eMemory Technology Inc. (3529) Earnings Call Transcript & Summary

Good afternoon and welcome to eMemory's Third Quarter 2023 Webcast Investor Conference. Joining us today is our Chairman, Dr. Charles Hsu; President, Mr. Michael Ho; Head of IR, Ms. Li-Jeng Chen; and Director of the Finance Department, Ms. Teresa Kuo. The format of today's event will be as follows: First, eMemory's Chairman, Dr. Charles Hsu will give an opening remark. Afterwards, President, Mr. Michael Ho, will summarize our operations in the third quarter of 2023, followed by our business outlook. Next, Dr. Charles Hsu will give a talk titled OTP for SRAM repair and high-performance computing. Then we will go into the Q&A session where our management team will be ready to answer your questions. [Operator Instructions] As a reminder, this conference is being recorded and a webcast replay will be available after the conference is finished. For more information, please visit our website under the Investor Relations section. As usual, before we begin, we would like to remind everyone that today's presentation may contain forward-looking statements subject to the risk factors associated with the semiconductor and IP business. Please refer to the cautionary statement on Page 3 of today's presentation. Now I would like to give the floor over to eMemory's Chairman, Dr. Charles Hsu.

Good afternoon, everyone and thank you for attending our conference today. With a strong demand for licensing and accumulated new tape-out entering the mass production stage to contribute royalty, we are moving into the next multiyear growth cycle. Starting next year, we will see significant growth across all our technologies. For NeoBit 8-inch applications, the inventory adjustment phase has come to an end. We have seen customers placing wafers order one after another. The growth of 12-inch NeoFuse royalty will also be accelerated as more than 70 tape-outs in 16, 12 and 7 nanometer move into production. Our MTP related technologies, which we have been cultivating for more than 20 years, are now making their way into DDR5 and other mainstream applications. We expect licensing and the royalties to see significant growth in the coming years. In addition, our PUF-based security-related technologies have increased demands in various applications, particularly our PUF-based root of trust is adopted in Arm's confidential computing for mobile, auto and edge computing. PUF-based security will be a strong growth driver and we are very confident in our future growth. Next, I invite our President, Michael Ho to share our third quarter performance and the future outlook.

Okay. Thank you. Good afternoon, everyone. Now let's begin with our 2023 third quarter financial results. The third quarter revenue was TWD 787 million, up 13% sequentially but down 0.4% year-over-year. Operating expenses were TWD 370 million, up 12.8% sequentially, mainly attributable to the licensing in the -- to the increase in bonuses and rewards. Operating income was TWD 417 million with an increase of 13.1% sequentially but decreasing 6.3% year-over-year. Operating margin increased by 0.1 percentage point sequentially but decreased by 3.3 percentage points year-over-year to 53%. Net income was TWD 406 million, up by 15.4% sequentially but down 0.1% year-over-year. EPS for the quarter was TWD 5.44 and ROE was 57.2%. Next, let's move on to the revenue contribution by licensing and royalty. Licensing in the third quarter accounted for 32.9% of the total revenue, up 3.8% sequentially and 79.2% year-over-year. Royalties in the third quarter contributed 67.1% of the total revenue, increasing 18.1% sequentially but decreasing 18.3% year-over-year. Total revenue for the third quarter increased by 13% compared to the previous quarter but decreased by 0.4% compared to the previous year. With that, I will comment on our revenue contribution by specific IPs. NeoBit accounted for 27.4% of the total licensing revenue in the third quarter, increasing 25.5% sequentially and up 87.4% year-over-year. Its royalties accounted for 22.1% of the total royalty, down 14.5% sequentially and down 50.1% year-over-year. NeoFuse accounted for 38.4% of total licensing revenue in the third quarter, down 12.3% sequentially but up 48% year-over-year. In terms of total royalty revenue, NeoFuse royalties increased by 32.5% sequentially and 2.7% year-over-year, accounting for 75.6% of total royalties. PUF-based security IPs accounted -- contributed to 22.1% of licensing revenue, increasing 108.7% sequentially and increasing 189.4% year-over-year. Its royalties accounted for less than 1% of total royalties, down 91.5% compared to the previous quarter and down 97.8% compared to the previous year. MTP technology accounted for 12.1% of the total licensing revenue, down 39.9% sequentially but up 59% year-over-year. Royalty from MTP increased 39% sequentially but down 44% year-over-year, accounted for 2.3% of total royalties. Now let's look at the royalties for 8-inch and 12-inch wafers. For 8-inch wafers accounted for 36.3% of royalties, down 3.8% sequentially and down 36.5% year-over-year. For 12-inch wafers contributed 63.7% of royalties increasing 35.7% sequentially but decreasing 2.3% year-over-year. In total, 139 product tape-outs were completed in the third quarter. We will provide more information in the management report. In the next section, I will address our future outlook. We expect our revenue in the fourth quarter to grow further than in the third quarter. For licensing revenue, strong licensing demand will continue to drive the growth momentum of licensing fees. For royalty revenues, we expect the royalties to regain the growth momentum as the new process and the applications enter production. Moving on to the new IP technology and the business development. One, our development of a special process such as the HV, HK, BCD, embedded flash and emerging memory like RRAM and MRAM, continue to move into more advanced nodes, driving more applications and sustained growth in royalty revenue per wafer. Two, security-related applications are the focus of our development in advanced process. In addition to licensing to major foundries, we expect to complete the licensing to U.S. foundries this quarter. This concludes my comments. Next, I will pass the time to Charles. Thank you.

Okay. So next, I'm going to talk about the OTP for SRAM repair in high-performance computing. As you may know, our NeoFuse OTP has been used in 2x and 1x nanometer DRAM for the repairing functions. This quarter, I will be introducing a new topic, NeoFuse OTP has expanded its application into SRAM repair for high-performance computing chips. In high-performance computing, in order to achieve the high-performance computation, high-density SRAM cache memory is necessary. As the SRAM memory density is getting higher, the SRAM yield will decrease due to more defects in larger SRAM block. Instead of disposing the faulty bit, it makes more sense to repair it. So today, I will talk about how is the SRAM repaired and why NeoFuse OTP is the best solution for it. High-performance computing are the processors that performed at high speed applications using HPC increased computing capabilities and they require more SRAM memory and the increase in density. For example, smart sensing, automotive driving assistance, voice recognition and large language models, are applications that use HPC, including CPU, DPU, GPU, AI accelerator and more. And these applications require technology in advanced nodes and high-density embedded SRAM to deal with the faster computing and the larger training models. As more multi-core, high-performance CPUs are used for data processing, more SRAM is needed for cache memory. And the application used in HPC also have a larger chip area. So there's a higher chunks of manufacturing defects, low yield and the low reliability. Therefore, SRAM repair in advanced nodes is a must. SRAM repair is basically replacing bad memory bits with redundant memory bits. Redundant memory bits are actual bits reserved for the bad memory bits. The redundancy repair scheme illustrated on the slide is adding spare memory columns or rows to replace a bad memory bits. Usually during the testing phase after fabrication, the location of the bad memory bits is stored in the switch box, which is typically located in an OTP or in eFuse. The row and the column of the bad memory bits are switched out with the row or column containing the repair memory bits. And this diagram here shows a memory array on the left-hand side and the bad memory cells are marketed with a red X and redundant memory bits are located on the right. The location of the column containing the bad memory bits will be saved in the switch box. So the specific column can be identified and the redundant column can repair by replacing the bad memory bits. To repair SRAM, we need to store the location of the bad memory cells. If you consider the area cost effect, yield, reliability and the flexibility, then NeoFuse OTP is a better solution for storing a switch box as compared to a eFuse. NeoFuse OTP has a smaller cell size than eFuse. For high-density SRAM repair, the requirement for OTP density is also higher. The advantage of cost saving by using OTP is superior to eFuse since the size of OTP block is much smaller than eFuse block. Besides, the programming scheme of eFuse is to burnout the fuse. The yield for eFuse is much worse than OTP. Other than size and the yield, reliability is also an issue. Aside from OTP's excellent production record, eMemory's OTP has also passed rigorous testing to guarantee data retention for at least 10 years and which is not always the case for eFuse. For flexibility, OTP is also much more flexible than eFuse since it has better coverage for different form factors of memory, such as 3D or multichip packages and they can guarantee higher yield during the production. When it comes to SRAM repair, it is important to know that more advanced process node will more likely have memory defects due to the smaller feature size, which lead to a larger repair densities. Therefore, having a good OTP solution is important in ensuring we can meet the requirements for memory repair in advanced nodes. And this conclude my remarks. Next, I will -- we will enter the Q&A session.

[Operator Instructions] All of our questions will follow the format of answering the Chinese version first, followed by the English version.

[Foreign Language]

[Foreign Language]

So the question was Charles just gave a great talk about SRAM repair. What is the current solution for SRAM repair? And under what circumstances would SRAM repair use eMemory solution? And also how big is this market? Charles?

SRAM repair is many -- SRAM is mainly embedded in processes such as CPU, AP, GPU, et cetera. If the density of the OTP for SRAM repairment exceed 8K bits, eFuse is insufficient. So you will need to switch to antifuse such as NeoFuse. All processors represent a potential market. As SRAM density continue to grow, the potential market for adoption will also increase.

[Foreign Language]

[Foreign Language]

So the question was, what's your view on the penetration rate over the next few years? Michael?

We foresee an increase in foundry penetration rate due to our extensive development in technology platforms across multiple foundries. More than 100 process platforms are development -- under development each year, with an average of 400 new product tape-out growing to an average of 600 new product tape-outs per year, designed in our OTP, MTP and the PUF-related security technologies.

[Foreign Language]

[Foreign Language]

So the question was, there are ongoing concerns for operations in data centers to be secure and confidential. I heard that AMD, Google, Microsoft and NVIDIA recently worked together to develop a Caliptra initiative to define standards for hardware root of trust. Are you guys part of the standard? And a follow-up, what does this mean for eMemory? Charles?

Caliptra is an open-source standard that aims to integrate security mechanism into chips and play a crucial role in the Open Compute Project reference design. Its primary goal is to establish an open-source standard for the hardware rule of trust, which is essential for hardware-based security functions embedded in CPU, GPU, SoC, ASIC, network chips, SSD and more. As far as we know, this is the first time TRNG, OTP and the PUF are collectively addressed within a single standard, which happens to be our strengths and expertise. This is also means the hardware root of trust of the processor for data center must include hard macro, TRNG, OTP and the PUF as key components. However, since the hard macro are unavailable through open source, they must be obtained through a license from us. While Caliptra is still in its early stage, we foresee broader adoption for the hardware-based security in data center in the future, which opens significant opportunity for our PUF-based technologies.

[Foreign Language]

[Foreign Language]

China's mature manufacturing processes are severely oversupplied. Will foundry wafer prices decline in the future and affect your royalty revenue? Michael?

The OTP in mature process is a very stable cash cow for us due to the limited R&D resources. The leading edge and additional royalties from the PUF-based security solutions will drive the ASP for future royalties. On the other hand, because of the overcapacity foundries are allocating resources to develop specialty process to add value and differentiation, such as our MTP-related technologies, which witnessed substantial licensing growth this year. The royalty rate for MTP-related technologies is double that of OTP, which will also increase the penetration rate of mature process and the growth of royalties per wafer.

[Foreign Language]

[Foreign Language]

The next question is about Arm. So in your partnership with Arm, which areas are you targeting? And when will you start to see real contributions from them? Charles?

We are working with them on Confidential Computing Architecture and Corstone architecture, covering applications such as smartphones, automotive and IoT, particularly edge computing. There are license contribution from automotive and IoT customer already.

[Foreign Language]

[Foreign Language]

Are your IPs used in chiplets? And what is the current proportion of revenue? Michael?

Customers use our OTP for multi-chip repair because of traditional eFuse is unsuitable for post-packaging modifications. For example, in recent years, ISP that needs to be packaged with DRAM and CIS are driving customers to use our OTP for ISP. As a result, ISP has contributed to over 10% of our royalties this year. Furthermore, DRAM manufacturers have also licensed our technology for DRAM repair and after 3D or 2.5D chip packaging with the logic chips. We expect more adoption of our IPs for similar application in the future.

[Foreign Language]

[Foreign Language]

So there's questions about our recent announcement. So eMemory recently announced the purchase of new offices. Does this mean you also have major employee expansion plans in the works? Michael?

We have been using our current office for over 15 years but our employee count have increased by more than 50%. So the existing office space is not enough. We acquired a new office to provide employees with a better working environment and will be prepared for the expansion in the future.

[Foreign Language]

[Foreign Language]

Are your IPs being adopted for 7 nanometers and below? And when will you start receiving royalties? Michael?

Over 25 tape-outs from the past have now entered in production. We expect a significant growth in royalties beginning from the next year.

[Foreign Language]

[Foreign Language]

Will the U.S. government's latest ban on China's AI chips affect your business? Michael?

We don't have any impact since the banned chips don't use our IP yet.

[Foreign Language]

[Foreign Language]

As generative AI proliferates on edge devices, will this increase the adoption of eMemory IPs? Charles?

Edge devices, all need security protection so that they can securely connect to cloud services, particularly with generative AI embedded in edge devices, the AI model and the training data must also be protected. Our PUF-based security IP will have a big market in edge devices.

[Foreign Language]

[Foreign Language]

You experienced substantial growth in licensing revenue this year. May I ask what is the reason? And does this mean the future royalties will also increase? Michael?

The growth of our licensing revenue is mainly driven by the strong demand for all of our technologies, which will also drive the strong momentum of future royalties.

[Foreign Language]

[Foreign Language]

Well, changes in Gate-all-around and backside power delivery network affect our technologies? Michael?

No. This will not affect us. We are developing our IP in related process.

[Foreign Language]

[Foreign Language]

As foundries begin to build outside of Taiwan, will your expenses increase? Michael?

Our R&D is still based in Taiwan and we will not expand along with foundries expansion overseas. This is because we provide the technology and the design licensing, which doesn't need -- require on-site execution in the local foundries.

[Foreign Language]

[Foreign Language]

There have been many discussions recently regarding the use of Arm CPU and PCs. How will this affect your partnership with Arm? Charles?

Arm's CPU application in PC belong to their client business. Similarly, CPU in mobile also belongs to client businesses. Our PUFrt will be used in the Runtime Security Engine of Arm's mobile CPU. Once verified, it may also be used in CPU for PC.

[Foreign Language]

[Foreign Language]

What opportunities or threats do you face in the design of processors such as CPU, GPU, NPU? Charles?

Future computing processors will move toward confidential computing, which require high-level security functions. As we have successfully developed a PUF-based root of trust security IP, which provides very strong security functions for these processes, we have a great opportunity to license this IP to this confidential computing processors. Besides, in order to enhance the computation performance, these processors all need high-density SRAM and our NeoFuse OTP play an important role for SRAM repair. We expect our IP to cover a huge market in this processor-related applications.

[Foreign Language]

[Foreign Language]

So over the past 6 years, NeoMTP's licensing revenue was more than TWD 400 million. Last year, it was TWD 90 million and the year before, it even reached TWD 140 million. However, in the past 6 years, NeoMTP's royalty revenue was less than TWD 200 million. Can you explain the gap? And what will NeoMTP's royalty growth look like next year, particularly with your Korean business. Michael?

Okay. The licensing of MTP covers 5 technologies, it includes like NeoEE, NeoMTP, NeoFlash, RRAM and MRAM. So it's more than the NeoMTP only. Over the past few years, MTP's licensing revenues mainly came from the foundries. MTP's development and verification period is relatively long. Typically, exceeds 2 years. As a result it takes at least 5 years before customers even move from the tape-out to production, to royalty contribution. However, since these technologies have already been adopt into the mainstream applications, we anticipate accelerated growth in the royalty in the future.

[Foreign Language]

[Foreign Language]

According to Pages 28 and 29 of this presentation, actually, the number of cases in the production and development for 4, 5, 6 and 7 nanometers is very small. Is this problem risk related? And what is the future outlook for these 2 major nodes? Michael?

The 4,5, 6, 7 nanometers in the presentation refer to the each process node licensing and the development process, not the finalized number of the customer designs. For example, the N7 NeoFuse/NeoPUF and N6 NeoFuse/NeoPUF represent 2 process nodes, not the tape-out numbers. The actual tape-out numbers of accumulated for N7 and N6 exceeds 25. For precise tape-out numbers, please refer to the -- our management report instead.

[Foreign Language]

[Foreign Language]

TSMC mentioned that the 7-nanometer revenue declined this quarter in their earnings. Will this affect eMemory's royalty contribution? Michael?

Our TSMC 7-nanometer applications are just entering mass production but not -- but are independent of the application currently in production at TSMC. TSMC is actively developing a mixed signal, mixed mode-related technologies like RF to increase the foundry utilization rate of 7- nanometer, which will drive more applications to adopt our IPs.

[Foreign Language]

[Foreign Language]

Regarding Charles' talk, have you seen any customer adoption for SRAM repair? Charles?

We have AI-related tape-out which uses our OTP for SRAM repair due to high density -- high SRAM densities. With the increase in SRAM density in AI and the HPC processors, we foresee increasing demand of our OTP for SRAM repair.

[Foreign Language]

[Foreign Language]

This is also a question about Arm. So if the Arm partnership is very successful, what does this potentially mean for eMemory? Charles?

Arm is the leading CPU IP company with more than 95% of the market share in mobile. The successful co-development of PUF-based root of trust security into their CPU has the following implications. First, our PUF-based security root of trust solution is the best in the market. And the second, we can enter the mobile market with Arm, which currently hold over 95% of the market share. And third, we expect our market share in the confidential computing market will also be very big.

[Foreign Language]

[Foreign Language]

PUF-based technologies have seen substantial growth in licensing contributions this quarter. Who are some of your current customers and/or applications that have adopted PUF? Charles?

Other than CPU partner, we have also licensed to DPU, AI, edge computing, HPC, IoT and the ADAS customers.

[Foreign Language]

[Foreign Language]

So this question is about hit rate, which is the conversion rate from new tape-outs to royalty revenue. So this person is asking what is eMemory's typical hit rate? Michael?

The conversion rate from tape-out to mass production exceeds 95%. We have a high conversion rate because most of our tape-outs come from the mature applications of big companies, almost all these tape-outs will successfully move into production, resulting in the royalty revenue.

[Foreign Language]

[Foreign Language]

The question was, what is your progress on 3 nanometers and 5 nanometers? Michael?

This year, 5-nanometer technology will be adopt in the self-driving, data center and AI-related applications. Meanwhile, 3-nanometer technology is undergoing verification at the major foundries. In addition to our collaboration with CPU partners, we also have the -- like ADAS and the processor customers, as well as U.S. cloud server provider requesting our IP for the 3 nanometer. We are confident about the verification process and anticipate customer design wins next year.

[Foreign Language]

[Foreign Language]

The question was, do you only license your IPs to China and Taiwan? And a follow-up, do you have U.S. customers like Intel and GlobalFoundries? Michael?

Approximately 30 foundries worldwide have licensed our technologies and it include Intel and GlobalFoundries.

So in the interest of time, we will begin the closing comment. Charles, please proceed.

Okay. So thank you very much for attending our investment conference -- investor conference. For more information about our PUF-based security IP and the technology, we encourage you to visit our PUFsecurity website and check out our articles and the other materials. Thank you again for your patience and the support for eMemory. We will continue to work hard on technology and IP innovation and the PUF-based hardware security solution for our customers and bringing a higher return for our shareholders. Thank you.

Thank you, ladies and gentlemen. Before we conclude today's conference, please be advised that the conference recording will be accessible within the next 3 hours. Thank you, everyone, for joining with us today. We hope you will join us again next quarter. You may now disconnect. Goodbye and have a good day.