CATBeamTech has expanded its business beyond medical X-ray solutions into the semiconductor industry. Its core technology leverages Carbon Nanotube (CNT)-based X-rays to identify internal defects in High Bandwidth Memory (HBM).

CATBeamTech is currently conducting Proof of Concept (PoC) evaluations for its CNT X-ray equipment with domestic and international memory manufacturers. Upon receiving HBM sample wafers from memory makers, CATBeamTech images cross-sections to inspect for defects at specific locations. During performance verification, potential clients provide a mixed batch of normal and defective samples, and the equipment's performance is evaluated by comparing its classification results against the client's answer key.

HBM features a vertically stacked structure of DRAM chips in 4, 8, 12, or 16 layers. CATBeamTech identifies defects such as micro-bump failures connecting individual DRAM dies, die lifting, and warping.

As the number of stacked layers increases, early inspection becomes critical. Discovering a defect after completing the stacking process results in financial losses that compound across all preceding manufacturing steps. By identifying anomalies mid-process, CATBeamTech helps minimize these losses.

The observed defects are quantified: if the defect count exceeds normal thresholds, the system flags an anomaly for process management. "In the industry, there are rumors that certain memory makers' HBM yield is below 50%," said Je-hwang Ryu, CEO of CATBeamTech. "By tracking defect counts inside HBMs, we can help manage overall yield."

X-ray inspection is a non-destructive semiconductor testing method used to check for internal HBM defects. Through sampling inspection, process errors can be detected and addressed early. The system scans key points on a wafer to evaluate overall alignment and can trigger a secondary, high-precision inspection for suspicious areas. Companies like SEC and Zeiss also produce similar equipment.

CATBeamTech's primary advantage lies in its ability to build fully customized equipment. Parameters such as inspection speed and detectable defect size can be adjusted to client specifications because the company directly manufactures its own equipment, components, and CNT sources.

The equipment can detect defects ranging from 10 to 100 nanometers. According to CEO Ryu, competitors typically observe defects at the 500-nanometer level, giving CATBeamTech a precision advantage of 5 to 50 times. However, smaller defect detection involves a trade-off, such as longer inspection times. For commercialized products, CATBeamTech is developing technology optimized for 100-nanometer-level defect detection, balancing speed and resolution.

X-ray inspection works by transmitting electron-generated X-rays through the target object to observe its internal structure. CATBeamTech utilizes carbon nanotubes (CNTs) as its electron emitters. Traditional X-ray devices emit electrons by heating a filament, whereas CNTs emit electrons via an electric field without requiring thermal heating. Compared to filament sources, CNTs offer faster inspection speeds and are easier to miniaturize. They also enable precise output control, varying X-ray power based on the thickness and density of the target material.

With its CNT-based X-ray technology, CATBeamTech was selected by the Ministry of Small and Medium-sized Enterprises and Startups (MSS) as a promising semiconductor startup under the 'Deep-Tech Startup 1000+' project. Selected companies receive funding from the MSS for commercialization and R&D.

Q&A

Q: What is the Deep-Tech Startup Project?

A: It refers to sectors that the Ministry of Small and Medium-sized Enterprises and Startups (MSS) considers deep-tech, which includes semiconductors, defense, and healthcare. We were selected as a deep-tech company in the semiconductor sector, where they provide us with support. It allows us to further refine and systemize various technologies.

Q: Does the government give you money?

A: They give us funding, and they also serve as a bridge. They provide great guidance in terms of networking and connecting us with client companies.

Q: How much money do they give?

A: Around 1 billion KRW. We originally started with medical and component technology, so the industrial side was somewhat unfamiliar and was an area we didn’t know well. The project provided a great opportunity for us to enter this market rapidly.

Q: Were you selected because you already possess deep-tech technology? Or were you selected as one of the candidates to create deep-tech technology?

A: They look at both. Core technological capability is essential, and a company must possess materials or component technology. Furthermore, the derivative directions and applications must fall under deep-tech criteria to be selected.

Q: Could you please provide a brief introduction to your company and its proprietary technology?

A: Our company is a materials, components, and equipment company. For materials, we directly possess Carbon Nanotube (CNT) technology and synthesize CNTs in-house. Using CNTs, applications can extend to displays, X-rays, or even powder forms for other materials companies. However, our initial primary direction was chosen as X-rays.

We also manufacture X-ray components using electrons emitted from the tips of CNTs. Instead of importing X-ray sources, we possess various proprietary platforms. Because of this, just as bathroom, restaurant, and kitchen lightings are all different depending on their application, we can adapt our X-ray sources to fit specific needs—such as what semiconductor feature to look at, how high throughput needs to be, or how small or large of an object we need to observe.

Q: You previously appeared on our interview channel to introduce your CNT source X-ray technology and overall company background—viewers can refer to the link we’ll provide. You’ve used self-manufactured CNT X-ray sources to inspect internal structures and improve yields in sectors like batteries. I hear your recent focus is on HBM?

A: As the AI market grew dramatically, the world became covered in talks about artificial intelligence. With that, the need to look inside the HBMs used within core GPUs has grown significantly. The reason is that low yields are a widely known issue in this market, which also makes memory prices extremely expensive. Memory prices have risen considerably, but looking ahead, chips will be stacked up to 16 layers. If a problem occurs, the whole thing has to be thrown away.

If you build the 1st floor and inspect it, then build the 2nd floor and inspect it, and keep building if there are no issues, it works fine. But if there’s a problem on the 2nd floor and you keep building the 3rd, 4th, and 5th floors, that entire unit has to be discarded. That is why early inspection is currently so necessary.

Q: You compared it to an apartment, where each floor has structural components and plumbing connections. HBM is similar: if micro-bumps are poorly formed, cracked, or voided, electricity won't flow properly and causes failures. Your company's solution can observe this non-destructively, correct?

A: That is correct. Proper stacking means the micro-bumps are in their proper positions. If you look at a cross-section of the bumps using CT, you can count them. If they are misaligned, you won't be able to count the bumps normally in that cross-sectional view. Therefore, if this inspection goes well and allows normal counting across 2, 3, 4, up to 8 layers, the unit is confirmed as good. In short, processing the numerical values measured from images into statistical data—that is the direction our company is taking.

Q: Why do you process it statistically? Isn't it enough to just see if there is a defect?

A: Yes, seeing if there's a problem is the field called inspection. But in our field, metrology, numbers are key. Those numbers are ultimately converted into data linked directly to yield management. Usually, a separate team handles related tasks and manages yield by compiling statistics during device manufacturing.

Q: So inspection is "looking at it," and metrology is "counting it"?

A: Yes, exactly. Because it expresses exactly where and how a problem occurred as a number, it allows for proactive prevention. That difference is why we aim for metrology.

Q: You are a company that uses visual inspection to perform counting technology. One thing I wondered about is: isn't using X-rays to look inside HBM already a widely used technology? Companies like SEC and Zeiss are already in this market. Since everyone uses X-rays, I wondered what your differentiation is. Is it higher resolution, or are there other differences?

A: Obtaining X-ray images, building 3D volume data (CT data), and providing those results are the same. Our strength lies in directly possessing the X-ray source technology itself. It can be compared to a screwdriver: you need to use the exact right tool for the situation to identify problems quickly. This means our key advantage is providing solutions that perfectly match customer customization needs.

Recently, we have also been leveraging Artificial Intelligence (AI) technology to meet customer criteria—shooting images faster, lowering costs, improving visibility, and enhancing our overall customization capabilities. Companies like Zeiss mostly focus on general-purpose equipment. However, our advantage of being based in South Korea allows us to collaborate closely with clients here.

Q: You mentioned that holding the CNT source technology allows you to build equipment in whatever form the customer wants. What specific customer requirements might exist?

A: Even when looking inside HBM, requirements differ. In some cases, they only need to look at micro-bumps, while in others, they need to inspect die lifting or warping. Depending on the case, they might need to use a single X-ray source, five sources, or in extreme cases, around 50 sources. Accurately understanding customer needs and designing matching system structures to deliver solutions is our major point of differentiation from existing companies.

Q: How about speed and maintenance? Regarding lifespan, isn't CNT shorter compared to filaments?

A: Technology has advanced significantly, so lifespans are now at a similar level. Technically, filaments use a heating method, whereas carbon nanotubes can be viewed as digital X-ray sources. Filaments are analog technology, while CNTs are digital. Filaments are bulb technology, whereas CNTs are LED technology. A bulb setup uses only a single source because turning it on and off like digital equipment is difficult, so it stays continuously lit. In contrast, CNT technology can be turned on and off very easily.

Therefore, if multiple sources are required depending on the inspection target, they can be deployed as an array. If a certain area is thick or dense, we can use a stronger X-ray; if density is low, we can use a lower-output X-ray. Offering customized configurations is a massive advantage of CNT.

Q: For example, when getting a CT scan at a hospital, the machine rotates and takes time. Are you saying that instead of rotating the source, you can arrange sources in a circular array to capture images at once?

A: Exactly. X-ray sources used in hospital CT scanners are filaments, meaning there is only one X-ray source inside. Because it generates heat, it requires a cooling system. Then, large power modules called generators are needed.

It also requires high-grade bearings, and when all the mechanical components for maintaining stability are added up, it weighs several tons. When that technology transitions to digital, industrial applications gain the exact same benefits. Our company's starting point was originally to change the paradigm in the medical field, and we are now able to realize those same advantages in industrial fields.

Q: Is it true that your equipment can identify defects down to the 10-nanometer level?

A: We are a company holding technology solutions aimed at reaching the 10-nanometer domain. Competitors' typical specification is around 500nm, or 0.5 micrometers. We possess the core X-ray technology and focal point control technology mentioned earlier. Optical tech is required, and by harmonizing layout, magnification, and vibration control, we can observe down to 10 nanometers.

Q: Are you able to see down to 10 nanometers right now?

A: Yes, although 10 nanometers isn't visible with crystal-clear precision; it ranges between roughly 10 and 100 nanometers. Since other companies are usually around 0.5 microns, we can see much better. There is a market for high-end equipment that maximizes resolution, but equipment used in mass production lines has size limitations. Under those production constraints, 10 nanometers obviously cannot be seen.

Q: How long does it take to inspect a single HBM?

A: It depends on whether it is a 4-layer or 8-layer stack, but inspecting a single point usually takes about 10 to 60 minutes.

Q: What does "a single point" mean?

A: An HBM cross-sectional sample is about 10mm × 10mm in size, and we are looking at one specific point on that sample.

Q: Then clients cannot use this for 100% total inspection; they would use it for sample testing?

A: Correct. If it's a 300mm wafer, it has four quadrants. They can inspect them one by one or two by two to obtain information that represents the overall alignment state.

Q: So it would mainly be used during the development process?

A: Right. Some teams look at it during development, and some look at it at the final stage. I believe it must be used during development to catch yields effectively.

Q: Do you mean they take a rough look first, and if there seems to be a problem, they bring it over to take a precise image of that exact spot?

A: Exactly. The 1st step is a screening inspection, and the 2nd step is a precision inspection. In screening, they might use a fast CT scan or look only with basic X-rays.

Q: If a suspicious area appears under X-ray, do you perform statistical processing on that region?

A: We inspect only the suspicious area. We inspect the representative point, check a certain surrounding area, and then move on to the next process.

Q: Is there high market demand for this?

A: As known in the industry, current (HBM) yields are quite poor. Some companies are rumored to be below 50%. It's not great from a resource efficiency standpoint.

Q: Speaking as a non-expert, wouldn't doing 100% total inspection be more effective for boosting yield? Is total inspection not possible with X-rays?

A: It is possible. However, due to the current AI boom, even if chips are discarded, semiconductor makers' customers have enormous product demand and keep asking for continuous supply. That's why the current ecosystem was formed. But looking ahead at the next 10 years, normal manufacturing processes must include yield management. That way, our great domestic semiconductor makers can maintain their gap over China and latecomers while preparing for the future. We want to contribute with our core analysis and yield management technology.

Q: So you're saying they currently don't do total inspection. Is it because they can't, or because they choose not to?

A: Both. Defects aren't easily visible, and inspecting takes too much time.

Q: But your company's solution isn't a total inspection method either, is it?

A: Right now, we are at the PoC stage—demonstrating the concept and establishing the inspection logic.

Q: What point are prospective clients most interested in? Are they drawn to seeing finer details than others, or is it your proprietary source tech and faster inspection speed?

A: Clients have teams closer to R&D labs and teams closer to production factories. Naturally, R&D teams value precision even if speed is slower, while factory teams say, "It can be less precise, just make it fast." Both needs exist.

Speeding up the inspection makes images slightly less clear or blurry, while spending more time yields clearer images. It's a trade-off.

Q: Who are you currently talking with?

A: We are speaking with all major semiconductor clients, including Korean and US companies.

Q: They must be memory makers. When did discussions regarding HBM begin?

A: We started last year.

Q: How much does a single unit cost?

A: It depends on inspection speed and logistics integration, but prices generally range between 500 million and 3 billion KRW. If it needs to shoot fast while maintaining crisp clarity, it's 3 billion KRW; if it shoots relatively slowly with lower resolution, it's 500 million KRW.

Q: Are PoCs currently underway?

A: Yes.

Q: What does running a PoC entail? Are you receiving wafers and chips to inspect them, or is equipment installed on-site somewhere?

A: Customer requests are like: "We'll give you samples, show us what comes out." They ask for visualization data first.

Q: Do they tell you which specific points to capture?

A: They ask things like: "What can you see?", "Can you count the numbers?", etc.

Q: Can results be improved using AI as well?

A: We need to guide it toward enhancing both inspection precision and speed.

Q: So you receive samples, inspect them, send back the results, and continue that cycle?

A: Yes, we are continuously exchanging data. Inspecting HBM extracted from commercially available GPUs can be freely promoted. What we showed today is a sample from a disassembled Nvidia board we purchased, since client samples cannot be disclosed. Clients ask us questions while already knowing the correct answers.

Q: Meaning if it's a defective sample, the defect should show up?

A: They mix normal samples and defective samples together when handing them to us.

Q: It sounds like solving a puzzle. They evaluate whether you can see it or not?

A: Yes, so they already know the answer.

Q: Do they tell you how many points you got right?

A: Yes, they share the evaluation results with us.

Q: Are you getting 100 points right now?

A: Sometimes we get 50 points, and sometimes we get around 95 points.

Q: Why do the scores vary? Is it because points were captured incorrectly?

A: Sometimes they specify the points, and sometimes they don't. When inspecting at the 10-nanometer scale, there are countless points, so we scan an area of about 20 micrometers within a 100-micrometer field at a resolution capable of resolving 10-nanometer features.

If there's a void, we quickly catch that empty space, or if alignment is bad, we look for areas where bump counting fails. There are various parameters and elements to look at, so we tailor solutions to client requirements.

Q: What score satisfies the customer?

A: Customers unconditionally want 100 points.

Q: Does that mean they aren't satisfied even with 95 points? Why do scores come out at 70 or 90 points?

A: It requires a process. You have to start from 50 to get to 60, and go through 60 to reach 70. No matter how smart we are or how many solutions we have, that process is about getting familiar with customer demands and understanding each other. So even if we hit 50, 60, 70, and eventually 100 points, when they bring another application next time, we can climb the scoring ladder much faster.

Our ultimate target is 100 points. If hitting 100 points makes it impossible to meet speed requirements, we might compromise in the 90-point range.

Q: If it's 100 points, it will obviously be used as a good unit, but whether it scores 90 or 50 points, shouldn't defective units be discarded?

A: They have to be discarded, but the client knows best how much margin they can allow. Having a void doesn't automatically mean the product won't function at all. Looking at the Nvidia B200, the government mentioned acquiring 250,000 units last year. Those cost around 50 million KRW each, with cost of goods sold (COGS) at about 50%. Since eight HBMs are in a B200, dividing 25 million KRW by eight puts one HBM at roughly 3 million KRW.

Q: People don't usually calculate it quite like that, though.

A: Yes, it's a simplified calculation. If an 8-layer stack goes in, we used to buy individual DRAMs for PC memory for 50,000 KRW each. Though it's a different memory type, an 8-layer DRAM stack would technically be a lower-cost item on its own. But when packaged as an HBM unit, it comes to around 3 million KRW. Clients can simply calculate that discarding some units still leaves high profit margins.

Q: You are currently receiving chips, sending back data, and getting scored. When do you expect this to generate actual revenue?

A: I believe that even at 80 to 90 points, clients will adopt it and adjust it internally for application. Actual line deployment might differ.

Q: Are they not using it currently?

A: They do use inspection, but our equipment offers better resolution. That is where our competitive edge lies.

Q: Is it because you build customized systems tailored to their concepts?

A: Yes. They don't reveal exact details, but judging by their reactions, follow-up projects continue. That indicates a relationship we can build upon.

Q: If this leads to actual adoption, how many units could you sell?

A: Selling one unit leads to selling two, and once the client feels confident, they could purchase up to 50 units.

Q: When investors ask if things are going well, what do you tell them?

A: We tell them that we are actively exchanging data with current HBM clients.

Q: When do you tell investors revenue will materialize?

A: Working hard on HBM inspection has created derivative technologies and spin-off products, which are already generating revenue.

Q: What are those specifically?

A: Components that go into semiconductor manufacturing equipment, such as valves. Since these components also require non-destructive testing, we built and sold inspection equipment for gas nozzles and CVD equipment parts. To investors, we explain that while HBM inspection is our pinnacle goal, we have created intermediate step-by-step product milestones along the way.

Q: CEO Ryu, once you achieve clear revenue results, please visit us again to share the news.

A: Please invite me anytime. Two years ago we discussed batteries, last year we discussed carbon nanotube materials, and now we aim to become a company specialized in HBM inspection.

I’d like to leave one final thought: the battery and display markets have largely shifted toward China, where competition is intensifying. In contrast, South Korea holds a dominant position in the HBM market. To protect this leadership, yield inspection technology is vital. Proper yield inspection will allow Korean semiconductor makers to maintain their gap over Chinese competitors for a long time. We want to contribute to that effort with our core analysis and yield management technologies.

Global Technology said it will accept subscriptions from retail investors for two days starting Sept. 16.

Global Technology is scheduled to list on the KOSDAQ on Sept. 29. Following five trading days of bookbuilding with institutional investors starting Sept. 7, the company finalized its offering price at 10,000 won on Sept. 15.

"Our fundamental technological capabilities and earnings growth potential remain unchanged," Global Technology said. The company has low-power, constant-current time-division driving technology that reduces the number of driver integrated circuits (ICs) required while maintaining the same image quality. Cumulative IC sales have exceeded 200 million units.

Global Technology is a fabless semiconductor company. It develops and supplies both ICs and modules. The company also mass-produces module-level products, including display modules and backlight units (BLUs) for automotive customers. Revenue reached 28.8 billion won in the first half of this year, approaching its full-year revenue of 29.7 billion won last year. The company also swung to a profit of 1.155 billion won.

The company said it has also minimized the potential overhang. Shares available for trading immediately after Global Technology's listing will account for 22.19% of its total shares. The remaining 77.81% will be subject to lock-up periods ranging from one month to as long as four years. That compares with 37% for fabless company Haechitech, which listed last month.

"We minimized the number of shares available for trading on the listing date to protect investors," Global Technology CEO Kim Min-sun said. "After the listing, we will drive corporate value by delivering tangible results through diversification of next-generation mobility products."

Rebellions said on Sept. 16 it will supply racks to Japanese artificial intelligence (AI) company ai&’s data center. The company plans to expand the supply to more than 100 units.

Rebellions will deploy its heterogeneous infrastructure system “REBELRack” at ai&’s data center. An initial batch will be installed at ai&’s Tokyo data center, with additional units to follow. Rebellions declined to disclose the volume or contract value, citing a nondisclosure agreement (NDA) with the customer.

Heterogeneous infrastructure combines different computing devices, including neural processing units (NPUs), graphics processing units (GPUs) and central processing units (CPUs). Rebellions will provide ai&’s data center with a heterogeneous infrastructure solution focused on power efficiency and lower operating costs.

REBELRack is Rebellions’ fully integrated rack-scale AI infrastructure. It supports the open-source frameworks used by ai&, allowing the system to be deployed without a large-scale infrastructure transition. Rebellions expects the system to begin operating at ai&’s data center on the day of delivery.

The system supports up to 32 Rebellions accelerator cards and eight AMD EPYC 9355 CPUs. It has 6 terabytes of DDR5 system memory and delivers 64 petaflops (PFLOPS) of performance using 8-bit floating-point (FP8) operations.

Rebellions said a configuration of 100 REBELRack units would have maximum power consumption of 2.8 megawatts (MW) for the rack equipment alone. The company did not disclose when delivery of all 100 units would be completed. ai& plans to build out its data center to a capacity of 40 MW by the end of 2027.

Wonik Robotics is seeking to supply humanoid robots to Samsung Electronics' semiconductor plants.

"We are currently conducting proof-of-concept (PoC) testing of wheeled robots with Samsung Electronics and other major semiconductor companies," Wonik Robotics Executive Director Kim Min-cheol said at the Korea Semiconductor Industry Association's Advanced Sensor Forum held at Josun Palace in Seoul's Gangnam district on Sept. 15.

Samsung Electronics' evaluation consists of three stages. It first checks whether the robot operates properly and then evaluates its reliability. During reliability testing, Samsung examines whether the robot can handle variables that may arise when handling wafers, including minute vibrations and dust. Semiconductor plants are highly sensitive to vibration and dust, creating a high barrier to the adoption of humanoid robots.

Even if the robot passes Samsung Electronics' final evaluation, the initial supply volume is expected to be limited. "If we pass, it will be meaningful in that we will have secured a major corporate customer," Kim said. He did not disclose the expected supply volume or timing.

Wonik Robotics' robot is wheeled. Humanoid robots can be divided into wheeled and bipedal types based on their method of locomotion. Wheeled robots have a human-like upper body but use wheels instead of legs for their lower body.

The robot is designed to transport wafers and other materials inside semiconductor plants, replacing tasks currently performed by workers. Wonik Robotics has previously supplied robotic hands for automation at Samsung Electronics' smartphone inspection lines.

Wonik Robotics develops the robot's body, including its hands, arms and lower body, in-house. The company is considering eventually supplying only the robotic hands while sourcing the remaining components externally. Robotic hands are one of the biggest technological bottlenecks in humanoid robots because they require the most precise movements. They are also among the most expensive robot components.

Wonik Robotics uses a five-finger robotic hand modeled after the human hand. The company also considered a four-finger design excluding the little finger, but opted for five fingers to ensure a stable grip on objects.

A key technology in the robotic hand is artificial intelligence (AI)-based object recognition. Sensors embedded in the hand detect an object's type, material and weight. The robot adjusts its grip strength accordingly, applying more force to heavier objects and less to fragile ones. AI assists in recognizing objects. Wonik Robotics develops its AI models in-house and uses proprietary data accumulated over more than a decade in the robotic hand business to train them.

Wonik Robotics' new plant in Wanju, North Jeolla Province, is expected to serve as its mass-production base for robots. The company has its headquarters in Seongnam and a research center in Anyang, both in Gyeonggi Province. Last month, Wonik Robotics said it would secure 350 billion won in funding from the government for projects including construction of a robotic hand production plant in Wanju.

Global Technologies has set its final initial public offering price below its proposed range ahead of its KOSDAQ listing.

Global Technologies said Sept. 15 that it set the final offering price at 10,000 won per share. The company had proposed a price range of 13,000 won to 15,000 won. The final price was as much as 33.3% below the proposed range. The offering will raise a total of 40 billion won, with the company expected to have a market capitalization of about 196.5 billion won.

Global Technologies said it decided on the price after taking into account "the subdued IPO market and increasingly cautious investor sentiment," adding that it sought to build medium- to long-term trust with investors rather than focus on short-term demand for the offering.

The company conducted bookbuilding for institutional investors over five trading days from Sept. 7 to Sept. 11. A total of 961 domestic and overseas institutions participated, resulting in a subscription ratio of 123.19 to 1.

Global Technologies designs light-emitting diode (LED) driver integrated circuits (ICs). While it operates as a fabless chip company, it has a pilot line for microLED-related products in Pyeongtaek. The company supplies not only ICs but also products in module form, including backlight units (BLUs). Its main customers are in the TV and monitor markets. It has recently been expanding into the automotive LED market.

Global Technologies passed the KOSDAQ preliminary listing review in July. It is pursuing a technology-special listing. Public subscriptions will be held Sept. 16-17, with the listing scheduled for Sept. 29. Proceeds from the listing will be used from this year through 2028 for wafer costs, research and development (R&D), and a microLED development line.

The company is targeting revenue of 169.5 billion won in 2029. Revenue in the first half of this year was 28.8 billion won, equivalent to 97% of its full-year revenue of 29.7 billion won last year. It also swung to a profit of 1.155 billion won during the period.

SK keyfoundry said Sept. 9 it signed a long-term agreement to provide 130-nanometer foundry services to automotive semiconductor company Elmos Semiconductor.

SK keyfoundry recently held a signing ceremony in Frankfurt am Main, Germany, and will provide foundry capacity to Elmos through 2037. Elmos has used SK keyfoundry’s 350-nanometer process technology for the past 18 years.

“Elmos is an outstanding company in the automotive semiconductor market,” an SK keyfoundry official said. “Securing a long-term supply agreement is meaningful not only in terms of revenue but also for our technological growth.”

Wafer supply prices may be adjusted under the agreement. The 200-millimeter wafers supplied by SK keyfoundry are experiencing a global shortage as major chipmakers including TSMC and Samsung Electronics reduce 200-millimeter capacity and shift toward 300-millimeter production.

SK keyfoundry and Elmos will jointly develop the E550.01 Smart 4-Channel eFuse Controller. The chip controls the supply of power inside a vehicle across four separate paths and cuts power to protect electrical systems when abnormalities such as overcurrent occur. Mass production is targeted for 2027.

SK keyfoundry will also provide Elmos with deep trench isolation (DTI) capability. The company said the technology enables highly integrated, high-voltage designs with low substrate current.

“SK keyfoundry has been a trusted partner for many years,” Elmos Chief Executive Officer Arne Schneider said. “It will support our continued growth.”

“We will accelerate our automotive semiconductor supply, starting with Elmos,” SK keyfoundry CEO Lee Dong-jae said.

DB HiTek is in the final stages of preparing for mass production of its 8-inch silicon carbide (SiC) wafer foundry business.

DB HiTek said Sept. 9 that it had completed reliability verification of a 1,200-volt SiC MOSFET. The company plans to provide customers with its third-generation process design kit (PDK) in November as it moves to launch its foundry business.

The third-generation PDK targets specific on-resistance of 2.3 mΩ·cm² at an 18-volt gate voltage, down from 2.5 mΩ·cm² for the second generation. It is also designed to allow the device to withstand a sudden short circuit for 2.5 microseconds without damage. DB HiTek said the third-generation PDK could shorten semiconductor design time by more than a year.

“We have secured the core technology and mass-production foundation for what will be the world’s first 8-inch SiC foundry service,” DB HiTek said. “We will prepare for mass production as planned and move to establish an early lead in the market.”

Mass production is targeted for the first half of next year. Rival SK keyfoundry is also expected to begin mass production of 8-inch SiC foundry products in the first half of next year. SiC foundry operations require production lines to be developed and established jointly with customers, meaning both companies have effectively secured customers ahead of mass production. Samsung Electronics also has an 8-inch SiC production line, but does not offer it to external foundry customers.

The SiC semiconductor market currently centers on 150-millimeter wafers, while mass production using 200-millimeter wafers remains rare globally. SiC semiconductors use specially processed wafers that differ from conventional silicon (Si) wafers. Due to the technology’s lower maturity, production yields for 200-millimeter SiC wafers are lower than for 150-millimeter wafers, resulting in lower output.

DB HiTek will operate its SiC foundry at its Sangwoo Campus in Eumseong, North Chungcheong Province, converting the current development line into a mass-production line. Initial capacity will be relatively small at 800 wafers a month, about 1.3% of the Sangwoo facility’s total monthly capacity of 63,000 wafers. The company has established an integrated process covering design, manufacturing, characterization, reliability verification and epiwafer supply.

For now, DB HiTek is operating the development line primarily for customers it has already secured. It plans to add customers after establishing the mass-production system and gradually expand capacity in line with market demand.

Meta and Panmnesia have proposed a next-generation artificial intelligence (AI) data center architecture designed to make an entire data center operate like a single chip. The approach could address latency problems that emerge as AI models grow larger.

Panmnesia said Sept. 8 that it and Meta introduced technology in a Nature Reviews journal that connects central processing units (CPUs), accelerators and memory through Compute Express Link (CXL) and scales the architecture across a data center. Researchers from Meta and Panmnesia co-authored the paper.

The approach uses CXL to reduce variations in latency between racks, allowing a data center to operate predictably like a single chip. CXL is a high-speed interface that connects different devices including CPUs, graphics processing units (GPUs) and memory.

Racks, the building blocks of AI infrastructure, exchange data rapidly within each rack using high-speed link technologies. Standard interfaces such as Ethernet and InfiniBand are used to connect one rack to another.

The problem is that as AI models expand to trillions of parameters, latency with conventional interfaces increases sharply. AI companies have increased the number of CPUs and GPUs to train on massive datasets, inevitably resulting in higher latency.

Meta and Panmnesia addressed latency variation by connecting racks through CXL. They added three types of hardware: high-fanout switches, fabric controllers and link acceleration units. They also proposed optical CXL to reduce the physical distance over which signals must travel.

The number of accelerators that a single CPU can manage increased from two to 16, while up to 960 connected domains can be managed. Latency was reduced by as much as 90%. Panmnesia said the CXL-based technology demonstrated that models larger than today's AI models can be trained without interruption.

"CXL presents a direction for next-generation AI infrastructure in which a data center operates as a single computing system," Panmnesia CEO Jung Myoungsoo said.

Arm is preparing for the spread of agentic artificial intelligence (AI) across mobile devices, data centers and robotics.

Arm unveiled “CSS for Mobile 2,” its latest compute subsystem (CSS) for mobile devices, on Sept. 7. CSS for Mobile 2 combines the Mali G2 Ultra NX graphics processing unit (GPU) with the C2 CPU Cluster.

The platform is optimized for edge environments as agentic AI evolves. Arm expects agentic AI workloads to increasingly shift from data centers to edge devices such as desktops, laptops and mobile devices, driven by high token costs, long latency and security concerns. The company expects on-device computing, which operates physically closer to users, to become increasingly important in addressing these issues.

The Mali G2 Ultra NX features a dedicated neural accelerator. It supports neural super sampling, which renders graphics at a lower resolution and then uses AI to reconstruct them at a higher resolution, extending technology previously used on desktops to mobile environments. Arm placed the neural accelerator close to the GPU to enable neural graphics at about 1 watt of power on mobile devices. Frame rates can increase by as much as fourfold at the same power level.

The C2 CPU Cluster enhances Scalable Matrix Extension 2 (SME2) capabilities for AI computing. While the previous C1 supported one SME2 unit, the C2 can be configured with up to two. SME2 performance in the C2 reaches 5 to 7 trillion operations per second (TOPS). General CPU performance has also improved, with single-thread performance up 15% and multi-thread performance up 12% from the previous generation. Processing time for agentic AI workloads is reduced by as much as 24%.

Arm also introduced “Neoverse CSS N4” for data centers. The platform is designed to accelerate development of custom systems-on-chip (SoCs) for general-purpose computing, networking and storage. It supports enhanced security, higher memory bandwidth, chiplet interfaces, LPDDR6 and PCIe Gen7. Arm increased design flexibility so customers can configure core counts, cache capacity and input/output (I/O) according to their requirements. Performance per socket is more than double that of the previous generation, while performance per watt improves 25% and memory bandwidth increases 75%.

Arm also provided an update on its first silicon chip, the “AGI CPU,” announced in March. The company said OpenAI, Meta, Cloudflare, Oracle, SAP and Lenovo are developing AGI CPU-based solutions following its launch.

Related products are expected to be released in the coming months. Customers that design their own SoCs can receive the technology as intellectual property (IP), while those seeking to rapidly expand computing infrastructure can choose finished products.

Arm is also strengthening its push into the physical AI market, including autonomous driving, robots and drones. The company estimates the physical AI computing market is currently worth $25 billion and expects it to grow to more than $200 billion after 2030.

Arm also announced the launch of Total Design for Physical AI, with more than 80 key companies including AWS, Hugging Face, NXP, Siemens and Unitree participating. The initiative links the entire development cycle — including silicon, cloud, sensing, security and system validation — with the physical AI ecosystem to support product development. Arm is also developing a common framework for robotics platforms, aiming to rigorously standardize robot capabilities by level, similar to the Level 1-to-5 system used for autonomous driving.

Synopsys is expanding its semiconductor design platform following its acquisition of Ansys, targeting further growth in South Korea’s semiconductor market.

Shankar Krishnamoorthy, chief product development officer (CPDO) at Synopsys, told a press conference at the Grand InterContinental Seoul Parnas in Gangnam on Sept. 8 that the company will “accelerate technological innovation in South Korea through agentic artificial intelligence (AI), multiphysics-based design and a silicon-to-systems integrated engineering approach.”

The importance of the South Korean market is growing. According to Synopsys’ own research, semiconductors are expected to account for 42.3% of the country’s total exports this year, while high-bandwidth memory (HBM) exports are projected to increase 153%.

Most of Synopsys’ revenue in South Korea comes from the memory sector, with Samsung Electronics and SK hynix among its major customers. The two companies use Synopsys’ semiconductor design tools and other technologies to develop memory chips.

Synopsys is also expanding its revenue base into system semiconductors. “Beyond memory, we work with Samsung Electronics’ foundry and System LSI businesses,” Krishnamoorthy said. “We also collaborate with Rebellions and FuriosaAI across all areas, including electronic design automation (EDA) tools.”

Synopsys’ strength lies in the breadth of its design environment. The company acquired engineering simulation software company Ansys in July last year, expanding beyond its existing EDA and intellectual property (IP) portfolio to add system design and simulation capabilities. This allows chips to be designed with consideration for the systems in which they will operate, including rockets, aircraft and data centers.

The company is also addressing the growing complexity of AI chip design. As AI chips incorporate more dies and HBM, interconnect density increases, requiring designers to consider signal integrity, power, thermal effects and stress simultaneously. Synopsys combines EDA and multiphysics technologies to support advanced designs including three-dimensional integrated circuits (3D ICs) and co-packaged optics (CPO).

Synopsys is also applying agentic AI to EDA. The technology assists engineers and autonomously performs lengthy tasks. In an agentic AI design verification process with Nvidia, the time required to obtain register-transfer level (RTL) results was reduced by a factor of 50, while verification coverage increased 20%. Productivity in analog and mixed-signal design improved by as much as threefold.

Synopsys also supports its own application programming interface (API). The company said it requires fewer tokens and lower token costs than competing solutions while significantly reducing latency.

SEMIFIVE has secured its first mass-production project using a 4-nanometer process. The company did not disclose the size of the supply deal, but said it exceeds 10% of last year's revenue of about 121 billion won.

SEMIFIVE said Sept. 8 that it will begin mass production of a 4nm chip for South Korean fabless semiconductor company HyperAccel. Production has been delayed by two quarters from the original schedule. In 2024, SEMIFIVE said it would design HyperAccel's "Bertha" using Samsung Electronics' 4nm process, with mass production initially targeted for the first quarter of 2026.

The company did not disclose the contract value due to a non-disclosure agreement (NDA) with the customer. The deal exceeded 10% of last year's revenue, making it subject to regulatory disclosure. The contract runs through June 15, 2028. Revenue will be recognized in stages as products are supplied, with a 10% upfront payment, 60% interim payment and 30% final payment.

Bertha is specialized for artificial intelligence (AI) inference workloads, including large language models (LLMs). The chip has a die area of more than 500 square millimeters. Naver Cloud is reportedly a potential customer.

As a design house, SEMIFIVE designed Bertha for optimization with Samsung Electronics' semiconductor foundry process. It handled front-end design, verification, packaging, testing and mass production on a turnkey basis. Under the contract, SEMIFIVE generates revenue each time the jointly designed chip is produced.

SEMIFIVE's mass-production orders totaled about 42.3 billion won in the first half of this year, roughly double the 21.2 billion won recorded last year.

"We will continue the growth in mass-production revenue demonstrated in the first half into the second half," SEMIFIVE CEO Cho Myung-hyun said. The company expects revenue to exceed 200 billion won this year. Company-wide revenue totaled about 94.7 billion won in the first half.

Siemens EDA outlined how artificial intelligence (AI) can reduce semiconductor design time by as much as tenfold, highlighting multi-agent AI as a key technology for addressing the growing complexity of chip development.

Ankur Gupta, Senior Vice President of IC Products at Siemens Digital Industries Software's Electronic Design Automation (EDA) business, presented the company's AI-native design strategy during a media briefing held at Lotte Hotel World in Seoul on Aug. 11.

"Last month, we integrated a broad range of agentic AI capabilities into our EDA platform, including multi-agent workflows, self-verification and long-duration autonomous execution," Gupta said.

Multi-agent AI enables multiple AI agents to divide and coordinate semiconductor design tasks. After engineers specify design parameters such as manufacturing process, voltage and temperature conditions, the agents automatically perform verification, debugging and design modifications. Siemens introduced its AI-powered EDA system, Fuse, in 2025.

According to the company, multi-agent AI can accelerate transistor characterization workloads by up to ten times while reducing token costs by roughly 90%.

"Engineers at large companies are restricting or even suspending the use of large language models because token budgets have become unsustainably high," Gupta said. "The EDA industry recognizes this as a critical challenge and is developing technologies that deliver much higher token efficiency."

Siemens is also collaborating with Nvidia to develop AI-enabled EDA tools. Earlier this year, the two companies announced a strategic AI partnership aimed at integrating AI throughout the semiconductor design flow—from register-transfer level (RTL) design through verification. Under the collaboration, Nvidia's AI infrastructure and Nemotron models will be incorporated into Siemens' EDA software.

Gupta said modern semiconductor designs are reaching unprecedented levels of complexity.

"Today's chips contain more than 200 billion transistors and exceed thermal densities of 100 watts per square centimeter," he said. "Conventional design approaches can no longer keep pace with the speed of the market."

He added that even modest improvements in chip efficiency can have a substantial impact.

"A 5% reduction in power consumption translates into significant energy savings at the system level, making it an important factor in building a more sustainable industry," he said.

Other leading EDA vendors, including Synopsys and Cadence Design Systems, are also incorporating AI capabilities into their design platforms.

Gupta said Siemens differentiates itself through its focus on design verification, emphasizing that the company provides a comprehensive integrity platform.

"In semiconductor design, a single minor error can result in design failure or lower manufacturing yields," he said. "Because such failures can cost hundreds of millions of dollars, building software that customers can trust is the industry's highest priority."

Siemens also said AI can significantly reduce verification time, one of the most time-consuming stages of semiconductor development. According to the company, its EDA tools can accelerate automatic test pattern generation (ATPG) simulation by up to five times for certain workloads. ATPG simulation is used to validate test patterns in virtual environments before they are applied to detect manufacturing defects in semiconductor devices.

South Korean semiconductor company Fadu announced on Aug. 10 that it is recruiting junior-level engineers, ranging from new graduates to professionals with up to eight years of experience.

The company will accept applications through Aug. 23 as part of an effort to secure top talent and strengthen its future technology competitiveness.

Applicants will undergo a coding test later this month, followed by interviews in September. Successful candidates are expected to join the company in November and work at Fadu's headquarters in Seoul's Gangnam district.

The recruitment program is expected to hire more than a dozen employees. Eligible candidates include bachelor's degree graduates as well as engineers with up to eight years of experience in hardware or embedded software development.

Open positions include firmware development, firmware verification and test engineering roles.

The selected employees will analyze customer requirements for Fadu's enterprise solid-state drive (eSSD) products and help incorporate those requirements into product development. Responsibilities will also include analyzing eSSD test results, identifying root causes of issues and proposing improvements.

Iron Device said on Aug. 10 that cumulative sales of its SMA1305 smart power amplifier for Samsung Electronics mobile devices and tablets have exceeded 100 million units, reaching the milestone four years after shipments began.

The company began supplying the SMA1305 in January 2022. According to its latest tally, cumulative shipments surpassed 100 million units as of March this year, equivalent to an average of approximately 25 million units supplied annually.

On average, two SMA1305 chips are installed in each mobile device, while tablets typically use four units per device.

"The achievement demonstrates that our technology is being recognized in the global market," said Lee Young-taek, a director at Iron Device. "We also have high expectations for sales of follow-up products."

The SMA1305 is a mixed-signal semiconductor used in smartphones and tablets. The chip amplifies audio signals to drive speakers and incorporates Iron Device's proprietary speaker protection algorithm, which the company says improves performance and reliability compared with competing products. The device also integrates low-noise, low-power design and audio signal processing technologies to maximize system efficiency.

Iron Device develops chips that combine analog, digital and power technologies on a single device. The company has recently expanded into the on-device AI market, targeting applications in smart glasses and humanoid robots.

"Audio products are evolving beyond simple sound input and output devices into AI sensors capable of understanding their surroundings," the company said. "We define audio as a core interface for interacting with the real world and are developing technologies accordingly."

LX Semicon has begun supplying automotive semiconductors to Hyundai Motor and Kia, marking the first major achievement from a new business initiative launched four years ago to diversify beyond display driver ICs (DDIs).

The company said on Aug. 10 that it has been supplying its automotive microcontroller unit (MCU), code-named LX61101, to Hyundai Motor and Kia vehicles since July. The chip has already been installed in vehicles produced from July, and the scope of adoption is expected to expand. Specific vehicle models and shipment volumes were not disclosed due to non-disclosure agreements (NDAs) with customers.

The LX61101 is a motor-control-focused MCU based on a 40 MHz Arm Cortex-M3 core. It is optimized for motor control in automotive body electronics systems. When installed in vehicle doors, the chip monitors and controls motor speed, position and rotation direction in real time. For example, if an obstacle is detected while an automatic door is closing, the system can reverse operation and reopen the door.

LX Semicon said South Korean companies were involved across the entire process, including foundry manufacturing and back-end supply chain management (SCM).

The company has been developing automotive MCUs as a future growth engine since 2022. In 2023, it participated in Hyundai Motor and Kia's supplier selection process for new business projects. Following three years of development and testing, LX Semicon has now entered mass production of the LX61101.

The chip also meets key automotive semiconductor standards, including the AEC-Q100 reliability qualification standard and ISO 26262 functional safety requirements.

Telechips returned to profitability in the second quarter of 2026, driven by increased sales of in-vehicle infotainment (IVI) chips.

The company said on Aug. 6 that second-quarter revenue rose 33.6% year over year to 59.2 billion won. Operating profit came in at 3.18 billion won. Operating margin improved to 5.4% from negative 8.0% a year earlier.

Growth was primarily driven by the automotive IVI chip business. Increased shipments of products supplied to existing customers supported the company's earnings improvement during the quarter.

Revenue recognition from a system-on-chip (SoC) development contract worth approximately 77.2 billion won, signed in October 2025, also contributed to results. The contract began contributing revenue in the fourth quarter of 2025 and has supported the company's top-line recovery since then. However, the impact was smaller than in the first quarter. A Telechips representative said revenue recognized from the contract exceeded 10 billion won in the first quarter, while the amount recorded in the second quarter was lower.

Net loss narrowed 27.3% from a year earlier to approximately 4.1 billion won. The company said the loss was affected by a decline in the valuation of its stake in Chips&Media, in which Telechips held a 7.25% stake as of the first quarter. Chips&Media's share price fell 31.4% over the 12-month period, declining from around 16,900 won at the end of June 2025 to 11,600 won at the close on June 30, 2026.

"We will continue to improve profitability in the second half by expanding our existing customer base and increasing mass production for new projects," a Telechips representative said.

DB HiTek announced on Aug. 6 that it has transferred semiconductor process equipment to MoaFab, South Korea's national integrated nano-fab platform, as part of efforts to strengthen the country's semiconductor research and development infrastructure.

The company held a semiconductor equipment transfer ceremony at its Sangwoo Campus in Eumseong, North Chungcheong Province, together with the Ministry of Science and ICT and MoaFab. The initiative is aimed at supporting microelectromechanical systems (MEMS) research and development.

MEMS refers to technologies used to manufacture miniature precision mechanical devices at the micro- and nanoscale.

Under the agreement, DB HiTek is donating a total of three pieces of equipment, including one high-temperature thermal processing system and two etching systems.

The equipment will be used by the National NanoFab Center and the Nano Convergence Technology Center for research and development activities as well as prototype production.

In March, DB HiTek signed a memorandum of understanding with the Ministry of Science and ICT, Samsung Electronics, SK hynix and other organizations to strengthen the competitiveness of MoaFab.

MoaFab is a national nano-fabrication support platform established by the Ministry of Science and ICT and the National Nano Infrastructure Council. The platform integrates and manages nano-fabrication equipment and facilities across South Korea to support nanotechnology research, education and prototype development.

FADU plans to begin supplying samples of its seventh-generation solid-state drive (SSD) controller in 2028, according to a technology roadmap unveiled at FMS 2026.

Nam Yi-hyun, Chief Executive Officer of FADU, disclosed the plan on Aug. 4 during the Future of Memory and Storage (FMS) 2026 conference in Santa Clara, California.

The company did not provide detailed specifications for the seventh-generation controller. However, FADU previously stated in December that it aims to improve SSD performance by more than tenfold compared with the previous generation.

Its long-term target is to achieve random read performance of 100 million input/output operations per second (IOPS) for 512-byte workloads.

At FMS 2026, FADU operated an exhibition booth featuring demonstrations of its sixth-generation SSD controller.

The Gen6 controller, which was recently completed, is scheduled to begin shipments in the second half of this year.

According to the company, the new controller delivers more than twice the performance of its fifth-generation predecessor while helping reduce total cost of ownership (TCO) for data center operators.

FADU is also pursuing a strategy of supplying SSD controllers directly to server customers rather than relying exclusively on memory manufacturers as intermediaries.

The company is seeking to expand its roster of hyperscale customers as part of that effort.

Meta is among FADU's existing hyperscale customers, and the company aims to secure additional large-scale cloud and data center operators as demand for high-performance storage infrastructure continues to grow.

FuriosaAI will supply its neural processing units (NPUs) for a large-scale AI data center project in Sweden, marking a significant expansion of the company's presence in the European AI infrastructure market.

The company announced on Aug. 5 that it is participating in an AI data center development project in Stockholm alongside US AI infrastructure company I/ONX and data center operator Velox.

The facility is planned to reach a total capacity of 15 megawatts and is expected to deploy more than 8,800 units of FuriosaAI's Renegade NPU.

The project will begin with the launch of a 2-megawatt facility early next year, followed by an additional 8 megawatts later in the year. Capacity will eventually expand to 15 megawatts.

FuriosaAI plans to initially supply approximately 1,800 Renegade processors before increasing shipments as the project scales.

The company will provide both Renegade hardware and its accompanying software stack. I/ONX will be responsible for system integration and AI platform services, while Velox will oversee data center development and operations.

According to FuriosaAI, Renegade will operate alongside graphics processing units (GPUs) within the facility. GPUs will be used for AI model training and selected compute workloads, while Renegade processors will handle inference tasks.

The company said the hybrid architecture is designed to improve both performance and power efficiency.

FuriosaAI expects the project to serve as a foundation for expanding AI inference services across Europe.

The company recently established its European subsidiary in Lisbon, Portugal. Through the new operation, FuriosaAI aims to expand its presence in the region's sovereign AI and AI infrastructure markets by promoting its inference-focused computing platform.

Haechitech will begin supplying temperature sensor integrated circuits (ICs) for enterprise solid-state drives (eSSDs) to a major South Korean memory manufacturer this year, marking its entry into a market previously served exclusively by Texas Instruments (TI).

Speaking at the company's initial public offering (IPO) event in Seoul on Aug. 5, CEO Choi Sung-min said Haechitech had secured a supply position for temperature sensor ICs that had previously been supplied solely by TI to the memory company.

"Haechitech's products will be applied to newly launched eSSD products," Choi said.

He added that Haechitech's sensor performance exceeded that of TI's competing products. According to Choi, TI's temperature sensing accuracy is rated at ±0.4°C, while Haechitech's solution achieves ±0.2°C.

Temperature sensor ICs monitor thermal conditions inside eSSDs. Accurate temperature management is critical because heat control directly affects performance, reliability and operating costs in enterprise storage systems.

Despite the design win, Haechitech does not expect temperature sensor ICs to become a major revenue contributor in the near term.

"We significantly lowered pricing to enter as a new supplier," Choi said. "The business is more about expanding our customer portfolio and securing new accounts."

The company expects geomagnetic sensor ICs to remain its primary business.

Choi said revenue from geomagnetic sensors will continue to exceed that of temperature sensor ICs as demand for mid-range and entry-level smartphones grows.

According to the company's securities filing, Haechitech projects temperature sensor IC revenue of approximately 7.9 billion won in 2028, compared with about 24.9 billion won for geomagnetic sensor products.

Geomagnetic sensor ICs detect the Earth's magnetic field and are commonly used in smartphones as electronic compasses. In 2025, geomagnetic sensor sales totaled approximately 9 billion won, accounting for 63.7% of the company's total revenue of roughly 14.1 billion won.

Key customers include Samsung Electronics, OPPO and Vivo. Manufacturing is outsourced to SK keyfoundry.

Haechitech was established in 2017 after being spun off from Magnachip Semiconductor's sensor business division. The company acquired product rights for OPPO and Vivo-related sensor products at the time of the separation.

After generating revenue of 3.9 billion won in 2022, Haechitech recorded an average annual revenue growth rate of 53%, supported by early supply relationships with major electronics companies.

The company plans to list on the KOSDAQ market through the technology-special listing track on Aug. 25.

Haechitech will offer 1 million shares, representing approximately 18.13% of its total outstanding shares after listing. The indicative IPO price range has been set at 23,000 won to 28,000 won per share, with the final offering price scheduled to be determined by Aug. 7. DB Securities is serving as the lead underwriter.

The company forecasts revenue of 23.1 billion won in 2026, rising to 37.3 billion won in 2027 and 54.2 billion won in 2028. It expects to return to annual operating profitability in 2027, with projected operating profit of approximately 4.3 billion won.