Tingbo He, president of Huawei's chip-design subsidiary HiSilicon, announced at the IEEE International Symposium on Circuits and Systems in Shanghai last weekend that her company has developed a chip-optimization approach capable of closing China's 5-year lag in AI semiconductor performance. HiSilicon targets 1.4-nanometer equivalent chipmaking performance by 2031, three years behind TSMC's expected 2028 introduction of the same process but a significant reduction in China's current deficit.
The new approach, which HiSilicon calls Tau's Scaling Law, optimizes performance across chips, circuits, and entire computing systems rather than relying on geometric transistor shrinkage. He said the method has replaced Moore's Law as HiSilicon's guiding principle. Moore's Law, named for Intel cofounder Gordon Moore, dictates that computing progress depends on roughly doubling the number of transistors packed into a chip every two years.
US export controls prohibit Huawei from working with TSMC, the world's leading chip foundry, forcing the company to rely on China's SMIC, which uses older-generation lithography machines. The restrictions aim to limit China's ability to develop frontier AI using domestically produced silicon. By some estimates, China lags more than 5 years behind the leading edge in AI chip development.
Key facts
- 01HiSilicon targets 1.4-nanometer equivalent chipmaking performance by 2031, three years behind TSMC's expected 2028 introduction of the same process.
- 02China's chip industry lags more than 5 years behind the leading edge due to US export controls blocking access to advanced lithography equipment.
- 03HiSilicon's Tau's Scaling Law optimizes performance across chips, circuits, and systems rather than relying on geometric transistor shrinkage.
- 04Tingbo He, HiSilicon's president, promised a proof-of-concept chip before winter 2026.
- 05Geometric scaling plateaued for HiSilicon six years ago, forcing the company to pursue alternative approaches to Moore's Law.
HiSilicon's technical approach includes LogicFolding, a method that reduces the time required to perform key logical operations within a circuit, and design optimizations that account for nanoscale electronic phenomena. The company is also developing interconnects that speed chip-to-chip communication, a critical bottleneck in training large AI models.
He promised a proof-of-concept demonstration before winter 2026. The company plans mass production of chips using the new techniques starting in 2027. HiSilicon's announcement suggests US sanctions aimed at kneecapping China's chip industry have spurred domestic innovations that could eventually erode America's technological edge.
The chip industry has begun running into the limits of Moore's Law as transistors approach the 2-nanometer width threshold. When transistors are just a few nanometers wide, quantum effects interfere with their normal functioning. Many chips are already made with workarounds — Apple's most powerful processors, for example, stitch two chips together to make a more powerful single one.
HiSilicon's strategy involves hybrid bonding and 3D chip stacking to improve performance beyond what geometric scaling alone can deliver. The company says its innovations reduce not just compute time but also the time data spends moving between chips and inside a chip, addressing a key bottleneck in both AI training and inference workloads.
“For both [AI] training and inference, the win is not just in shortening compute time. It is in shortening the time that data spends moving, between chips and inside a chip.”— Tingbo He, President of HiSilicon
Lennart Heim, an independent semiconductor and AI policy analyst, says Huawei's strategy suggests the company is running into limits on how much more performance it can squeeze out by shrinking and densifying chips alone. He notes the increasing reliance on techniques like hybrid bonding and 3D chip stacking.
Whether HiSilicon can execute on He's timeline remains uncertain. The company must demonstrate that Tau's Scaling Law can deliver measurable performance gains without access to the cutting-edge lithography equipment that TSMC uses. The promised winter 2026 demonstration will test whether HiSilicon's innovations are viable at scale or whether the company is overstating its progress to reassure domestic stakeholders.
Huawei's announcement complicates the assumption that US export controls will indefinitely preserve Western chip dominance. If HiSilicon can deliver on its targets, China gains a credible path to competing in AI infrastructure without relying on TSMC or other sanctioned suppliers. That shifts the timeline for when China might field domestically produced chips capable of training and deploying frontier models — and forces Western chipmakers to account for a competitor that's no longer constrained by Moore's Law orthodoxy.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
