As U.S. export restrictions tighten on advanced semiconductor technology, China is exploring innovative strategies to bridge the performance gap with industry leaders like Nvidia. One such approach gaining attention is chip stacking, which aims to maximize the capabilities of older, domestically producible chips by integrating them into more sophisticated systems. But can this method truly rival the performance of cutting-edge GPUs? Let’s delve into what this strategy entails and its potential implications.

Understanding China’s Chip Stacking Strategy

The core idea behind chip stacking is to build upward—combining multiple, less advanced chips to create a system with enhanced performance. Chinese researchers, such as Wei Shaojun from Tsinghua University, have proposed integrating 14-nanometer logic chips with 18-nanometer DRAM modules using advanced 3D hybrid bonding techniques. This process involves stacking chips vertically and establishing direct copper connections at micron-scale pitches, reducing data transfer bottlenecks common in traditional architectures.

This approach targets the specific technological restrictions imposed by U.S. export controls, which limit access to manufacturing processes below 14nm for logic chips and 18nm for DRAM. By working within these boundaries, China aims to develop high-performance systems without relying on the most advanced fabrication nodes.

Performance Claims and Challenges

Proponents like Wei Shaojun claim that such stacked systems could reach performance levels comparable to Nvidia’s 4nm GPUs, with figures suggesting up to 2 TFLOPS per watt and total performance around 120 TFLOPS. However, actual Nvidia A100 GPUs deliver significantly higher performance—up to 312 TFLOPS—highlighting a notable performance gap.

While the architectural concept of stacking older chips is innovative, it cannot fully replicate the advantages of advanced process nodes, such as improved power efficiency, higher transistor density, and better thermal management. The challenge remains whether the performance benefits of stacking can adequately substitute for more traditional, smaller-process chips.

Nonetheless, this strategy could offer cost-effective solutions and reduce reliance on foreign technology, making it a strategic priority for China’s semiconductor ambitions despite performance limitations.

Inspired by

• https://www.artificialintelligence-news.com/news/china-chip-stacking-strategy-nvidia/

Sources

• scmp.com