SMIC has squeezed more from old DUV kit, but the Kirin 9030 is not troubling TSMC just yet.
China’s biggest chipmaker has been forced to get creative because it cannot access EUV gear in any useful form. Huawei’s Kirin 9030 shows that SMIC can still advance its 7nm process.
According to Semianalysis, the interesting bit comes from the metal pitch. The Kirin 9030’s smallest metal pitch was 32.5nm, which is 10 per cent tighter than Intel’s 18A at 36nm on Panther Lake CPUs.
That sounds like SMIC has taken a swing at Chipzilla. It is even tighter than TSMC’s N6, which used EUV and measured 40nm on MediaTek’s Helio G99.
The problem is that density alone does not make a world-class chip, no matter how loudly anyone waves a process-node label around.
SMIC pulled this off with multi-patterning, design technology co-optimisation and complicated integration. That gave the Kirin 9030 better area figures, but the price was paid elsewhere.
Performance and efficiency are where the story gets less heroic. If you believe the Tame Apple Press the Fruity Cargo Cult’s efficiency cores reportedly deliver 20 per cent higher integer performance than the Kirin 9030’s prime core.
That would already sting, but Job’s Mob does it while drawing about 1W. Huawei’s prime core needs roughly 4.5W, which is not exactly sipping power.
SMIC can claim its 7nm N+3 process looks close to TSMC’s N6 in some ways. Unfortunately, N6 is an older node, so this is not the flex some boosters seem to think it is. The Kirin 9030’s prime core lands around Cortex-X2 territory for efficiency and performance. That Arm design arrived in 2021, making it old enough to have seen several AI hype cycles come and go.
The voltage-frequency curve is another headache. SMIC and Huawei do not have the transistor budget enjoyed by outfits designing around TSMC’s more advanced nodes.
That gap will widen again as the first 2nm SoCs arrive. TSMC’s advantage is not just density, but power, performance, yield and the boring manufacturing bits that actually matter.
Future Kirin chips might keep chasing density through multi-patterning and design tricks. That could produce impressive microscope shots while still leaving power and performance lagging.
SMIC is not close to Intel, Samsung or TSMC across the whole process technology stack. It has shown skill, persistence and a refusal to be boxed in by sanctions, but physics remains a miserable git.
Huawei’s LogicFolding packaging may be the more interesting route. Stacking could recover density, shorten signal paths and lift performance, while SMIC keeps wringing every last drop from DUV.
