Summary of Key Points
Huang Renxun's statement that "Huawei's 'Tao Law' does not pose a threat" has sparked controversy, raising three critical issues: Is his claim that NVIDIA leads by ten years credible? What is the difference between "logic folding" and "chip stacking" in chip technology? Why is this development more significant for domestic chips than the "Tao Law" itself? Essentially, it demonstrates the innovative path taken by domestic chips under restricted circumstances, challenging the technological perceptions of international giants.
1. How Credible is Huang Renxun's Claim of a Ten-Year Lead?
Huang Renxun argues that NVIDIA's lead over its competitors stems from two main factors: the performance advantages of its GPU architectures (such as the latest Hopper) and the CUDA software ecosystem, which serves as a universal toolset used by AI developers worldwide and is difficult for others to replace. However, Huawei's Ascend chips have recently demonstrated performance close to that of NVIDIA's A100 in large-model training scenarios, and Huawei's MindSpore software is also improving rapidly, with more domestic developers adopting it.
The claim of a ten-year lead is more of a marketing tactic. While NVIDIA does have a strong ecosystem, Huawei has made significant progress in certain areas (such as domestic AI applications), and the gap is not as substantial as claimed. Huang Renxun's confidence likely reflects past advantages, but he did not anticipate Huawei's rapid catch-up under limited conditions.
2. What Exactly is the "Tao Law," and Why Did Huang Renxun Misunderstand It?
The "Tao Law" likely refers to Huawei's strategy of using advanced packaging technologies (such as chiplet stacking) to enhance chip performance after facing limitations in chip manufacturing processes. Instead of competing with others on the most advanced 3nm/5nm processes, Huawei focuses on leveraging mature processes combined with innovative packaging methods.
Huang Renxun may have underestimated the effectiveness of this approach, believing that chip stacking was not as sophisticated as NVIDIA's single-chip logic folding. However, Huawei's Ascend 910B, which combines multiple 7nm chips through packaging, has achieved performance comparable to the A100. His misunderstanding highlights that Huawei's strategy is not about technological backwardness but about using mature processes and innovative packaging to circumvent manufacturing bottlenecks—a approach well-suited to China's current situation.
3. Logic Folding vs. Chip Stacking: Two Approaches to Survival in the Tech Race
In simple terms:
- Logic Folding: This involves packing more computing units and circuits into a single chip to increase its power, requiring advanced processes (like 3nm) and facing higher costs and potential limitations (e.g., from ASML's lithography machines). NVIDIA has primarily relied on this approach.
- Chip Stacking: Similar to building with Lego bricks, this method combines multiple smaller chips (for computing, storage, communication, etc.) using high-speed interconnect technologies. Each small chip can be manufactured using mature processes (like 7nm), reducing costs and bypassing process-related constraints. Huawei is currently breaking through using this approach.
In other words, logic folding focuses on single-point breakthroughs, while chip stacking represents a combination of different components for a more adaptable and sustainable strategy.
4. The Greater Significance of This Development for Domestic Chips
This advancement is more important than the "Tao Law" for several reasons:
1. It dispels the misconception that without advanced processes, high-end chips cannot be made: Huawei's success with chip stacking shows that mature processes and innovative packaging can produce chips comparable to international standards, providing a clear path for domestic chip companies.
2. It challenges the notion that international giants' technological barriers are insurmountable: Huawei's progress has prompted NVIDIA and others to recognize the potential of Chinese innovation, indicating that domestic chips now have the capability to compete.
3. It drives industrial upgrading: This development will benefit domestic packaging companies (such as Changjiang Electronics and Tongfu Microelectronics) and chiplet technology firms, promoting the entire industry towards more advanced packaging methods and reducing dependence on foreign processes.
In summary, Huang Renxun's "blowback" event marks a shift for domestic chips from following others to taking a leading role in innovation. It demonstrates that we can develop our own innovative paths without relying on foreign technologies.