your cooperation with them been going?

Weizmann helped us to identify a bioinformatic as a viable market for us.

And we do demonstrate a -- we have covered all magnitude of improvement over what they were doing today, okay.

So what that -- that's helped us tremendous from our [semiconductor arc] capability.

And I think from the marketing point of view, we go to Weizmann that we can get in touch with a big biopharmaceutical company, and as part of our marketing campaign, to go into that market space.

That makes sense.

And then it comes to that expected contribution of the hardened product in 2020, what would you be -- just for modeling purposes, what would be a reasonable estimate, given the delay?

Hard to say.

We're trying to get a feel for that now because we're trying to understand when the next release would be.

So I don't want to put a number out there yet until I feel comfortable with it.

So I'm trying to get both a number and more of a time frame because this is all fairly new news.

So at this point, I'm going to hold off before I give you a number.

Okay.

Makes sense.

And one more question, just with regards to the APU, when it comes to modeling that for 2020, what else can you say?

Calendar year 2020.

No, I think the 2020, will be a most likely or will be our -- the benchmarking and all payment setting -- our capability to the market, okay.

So you are more like -- I think any volume, you will be also all in the assembling phases, okay?

I think the significant volume will come up more likely in the 2021 period.

So we've had customers that have already mentioned that they would like to buy the boards for the evaluation and for the testing internally.

So we will be selling individual boards for those purposes for basically trial kind of testing.

Hard to predict when the actual production will kick in.

Our next question comes from Kurt Caramanidis of Carl M. Hennig.

When will we get something?

And can you put some color on what you're seeing with your testing that you've got now?

You said by the end of the year, you'll have some more.

What are you seeing now on -- I'm talking on the APU.

And then, will we get a release in the next month or 2 on how that's tracking with your other testing?

Yes.

So I just -- I mentioned earlier, we are -- the present quarter, I think we can pretty much complete all the benchmark.

But right now, we're still expecting our fully functional silicon coming back.

So all the benchmark we are doing right now is we use our initial [silicon-rich steel package].

But even with the silicon, with the part, I think we can see a significant improvement.

I think, I will say in the tensile factor over the existing solution.

But for us to put the complete benchmark number, and I think we better to hold it back until we see that the real silicon to demonstrate our capability.

In other words, we don't want to publish a half-baked number, and then later on, we have to change the number again.

But anyway, in general, I definitely think we have a much better solution of a magnitude better than fiber solution earlier today.

All right.

And Kurt, and to add to that, besides the APU itself, we're are also putting them on boards.

And the first boards we're doing now is an off-the-shelf FPGA board, and we've created a daughter card for it, and that has limitations on performance on that off-the-shelf board.

We have designed our own custom board, which will be the one that we would use in production.

We do have samples of that production board in hand now, and we're going through a debug process.

So we also have to take that into consideration on the benchmarking what we release to the public because, as Lee-Lean said, we don't -- we want to make sure we show the best numbers.

The numbers on the off-the-shelf board, though the performance is limited because of the board, is still a very good improvement over current solutions.

But as far as what we're going to produce, we're going to decide internally, but we may hold off to release the full benchmarks based off of the production board.

So that wouldn't be this year?

It would be -- if it's not this year, it would spill into early next year.

Yes.

We still talk about late this year and early next year.

Right.

It's not very far away.

So are you testing your second silicon?

Or you're testing third silicon?

Is that why you're waiting?

Or you're using second silicon with the board that you're making, and that's kind of been a little bit of a delay?

No, there are 2 factors.

Well, to make this -- to get a more complete number picture, then we have a silicon, we have a CAST of both, and then we also have a software to write.

So right now, we are taking all [3 tests.] So I think the first report, we probably come out [would be from the second silicon] but with the production board, okay?

So that will be phase -- first phase of the data.

And then of course, a few months later, we'll come out with the first silicon.

The first silicon, we will get it back end of quarter, so we will probably -- more like end of next quarter, then we can get the data published.

Okay.

So when are you having your -- because there's a couple of things -- time lines -- when are you having the training at the office?

Because you're saying that January or so, you'll be having it for customers, the -- your board with the chip.

And then you're saying you're on schedule for similarity search training at the headquarters.

When is that going to take place?

So the training, we're going to be doing an internal run-through in the beginning of November.

And what we're targeting for the customers was either going to be the last week of November or the beginning of December.

It's dependent also on the folks that we're having in the first training.

So we'll be using obviously the off-the-shelf board for that, which is going to be fine for the training.

I mean again, for the benchmarking, we don't want to have the performance limitations of the off-the-shelf board, but for training and workshops, that board is fine.

So that you'll see, or will -- I'm not saying we'll see, but we'll have the workshops sometime, like I said, best case will be end of November but it's looking like the first or second week of December right now.

And how many parties are interest -- expressed interest in that?

I think it was a handful a few months ago, has that increased?

Or how is that looking?

Yes.

It's more than a handful.

So we'll -- the room that we're going to be doing the training in holds 32, and it'll probably have a mostly, capacity.

Okay.

And I guess I would go back in queue, but I'm not sure if there's anyone waiting.

So I apologize, if you want, I can do that.

But facial recognition is something recently that was on in the news, that there are some areas that it doesn't work very well.

Have you gotten any indication that your chip may be able to do better for facial recognition?

Detroit comes to mind, there was a hullabaloo with the Detroit police chief and the accuracy of the facial recognition software.

Do you have any commentary around that because I think that was an area you were looking at?

Or is it too early to tell?

So we do have a partner in that space that we're working with, kind of like a Weizmann for the cheminformatics.

And right now, what we've done is we actually haven't -- we've done a quick demo on our first silicon, which as you all know, is very buggy.

We haven't done a demo with them on the second silicon.

What we have done though, is actually used some of our algorithms on their current hardware platforms to show that not only do we have -- again the upcoming hardware, which we'll show them, but we also a very strong software solution as well.

So we have done a demo based off of our algorithms, which impressed them.

So we're at a point now where we're shortly going to be showing them a demo on the second silicon.

But we haven't gotten into the level of detail that you're talking about of how does that improve accuracy.

Certainly, they have an accuracy target and they have a performance target, and we certainly feel like we're going to hit both of them very easily.

Okay.

And then can you just talk about quickly Gemini-2?

So Gemini-1 is one board, one chip.

What would Gemini-2 do?

And then what's the magnitude difference?

Gemini-2, okay, first of all, we -- this is the second chip we are doing, so we have a lot of learning curve going along the way.

So we think we can be a -- the performance can be improved tremendously, probably more like 1 load of magnitude, and of course, we can do the better integration with Gemini-2 because -- no, no, the Gemini-1, we have a FPGA.

It's all interfaced, so Gemini-2, will integrate that.

So just another load of magnitude, the cost improvements.

So I think Gemini-2, from the cost -- peak performance cost point of view, you will be probably a couple load of magnitude better than the Gemini-1.

But of course, the schedule-wise, hopefully Gemini-1, we can establish the market position, and then the Gemini-2 is the much better cost performance to follow that.

Okay.

Would you say you've proved out concept?

I mean that's kind of what your goal was with Gemini-1.

Do you feel that you've proved the concept?

Or is it still too early that it's working?

I mean even though we think the Gemini-2 is a much better product, but the Gemini-1 compared to the current solution already in the market, we have this cover all the magnitude, better than the viable solution out there.

So even Gemini-1 itself is a very revolutionary product in our mind.

So hopefully, we can demonstrate the capability and then establish the Gemini road map among the customer.

So we have done some benchmarking, and again, it's on the limited board -- limited off-the-shelf FPGA board, so we're not releasing those numbers.

But the benchmarking does show that, yes, our technology and architecture is unique and certainly has much better performance.

And it's not just on a pure performance level but also on the power side.

The power side, we are much lower power, which is something that's also important to this market space.

So the answer is, yes, we do feel like we've proven that out.

We just haven't released those benchmarks yet, just as I mentioned because we're waiting to do it with our production board, which is not limited in performance.

Okay.

Fair enough.

Finally, will you be getting a website soon that matches the new technologies?

It's not really doing that at this point.

Do you have something in the works that can be more applicable to what you're actually doing?

I do know we're going through some upgrades to the website now.

I don't have all of those details, but there will be some changes coming.

Okay.

And certainly, it doesn't really do much for the AP -- for Gemini, which I think it certainly should.

I saw you got a new marketing person in your Board.

And certainly, there's lots of opportunity with the website for improvement to more -- show more about what -- where you're kind of going.

Our next question comes from Jeff Bernstein of Cowen.

Just wanted to get an update on your Nokia outlook.

So Nokia, this quarter will be down, which also explains why our quarter will be down.

We've received the forecast for them.

They usually give us a 12-month rolling forecast.

And for the next 2 quarters, it's really what's -- I guess solid.

So the next 2 quarters, which would be the calendar first and calendar second quarter of 2020, it looks like it's going back to the normal run rate.

So they have forecasted that this quarter would be down, and that will be returning to normal run rates in first and second quarter of next year.

Our next question comes from [Sharran Ty], private investor.

I have a question regarding the training.

So is it going to be only training in your office?

Or is it going to be placed like you can -- anyone else can play it online by using mainly, then you can have a solution on cloud or somewhere, anyone can play it?

How is it going to be?

Yes.

It will be both.

So the workshop will be a hands-on workshop, but we understand the limitations of attendance.

So I can't tell you if the first workshop will be through the cloud or not but certainly not too long down the road, we will be having training that will also be cloud-based.

And what about the software solution you're talking about?

Is it going to go open-source software?

Is it like -- how do we -- how it will be?

So the software, it's libraries and algorithms, specifically for a device and depending on application.

As Lee-Lean mentioned earlier, it can be from the very low level all the way through the Python.

So it's certainly -- it's more library and algorithm-based.

And then it's part of the workshop -- I'm sorry, so that'll be part of the workshop too, is to train folks on how they can write their own libraries specifically for the device.

You've mentioned that the software packages you have are performing well, even with the current -- I mean with the current hardware, right?

So are you going to be open sourcing the software?

Is it going to be solution from your side?

Because if I (inaudible)

Yes.

So the example I gave was an algorithm.

It wasn't the libraries.

So the libraries that we have are specific for the APU/Gemini.

So those cannot be used on a separate hardware platform.

The algorithm, however, could be.

And that's -- that was -- so that's a different -- it's not apples-to-apples.

As far as the software package, we'll have a package that's deliverable with the device.

But as I mentioned, it's not going to have every possible library, every possible algorithm but we can show folks how to write their own.

Okay.

What about going for the market?

How -- what is your future plan?

Is it going to be cloud-based solution, like for example, the [Native Glass or Azure]?

Or what is your market strategy?

It would be market- and customer-driven.

So there are going to be folks that are going to want an on-prem hardware solution and there are going to be folks that are going to be going through the cloud.

So it will be a combination depending on the customer and the market.

Our next question comes from Kurt Caramanidis of Carl M. Hennig, Inc.

Like you haven't had enough.

I forgot one.

Collaborations, do you see any collaborations with other larger companies?

Is it too early, meaning someone that has a large IT department and your hardware can work together?

And do you see any service revenue opportunities as you're kind of starting to engage?

So the answer is yes.

The collaboration that we're actually focusing on initially isn't probably what you're talking to.

So we are looking to do some collaboration more right now in the cheminformatics.

As we mentioned, Weizmann Institute, with their solution right now, they're using software packages specifically for this market.

So we're working with -- there's a couple of known folks in those markets that we're starting to work with now to make sure that we can kind of integrate and collaborate with them.

What you're talking about, I mean, we've started having internal discussions about that, but it's way too soon to talk about what collaboration we may do as far as the data center type of approach.

It appears there are no further questions at this time.

I'd like to turn the conference back to you for any additional remarks.

Thank you all for joining us.

We look forward to speaking with you again when we deliver our third quarter fiscal 2020 results.

Thank you.

This concludes today's call.

Thank you for your participation.

You may now disconnect.