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

Good afternoon, and welcome to eMemory's Third Quarter 2021 Webcast Investor Conference. Present today, we have 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 2021, followed by the outlook of our business. Next, Dr. Charles Hsu will introduce the topic titled Confidential Computing Architecture. 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 within 3 hours after the conference is finished. Please visit the company's website at www.ememory.com.tw 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 that are subject to risk factors associated with the semiconductor and IP business. Please refer to the cautionary statement as shown on Page 3 of today's presentation. Now I would like to give the floor over to eMemory's Chairman, Dr. Charles Hsu.

Okay. Good afternoon, everyone. Thank you for attending our conference call today. This quarter, our new President, Michael Ho, has officially taken office. Michael has nearly 20 years of management experience with eMemory and holds over 30 patents in the semiconductor technology field. Before becoming President, Michael was the Senior Vice President and oversaw sales, product marketing, global business development and the customer engineering services. During his tenure as Senior Vice President, Michael successfully expanded the global business, making eMemory one of the world's top 10 silicon intelectual property providers. We are confident that his professional knowledge and expertise will continue to help drive the company's growth. As mentioned in the previous quarters, eMemory has entered a multiyear growth cycle. We have started to receive royalties for 16-nanometer and 7-nanometer. With this production record, we expect more customers' adoptions to come and will be our next driving force after 28-nanometer. In addition, customers' adoption of crop-based solutions and emerging memory will increase license than royalties, which will drive our growth beyond the next 3 to 5 years. And later, I will explain what role PUF plays in confidential computing, which is considered one of the next big trends of the decade. With this, next, I would like to invite our President, Michael Ho, to share our third quarter performance. Michael, please?

Thank you, Charles. Good afternoon, everyone. First, I will begin with our third quarter results. The third quarter revenue was TWD 594 million, up 9.8% sequentially, and up 34.7% year-over-year, or up 11.1% sequentially and up 42.8% year-over-year in U.S. dollars. The operating expense were TWD 278 million, up 7.3% sequentially and up 13.8% year-over-year, mainly attributable to the increase in salary and other related human resource expenses, such as increases in bonus and rewards. This brings us to the operating income of TWD 317 million, with an increase of 12% sequentially and 60.5% year-over-year. Therefore, the operating margin increased by 1.1 percentage points sequentially and by 8.6 percentage points year-over-year to 53.3%. Overall, our third quarter EPS was TWD 3.72 and ROE was 53.2%. For the first 3 quarters of 2021, the revenue was TWD 1.732 billion, up 35.3% year-over-year. The operating expense increased by 14.8%, and the operating margin was 5.1%, with an increase of 8.2 percentage points. EPS is TWD 10.92 and ROE is 52.1%. Now let's move on to revenue contribution by licensing and royalty. Licensing in the third quarter accounted for 31.7% of the revenue, up 8.1% sequentially and up 69.8% year-over-year, or up 9% sequentially and up 80% year-over-year in U.S. dollars. Royalties in the third quarter contributed 68.3% of total revenue, increasing 10.6% sequentially, and increasing 22.9% year-over-year, or up 12.1% sequentially and up 30.3% year-over-year in U.S. dollars. In the first 3 quarters of 2021, the total revenue grew 35.3% as compared to previous year. Licensing and royalty have a growth of 61.0% and 26.2%, respectively. In terms of U.S. dollars, the total revenue increased 43.4% year-over-year, with licensing and royalty both increasing 70.8% and 33.6%, respectively. In terms of revenue contribution by specific IPs, the results are as follows: NeoBit accounted for 20.7% of the total licensing revenue in the third quarter, decreasing 22% sequentially, but increasing 56.5% year-over-year, mainly due to new applications in 8-inch, such as automotive and IoT related. Its royalty accounted for 47.3% of total royalty, up 11.2% sequentially, but down 9.5% year-over-year as some PMIC and DDI customers migrated to 12-inch and used the NeoFuse instead. Many new NeoBit tap-outs accumulated in the past years, such as legacy PMIC, automotive and IoT-related products will drive the growth of NeoBit royalty once it enters into mass production. NeoFuse accounted for 49.5% of total licensing revenue in the third quarter, down 8.7% sequentially, but up 36.2% year-over-year. Its royalties increased 10.9% sequentially and 88.7% year-over-year due to increasing production from existing and new products, such as WiFi 6/6E, OLED, DTV, DRAM and others. This brings the royalty of NeoFuse to 49.7% of total royalties. Our PUF-based security IPs contributed to 1.5% of licensing revenue and have royalty contribution in the third quarter. Our partnership with global customers is actively ongoing. We expect more contribution in the future. As to MTP technology, licensing revenue accounted for 28.3% of total licensing revenue, increasing 164.6% sequentially and 367.6% year-over-year, mainly due to contribution from China customers. Royalty from MTP decreased 2.7% sequentially, but up 6.3% year-over-year to contribute 3% of total royalties. In the follow -- in the first 3 quarters of 2021, for NeoBit, the licensing revenue increased 67.7% year-over-year, but royalty decreased 5.5%, accounting for 40% of the total revenue. For NeoFuse, the licensing and royalty revenue grew 26.2% and 92.5% year-over-year, contributing to 50% of the total revenue. For PUF-based security IP, licensing revenue increased 8% year-over-year, about 0.6% of total revenue. For MTP technology, the licensing and royalty revenue increased 328.7% and 10.2% year-over-year, accounting for 9.4% of the total revenue. Now looking at the royalties for 8-inch and 12-inch wafers. 8-inch wafers, which accounted for 52.3% of royalties, increased 9% sequentially and 6.7% year-over-year due to applications expanding into automotive and IoT-related applications. 12-inch wafers contributed to 47.7% of royalties, increased 12.3% sequentially and 47.4% year-over-year, mainly due to 28-nanometer related product ramping up production. There were 149 product tape-outs completed in the third quarter, which reflects that our IP demand remains strong. We will provide more information in the management report. In the next section, I will address our future outlook. We expect the growth of revenue to continue in the last quarter of 2021 and beyond. For the licensing revenues, as the demand for all our technologies continues to remain strong, we expect licensing revenue to continue growing in the fourth quarter and beyond. For the royalty revenues, 8-inch and 12-inch royalties will continue their growth momentum. 8-inch royalties will grow due to demand and content increase for automotive and IoT, for example, PMIC, MCU, fingerprint and sensor-related applications. 12-inch royalties will have a strong growth since customer productions are increasing for TDDI, OLED, ISP, DTV, set-top-box, WiFi 6/6E, bluetooth, ethernet, switch, TWS, DRAM and others. In addition, 12, 16-nanometer and the 7-nanometer royalties already started to contribute in the fourth quarter. Now looking at the new business development. Our new applications are centered on the business development of hardware security. NeoFuse, in advanced processes, is being adopted for AI, HPC and automotive applications, especially in 7-nanometer and 6-nanometer process nodes. Business activities for PUF-based security solutions are in progress in IoT, industrial IoT, AI, blockchain, FPGA, Data Processor Unit, DPU, Mobile Storage UFS and automotive applications. Our PUFrt and PUFcc have been adopted by several customers across various applications. Our collaboration with ARM will expand from IoT to CPU security architecture. eMemory has received the TSMC OIP Partner of the Year Award for the 12th year in a row, demonstrating the trust of our partners have in us. For new IP technology development, in Q3, NeoFuse has been qualified in TSMC's N6 process. This security-enhanced OTP combines NeoFuse with NeoPUF to protect data, targeting mobile, consumer AI, networking, 5G infrastructure, GPU and high-performance computing. Some customers have already completed design-in, and 5-nanometer is ongoing reliability verification. Our ReRAM IP has been qualified in UMC's 40-nanometer process. We are one of the world's first companies to provide this emerging memory and offer more comprehensive solutions for automotive, edge computing, AI and IoT markets. We will extend ReRAM technology to more leading process nodes and specialty processes such as BCD and high voltage. PUF-based solutions are under development for implementation in WiFi 6/6E for AIoT edge devices and AI image processors. Now I will pass the time to Charles.

Okay. Since there are many articles talking about the confidential computing, so I would like to introduce this confidential computing architectures and also tell you why and how our PUF can be used in the confidential computing. So please turn to Page 15. With the rise of IoT and AI applications, a tremendous amount of data is generated. The processing of this huge amount of data requires not only high-speed computing, but also secure computing such that the sensitive data is safeguarded. The processing of data includes data at rest, data in transit and data in use. And for the data at rest and in transit, we can encrypt the data so that they cannot be compromised. However, for the data in use, such as the execution of program code and its related instruction set, the data cannot be encrypted and need to be plain data for CPU to execute. So in the current computer systems, data in use are brought into virtual memory, which is shared by different applications and is, therefore, not protected. So this has been resulted in attacks on CPU, such as the Meltdown and the Spectre attack on Intel's CPU, which was reported last year or the year before. And so to solve these kind of issues, confidential computing architecture is needed. So in the following slide, I will introduce what is confidential computing. So please turn to Page 16. So in this page, you can see the nonconfidential computing in traditional architecture is illustrated in Figure 1. The CPU and the memory, they take care of the application. And since a single CPU can only run one application at a time, so in this case, the CPU will be -- will solely use the memory. And to improve the security of such structures, the CPU's security function can encrypt and manage the memory as shown in figure 2. Please turn to Page 17. However, as CPU moves to multicore structures, which can tackle many applications simultaneously, the architecture will evolve to the one shown in Figure 3. The applications are divided into two sections. One is secure section, and which will be operated in a trust zone, and the other is a nonsecure section. And in the trust zone, the operation of the application happens inside the so-called security enclave, which means data and code are all protected by its security functions. But on the other hand, for the nonsecure sections, data and code are stored in the virtual memory, which is shared by many applications. So in Page 18, as more applications are required to operate in the trust zone, the dedicated memory, which was supposed to be solely used by its corresponding CPU, becomes inefficient. To improve such case, the secure applications will also need to use the virtual memory to increase its storage capacity and also flexibility, as shown in Figure 4 on the left-hand side. In order to secure these applications by using virtual memory, the data and the code will be tagged for the specific applications, such that other applications are not able to access them. Even if it can be accessed, the data is not meaningful since they are tagged or encrypted. And so our PUF is needed for such functions since it can efficiently scramble the data and the code and also descramble using the same PUF numbers. But however, in those applications, the trust zone will still be isolated. And because those applications will be mainly used for the payment-related applications, so -- which is needed to be much more secure, but the rest of the applications, which also need security, but they can use the memory-taking technology to protect their program code and the data in the virtual memories. Okay. So in Page 19. So how can our PUF help in the confidential computing architectures? So recently, like the new rise of DPU, which actually is used to unload the data storage and security functions of the CPU, and if you -- is targeted at helping CPU architecture move towards confidential computing. So we are working with some companies to implement our PUF into their security architectures. So by applying a PUF to protect the data and the code in the virtual memory, PUF can enable the computing to become confidential computing, and can help protect all the devices of the Internet and the cloud. So just to conclude my introduction for the confidential computing.

[Operator Instructions] Our first question is, compared to the past, has the development process for advanced nodes, from initial development to mass production, accelerated or slowed down? Take, for example, the 28-nanometer process node.

Okay. Compared to the past, the complexity of the design, the number of masks, production processes and the functional verification for advanced manufacturing processes have greatly increased. Therefore, the time from development to mix production will be much longer. However, because of our credibility from the past, once we have production record for new technology and new process nodes, customers will speed up the adoption of our IPs. In addition, our license fee and royalties will also increase because of the wafer ASP of advanced process is much higher than that of a mature process. For example, the foundry wafer price and the license fee of 28-nanometer will probably be 4 or 5x more than that in 0.18 micro, which will definitely increase our revenue. Thank you.

Our second question is, will emerging memory MRAM and ReRAM replace eMemory's current OTP and MTP markets?

It's a good question. For emerging memory like MRAM and ReRAM target at the traditional embedded flash market. Compared to embedded flash, ReRAM and MRAM require fewer mask layers and are much easier for process integration, which can be maintaining the existing device model. Furthermore, they are applicable to more leading process nodes, especially after 28-nanometer. Our OTP can work as a security and repair function to protect and correct the data stored in the emerging memory. Therefore, these emerging memory help extend our technology portfolios, but not replace our existing technologies. Thank you.

Our next question is, along with traditional applications in DDI, PMIC, DRAM and multimedia, what other areas of applications can NeoFuse 12-inch grow significantly? What is the target market share of 12-inch in the future, excluding NeoPUF?

Okay. Besides those traditional applications, we also have NeoFuse OTP adopted in 12-inch wafer process like ISP, CIS, TWS, WiFi, network IC, SSD controller, AI SoC, ADAS and other applications. We expect they will have obvious growth in the near future. It will also help increase our royalty ASP and bring in more design license fees as well. Thank you.

How are the certification progress and implementation status for MRAM and ReRAM? Does eMemory own IP? Is the business model still charging royalties?

As mentioned in our joint press release last week, our ReRAM IP was qualified on the UMC 40-nanometer process. We are also working on 22-nanometer process node for both ReRAM and the MRAM technology development and the designs in ongoing projects. Fab customers license memory cells from patent owners, for example, IBM for MRAM and Panasonic for ReRAM. Based on the memory cell, we add the circuit design to become a complete IP. Our business model remains the same. We will charge the negotiated royalty rate of MRAM, ReRAM according to wafer price after the initial license. Thank you.

Our next question is, does eMemory expect NeoFuse and NeoPUF to significantly mass-produce and increase the revenue and royalties in 2023 or 2024, or will it take longer?

Thanks for the question. NeoFuse royalty has increased significantly already. We have also started to receive some royalty income from NeoPUF technology. With the production record, we expect more customer adoption to come and drive royalty growth. Thank you.

Our sixth question comes from -- are eMemory IPs applicable to Metaverse-related applications?

Okay. It is a very hot topic recently. Yes, VR/AR devices will use DDI, power, sensor, low-power embedded memory MCU, and the connection speed will require at least WiFi 6E, which will use our IP. For example, our IPs are already applied into HoloLens. After that, there will be security requirements which will require PUF. Thank you.

Yes. I can also add a comment on these new applications. Since this is new applications and eMemory, we also have the new IPs such as [indiscernible] security IP and PUF IP. And when a customer has new applications, they will have a new product. So the new product always would like to use the new IP to add the value on their applications. So I think with this new application in Metaverse, we will have more IPs to support these applications. Thank you.

Our next question is also related to Metaverse. The concept of Metaverse requires information security. What potential role can eMemory play in Metaverse?

As I mentioned earlier, besides the fact that our IPs will be applied in various chips, eMemory's role in Metaverse is mainly to provide chip hardware root of trust and the co-processor solution, and help servers and connected devices by providing the unique identity, key generation and authentication and encryption and decryption functions required for Metaverse's security network connection. Thank you.

Next question is, what are the target applications for automotive?

Thanks for the question. Our logic NVM and security IPs are used in automotive applications like ADAS, various kinds of sensors, for example, like TPMS, CIS, temperature, light, motion sensors, PMIC, infotainment display and security protection applications. We offer IP solutions from 0.25 micron to 7-nanometer, which all have customer adoption already. Thank you.

Your next question is, top global companies like Apple, Tesla and Nvidia have split their stock prices after their share prices have increased exponentially in recent years. Does our company consider evaluating a potential stock split?

Our focus is to innovate and execute our technology for our customer in the market, and aim for a sustainable long-term growth model for our shareholders. Whether or not stocks are split does not change the fundamentals of a company, and is therefore not within our consideration.

Next question is, looking at the history of eMemory, the business model has evolved. In the beginning, no one guessed that eMemory's technology could play such an important role in semiconductor applications. Can you discuss the company's development direction for the next 10 years and beyond?

eMemory's business model hasn't changed since we launched our IP business. In the early days, our IPs were widely used in analog related chips. With our innovation into security area, for the next 10 years and beyond, we will continue developing technology and extending it to every IC. Security will be tremendous driving force beyond that. Thank you.

Next question is, how does the current shortage of automotive chips affect eMemory's revenue? What is automotive's proportion in eMemory's revenue? Do you expect future growth in the automotive application?

We didn't classify the automotive segment as some of our existing customers supplied their chips to all smartphone-s, industrial and automotive applications, for example, DDI for infotainment, sensor and MCU or PMIC for EV cars. In addition, as our technologies are ready in leading edge process nodes, we have expanded our IPs into autonomous driving platform, which will be applied into ADAS and various networking chips. We do expect automotive contribution to be one of the driving forces for our growth in the future. Thank you.

Twelfth question is, in the past, management had predicted the market share for the 8-inch and 12-inch market. Can management also offer a prediction for eMemory's possible market share in hardware security?

[indiscernible] question. For 5G and IoT relevant security needs, security is essential for particular processors, which are mainly in production on the 12-inch leading process node. Therefore, we believe our market share in 12-inch will be higher than in 8-inch as our solution is much better and more secure than existing eFuse solutions. Thank you.

Our next question is, Apple, Intel and Nvidia will all be developing chips with RISC-V architecture. Besides cooperating with Andes on RISC-V, does eMemory plan to collaborate with other companies?

Yes. We already have a joint development and press release with Andes on their RISC-V CPU core, along with our PUF-based root of trust and crypto coprocessors. We are also working with several other leading RISC-V-based CPU core companies, such as Sifive, OpenFive, StarFive. Some of them are in U.S. and China and other regions, too. We're also working with the Verisilicon and Alchip to provide our OTP and PUF-based IP solutions to fulfill their customers' security needs. Thank you.

Our next question is, is there any progress in the FIPS 140-3 standard?

Standard? FIPS is the certification for the CMVP encryption module required for the product design of eMemory's customers. In short, the FIPS basically certifies the products, not the IP. But however, our IP, which includes cryptographic algorithm developed by us, has passed the CAVP certification, and -- which can help customers accelerate their CMVP or FIPS certifications. Thank you.

Our next question is, based on the company's current partnership with PUF customers, what applications and process nodes are these customers targeting?

As mentioned earlier, our PUF-based security solutions are under progress in applications of IoT, industrial IoT, AI, FPGA, Data Processor Unit, WiFi and automotive fields. And their relevant process nodes ranges from 40-nanometer to 7-nanometer.

Our next question is, what is the relationship between eMemory's IP and virtual currency? What can it be used for?

We have crypto-mining customers who adopted our NeoPUF and also the PUFrt, mainly for the security functions. In addition, using NeoFuse or the PUF root of trust, we can provide this as a security function. We can provide the private key for storing cryptocurrency, is the most secure method, making PUF an ideal choice for the implementation across various digital wallets. Furthermore, using the private key generated by PUF as a digital signature in different blockchain applications is simplest and also safest and the most convenient security method to -- for the cryptocurrency protections and the transactions. Thank you.

Our next question is, how does PUFsecurity plan to quickly educate its customers on how to use PUF and PUF-based solutions? What is the current progress of PUFsecurity's software development? How can the collaboration with ARM expand the product application base?

You can -- if you look at our website, also our [indiscernible] and also in many different medias, PUFsecurity is regularly publishing new product information and security technology design white papers through this -- social media and webinars, also with various application subjects. In addition, there's also -- we also have PUF Academy, which provides several hardware security-training courses to educate the market and help chip designers to implement our solution into their design. And for the software development, which we are doing, is mainly aimed at developing and verifying various software and firmware used for the security function and also the driver and the APIs, which are required to use the PUF-based IP solutions. And our subsidiary, PUFsecurity, is currently working with CPU IP vendors to develop the SoC chip security design architecture for their customers. Thank you.

Our next question is, last year, eMemory mentioned transforming into a Security as a Service company. Is the goal of PUFsecurity to become a Security as a Service company, or do they have other plans?

I would like to clarify that this is not for the business transformation. The reason that eMemory -- we established the PUFsecurity basically to extend our business by using the PUF technology invented by eMemory. So it is to extend our new business instead of transformation of our business because our current business, IP business, still has a lot of room to grow, but we would like to extend our business into the security because the PUF technology is based on eMemory's original OTP technology. It currently is the best PUF in the world. And we can provide an integrated one-stop shopping for the security solution to customers.

Our next question is, we know that security is essential in the 5G era. What is the estimated total market value for security in 2025? Moreover, what percentage of this total would you estimate would be for hardware security?

The connected world from 5G and IoT will need the security IPs for the unique ID and authentication and also for the secure communication functions. For 5G and the IoT markets, they are in the order of several billions of devices per year. And we believe that hardware security's protection functions will perform much better than the software security. So hardware security will also play a major part in the long run, for the -- for example, for the Zero Trust security, for the new applications such as Metaverse, and also confidential computes.

Our next question is, Apple, M1, AMD, Nvidia chips all emphasize the connection between processor and memory. Will eMemory's experience in memory development benefit from this trend?

Our OTP and PUF-based security IPs can be used for setting, configuration, encryption and data protection functions. We already have several projects ongoing with customers for relevant applications. Thank you.

Our next question is, what is the difference between NeoPUF and Synopsys' DesignWare tRoot H5 hardware security model?

To compare with Synopsys tRoot, our PUF-based solution is PUFsecurity core processor developed by our subsidiary, PUFsecurity. And the Synopsys' solution is a pure digital design, which contains security algorithms. And the key for root of trust is generated from the algorithms and needs to be injected through an external security environment, which is usually not very safe. Instead, our PUFsecurity core processor, we have included PUF inside. And also, we also have the secure OTP. We also have true random number generator, and also put this all into the anti-tempering [indiscernible]. So for our PUF-based root of trust, the keys inborn by the chip itself, which is more secure and cost-efficient than Synopsys' solutions. Thank you.

Our next question is, Palo Alto Networks can also use software to achieve a Zero Trust network access from the cloud and generate public and private keys. What are the advantages of eMemory compared to Palo Alto Networks?

Okay, again, similar question as the previous one. And so the answer is, our NeoPUF is chip fingerprint that comes from the chip itself. And this fingerprint can be used to generate the chip's genetic private key and also public key. So if the key pair is generated by software solutions, such as what Palo Alto Networks does, it needs to be stored into the chips. So again, to do such, it must inject through an external very secure environment, which is usually very high-cost and is actually not very cost too -- very secure too because you also need to have the operator to do these operations. So in comparison, a genetic root of trust based on PUF -- our NeoPUF, is more secure and cost-efficient. Thank you.

Our next question is, for advanced packages, such as CoWos, InFo and other 2.5D or 3D packages, what functions or benefits can eMemory's IPs provide for customers who adopt?

For multichip packaging, eMemory's OTP IP provides in-package repair function for DRAM, calibration function for power management, display and image sensors. Therefore, in this aspect, our IP improves the yield of multichip package. Second, PUFsecurity IPs are our security functions in the security chip of multichip package. In this aspect, our IPs enable the security of the subsystem. Thank you.

Out next question is, are there any network effects with adoption of NeoPUF? That is, are hardware security protocols easier to implement if everyone uses the same PUF, such that if one leading company adopts NeoPUF, then others will follow, and NeoPUF will effectively become an industry standard?

This question is asked by David Manulife. I think this is a very good question. Yes. I think, from our past history, we always create new IP, new solution to disrupt the traditional usage of the technologies. So again, we -- in the NeoPUF, actually, this is also very new to the industry and will be used for the securities. So if this -- a leading company, which adopts these solutions, and of course, because this solution will give a much better security solution to their hardware. So if the leading company, or they are a couple of companies, begin to use this solution, then the rest of the industry will start to follow. It's just like -- those leading companies will be just like early adopters. And when the early adopters start to adopt these solutions, and there are a couple -- there are many more companies, they are -- we call it early majorities, they'll follow with these early adopters to adopt this solution. I think this is what you call the network effect. It's definitely yes.

Next question is, for quarter 3, the 28-nanometer royalty, what is the ratio for the 28-nanometer royalty?

In the third quarter of 2021, the proportion of 28-nanometers is 18%, compared to last year's 7%.

The next question is, for quarter 3, 0.11, 0.13 and 0.18 micrometer licensing suddenly increased. Why is that?

The third quarter, in the legacy node, we also received some NTOs from like automotive-related, IoT-related, these kind of applications. So we see the NTO number increased in the third quarter.

I think one of the major reasons is from China because China wants to localize some of the -- actually is our main location, especially power management and driver IC. So our licensing from China actually increased a lot, okay? And this is one of the reasons Dave asked about, the MTP licensing increase, which is like putting interest on all kind of emerging memory. And we are working with the China government-based fab for MRAM development. Thank you.

Our next question is, why has PUF licensing revenue declined year-over-year in Q3? How do you estimate the potential total addressable market size for PUF? And can the root of trust chip -- and what is the root of trust chip in hardware security market size?

I think this is the very beginning of the PUF. So we do have a lot of activity, but every case is changing about the security architecture. So if you change about architecture, which should take a much longer time for the engagement. So you pretty see some kind of lumpy number at the very beginning stage, just at the beginning, what we are seeing on the NeoPUF. This is quite normal. And once we have the class effect, so they will become a pretty steady income, that's the reason. So this is the beginning of the -- especially right now, we have the production record from our first PUF customer, which you just had a question asked, is a 7-nanometer FPGA? This is very important for us. [indiscernible] because most of the customers will ask if you have a production record or not. So we do work with many customers, and we -- they're all waiting on the production record, okay? So in the future, especially next year, you're going to see a much more significant growth moving forward. Thank you.

[indiscernible] comments. For the -- what process node does PUF work best? Actually, the range of process node, which all require a PUF for the security. For example, in the process node 14-nanometer, that's basically the IoT applications. And for the 55-nanometer, that's basically for the MCU applications. And for the 28-nanometer, that will be for the AI applications. And for 16, that's also AI. And for the -- down below 7, there are -- some of the customers is working on the GPU and FPGA and also the DPU. And also for -- after that will be 5-nanometer and the 3-nanometer. I think that will be -- mainly will be for the confidential computing. So there's a whole range of the process nodes and for different applications. As long as they require securities, the PUF will be before those applications.

What are the key risks to your continued growth success? Any emerging competitors?

For the -- I think, for our business model, and we also have continued to innovate new technology to meet the industry's requirements. So the risk for us, I think, is global economy. If the global economy is not good and the demand on the electronic devices decreases, I think that will impact us, too, because then -- if there's no chip is produced then because we licensed our technology to those chip makers, right? And so for the -- that's the risk we have seen currently so far. And for the emerging competitors, currently, I think the reason that we are not afraid of competition is because our security basis is actually on our OTP technologies. And in the [indiscernible], we have more than 400 technology -- OTP technologies. So because the PUF technology is for -- is a hard IP, PUF is a hard IP and the PUF base root of trust is IP. And they are all based on our previous OTP technologies, and -- which create the root of trust IP. And because we are talking about the hardware security, so the hard IP is very important. And for any hard IP, you have to be verified into -- in the foundry at every process the customer wants. So this will take a long time if you start from the scratch. But since eMemory has been working in the industry for 20 years and we have built more than 400 process platforms, so in any of these platforms, we can implement the PUF-based security very easily. So for this part, we don't think that even there's emerging competitors, so we are not afraid of. I think for this PUF-based security, we do have a very big confidence that we will be the major player in the market in the future.

Next question is, what is the advantage of your OTP-based PUF versus SRAM PUF?

This is a question we have been always asked by our customers because SRAM PUF has been in the market for more than 10 years. But unfortunately, because SRAM PUF is not easy and they're not good to use, so that's why they switched to our PUF solutions. Because SRAM PUF is basically based on the SRAM mismatch. I mean, because SRAM is actually back-to-back inverters. So if there's a mismatch in these back-to-back inverters, so you will show up the random numbers. But because as the technology continues to scale down and the operating voltage will also scale, right? So when the voltage is scaled down to like a 0.8 volt and the threshold voltage control has to be very tight. So because threshold voltage will be very uniform and very tight, so the mismatch of this -- the mismatch in the SRAM will be getting very small. So when the mismatch become very small, the random number created in SRAM will not be stable. So this is already a risk for the customer. If they are using the secret key, which is created by SRAM, and it is not stable, that means you will lose your key. So if the system lost the key, then the system will crash, okay? So that's the reason that SRAM will not be the good PUF for the device to use. But on the contrary, our PUF is based on our OTP technology. And our OTP has been proven so far. Every chip uses our OTP, has been satisfied and more than 10 year's lifetime. We are -- we have qualified our technologies, so there's not any issue on the stability of the numbers created in the PUF. So our PUF is much more stable. And that is the most advantage for our PUF. And in order to be used as a private key, the number created in the PUF has to be very -- has to be random, unique and also, most important, stable. So I think that's our advantage.

[indiscernible] is regarding MRAM. What applications is it targeted for Chinese customers?

I think there's a big market in China is automotive. And we all know that for the technology in 28 or below, the traditional embedded flash is very high-cost and also very difficult to be successful in the technology 28 and below. So that's why the emerging memory such as MRAM is taking off because of the -- to replace the traditional embedded flash. And one of the big advantages of MRAM has -- is it is subject to the high-temperature stress because the charge loss in MRAM is only be affected by the so-called -- because MRAM is a magnetic RAM, right? So magnetic RAM, when the magnetic -- the number in the magnetic RAM change, will be subject to the magnetic field change is not -- so the temperature, even the temperature change, it will not change the privatization of the magnetics. So the -- in China, they are targeting to use the embedded MRAM in the high-end MCU to replace the current embedded flash because automotive needs a chip to be operated at higher temperatures than consumer device. Thank you.

Out next question is, please comment on recruitment.

The question is mainly -- probably concerning about the -- now the recruitment in Taiwan because a lot of -- because of a shortage of talents, so a lot of companies is giving a very good deal for the engineers. I think that -- I think there are a couple of things. One thing is the -- basically, now there are a lot of recruiting is to recruit the digital designer. For example, the -- like all the big companies in Taiwan, MediaTek, NovaTek, RealTek, they are basically SoC companies, and many require a lot of digital designers and also the software designers. But for eMemory, we are the memory companies, okay? So because -- particularly our memory is very special memories. So far, we have not suffered any -- the engineering loss issues. And actually, our turnover rate is very low. And -- but one thing is, we are very happy is that some of the engineers, they actually would like to change their original initial working environment to our environment because I think quite a few of engineers, they -- after they work for the same company and they are doing the routine jobs, they would like to change their working environment to look for the environment who are -- which are more innovative. And -- but as you -- everybody knows, in order to have an engineer to have more innovation, you have to give them some more room to think. So I think that's our culture. So for the recruiting, actually we have any -- not any problems to hire any new talents. Thank you.

This time, we will now begin the closing comments. Charles, please proceed.

Sorry, I didn't turn on the microphone just now. So thank you, everyone, once again, for your patience and support for eMemory. And also, thank you for quite a few questions you asked. And we will continue to work hard on our IP innovations and the security solutions for our customers and bring a higher return for our shareholders. Thank you. Bye.

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