The Huawei Kirin 9050 has raised concerns, as it signals a significant shift in HiSilicon’s strategy. This is all tied to Huawei’s new approach, where they aim to compensate for the lack of access to cutting-edge manufacturing processes with 3D packaging to increase density, performance, and efficiency within SMIC’s current limitations. The chip is expected to launch this year, and most impressively, it could outperform the Apple A18 Pro.
The core of this news is that Huawei won’t be competing against Apple or Qualcomm from the same industrial standing. Apple manufactured the A18 Pro using TSMC’s second-generation 3nm technology. This SoC features a 6-core CPU, a 6-core GPU, a 16-core Neural Engine, increased memory bandwidth, and efficiency improvements over the A17 Pro. Huawei, on the other hand, remains constrained by restrictions on EUV technology and its reliance on SMIC, whose demonstrated capability in recent Kirin chips has hovered around 7nm-class processes.
The miracle for Huawei to lead China lies not in lithographic processes, but in ‘LogicFolding’
This context aligns with Huawei’s broader announcement regarding the Tau Scaling Law and LogicFolding, a strategy aimed at improving performance by reducing internal distances, latencies, and data movement within the chip. This approach focuses on optimization rather than solely pursuing smaller transistors. Huawei plans to implement this architecture in its upcoming Kirin smartphone chips this year. However, its long-term ambition is to achieve densities equivalent to 1.4nm processes by 2031 through 3D packaging.
To claim that the Kirin 9050 SoC, destined for the Huawei Mate 90, would surpass the Apple A18 Pro is a significant assertion. Currently, there are no reported results demonstrating this to be the case. The specific benchmarks used are not detailed, nor is it specified whether the comparison pertains to CPU, GPU, AI, multi-core performance, synthetic workloads, or sustained performance. Furthermore, power consumption, a critical factor in smartphones, is not mentioned. Nevertheless, this is Huawei we’re talking about. They went from being the world’s largest smartphone manufacturer to nearing bankruptcy following US sanctions. Today, not only have they recovered, but they are already competing head-to-head with NVIDIA in AI chips within China. The company has regained its composure and is stronger than ever.
Huawei is striving to demonstrate that US sanctions do not necessarily prevent progress, although they do necessitate navigating more complex paths. Unable to directly compete in leading-edge lithography against TSMC, the company is seeking performance gains through advanced packaging, vertical integration, stacked chips, and data flow optimization. This strategy aims to shift away from the traditional lithographic scaling race towards system-level efficiency. While 3D packaging can enhance density and reduce latency, it introduces challenges related to heat, cost, design complexity, manufacturing yield, and EDA tools. Indeed, Huawei acknowledges that overheating and the need for new design tools remain significant obstacles. In mobile devices, this is crucial: it’s not enough to win a short benchmark; sustained performance, battery life, and thermal management are essential.
