• Huawei expands LogicFolding as chip limits persist.
• Kirin and Ascend share limited chip capacity in China.

Huawei has introduced a new chip architecture across its latest Mate 90 smartphones as the company looks for ways to improve processor performance while access to advanced semiconductor manufacturing remains constrained.

The higher-end Mate 90 models use Huawei’s Kirin 9050 Pro processor, which incorporates a design approach called LogicFolding. Huawei says the architecture shortens signal paths inside the chip by restructuring parts of the circuitry across vertically connected layers.

Huawei first commercially introduced LogicFolding in the Kirin 9050 Pro used in the Mate XT 2 in September last month. The Mate 90 extends the architecture into Huawei’s flagship Mate series.

LogicFolding moves into commercial chips

The approach forms part of what Huawei calls its Tau Scaling Law, which it presented at the IEEE International Symposium on Circuits and Systems in May this year. Instead of relying only on shrinking transistor dimensions, the method focuses on reducing signal propagation delay across devices, circuits, chips, and computing systems.

LogicFolding distributes selected logic circuits across vertically connected layers, replacing some longer horizontal connections with shorter vertical interconnects. Huawei says this reduces critical signal paths and allows it to increase transistor density without relying only on smaller process nodes.

Richard Yu, executive director of Huawei and chairman of its Consumer Business Group, said the technology is already being used in the Kirin 9050 Pro.

“This technology vertically stacks the logic layers of chips and restructures the circuit layout through inter-layer high-density interconnect to shorten critical paths and reduce latency,” Yu said. “This technology was adopted on our Kirin 9050 Pro chip.”

Huawei says the Kirin 9050 Pro reaches about 238 million transistors per square millimetre. The company has cited different percentage gains depending on the comparison baseline: earlier LogicFolding materials put the increase at 55% against a previous planar design, while Yu said at the Mate 90 launch that transistor density was about 28% higher than the immediate previous-generation chip.

The two figures, therefore refer to different comparison points rather than conflicting measurements.

The company also claims lower power consumption at equivalent performance compared with its previous planar design. Those density and efficiency figures have not been independently verified.

Huawei had outlined the technology several months before its first commercial use. At its May presentation, the company said Kirin chips due in the second half of 2026 would be the first to adopt LogicFolding.

The Mate 90 Pro Max is now one of the devices using the Kirin 9050 Pro. Huawei says the top-end Mate 90 models provide a 31% overall performance increase compared with the previous flagship generation, although the figure has not been independently verified.

Yu said Huawei also designed several supporting components used in the Mate 90, including the Balong modem, image signal processor, security processor, CPU, GPU, and neural processing unit.

“They offer higher energy efficiency and deliver much-improved overall system-level performance, even though we have limited access to advanced semiconductor processes,” Yu said.

Yu said Huawei invested more than CNY1.8 trillion in R&D between 2007 and the end of 2025, with R&D spending rising to more than 20% of annual revenue in recent years.

China’s advanced chip capacity remains limited

The architectural changes do not remove the manufacturing constraints facing Huawei. Reuters has reported that the three-dimensional architecture requires more wafers to produce, while advanced semiconductor production capacity in China remains limited.

Yu confirmed that Huawei’s Ascend AI processors use the same advanced manufacturing capacity as its smartphone chips.

“Currently, the production capacity for advanced semiconductors in China is indeed limited. Huawei’s Ascend chips for the AI domain also use such capacity,” Yu said.

Huawei has developed an internal requirements plan covering both businesses, according to Yu. He said current production remains tight but is generally sufficient to meet the company’s shipment requirements.

“We hope that production capacity will ramp up in China’s semiconductor industry so that we can manufacture more advanced products for consumers in the Chinese market as well as the overseas market in the future,” Yu said.

The manufacturing constraint is also tied to China’s continued reliance on deep ultraviolet, or DUV, lithography for advanced chip production. Yu said extreme ultraviolet, or EUV, equipment remains important for producing more advanced semiconductors.

“EUV tools are certainly key to manufacturing advanced chips. China is currently working on these tools, but right now we’re still relying on DUV domestically,” Yu said.

He also directly linked LogicFolding to those restrictions.

“Our LogicFolding technology is just an approach we adopt to improve our capabilities when our access to certain technology is restricted,” Yu said.

LogicFolding is one part of a broader approach. Yu said Huawei also relies on software efficiency, HarmonyOS, its app ecosystem, cloud services, and system-level optimisation to improve performance when access to more advanced semiconductor processes is limited.

He also pointed to cooling technology as another area Huawei has worked on to reduce power consumption and improve efficiency in the Mate 90 series.

Extending DUV to smaller geometries generally requires more complex patterning and additional process steps.

Huawei has not disclosed which foundry manufactures the Kirin 9050 Pro. SMIC is widely associated with Huawei’s recent Kirin processors, but neither company has publicly confirmed the manufacturer of the latest chip.

Earlier Huawei processors provide some indication of how far SMIC has pushed DUV-based manufacturing. TechInsights found that the Kirin 9000S processor inside Huawei’s Mate 60 Pro was manufactured by SMIC using its second-generation 7nm process, known as N+2.

TechInsights later found that the Kirin 9030 used SMIC’s N+3 process, which it described as a scaled evolution of the foundry’s 7nm-class technology produced without EUV lithography.

Huawei has also been rebuilding more of its smartphone supply chain around domestic technology. Yu said Huawei resumed domestic production of smartphone chips in 2023 as it worked more closely with China’s domestic supply chain.

“We have worked with the domestic supply chain to manufacture advanced chips by relying on homegrown technologies and capabilities,” Yu said. “In 2023, we started to manufacture the chips used by smartphones domestically again.”

Yu said Huawei had reduced its dependence on US technology through work on domestic chips, its operating system, and foundational software, while the company continues to acknowledge limits around advanced semiconductor manufacturing processes and equipment.

Huawei is also using software and system-level optimisation alongside its chip development. Yu said HarmonyOS, application software, cloud services, and other parts of the device stack are being used with hardware improvements.

“At the same time, we leverage the higher efficiency brought by software, HarmonyOS, and the app ecosystem, and the synergy with cloud services to improve the overall experience,” Yu said.

He noted that while a smartphone’s main processor requires advanced manufacturing processes, many other components do not have the same requirements.

“Therefore, we have adopted a systems engineering approach to continuously optimise overall performance and compensate for not having access to more advanced processes,” Yu said.

Smartphone and AI chips share capacity

Those manufacturing limits also affect Huawei’s AI chip business. TechInsights has said that China can manufacture scaled mobile logic using DUV, but identified yield as a larger constraint for data-centre AI processors because those chips use larger dies.

That distinction matters because Huawei’s Kirin smartphone processors and Ascend AI processors draw on the same limited pool of advanced semiconductor capacity.

The semiconductor requirements of smartphones are also increasing as more AI processing moves onto devices. Yu said local AI workloads require greater computing power, which places higher demands on advanced semiconductor processes.

“In the AI era, more powerful on-device AI computing power is certainly very valuable for rapidly processing local tasks,” Yu said. “This means increasing requirements for the computing power of chips, which requires more advanced processes.”

Huawei says the Kirin 9050 Pro can support a 30-billion-parameter mixture-of-experts model running on higher-end Mate 90 devices. The company has not provided independent benchmark data showing the model’s performance, memory use, or power requirements on the device.

Vertical integration is not unique to Huawei. Belgian research institute imec and EV Group demonstrated wafer-to-wafer hybrid bonding with a 200-nanometre interconnect pitch in May 2026 for high-density logic-to-logic and memory-to-logic stacking.

Huawei’s LogicFolding architecture is a separate implementation, but it uses the same broader principle of moving parts of chip integration vertically rather than relying only on conventional two-dimensional scaling.

Huawei has set a longer-term target of using Tau-based designs to reach transistor densities comparable with a 1.4nm process by 2031. The figure is a company roadmap target rather than a demonstrated manufacturing result.

Asked whether LogicFolding would appear in additional smartphones, Yu confirmed that Huawei intends to expand its use following its introduction in the Mate XT 2 and subsequent deployment in the Mate 90 series.

“Yes, more of our phones will definitely use these LogicFolding chips in the future,” Yu said.

Any wider rollout will still depend on the advanced semiconductor capacity available to Huawei in China.

Want to learn more about AI and big data from industry leaders? Check out AI & Big Data Expo taking place in Amsterdam, California, and London. The comprehensive event is part of TechEx and is co-located with other leading technology events, click here for more information.

Tech Wire Asia is powered by TechForge Media. Explore other upcoming enterprise technology events and webinars here.