虎嗅

Beyond the Hype: Explaining the Mathematical and Physical Principles Behind Huawei's Tao Law

原文:超越炒作:科普华为韬定律背后的数学与物理原理

Summary of Key Points

This article uses the analogy of a “dark, automated factory” to explain how traditional two-dimensional chips are reaching performance bottlenecks due to issues such as wiring delays and heat dissipation. Huawei’s proposed “logic folding” (three-dimensional chip architecture) overcomes these limitations by stacking components in three dimensions, thereby shortening signal transmission paths. The article analyzes both the mathematical principles and advantages of this technology, as well as practical challenges like heat dissipation. It concludes that this represents an inevitable innovation in the semiconductor industry beyond Moore’s Law, serving both as a technological breakthrough and a strategy to overcome geographical constraints.

I. The Dilemmas of Two-Dimensional Chips

Imagine a chip as a fully automated, dark factory:

  • Transistors = Robots: Responsible for processing data;
  • Logic Units = Workshops: Composed of numerous robots performing specific tasks (e.g., addition);
  • Electrical Signals = Transport Vehicles: Transfer processed components (data) between workshops;
  • Global Clock = The Scheduling Signal: Controls the simultaneous start and stop of all workshops.

Over the past few decades, efficiency was improved by reducing the size of transistors (robots), allowing more of them to fit within the chip. However, this created new problems: as the number of logic units increased, the chips had to be laid out in a flat plane, forcing electrical signals to travel long distances, which often took longer than the actual processing time.

In chips:

  • The switching speed of transistors is almost negligible, but the time it takes for electrical signals to travel through copper wires accounts for 75%-80% of the total delay.
  • Longer wires result in higher resistance and electromagnetic interference (parasitic capacitance), with delays increasing exponentially with distance. For example, doubling the length doubles the delay.
  • Although smaller manufacturing processes (e.g., from 14nm to 5nm) reduce wire lengths, the decreased thickness and increased spacing lead to increased resistance and interference, offsetting the benefits of shorter wires.

This is where two-dimensional chips reach a dead end: no matter how transistors are miniaturized, the bottleneck of signal transmission cannot be overcome.

II. Huawei’s “Three-Dimensional Solution”

Huawei’s approach is simple: transform the flat chip factory into a three-dimensional one!

  • In a traditional two-dimensional chip, signals had to travel long distances (e.g., 40mm diagonally).
  • With three-dimensional stacking, signals can use vertical channels to directly reach the desired destination, significantly reducing travel times (e.g., from 40mm to 9mm).

The mathematical advantage is clear:

  • In two-dimensional chips, the side length of the chip is proportional to the square root of the number of transistors; in three-dimensional chips, it’s proportional to the cube root.

For example, if the NVIDIA H200 chip were made using 14nm technology in a flat layout, its side length would be over 53mm (beyond the limitations of current lithography machines). With three-dimensional stacking, the side length would only be 5.2mm, reducing key path delays by 99%.

This is the core of Huawei’s “τ scaling law”: using three-dimensional layouts to minimize signal transmission times and compensate for the slower speeds of older manufacturing processes.

III. Benefits and Challenges of Three-Dimensional Chips

Benefits:

1. Performance Improvement: Shorter signal paths allow chips to operate faster.

2. Lower Power Consumption: Reduced voltage requirements (power consumption is proportional to the square of the voltage); for instance, 14nm three-dimensional chips may consume less power than 5nm two-dimensional chips.

3. Overcoming Process Limits: High-performance chips can be produced using existing, mature manufacturing processes (especially important for Huawei).

Challenges:

The biggest challenge is heat dissipation. While two-dimensional chips can use backside cooling pads, three-dimensional chips with multiple layers face significant thermal issues. Recent research from KAIST in South Korea (using embedded liquid cooling) could potentially solve this problem.

IV. Why Previous Approaches Failed

The industry has attempted to bypass the limitations of two-dimensional chips, but they all had fatal flaws:

1. Pipeline Segmentation: Dividing long signal paths into segments with buffers (registers) increased delays and wasted space and power.

2. Asynchronous Architecture: Removing a global clock allowed workshops to operate independently, but this led to data bottlenecks (fast workshops blocking channels, slowing down others).

These methods only provided temporary solutions and did not address the fundamental issue of wiring delays.

V. The Dawn of a New Era: What Does Huawei’s Breakthrough Mean?

Moore’s Law (doubling transistor counts every 18 months) is reaching its limits, as transistors can no longer be miniaturized significantly. Future chip improvements will rely on architectural innovations, with three-dimensional stacking being a key direction.

For Huawei:

  • Geographical constraints prevent the use of advanced 5nm and below processes; therefore, three-dimensional architecture is a necessary alternative.
  • Solving heat dissipation issues would enable Huawei to produce high-performance chips using mature processes.

For the industry:

  • Three-dimensional chips are an inevitable choice for the post-Moore era, regardless of geographical constraints.
  • Huawei’s innovation could accelerate the adoption of three-dimensional technology across the industry.

In conclusion, Huawei’s logic folding architecture represents a genuine technological breakthrough, supported by both mathematical and physical principles. It is also a clever marketing narrative that simplifies complex concepts. However, the real value lies in its ability to solve the core limitations of two-dimensional chips, marking a necessary step forward for the industry.

Conclusion

The era of flat chips is coming to an end, and the dawn of three-dimensional chips has begun. Huawei’s logic folding architecture is not only a strategy to overcome geographical barriers but also a key to unlocking new possibilities in computing power. If heat dissipation issues can be resolved, three-dimensional chips will completely transform the semiconductor industry.