Huawei is questioning the path the semiconductor industry has been following for decades. According to Liao Heng, Huawei’s chief chip scientist, manufacturers like Nvidia will eventually run up against fundamental physical limitations. Huawei is therefore opting for a different approach, as Techzine previously reported.
During a four-hour interview with Bloomberg, Liao outlined his vision for the future of chip development. In his view, continuously shrinking transistors is not a strategy that can be pursued indefinitely. Nvidia, on the other hand, is constantly scaling up its AI processors by combining more computing cores with ever-larger amounts of high-bandwidth memory (HBM). However, according to Liao, that model has an endpoint. “The industry will keep going, but once that physical limit is reached, an avalanche will follow,” he said.
A new path for chip development
Huawei is therefore working on an alternative design principle that the company calls the Tau Scaling Law. Instead of emphasizing further miniaturization, this approach focuses on faster communication between the various components of a computer system. As the efficiency of smaller transistors continues to decline, system architecture becomes more important than pure miniaturization.
Later this year, we’ll see just how effective that strategy is. Huawei is preparing to launch its first smartphone chip developed according to this design philosophy, using a technique called LogicFolding. Liao expects research firms such as SemiAnalysis and TechInsights to conduct an in-depth analysis of the chip after its launch, which will reveal the extent to which Huawei has narrowed the gap with its international competitors.
Sanctions drive innovation
These statements align with Huawei’s strategy to independently develop new chip technology despite U.S. export restrictions. While competitors such as TSMC, Intel, and Samsung base their future production plans on increasingly advanced lithography machines from ASML, Huawei does not have access to that equipment. According to Liao, this forces the company to make fundamentally different choices.
He also stated that the global semiconductor sector is increasingly splitting into two distinct ecosystems. Due to geopolitical tensions, he said, both China and the United States are building fully independent production and supply chains.
DeepSeek as an example
During the interview, Liao also praised the Chinese AI developer DeepSeek. According to him, the company proved that smart architectural choices can compensate for some of the lack of computing power. Precisely because DeepSeek did not have unlimited hardware at its disposal, it was forced to develop more efficient models.
Huawei aims to apply the same concept to its AI processors. Through a so-called co-design approach, chips and AI models are designed simultaneously, so that more efficient software can offset some of the hardware’s limitations.
Liao’s statements show that Huawei intends to compete not only with more powerful chips, but above all with a different design philosophy. Whether this strategy truly offers an alternative to the current trend toward scaling up in the AI chip market remains to be seen with the launch of Huawei’s new generation of chips later this year.