Huawei’s chips are currently divided into six main categories, covering key areas such as terminals, communication, computing, and AI.
Huawei has built a full-stack domestic computing ecosystem with four major technology systems: Ascend (AI chips), Kunpeng (server chips), Euler (operating system), and Pangu (large models), with performance closely following international giants like NVIDIA’s GB200.
Huawei’s chips are currently categorized into the following six types, covering key areas such as terminals, communication, computing, and AI:
1. Smart Terminal SoC Chips (Kirin Series)
As Huawei’s most well-known chip product line, the Kirin series integrates CPU, GPU, NPU, baseband, and other modules, specifically designed for smart terminals such as smartphones and tablets. Its technical features include:
– Architectural Innovation: Utilizing ARM architecture and optimizing energy efficiency, such as the Kirin 9030 based on 5nm process, with a 30% performance improvement over the previous generation, supporting HarmonyOS 6.0.
– Communication Integration: Early integration of Balong 5000 baseband for dual-mode 5G support, with the latest products like Kirin 9020 further optimizing 5G-A technology, achieving a 100% increase in download speed.
– Imaging and AI Capabilities: Enhancing AI computing power through self-developed Da Vinci architecture NPU, combined with Kirin ISP for DSLR-level image processing, such as the Kirin 9030 supporting multi-focal length periscope continuous zoom technology.
2. AI-Specific Chips (Ascend Series)
The Ascend series focuses on artificial intelligence scenarios, divided into two main directions: cloud training and edge inference:
– Cloud Training: The Ascend 910B adopts the Da Vinci architecture, with a single card computing power of 320 TFLOPS (FP16), competing with NVIDIA’s A100, already used in large model training such as Baidu’s Wenxin Yiyan.
– Edge Inference: The Ascend 310 has a power consumption as low as 8W, suitable for edge devices like industrial robots and smart cameras, supporting full-scene AI deployment.
– Ecological Collaboration: Deep integration with the MindSpore AI framework to achieve “one-time development, multi-end deployment,” with plans to launch the next-generation Ascend 920 chip in 2025 to capture the NVIDIA H20 market.
3. General Server Processors (Kunpeng Series)
The Kunpeng series is based on ARM architecture, targeting cloud computing and data centers:
– High-Performance Computing: The Kunpeng 920 supports 64-core parallel computing, integrating 100G RoCE Ethernet cards, suitable for enterprise applications like virtualization and storage, replacing x86 servers.
– Domestic Replacement: Building a self-sufficient ecosystem through the openEuler operating system and openGauss database, with plans to develop over 1,000 native development partners by 2025, covering core industries like finance and electricity.
– Technical Collaboration: Forming heterogeneous computing with Ascend chips, where Kunpeng handles general tasks and Ascend focuses on AI computing, enhancing overall data center efficiency.
4. Communication Chips (Balong, Tiangang Series)
Covering two major communication scenarios: terminals and base stations:
– Terminal Baseband: The Balong series is integrated into Kirin SoCs, such as the Balong 5000 supporting NSA/SA dual-mode 5G, with download speeds reaching 1.4Gbps, and the latest products further optimizing 5G-A technology.
– Core Chips for Base Stations: The Tiangang series is used for 5G base stations, supporting large-scale MIMO and ultra-dense networking, reducing size by 50% and power consumption by 20%, promoting global 5G network deployment.
– Satellite Communication: The latest Kirin chips integrate Beidou satellite messaging functions, such as the nova 14 Ultra supporting bidirectional satellite communication, expanding emergency communication scenarios.
5. IoT Chips (Lingxiao Series)
Focusing on home and industrial IoT:
– Home Access: Lingxiao chips are used in routers and smart home devices, such as Hi5651 supporting 256 node connections, with built-in IPv6 hardware acceleration engine to enhance network stability.
– Industrial Applications: NB-IoT chips support narrowband IoT communication, with power consumption as low as micro-amps, already used in smart metering and logistics tracking scenarios.
– Short-Distance Communication: Combining SparkLink protocol to achieve low-latency, high-reliability device interconnection, such as the MatePad Pro 11 2024 supporting SparkLink connection and the third-generation M-Pencil stylus.
6. Other Specialized Chips
– Automotive Chips: The Kirin 9610A has passed automotive certification, with a computing power of 200K DMIPS, supporting L3-level autonomous driving, already applied in models like the Aito M9.
– Video Processing Chips: IPC SoC integrates ISP and video codec technology for smart surveillance cameras, supporting 4K HDR real-time processing.
– Ternary Logic Chips: Patent technology announced in 2024 can reduce the number of transistors by 40%, lowering power consumption to one-third of traditional binary, laying the foundation for future chip architecture innovation.
– The Huawei HarmonyOS smart home central control screen and Xiao Yi speaker use different chip solutions:
1. Smart Central Control Screen: The Huawei HarmonyOS smart central control screen (such as the Smart Central Control Screen S2, Central Control Screen MINI, etc.) is equipped with Kirin series chips similar to those used in Huawei tablets, such as Kirin 990. These chips have strong computing capabilities, supporting multitasking, high-definition display, and smooth operation of the HarmonyOS. Some high-end models also integrate AI Pangu large models, further enhancing voice interaction and intelligent decision-making capabilities.
2. Xiao Yi Speaker: The Huawei Xiao Yi speaker (such as Xiao Yi AI Smart Speaker, Xiao Yi Speaker 2e, etc.) mainly uses MediaTek MT8516 series chips. This chip is designed specifically for smart voice devices, integrating a quad-core ARM Cortex-A35 architecture, with a frequency of 1.3GHz, supporting multi-channel microphone arrays and Bluetooth/Wi-Fi communication, efficiently processing voice commands and optimizing sound quality. Additionally, some functions (such as AI large model interaction) may rely on cloud computing power support from Huawei’s Ascend series chips, but the terminal devices themselves primarily use MediaTek solutions.
It is important to note that different models of central control screens and speakers may adopt different configurations due to positioning differences, and specific chip information should be based on official product specifications. Technical collaboration and ecological layout.
Summary:
Huawei achieves “end-edge-cloud” full-scene computing collaboration through a unified technical base (such as Da Vinci architecture, HarmonyOS) and development toolchain (MindSpore, openEuler). For example, Kirin chips drive terminal devices, Ascend empowers edge AI, and Kunpeng supports cloud data centers, forming a chip matrix covering consumer electronics, communication, cloud computing, smart vehicles, and other fields. This full-stack layout not only enhances Huawei’s product competitiveness but also promotes the self-sufficiency development of the domestic semiconductor industry chain.