Good afternoon, everyone. Welcome to eMemory's first quarter 2026 webcast investor conference.

Joining us today is our Chairman, Dr. Charles Hsu, President, Mr.

Michael Ho, Head of IR, Ms. Li-Jeng Chen, Director of the Finance Department, Mr.

Joseph Hsia, and Head of Digital Marketing, Dr. Felix Hsu.

The format of today's events will be as follows: first, eMemory's Chairman, Dr. Charles Hsu, will give an opening remark.

Afterwards, our Financial Officer, Mr. Joseph Hsia, will present a review of our financial results.

Following that, President, Mr. Michael Ho, will share our business outlook.

Next, Head of Digital Marketing, Dr. Felix Hsu, will give a talk titled "Hardware Security for AI Agents".

Then, we will conclude today's conference with the Q&A section, where our management team will answer your questions. Please feel free to submit your questions in the input box on the webcast window throughout the conference.

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 the company's website under 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 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.

Thank you. Good afternoon, everyone.

In the first quarter of this year, eMemory continued to deliver a record-high operating performance. In addition to sustained growth in 3nm licensing for AI server CPU applications, we also successfully expanded into new areas, such as high-speed interface applications.

Meanwhile, as AI infrastructure rapidly evolves, industry demand for hardware security is undergoing a fundamental upgrade. The Open Compute Project's Foundation Chiplet System Architecture, introduced in 2025, explicitly requires every chip within a chiplet system to incorporate a hardware root of trust.

This signified that every chip in the future AI system will require built-in hardware security mechanisms to ensure system integrity and data security. Furthermore, the rapid advancement of agentic AI, together with increasing government requirements worldwide for post-quantum cryptography, is driving what we believe to be the most significant hardware security upgrade cycle in the past two decades.

In response to this trend, we are actively evolving from an IP provider into a system-level security solution provider. Our internally developed hardware security module product is expected to be officially licensed for customers' mass production deployment upon completion of the US NIST, National Institute of Standards and Technology, verifications.

We believe eMemory will be one of the key beneficiaries of this global hardware security upgrade trends. At the same time, the core technological foundation of eMemory remains in logic non-volatile memory.

We have long focused on enabling non-volatile memory technology through standard logic processes. In addition to being embedded into SoC in IP form, this technology also have the potential to evolve into standalone memory products.

Recently, global flash memory supply has remained constrained. eMemory's key advantages lies in our ability to directly leverage mature logic process capacity to produce non-volatile memory, avoiding the capacity bottlenecks associated with conventional, dedicated flash manufacturing lines.

We are currently working closely with multiple foundries to advance the related technology and customization effort. This development not only carries strong strategic value for foundries and end-system customers, but also has the potential to become the next major growth driver for eMemory's revenue and profitability, following OTP and PUF security solutions.

Next, I would like to invite our CFO, Mr. Joseph Hsia, to present the operating results for the first quarter of 2026.

Good afternoon, everyone. Thank you for joining us.

Now allow me to walk you through our financial results for the first quarter of 2026. 2026 Q1 revenue was NT$1.09 billion, increasing 4.4% quarter-over-quarter and 20% year-over-year.

Operating expenses were NT$432 million, up 4.6% sequentially and up 10.8% year-over-year. As a result, our operating income reached NT$662 million, representing an increase of 4.2% quarter-over-quarter and an increase of 26.8% year-over-year.

Operating margin was 60.5%, relatively flat sequentially, down by 0.1 percentage point, but increased by 3.2 percentage point year-over-year. Net income for the quarter was NT$596 million.

It experienced an increase of 5.9% sequentially and an increase of 29.1% year-over-year. EPS for this quarter was NT$7.98.

Next, let's move on to revenue mix by licensing and royalty. Licensing in the first quarter accounted for 34.8% of the total revenue, up 9.9% sequentially and up 58.6% year-over-year.

On US dollar basis, licensing grew 7.8% quarter-over-quarter and 64.4% year-over-year. Royalties, on the other hand, contributed 65.2% of the total revenue, increasing 1.6% sequentially and increasing 6.2% year-over-year.

On US dollar basis, there was a 1.2% decrease quarter-over-quarter, but 11% increase year-over-year. Overall revenue increased by 4.4% quarter-over-quarter and 20% year-over-year.

On US dollar basis, the growth was even stronger, at 1.8% quarter-over-quarter and 25.2% year-over-year. With that, I will break down the revenue contribution by specific IPs.

NeoBit. NeoBit accounted for around 20% of the total revenue in the first quarter.

Its licensing revenue decreased 28.4% sequentially and decreased 13.7% year-over-year, while royalties increased 3.1% sequentially, but down 4.2% year-over-year. Secondly, for NeoFuse.

NeoFuse accounted for around 59.8% of the total revenue in the first quarter. Its licensing revenue increased 39.8% sequentially and increased 20.7% year-over-year.

In terms of royalty revenue, NeoFuse royalty increased by 0.5% sequentially and increased by 7.4% year-over-year. For PUF-based security IPs, it contributed around 12.2% of the total revenue in the first quarter.

Its licensing revenue increased 21.5% sequentially and increased by 606.9% year-over-year, while its revenue account for around 1% of the total royalties in the first quarter. Lastly, for MTP technology, it accounted for around 8% of the total revenue in the first quarter.

Its licensing revenue decreased by 5.5% sequentially, but increased 37.9% year-over-year. Royalty from MTP was down 3.1% sequentially, but increased by 45% year-over-year.

Now let's look at the royalties for 8-inch and 12-inch wafers. 8-inch wafers accounted for around 33.7% of the royalties in the first quarter, down 5.5% sequentially and down 15.9% year-over-year.

On US dollar basis, this represents a sequential decrease of 8.2% and a year-over-year decrease of 12%. On the other hand, 12-inch wafers contributed 66.3% of the total royalties in the first quarter, up 5.6% sequentially and up 22.5% year-over-year.

On US dollar basis, this represents a sequential increase of 2.8% and a year-over-year increase of 28%. A total of 139 licensed product tape-outs were completed in the first quarter of 2026, and further details will be provided in our management report, which will be released shortly after this earnings call.

Next, I would like to pass the floor to our President, Michael Ho, to share more about our future outlook. Thank you.

Good afternoon, everyone. In the following section, I will address our future outlook.

For licensing revenue, it's driven by an increase in the number of license from both foundries and fabless customers, as well as higher average selling price per license. Demand remains strong for advanced-node technologies, AI-related applications, security-related solutions, and next-generation flash technologies.

For royalty revenue, continue to accelerate, driven by upgrades of existing products, higher ASPs from the new advanced-node applications, expanding PUF royalty contributions, and a growing mix of a higher royalty rate MTP products. For the new IP technologies: one, continued development of next-generation OTP and PUF-based hardware security solution, targeting sub-2nm process nodes.

Two, NeoFlash expanding in both embedded and standalone 1T flash applications, leveraging logic process-based architecture for improved scalability and cost efficiency. For business development platforms: one, chiplet security platform.

The company is collaborating with ecosystem partners to develop end-to-end chiplet security solutions, covering key technologies, including the supply chain security, chiplet authentication, secure communication, and hardware root of trust. These solutions are designed to address the licensing security requirement driven by AI and heterogeneous integration architectures.

Two, for data center Caliptra platform, targeting data center and AI server applications, we are developing a comprehensive IP platform that combines a PUF-based root of trust with a security subsystem architecture. This platform enable customers to accelerate the adoption and mass production of hardware security solutions, complying with the Caliptra framework and future data center security standards.

Three, root of trust and CPU platform collaboration. Our hardware security technologies have been successfully integrated into the latest-generation AI/AGI CPU platforms of major CPU vendors, providing chip-level root of trust, secure boot, device identity, and a secure key protection capabilities.

These technologies enhance trustworthiness and secure across the AI system from the chip to system level. Four, HSM edge server platform.

Our PUF-based HSM edge server is positioned as a key infrastructure component for future Security-as-a-Service, SECaaS, platforms. The platform supports critical applications, including secure OTA updates, privacy protection, device identity management, and post-quantum cryptography migration for the automotive, medical, industrial control, edge AI, and other high-security systems, helping the customer build a sustainable, next-generation security architectures.

This concludes my comments. I will now hand over to Felix, our Head of Digital Marketing.

Thank you.

Good afternoon, everyone. Thank you for being here.

The AI industry has spent the last few years building agents that can act on the world, but not enough time building the foundation that makes those actions trustworthy. That serious need for security is what we're going to talk about today.

An AI agent is not just a chatbot. A chatbot answers; an agent acts.

It runs in a loop. First, it gathers context; then, it reasons; it decides; it executes in the real world; and it records the result.

It provides status update, then it does it again and again. There are three key things we need to know about this loop.

First, it's autonomous. The agent decides for itself; there's no human pressing "Approve" between steps.

Second, it runs at machine speed. We're talking over a thousand decisions per minute.

No compliance officer, no security team, no human reviewer can keep up with that in real time. It is mathematically impossible.

Third, this is an important point, different steps carry vastly different stakes. When the agent is reasoning internally, a mistake is recoverable.

The moment that the loop touches the external world, which involves money, contracts, identity, infrastructure, the stakes become absolute. There's no "Undo" button on a wire transfer, and there's no rollback on a signed contract.

At every single step of the agent loop, there are two distinctive categories of risk. The first one is integrity risk: whether the agent is thinking with the right information, whether it's being fed false data, whether its reasoning engine is the one we deployed or has been silently swapped, whether its memory has been tampered with.

The second one is authorization risk: whether the agent is allowed to do what it's about to do, whether the action was approved by a human, whether the agent is even who it claims to be. They are fundamentally different problems, and here's what the market has missed, because most of the security industry today only address one of them.

Usually, it's integrity through software signing and model verification. However, the authorization side, the matter who has the right to act, is largely being handled by software policies that an attacker or even a misaligned agent can route around.

You cannot have governance without solving both. A perfectly authentic model executing an unauthorized transaction is still a catastrophe.

An authorized agent running tampered logic is still a catastrophe. Both columns have to be risk-free.

Today, in production AI deployment, they're not. Every major regulatory framework now being written, including the EU AI Act, the emerging US governance frameworks, the financial sector guidelines, they all converge on the same four requirements.

First, it's a human-in-the-loop approval for high-stakes decisions. Then, there's tamper-proof audit logs, protection of models from theft, and finally, zero-trust authentication on every interaction.

Every AI vendor in the world will tell you they offer these things, and every one of those promises in current implementations has the same fatal weakness: it lives in software. In the case where PUF is absent, having software is not sufficient.

As shown in the right column, software signatures can be forged, software audit trails can be rewritten by anyone with sufficient privilege, software keys can be extracted, software identity can be impersonated. What this means in practice is that today's AI governance is, to use a phrase we keep hearing from CISOs, "governance theater".

It looks compliant on paper; it does not survive contact with a determined attacker. The only way to make these promises enforceable rather than aspirational is to follow the middle column: knowing what governance requires, and anchoring the promises in something that we know the exact identity of, with signatures that cannot be copied or extracted, keys that cannot be stolen, which, essentially, is describing hardware.

In other words, it has to be a property of the silicon itself, unique to each chip, generated at the moment of need, and never stored anywhere it can be stolen. All arrows are pointing to PUF.

Thus, we, eMemory, provide the foundation layer, the hardware root of trust, and on top of that foundation, three properties become possible, the properties that no software-only stack can deliver. First is identity.

Every AI agent, every device, every chiplet, every endpoint gets an unforgeable identity that is intrinsic to its silicon. It cannot be cloned because the physical randomness it derives from cannot be replicated, even by the foundry that made the chip.

Next is integrity. Models, prompts, and agent logic stay sealed to the hardware that's authorized to run them.

If you steal the model file, then it's useless, blob of encrypted bytes. There's no key file to steal alongside it because the key is never written down.

Then, finally, there's authority: human-in-the-loop approval, audit logs, policy enforcement. All of these regulators are mandating become cryptographically enforceable rather than procedurally promised.

The audit log is no longer what the system says happens; it is provable in court, if necessary, by anyone with the public verification keys. The AI governance platforms, the agent orchestration tools, the compliance dashboards, these are all valuable, and they are all building real businesses.

EMemory is positioned at the hardware layer that everything else depends on. In summary, by moving security from the software layer to the silicon itself, eMemory enables agentic AI to operate with the level of autonomy and authority required for high-stakes industries like finance, healthcare, and autonomous infrastructure.

That includes my talk. Thank you for your time.

Thank you, Felix. This concludes our prepared statements.

Next, we will enter the Q&A section. We will now begin the Q&A section.

Please submit your questions in the input box on the webcast window. All of our questions will followed by the format of answering the Chinese version first, followed by the English version.

We will now collect the questions and begin our Q&A section. 第一題是，就是公司4月營收公告，那個權利金的部分比前一季，呃，就是應該跟1月比，呃，有掉一點點。那foundry一般來講，就是我們看一般那個公告的foundry的營收，它第一季都大部分都是比前一季增加，那也有聽到漲價，那請問這有什麼特別的原因嗎？請Michael回答。

我們目前聽到foundry漲價的部分，一般都是從第二季或下半年才開始，那在第一季還尚未反映出這個部分。那在我們第4月份權利金的下滑，那主要是受到某一家台灣代工廠的客戶，他賣廠調整產能，那那個廠剛好又是我們終端客戶的一個應用產線，所以造成了一個短暫性的權利金下滑。那這個部分隨著我們的晶片客戶移往其他的廠去做量產，那就會慢慢去恢復了。謝謝。

Royalty revenue in April declined compared with the previous quarter, while foundry revenues generally increased quarter-over-quarter in Q1. Together with the potential impact from foundry price increases, is there any particular reasons for this difference?

Michael, please.

Based on what we have heard, foundry price increase generally started from the second quarter or expect to take effect in the second half of the year. Therefore, the impact was not yet reflected in the first quarter royalty.

The decline in royalty revenue was mainly due to a temporary disruption caused by the sale of one Taiwan-based foundry customer. That foundry happened to be the production site for certain end-customer applications, which led to a short-term decline in royalty contribution.

We expect this impact will be temporary as the chip customers shift production to other foundries, the related royalty revenue should gradually recover.

好，Q1的權利金反映的是去年Q4在代工廠的這個出貨，那目前他們還是沿用去年的代工價格。而我們的這個晶圓代工廠的客戶的調漲，是從今年第二季陸續開始的，所以如果要反映到我們的權利金的話，應該會是在下半年。

What was the reason behind the decline in 8-inch royalty revenue in the first quarter? Will the recent increase in 8-inch foundry pricing provide support going forward?

Michael, please.

The royalty revenue recognized in Q1 reflect the foundry shipment from the fourth quarter of last year, and therefore still reflect last year's foundry pricing. Since foundry price increase are being implemented gradually this year, the impact of our royalty revenue is expected to be reflected in the second half of the year.

SRAM重要性越來越高加上die size變，就是它的density變很大，所以應該是更有誘因用我們這種IP來repair。那請問公司在這領域AI相關案子的發展狀況怎麼樣？那手邊大概有多少案子？那今年有沒有可能看到權利金的收入？我們請Michael回答。

我們的OTP在SRAM repair裡面一直都是扮演很重要的個角色。那之前也提到我們有跟西門子合作，整合他們的Tessent的軟體，那推出驗證過的一個解決方案。那這個可以簡化設計流程，加快客戶的導入。那目前公司在AI相關的SRAM repair已經有多個design-in，那主要集中在像AI data center的processor、memory controller、CXL等相關的晶片，而這些都是在先進製程的平台。隨著這些設計定案，慢慢進入量產，那我們相信權利金的貢獻會越來越明顯。

As SRAM becomes increasingly important and die size or density continue to grow, there is a stronger incentive for customers to adopt repair IP. Could you share the company's development progress in AI-related projects in this area?

How many projects are currently in the pipeline, and could we start to see royalty contributions in the coming years? Michael, please.

Our OTP has long played an important role in SRAM repair applications. We have already worked with Siemens to integrate our technology with their Tessent software, offering the validated solution that helps simplify the design process and accelerate customer adoption.

Currently, the company has multiple AI-related SRAM repair design-ins, mainly focus on the advanced-node chips, such as AI data center processors, memory controllers, and CXL-related devices. As these designs are finalized and move into mass production, we expect the related royalty contribution to become increasingly visible.

可以多介紹公司1T flash架構跟技術的發展，是包括NOR跟MLC都可以用這個技術生產嗎？那有沒有相關代工廠跟IC設計客戶正在跟公司接洽？我們請Charles回答。

好，我們最近提出來的這個1-transistor的NeoFlash是採用這個band-to-band high electron program的架構，那我們把這個select transistor移除掉，所以就變成一個1-transistor的這個flash cell，藉此來顯著的縮減這個記憶體的單位面積。那基本上最重要的是這樣的一個flash是利用這個logic process做出來的，那它的好處是用logic process做出embedded flash跟standalone flash的應用。那因為是邏輯製程，所以這個development cycle也很短，然後成本低，然後良率高，而且那個產能大。這樣的優點基本上是一個很突破性的創新。如此一來，那個embedded flash還有standalone flash都很容易可以取得，而且可以用現有的logic process就可以做出NOR flash，還有這個2D的這個NAND flash，還有這個multi-level的flash cell。目前已經進入實質的開發階段，而且已經有這個幾家foundry還有fabless正在進行合作。

Could you provide more details on the development of the company's 1T flash architecture and technology? Can this technology be used for both NOR and multi-level cell or MLC applications?

Are there any foundries or IC design customers currently engaging with the company? Charles, please.

1T NeoFlash adopts an innovative array architecture that overcomes the limitation of the conventional 2T design and remove the address select component, significantly reduce memory cell area. Essentially, this is a flash technology built on the standard logic processes, and its key advantage include the ability to support both embedded flash and standalone flash, a shorter development cycle, low cost, and higher yield, and a greater manufacturing capability.

We believe this represents a significant innovation as both embedded and standalone flash can be more easily implemented using existing logic process platforms. The technology can support NOR flash, 2D NAND flash, and multi-level flash cell and all MLC applications.

The technology has now entered the active development stage, and we are already cooperating with several foundries and fabless companies.

請問做embedded的記憶體的技術跟standalone記憶體最大的差別在哪裡？因為我們都知道我們20幾年來都是在做embedded。那這樣的know-how它可以直接apply到做standalone嗎？那另外potential的客戶群是哪些？那我們請Charles回答。 好，就是好，那沿著那個上次的上一題的回答，那對於這個embedded跟standalone

memory的差別在哪裡？那主要是對standalone的memory最重要就是那個bit cell要小，然後那個製程要簡單，然後良率要高，因為它跟這成本有關係。那力旺電子長期以來以這個logic non-volatile memory作為核心技術平台，協助客客戶在先進以及成熟製程當中同時同時達成這個快速開發、提升良率還有降低成本等關鍵的目標。那我們技術平台的核心優勢呢，包括快速開發，因為採用這個標準邏輯製程，即可實現非揮發性記憶體，不需要大幅修改製程製程的流程，也不需要改變既有的元件的模型，因此可以大幅的縮短開發與導入的時程。那第二點就是可以提升良率，因為完全相同相容於這個標準的邏輯製程，不增加複雜特殊元件以及這個額外關鍵步驟步驟，因此可以更容易維持高良率跟穩定的這個量產能力。那第三點這個降低成本，因為製程相對簡單，cycle time短，然後良率會比較高，可以有效降降低整個那個晶圓製造跟系統的成本。那在這個1T的這個logic flash的技術平台上面，我們更能以相同的核心記憶體技術同時可以支援這個embedded flash跟這個standalone flash兩大這個產品方向。事實上，embedded flash跟standalone flash在memory array的核心設計相當接近，最最大的差異在於這個系統的介面。Embedded flash的介面會依SOC系統需求進行客製化的設計，那standalone flash則是採用業界標準化的介面，那以支援各種系統平台以及這個產品應用。因此，我們的這個1T的記憶體的技術平台不僅能夠同時支援embedded flash跟standalone flash，也具備發展NOR flash跟那個NAND flash的架構的潛力。更重要的是，由於這個技術呢可以直接利用現有的邏輯製程的產能進行生產，不需建立傳統的flash專用的產線跟額外大規模資本支出，因此在全球flash供應鏈重組以及這個AI時代來臨之際的時候呢，具有非常大的這個發展潛力跟這個策略的價值。未來應用市場也將會十分廣廣泛，包括AI server、edge AI、智慧手機、電動車、工業控制、網通設備以及各類高安全性及高效能的運算系統。

What are the key differences between embedded memory technology and stand-alone memory products? Can the know-how we have accumulated in embedded memory be directly applied to stand-alone memory?

Also, who are the potential customers? Charles, please.

eMemory has long built its core technology platform around logic non-volatile memory. This platform helps customers achieve a faster development, higher yield, and lower cost across both advanced and mature process nodes.

The key advantages of our technology platform include three area. First, faster development.

Since our technology can be implemented using standard logic process, it does not require major process change or modification to the existing device model. This helps significantly shorten the development and also adoption cycles.

The second is higher yield because the technology is fully compatible with standard logic process and does not require complex special device or additional critical process step. It is easier to maintain higher yield and stable mass productions.

The third point is low cost. With a simpler process flow and a shorter cycle time, and also higher yield, overall wafer manufacturing and system costs can be effectively reduced.

On our 1T logic flash platform, the same core memory technology can support both embedded flash and also standalone flash. In fact, the core memory array design is quite similar between embedded and also standalone flash.

The main difference lies in the system interface. For embedded flash, the interface is customized based on the requirement of each SoC system.

For standalone flash, the interface follows industry standard in order to support a wide range of system platform and also end applications. Therefore, our 1T memory platform can support both embedded and standalone flash, and also has the potential to extend into both NOR flash and also NAND flash architectures.

More importantly, because this technology can leverage existing logic process capacity, it does not require customers or partners to build dedicate traditional flash production line or make large-scale additional capex investment. This gives the technology significant development potential and strategic values, especially as the global flash supply chain evolves and also AI-related demand continues to grow.

The potential application are very broad, including AI servers, edge AI, and smartphones, electrical vehicle, industrial control, networking equipment, and various high-security and also high-performance compute systems.

傳統的IP公司像是Arm、Rambus、Alphawave都朝向做晶片產品。請問公司做standalone的記憶體是否也是相似？那這會是未來的趨勢嗎？Charles。 我想對力旺來講的話，我們會持續做這個IP的business，我們不會進入這個做產品。但是即使做產品也不是去賣產品，而是把這個產品變成這個IP。那至於剛上面上述講的這些IP公司呢，他們都基本上是屬於IP設計公司，主要是基於既有的製程跟現成的電晶體架構來開發IP產品。但是力旺電子的核心的競爭力則是在這個我們的記憶體技術本身的發明跟這個平台化的這個能力。我們不只是設計IP，而是更從是從這個底層的記憶體元件，還有製程整合到架構技術的創新，建立完整的技術平台。因此，我們除了可以自己開發並銷售IP之外，更可以將核心記憶體技術授權給其他家公司，由合作夥伴進一步的發展成為IP或是終端產品。在商業模式上面，IP業務部分我們會持續自行經營，以維持我們技術領先跟這個市場影響力。而在產品端，我們則會透過技術授權的方式，讓更多的合作夥伴參加與這個產品化跟量產。我們則依據這個產品出貨收取權利金。由於產品授權的價值鏈更接近這個終端市場，其權利金的規模通常也會明顯高於傳統的IP授權模式。這也是力旺一直以來成長的重要方向，也就是我們持續發展成為具備核心技術平台與產品生態系影響力的這個技術授權公司。

Traditional IP companies such as Arm, Rambus, and Alphawave have been expanding into chip products. Is eMemory moving into stand-alone memory similar?

Could this become a future industry trend? Charles, please.

The companies mentioned mainly above are mainly IP design company. They typically develop IP product based on the existing process technology and standard transistor architectures.

The eMemory's core competence, however, lies in the invention of and platformization of the memory technology itself. We are not only design IP; we innovate from the underlying memory device, process integration, and architecture level to build a complete technology platforms.

As a result, in addition to developing and sell IP by ourself, we can also license our core memory technology to other companies, allowing partners to further develop them into IP or end products. From a business model perspective, we will continue to operate our IP business directly in order to maintain technology leadership and the market influence.

On the product side, we can enable more partners to participate in the productization and mass production through a technology licensing, while eMemory collects royalty based on the product shipments. Since product licensing is closer to the end market in the value chain, the potential royalty scale is typically much larger than that of a traditional IP licensing.

This has always been an important growth direction for eMemory. We are focus on the evolve into a technology licensing company with a strong core technology platform and a broader influence across the product ecosystem.

今年營收成長主要會是來自於哪方面？請Michael回答。 今年除了既有的應用做upgrade所可以帶來這個權利金單價的提升，譬如像我們在OLED的DDI已經進到16奈米FinFET提供給像摺疊式手機使用，那在PMIC的話也從主要的8吋轉到像55奈米的BCD。那另外新的應用如手機的modem的相關的晶片，那AI

CPU、BMC、embedded controller、SSD controller、CXL controller、DIMM，那跟高速網通介面相關的應用也會陸續的進入量產來帶動。

What are the key drivers of revenue growth this year? Michael, please.

Revenue growth this year is driven by both upgrades of existing applications and the ramp-up of new applications. For existing applications, product upgrades have led to a higher royalty ASPs, such as OLED DDI for foldable smartphone at 16nm, and the PMIC migration from 8-inch process to 55nm.

At the same time, several new applications are gradually entering mass production, including the smartphone modem-related chips, AI CPUs, BMCs, embedded controllers, SSD controllers, CXL controllers, DIMMs, and high-speed networking interface applications. These new ramps are expected to further support the revenue growth.

成熟製程代工廠紛紛漲價，但是消費手機市場需求還是很疲弱。那請問公司對價格是否能有效上漲，還有持續性有什麼看法？我們請Joseph回答。 好，那這個成熟製程價格的調整我們認為是有一定的結構性因素。那比如說這個龍頭的晶圓廠把資源投在這個最先進的製程以及封裝相關，那這會間接的去排擠及減少成熟製程的一些有效產能。那除此之外，AI相關不管是像data

center、edge AI或是physical AI裡面其實還有非常多的產品是在成熟製程生產，那比如說我們的這個高壓大電流的power，這也需要大量的這個成熟製程的產能。那還有就是最近大受矚目的這個CPO的架構中，通常是只有最核心的交換器晶片或是DSP為了極致的這個運算效率會採用5奈米、3奈米這樣子的先進製程，而周邊還是有非常多的光學以及類比的組件則是維持在這個成熟製程來進行量產，那以達到比較平衡的這個性能以及成本。那比如說我們在以色列的一個晶圓廠客戶，就把大量的這個power相關的業務轉介給中國的這個三線晶圓廠，那就把資源去集中投資在這個開發CPO相關的製程。

As mature-node foundries raise prices while consumer smartphone demand remains weak, how does the company view the effectiveness and sustainability of this price increase? Joseph, please.

We believe mature-node price adjustments are supported by structural factors. First, leading foundries are allocating more resources to the most advanced process nodes and advanced packaging, and this indirectly reduces the effective capacity available for other mature nodes.

Second, AI-related applications still require a significant amount of mature-node capacity, whether in data center, in edge AI, or in physical AI. Many supporting components, such as high-voltage and high-current power devices, are still being manufactured on mature nodes.

I think a good example is the recent development of CPO architecture. In a CPO architecture, usually only the most critical components, such as switch ASICs or DSPs, they adopt advanced nodes like 5nm or 3nm for maximum computing efficiency.

Many surrounding optical, analog, and power-related components, they continue to use mature nodes to achieve the best balance of performance, power, area, and cost. We have also observed that some foundry customers are reallocating capacity and resources.

For example, a certain power-related businesses are being shifted to other foundry partners, while more resources are being directed towards developing CPO-related processes.

隨著先進製程跟高效能運算需求提升，公司在Arm的平台上的相關應用是否有所否有看到更多的導入機會？請Charles回答。 好的，謝謝。就隨著這個AGI

CPU的量產，先進製程CSS，就是Compute Subsystem的客戶對於這個root of trust的需求明顯提升，那公司已經有多個客戶導入相關的這個解決方案。

With the continued advance of advanced nodes and growing demand for high-performance computing, are you seeing more adoption opportunity for your solutions on Arm-based platforms? Charles, please.

As AGI CPUs move into mass production, demand for root of trust solution among advanced-node CSS customers has increased significantly, and we have already seen multiple customers adopt related solutions.

可否分享公司跟Intel Foundry的合作進展？Joseph。

我們和Intel Foundry的合作目前主要聚焦在先進製程的OTP還有PUF-based security IP。那我們和Intel現在正在積極的導入PUF-based IP到Intel的18A，那去因應這個美國政府對於供應鏈安全的一些需求，謝謝。

Could you share the progress of the company's collaboration with Intel Foundry? Joseph, please.

Yes, our collaboration with Intel Foundry is mainly focused on advanced-node OTP and PUF-based security IP. We are actively working with Intel to bring our PUF-based IP onto Intel 18A to address the supply chain security requirements from the US government and the related markets.

Thank you.

DRAM模組相關應用的潛力如何？那大概是在哪些產品線？Joseph。 DRAM模組相關的應用對於我們的OTP、MTP以及security

IP都是蠻重要的一個成長機會。原因是AI server帶動DDR5、CXL memory、HBM以及這個高速記憶體模組需求的一個增加，而這樣子產品都需要更高的這個良率管理、模組設定以及安全的保護。那例如說我們OTP可以主要應用在這個修復以及識別，那例如說做這個memory的repair；那MTP可以主要用在這個更新、設定以及模組的管理，例如說DDR5 DIMM的SPD Hub以及PMIC相關的設定；那這個security IP主要就是用在資料安全以及這個硬體的信任根。那DRAM的模組不再只是記憶體，是會變成這個AI server或是CXL memory pool以及資料中心架構中的一個關鍵節點，那因此他們的安全需求也會相對的提升。

What is the potential from DRAM module-related applications? Which product lines would benefit the most?

Joseph, please. Yes,

DRAM module-related applications represent an important growth opportunity for our OTP, MTP, and security IP businesses. This is mainly driven by AI servers, which are increasing demand for DDR5, CXL memory, HBM, and high-speed memory modules.

These products require stronger yield management, module configuration, and security protection. For our OTP, the main application are repair and identification, such as memory repair.

For MTP, the main applications are updatable configuration and module management, such as SPD Hub and PMIC setting in DDR5 DIMMs. For security IP, the main applications are data security and hardware root of trust.

Going forward, DRAM module will no longer be viewed simply as memory component. They will become key nodes in AI servers, in CXL memory pools, and in data center architectures.

Therefore, as a result, the security requirement for DRAM will continue to increase. Thank you.

看公司發展這麼多的技術，是否需要大幅投入研發資源而導致營業費用大增？Joseph。 我們公司的技術都是靠這個長時間的這個積累，那而且是基於我們既有的核心基礎技術往下具延伸或是拓展。因為我們的技術都是需要這個代工廠以及晶片公司來實現，會由我們這個合作夥伴來支付這個授權金，cover我們研發相關的一些費用。那研發過程的一些資本支出，例如說測試晶片及相關設備的投入也是由客戶間接的來負擔。那我們過去就是一直以這樣子的商業模式來進行，那這個也彰顯我們有很強的技術核心競爭力，而且我們的技術必須要為客戶帶來更大的一個價值，那這樣子的商業模式才可以成形。那由於現在開展的技術以及研發還有專利權都已經完成，那後續相關的費用則是這個專案執行相關的這個研發人員，那授權案會帶動我們的授權金，那是足以cover這個相關的研發投入，那我們認為營業費用還是在非常可控的一個範圍之內。

As the company expands into more technologies, will this require significant R&D investment and lead to a sharp increase in operating expenses? Joseph, please.

Yes, most of our technologies are built on years of accumulated know-how and are extensions of our existing core technology platforms. Our technology are ultimately implemented by foundries and chip customers.

Under our business model, customers and partners, they pay licensing fees to help cover the cost of our R&D teams. In addition, capex during the development process, such as test chips and related equipment, are typically borne by customers indirectly as well.

This has been our long-standing business model. It also reflects the strength of our core technology and the significant value our technology bring to the customers.

Otherwise, this model would not be sustainable at all. For the technologies we are currently expanding into, much of the core R&D and patent development has already been completed.

Going forward, the additional expenses will mainly come from execution-related R&D headcount. As licensing projects increases, the related licensing revenue should also be sufficient to support the required R&D investment.

Therefore, we do believe the OPEX will remain within a manageable range. Thank you.

公司在新一代AI運算平台，比如說Vera Rubin

AI platform相關應用的進展如何？Joseph。 這個Vera Rubin平台是第一個導入2024年3月OCP所發布的這個Caliptra 1.0的運算世代，那裡面是第一次把OTP跟PUF放到設計規範裡面。那也因為這個規範，我們開始把我們的技術導入CPU、BMC、SSD controller、CXL controller以及網通相關的一些晶片。那OCP的規範其實也不僅限於這個Vera Rubin的一個架構，它其實也是用在這個雲端的服務商啊或是這個電信業者以及大型企業的一些私有雲的資料中心。那在這個之後，OCP又發布了包含這個PQC的Caliptra 2.0，以及開場我們董座所提到的這個Foundation Chiplet System Architecture，都有明確的去要求這個chiplet系統中的每一顆晶片都必須要具備這個hardware root of trust，那來因應市場對於統一的硬體安全架構、持久性的資料管理相關的一些需求提升。那我們認為在這樣子的趨勢之下，包含像晶片修復啊、這個資料儲存啊或是硬體信任根的功能都會逐步的去導入到不同的運算元件當中，那去朝向一致的一個架構來實現。那我們可以預期這個我們在AI領域的一個滲透率會隨著這個世代的演進逐步的去放大。

What is the company progress in next-generation AI computing platforms such as the Vera Rubin platform? Joseph, please.

Yes, the Vera Rubin platform is the first computing generation to adopt Caliptra 1.0, which was released by OCP back in March of '24. This is also the first specification to include OTP and PUF into the design requirements.

Based on this specification, our technologies have started to be adopted in CPUs, in BMCs, in SSD controllers, CXL controllers, and other networking-related chips. More importantly, OCP's specifications are not limited to Vera Rubin architecture itself.

They are also applicable to other cloud service providers, to telecom operators, to large enterprises building their own private cloud data centers. After that, OCP further released Caliptra 2.0, which includes PQC, as well as the Foundation Chiplet System Architecture, which was mentioned by our Chairman in the beginning of this earnings call.

That clearly requires a hardware root of trust in their standards as well. This reflects the market's growing need for a unified hardware security architecture and persistent data management.

Under this trend, functions such as chip repair, data storage, and hardware root of trust are gradually being adopted across different computing components, and that we believe moving forward towards a more consistent architecture. As each new generation of AI platform evolves, we do expect our penetration in AI-related applications to continue to expand over time.

Thank you.

公司security往系統級發展，請問business model為何？哪些是潛在客戶？Joseph。

在這個硬體安全系統的發展上，我們核心的策略主要有兩個方向：那第一個是以PUF為核心的基礎技術，那發展新世代的hardware security module，我們把它叫做這個HSM；那第二個部分是進一步的以HSM為安全的平台，那發展我們Security-as-a-Service相關的一些應用以及服務。那通過這樣的架構，客戶可以去adopt這個具備高度安全、高可信度的硬體安全系統，那不僅能夠保護裝置以及系統本身的運作安全，那也可以延伸到像AI通訊、這個資料保護、身分認證與雲端服務等等各種應用層面的安全需求。那在這樣子的商業模式方面，力旺電子將持續的以技術授權為核心，那授權HSM平台相關的安全技術給OEM或是這個系統廠進行產品化以及量產，那並依據這個終端產品的銷售金額來收取相應的一個權利金。那我們目標的客戶將會涵蓋這個電信公司、硬體資安設備的廠商，或是edge AI OEM廠商、工業電腦、AI基礎的建設的供應商，那以及未來對於這個PQC以及硬體root of trust有高度需求的各類AI以及雲端系統的一些業者。謝謝。

As the company expands its security business toward system-level solutions, what would the business model look like? Who are the potential customers?

Joseph, please.

Yes. In the development of hardware security systems, our core strategy has two main directions.

First, we are using PUF as the core foundational technology to develop our next-generation hardware security modules, which we call the HSM. Second, we plan to use HSM as a security platform to further develop Security-as-a-Service-related applications and services.

Through this architecture, customers will be able to build hardware security systems with a high level of security and trust, and these systems can protect not only the operation of devices and systems themselves, but also broader application-layer security needs, including things like AI, like communication, data protection, identity authentication, and cloud services. In terms of business model, eMemory will continue to focus on technology licensing.

We plan to license the HSM platform and related security technologies to OEMs and to system companies for productization and mass production, and we'll be collecting royalties based on the end-product sales. Potential customers include telecom operators, hardware security equipment providers or edge AI OEMs, industrial PC, AI infrastructure providers, and various AI and cloud system operators that will be having a growing demand for the PQC and the hardware root of trust.

Thank you.

In the interest of time, we'll begin the last questions. 公司在系統級硬體安全業務的發展現在進展是怎麼樣？那未來hardware secure

module或者是security as a service相關服務我們應該要如何看待？Joseph。 好，那就延續上一題繼續回答。那公司在這個硬體安全領域的一個發展，正在從既有的IP授權逐步去延伸到系統層級的安全解決方案。那整體架構可以分為三個層次來看：包含第一個是底層的HSM核心運算，第二個是中間的這個金鑰以及憑證管理，以及最上層的應用服務。那在底層的HSM核心運算方面，公司正在以PUF技術作為這個硬體信任根的基礎，去發展不同形態的HSM產品，那包含像是這個個人裝置、中小企業的合規需求，以及這個高階網路型的HSM應用。那未來相關的產品會去進一步的整合PQC以及PUF技術，那並以這個符合國際資安認證要求為目標，那就強化我們產品在高度安全的應用場景中的一個可信度以及市場的接受度。那第二部分是在這個中層的金鑰以及憑證的管理方面，那公司也正在與外部憑證以及PKI的生態生態系的夥伴進行合作，那就將我們底層的硬體安全能力以及憑證管理系統去進行個整合。那未來會提供更完整的這個HSM加PKI的解決方案。那這個將有助於客戶在像是身分驗證、金鑰管理、數位簽章以及系統認證管理上的應用方面，更快速的可以去導入我們這個硬體的安全架構。那最後是在最上層的這個應用服務方面，那公司已經完成了這個Signing-as-a-Service的平台開發，那未來可以支援這個數位簽章、裝置身分認證、資料的可信度驗證以及DID wallet等等的一些延伸的應用。那這也代表公司未來不僅可以提供硬體安全的一個IP，那也有機會進一步的去參與這個Security-as-a-Service相關的一個服務模式。那整體而言，HSM和Security-as-a-Service是公司從IP供應商往系統系統安全解決方案供應商延伸的一個重要方向。那短期我們仍然以這個技術整合以及應用驗證為主，那中長期則有機會去透過HSM的平台授權、PKI的整合方案以及上層的安全服務去形成新的一個商業模式以及成長的動能。謝謝。

What is the company's progress in system-level hardware security? How should we think about future HSM or Security-as-a-Service opportunities?

Joseph, please.

Yes, okay. The company's development in hardware security is gradually expanding from our existing IP licensing business into a system-level security solution.

Overall, we can look at the architecture in three layers: first, the HSM core computing layer at the bottom; the key and certificate management layer in the middle; and the application service layer on the top. At the HSM core computing layer, we are using PUF technology as the foundation for hardware root of trust.

Based on this foundation, we're developing different types of HSM products to support personal devices, compliance needs for SMBs, and high-end network HSM applications. Going forward, these products will further integrate PQC and PUF technologies with the goal of meeting international cybersecurity certification standards and requirements, and this will help strengthen the product credibility and market acceptance in high-security applications.

At the key and certificate management layer, we are also working with external certificate and PKI ecosystem partners. The goal is to integrate our underlying hardware security capabilities with certificate management systems and provide a more complete HSM plus PKI solution.

This will help customers adopt hardware-based security architecture more efficiently for applications like identity authentication, key management, digital signatures, and system trust management. Lastly, at the application service layer, we have completed the development of our Signing-as-a-Service platform.

In the future, this platform can support applications such as digital signature, such as device identity, trusted data verification, and DID wallets. This means the company can not only provide hardware security IP, but also has the opportunity and capability to participate in the Security-as-a-Service business model.

To conclude the things, overall, the HSM and Security-as-a-Service represent an important direction for the company as we expand from an IP provider into a system-level security solution provider. In the near term, our focus remains on technology integration and application validation.

Over the medium, long term, we see the opportunities to build new business models and growth drivers through our HSM platform licensing, through PKI integration solutions, and upper-layer security services. Thank you.

We will begin the closing comments. Charles, please proceed.

Hey, thank you once again for your patient and the support for eMemory. We will continue to work hard on technology and the IP innovation, and the PUF-based hardware security solution for our customers, and also bring a higher return for our shareholders.

Thank you.

Thank you, ladies and gentlemen. Please be advised that the conference recording 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.