China’s latest chip “breakthrough” is really about building a domestic lithography stack

TrendForce reports that Chinese chipmakers and equipment firms are hailing a new “breakthrough” in chip technology that supposedly paves the way for more advanced lithography and packaging, tied to AI-era memories like HBM, AI-DRAM, and AI-NAND. Strip away the nationalism and the press-release adjectives and the story is simpler: China is chipping away at foreign choke points in its semiconductor supply chain, one tool and one process node at a time.

What the “breakthrough” actually covers

The details are characteristically vague, but the themes are familiar: domestic tools for critical lithography steps, incremental progress on multi-patterning to stretch existing scanners further, and integration work around high-bandwidth memories and AI-oriented DRAM and NAND. None of this magically conjures a domestic EUV scanner overnight; the emphasis is on upgrading deep-UV capability, patterning tricks, and back-end technologies to close specific gaps exposed by export controls.

In practice, this usually means:

Better DUV patterning: Squeezing more out of 193 nm tools via advanced resists, multiple patterning, and tighter process windows.
Packaging and stacking: Building HBM-class stacks and AI-DRAM modules with local OSAT capacity, even if the front-end die are less advanced than leading-edge Western parts.
Toolchain substitution: Swapping banned foreign subsystems—lasers, optics, control electronics—with domestic or friendly alternatives, even at the cost of throughput and yield.

Why lithography is the hard part

Logic scaling is ultimately gated by lithography. Without EUV, you can push DUV a long way with clever patterning, but it gets expensive, power-hungry, and yield-sensitive. Export controls have not prevented Chinese fabs from shipping useful chips, but they do make it harder to compete at the very top end (leading-edge GPUs, premium smartphone SoCs) in a cost-effective way.

The current wave of announcements is about eroding the “no tools, no chips” argument. If domestic vendors can offer even mid-range lithography systems that are good enough for many AI and memory workloads, China reduces the leverage foreign governments have via tool bans.

HBM, AI-DRAM, and AI-NAND: why memory is in focus

Alongside the lithography messaging, TrendForce highlights Chinese interest in HBM, AI-DRAM, and AI-NAND. That choice is not random. High-bandwidth memory is a choke point for AI accelerators, and DRAM/NAND remain more forgiving of process lag than cutting-edge logic.

HBM: Even if local HBM lags SK hynix or Samsung by a generation, any domestic supply loosens dependence on imported stacks for AI accelerators and servers.
AI-DRAM: This is largely branding for DRAM tuned to AI workloads (higher bandwidth, better power efficiency, support for specific interfaces). It is easier to field at older nodes than a 2 nm CPU.
AI-NAND: Similarly, tuning NAND products and controllers for AI storage workloads—training data, retrieval-augmented generation—can be done on a wide range of process technologies.

How much is hype vs real progress?

There is genuine engineering work behind these announcements, but the messaging is crafted as much for domestic politics as for external investors. “Breakthrough” is a flexible word. The more relevant questions are:

Are domestic lithography tools shipping in volume to fabs, or only in pilot lines?
What are the actual line widths and cost structures compared to international peers?
How much of the HBM/AI-DRAM/AI-NAND stack is genuinely domestic versus assembled with imported critical parts?

Without those numbers, it is safer to treat this as another step in a long-term substitution process rather than a sudden leap to parity with ASML-equipped foundries.

Implications for the rest of the industry

More fragmented supply chains: As Chinese firms localise tools and processes, global chip supply becomes more bifurcated. Some products will be built mostly on Western/Japanese/ Taiwanese tools; others will lean heavily on domestically developed equipment.
Pressure on tool vendors: ASML, Tokyo Electron, and others enjoy a captive market today. Long term, even partially competitive Chinese tools will bring pricing and political pressure.
HBM competition: If Chinese memory vendors field credible HBM and AI-DRAM, Nvidia, AMD, and cloud providers get more supplier options—though qualifying those parts for mission-critical AI systems will not be trivial.

Editor’s take

The smart reading is that sanctions are working in the narrow sense—China is still behind at the absolute cutting edge—but they are also accelerating domestic tool and memory development. Everyone should assume that DUV-based scaling, clever packaging, and politically motivated capex will keep eating into the gap. The West is not standing still, but the days of assuming permanent reliance on a small set of tool vendors are numbered.