SK hynix Faces a U.S. Patent Fight Over HBM and 3D NAND

SK hynix faces a U.S. patent fight that reaches two strategic product lines: high-bandwidth memory and 3D NAND. The dispute is no longer just a lawsuit appearing in Google News results. It includes a federal trade investigation that can restrict infringing imports.

MonolithIC 3D alleges that certain SK hynix and Kioxia memory products infringe patents covering three-dimensional semiconductor structures. SK hynix has not been found liable, and the allegations remain contested. However, the requested remedy raises the stakes beyond any eventual damages award.

The conflict arrives as SK hynix holds a leading position in HBM, the stacked DRAM used beside AI accelerators. Samsung and Micron are pursuing the same expanding market. Patent enforcement now threatens to become another competitive variable alongside bandwidth, capacity, manufacturing yield, and customer qualification.

What changed in the SK hynix patent dispute

A private patent complaint has become a formal U.S. trade investigation with potential consequences for imported memory products.

The U.S. International Trade Commission announced the investigation on March 26, 2026. It followed a complaint that MonolithIC 3D filed on February 17 and later supplemented.

The USITC investigation names SK hynix, SK hynix America, Kioxia, and related entities as respondents. The commission describes the products at issue as certain NAND and DRAM memory chips, related products, and components.

A Section 337 investigation examines allegedly unfair practices involving goods imported into the United States. Patent infringement is one recognized basis for such a case. The commission can issue an exclusion order that directs U.S. Customs to block covered imports.

It can also issue cease-and-desist orders affecting inventory and sales within the country. Those remedies give an ITC complaint leverage that differs from a conventional damages lawsuit in federal court.

The commission emphasized that opening an investigation did not represent a decision on the merits. An administrative law judge must develop the evidentiary record, interpret the patent claims, and assess infringement and validity.

MonolithIC 3D also brought a patent case against SK hynix in the Eastern District of Texas. The public federal docket records a complaint for patent infringement filed in February.

The parallel proceedings serve different functions. A district court can award damages and enter other relief. The ITC concentrates on imported goods and moves through a specialized administrative process.

MonolithIC 3D’s allegations reportedly reach HBM2E, HBM3, HBM3E, and SK hynix’s 3D NAND portfolio. HBM stacks multiple DRAM dies and connects them through vertical pathways, allowing processors to access data at high aggregate bandwidth.

Three-dimensional NAND also depends on vertical construction, but it serves a different purpose. Manufacturers build layers of flash-memory cells to increase storage density without expanding a chip’s footprint at the same rate.

The presence of both technologies makes the complaint unusually broad in commercial terms. It does not mean one patent theory covers every aspect of both product families. Each asserted claim must still be mapped to specific structures or manufacturing steps.

A public version of the ITC complaint identifies HBM and 3D NAND products among the accused categories. That filing presents the complainant’s case, not an independent technical finding.

The case therefore changed in two important ways. The allegations moved into a forum capable of restricting imports, and the product scope reached components central to both AI computing and storage.

Those features explain why the story traveled across Google News technology feeds. They do not establish that an exclusion order is likely or that any accused product infringes a valid patent.

Why the fight matters to SK hynix now

The complaint targets SK hynix when HBM leadership carries greater strategic value than an ordinary memory-market advantage.

HBM supplies data to graphics processors and other accelerators used for training and running AI models. Its stacked design addresses a bottleneck that faster processors cannot solve alone. Computing units lose efficiency when memory cannot deliver data quickly enough.

SK hynix has built a strong position in that market. In a U.S. securities filing, the company cited IDC data showing a 56.4 percent HBM revenue share during the first quarter of 2026.

The same company filing says five suppliers accounted for more than 90 percent of first-quarter NAND revenue. That concentration makes disputes involving major vendors important to device makers, cloud operators, and storage customers.

Those figures describe market position rather than the merits of MonolithIC 3D’s claims. Still, they reveal why the targeted products matter. HBM is tied to the company’s AI growth, while NAND connects the dispute to a much wider storage business.

SK hynix said it began mass shipments of HBM4 during the second quarter of 2026 and planned to increase production during the second half. HBM4 is the generation succeeding HBM3E, with a wider interface and further changes intended to raise performance.

The company reported quarterly revenue of 79.3187 trillion won and operating profit of 60.5426 trillion won for that period. Its second-quarter results attributed the performance to AI demand and sales of higher-value products.

Those are company-reported results, and they should not be confused with proof about future HBM demand. They nevertheless show how closely the current business narrative depends on advanced memory.

An import remedy would matter because semiconductor supply chains cross national borders repeatedly. Design, wafer fabrication, packaging, testing, module assembly, and final-system integration can occur in different countries.

A restriction aimed at covered chips can therefore affect products farther downstream. Servers, accelerator boards, storage devices, and other systems may contain the disputed components when they enter the United States.

That possibility does not mean an investigation will interrupt shipments. Respondents can defeat infringement claims, invalidate patent claims, narrow product definitions, redesign implementations, appeal decisions, or negotiate licenses.

Yet customers must consider operational risk before any final ruling. AI infrastructure purchases involve long qualification cycles, fixed system designs, and large deployment commitments. Buyers do not want an unresolved component dispute to become a late-stage procurement problem.

The same concern applies to 3D NAND. Enterprise solid-state drives sit inside databases, cloud platforms, model-training pipelines, and inference services. NAND is slower than HBM, but it holds the datasets, checkpoints, software, and retrieved information feeding those workloads.

For engineers tracking the dispute through Google News, that distinction matters. The case does not target one interchangeable peripheral. It touches two memory layers with different roles inside the same computing environment.

A broad complaint can also increase defense costs and management attention. SK hynix must address patent construction, technical discovery, product mapping, prior art, and potential domestic-industry arguments while maintaining aggressive production schedules.

The company disclosed a second MonolithIC complaint filed in May and later supplemented. According to the securities filing, that matter names SK hynix and Kioxia and alleges infringement of five additional patents involving DRAM and NAND products.

Multiple proceedings can amplify uncertainty even when each claim remains unproven. They increase the number of patents, accused implementations, procedural deadlines, and possible outcomes requiring evaluation.

For SK hynix, the immediate pressure is therefore not a finding of wrongdoing. It is the need to defend a valuable product roadmap without giving customers reason to question supply continuity.

Patents collide with SK hynix’s HBM advantage

The central conflict is between MonolithIC 3D’s claimed rights in three-dimensional chip structures and SK hynix’s ability to sell stacked memory at scale.

HBM is not simply a faster version of conventional desktop memory. It combines multiple memory dies in a compact stack and places that stack near a processor through advanced packaging.

Vertical electrical connections help data travel between layers. A wide interface then transfers large volumes of data with lower energy per bit than many conventional arrangements.

Three-dimensional NAND takes another route to vertical density. Instead of stacking separate DRAM dies for processor bandwidth, it forms many layers of flash cells within a NAND structure.

Both technologies reflect the semiconductor industry’s response to the limits of two-dimensional scaling. When shrinking features becomes harder and more expensive, engineers use vertical integration to add capacity or improve communication.

That shared direction creates fertile ground for patent disputes. Patent claims can address device structures, layer relationships, fabrication sequences, interconnects, memory-cell arrangements, or combinations of those elements.

The legal question is narrower than whether MonolithIC 3D worked on three-dimensional semiconductors before SK hynix. A patent holder must identify valid claims and show that every required element appears in an accused product or process.

SK hynix can dispute that mapping. It can also argue that earlier publications or products invalidate the claims, that the patents cannot receive the proposed interpretation, or that other legal requirements are unmet.

The technical record will matter more than the broad product labels appearing in headlines. “HBM” describes a product category, not a single implementation. Different generations and vendors can use distinct structures, materials, packaging choices, and process sequences.

The same warning applies to “3D NAND.” A complaint covering an entire portfolio still requires evidence connecting particular patent claims to particular products.

This is why the case cannot be resolved from a Google News headline. Readers need to separate three layers of information: what the complainant alleges, what the commission has agreed to investigate, and what the evidence ultimately establishes.

So far, the commission has agreed only to investigate. It has not confirmed infringement, upheld the asserted claims, or approved an import ban.

MonolithIC 3D is described in reporting as a non-practicing entity, meaning patent licensing and enforcement form a significant part of its business rather than large-scale chip manufacturing. That status can shape public reaction, but it does not decide validity or infringement.

U.S. law permits patent owners to enforce their rights without manufacturing the accused products. At the ITC, however, the complainant must also satisfy a domestic-industry requirement tied to the asserted patents.

The public complaint says MonolithIC 3D relies on a licensee when addressing that requirement. Whether its evidence satisfies the legal standard will be one contested part of the proceeding.

The disagreement creates a clean opponent structure. MonolithIC 3D presents the products as implementations of protected inventions. SK hynix must show that its commercial memory does not unlawfully practice the asserted claims, or that those claims cannot support relief.

That conflict differs from the familiar contest among SK hynix, Samsung, and Micron. The memory manufacturers compete through product performance, manufacturing execution, customer access, and capacity.

A patent claimant can apply pressure through a separate channel. It can challenge the legal freedom to import a product even when that product has already passed customer qualification and entered mass production.

The strongest version of MonolithIC 3D’s case would connect valid claims directly to structural features that SK hynix cannot easily remove. The strongest defense would show a meaningful technical mismatch, invalid claims, or failure to meet an ITC requirement.

Neither position has been established publicly. Until the claim-construction and evidentiary process develops, confident predictions about an exclusion order would outrun the record.

What the patent claims do not prove yet

The size of the accused product portfolio makes the dispute important, but breadth alone does not make MonolithIC 3D’s case strong.

Patent complaints are designed to state a claimant’s allegations. They usually present the accused products and disputed technology through the plaintiff’s legal theory.

That format can create an impression of technical certainty. In practice, the most consequential questions often remain unresolved until the parties exchange evidence and the judge interprets disputed claim language.

A patent may use familiar engineering terms in a specialized way. Small differences involving layer order, electrical connections, processing steps, or memory-cell placement can decide whether a product falls within a claim.

Some evidence can also be difficult to obtain. Semiconductor manufacturing occurs inside tightly controlled fabrication facilities, while commercially available chips reveal only part of the process used to create them.

Technical discovery may require design documents, process information, source materials, expert analysis, and physical examination. The parties will likely disagree over what is relevant and how confidential information should be protected.

Validity presents another uncertainty. Three-dimensional integration has a long research and commercialization history, producing extensive patents, papers, prototypes, and products.

SK hynix and Kioxia can search that history for prior art, meaning earlier public material that may anticipate a claim or make it legally obvious. They can also seek Patent Trial and Appeal Board review where procedural requirements allow.

A challenge before the Patent Office does not automatically end an ITC case. Timing, institution decisions, claim scope, and the commission’s own schedule determine whether the parallel process changes the investigation.

The second uncertainty concerns remedies. Even a finding involving some patent claims does not necessarily remove every SK hynix HBM or NAND product from the market.

An order can be limited by the claims proved, the accused implementations, product generations, and downstream scope. Redesigned products can sometimes avoid the relevant elements, although redesigns require technical validation and customer approval.

Settlement remains another path. Patent cases regularly end in licenses, cross-licenses, supply arrangements, or confidential agreements before every legal question reaches a final judgment.

SK hynix has experience with large memory-patent disputes. That history gives the company an established legal and technical defense capability, but it does not predict the outcome here.

The competitive backdrop also limits simple conclusions. Samsung and Micron are not passive while SK hynix defends its position. Both companies continue developing and shipping advanced memory for AI systems.

Micron said in its quarterly results that HBM4E development was underway, with volume production expected in calendar 2027. Samsung has also been advancing its HBM product line and competing for accelerator qualifications.

A prolonged dispute can create an opening without producing an import ban. Customers seeking a second source may allocate more design work or future capacity to rival suppliers because uncertainty itself carries a cost.

That effect should not be overstated. HBM is not substituted as casually as a commodity component. Differences in packaging, thermals, performance, qualification status, and platform integration constrain quick supplier changes.

A customer cannot simply replace one HBM stack after reading a troubling item on Google News. It must validate the alternative within the accelerator platform and associated manufacturing process.

NAND offers more supplier diversity, but enterprise products also require firmware validation, endurance testing, controller integration, and reliability assessment. Switching can still consume considerable time.

Another unknown is the public-interest analysis. Before granting an exclusion remedy, the ITC can consider competitive conditions, consumer effects, public health and welfare, and the production of competing goods in the United States.

AI infrastructure importance could become part of that debate. However, public-interest considerations do not guarantee that commercially important products remain available after proven infringement.

The skeptical conclusion is straightforward. The dispute deserves attention because of its scope and forum, but the existing record does not justify declaring that SK hynix will lose access to the U.S. market.

The three signals to watch next

Claim interpretation, Patent Office challenges, and customer behavior will reveal whether this dispute becomes a supply problem or remains a licensing contest.

The first signal is the administrative law judge’s claim construction and investigation schedule. Claim construction determines what disputed patent language means for the infringement analysis.

A narrow interpretation can exclude major parts of an accused product. A broad interpretation can give the complainant more room, although broader claims may face stronger validity challenges.

The schedule matters because it establishes deadlines for discovery, expert reports, an evidentiary hearing, and an initial determination. It also helps customers compare the legal timeline with their HBM4 and future platform plans.

If early rulings align key claim terms with SK hynix’s implementations, the threat of an exclusion remedy will strengthen. If the judge rejects critical parts of MonolithIC 3D’s interpretation, the supply risk will weaken.

The second signal is whether SK hynix or Kioxia secures meaningful Patent Trial and Appeal Board review. An instituted challenge can put focused pressure on the asserted claims, especially when prior art reaches the same structures or processes.

Institution is not invalidation. It means the Patent Office found enough basis to conduct a formal review under the applicable legal standard.

Final invalidity findings would substantially weaken the corresponding enforcement theory. Denied petitions or surviving claims would strengthen MonolithIC 3D’s negotiating position without automatically proving infringement.

The third signal is customer and product behavior. Watch for risk disclosures, qualification changes, redesign references, licensing announcements, or unusual shifts in supplier allocation.

SK hynix’s shipment execution will provide useful context. Continued HBM4 expansion without disclosed customer disruption would suggest that buyers see the case as manageable.

A license or settlement could resolve the commercial uncertainty without revealing which side held the stronger technical argument. By contrast, disclosed redesign work would suggest that particular claim mappings deserve closer attention.

Rival announcements also matter, but they should remain supporting evidence. Samsung or Micron gaining an accelerator qualification does not prove that the patent dispute caused the decision.

Customers already seek multiple suppliers to improve resilience, negotiate terms, and secure capacity. Analysts must distinguish normal diversification from a response tied directly to litigation.

For developers and enterprise buyers, the practical lesson is not to predict a courtroom winner. It is to map which deployed systems depend on a single memory source and monitor whether vendors disclose a credible continuity plan.

Teams following this case alongside product specifications, court filings, earnings calls, and procurement notes need a consistent research trail. A searchable technical knowledge base can keep those materials connected as the record changes.

The next Google News alert will probably compress the dispute into a win, loss, or escalation headline. The more useful question is narrower: did a new filing change claim scope, product exposure, or the probability of interrupted supply?

Track those three signals before changing a hardware roadmap. If claim construction favors MonolithIC 3D, Patent Office challenges fail, and customers disclose contingency plans, the risk has clearly increased. If SK hynix narrows the claims, maintains shipments, and resolves the case through licensing, the dispute will look more like a manageable cost than a threat to its AI memory position.