Research Report on 35 Domestic Processor Chip (CPU/GPU/FPGA) Manufacturers

By : ilikephone / On : 12/03/2022

Last year, the AspenCore analyst team compiled and released the "Research Report on 30 Domestic Digital Chip Manufacturers", including 30 representative domestic manufacturers of CPU, GPU, FPGA and memory chips. In the past year, more companies with technical strength and growth potential have emerged in the domestic digital chip industry. We decided to treat memory as a separate category and compiled the "Investigation and Analysis Report of 30 Domestic Memory and Main Control Chip Manufacturers". On the basis of the original digital chip report, we expanded the camp of CPU, GPU and FPGA manufacturers, and compiled it into the "35 Domestic Processor Chip (CPU/GPU/FPGA) Manufacturers Research Report".

Domestic CPU

Domestic CPU processors are mainly for PCs, servers, embedded systems, mobile phones and tablets, security monitoring, automobiles, and video and multimedia processing and other application markets. The AspenCore analyst team has compiled 16 domestic CPU chip manufacturers, including:

Among these 16 domestic CPU chip companies, half of them are already listed companies. The latest companies listed on the Science and Technology Innovation Board are Loongson Zhongke and Guoxin Technology, and Haiguang Information IPO has also been accepted by the Shanghai Stock Exchange. The support of the capital market will further promote the development of domestic CPUs in the fields of Xinchuang, industrial and information security applications, and will also help CPU manufacturers to enhance their R&D technical strength and build and expand their respective ecosystems.

The Loongson series CPUs of Loongson Zhongke include the "Loongson No. 1" small CPU for industrial applications, the "Loongson No. 2" medium CPU for industrial control and terminal applications, and the "Loongson No. 3" large CPU for desktop and server applications. . In 2021, Loongson Zhongke released a completely independent instruction set architecture--LoongArch. Based on this architecture, the single-core performance of Loongson 3A5000 is improved by 50%, and the power consumption is reduced by 30%. Compared with domestic CPU products, it has obvious performance advantages.

Based on the open Loongson ecosystem, the company has established close cooperative relationships with board and machine manufacturers, as well as basic software and application solution developers, providing downstream enterprises with various development boards and software and hardware modules based on Loongson processors. Loongson Zhongke can provide 32-bit, 64-bit single-core, multi-core and different quality level processors and supporting chips. The two major system softwares, Loongnix and LoongOS, can be adapted to different application scenarios.

Based on self-controllable embedded CPU technology and chip customization services for three key application fields of information security, automotive electronics and industrial control, edge computing and network communication, Suzhou Guoxin Technology has designed and developed a series of independent chips and modules. product. Based on the three major instruction sets of M*Core, PowerPC and RISC V, the company provides more than 40 CPU cores in 8 series with independent intellectual property rights. The comparison between its main embedded CPU cores and Arm cores is as follows:

Domestic GPU

Compared with domestic CPUs, domestic GPUs have developed later, and there are fewer manufacturers involved in the research and development of GPU processors. The performance of domestic GPUs is still far from that of the world's leading GPU chips such as NVIDIA, AMD, and Intel. However, in recent years, the unique advantages of GPUs in AI applications, coupled with the pursuit of capital, have driven the entrepreneurial boom of domestic GPUs. GPU/AI chip start-ups such as Biren Technology, Hanbo Semiconductor and Moore Thread have raised up to 1 billion yuan, attracting technical talents from international manufacturers such as NVIDIA and AMD to join, and will also promote the high-performance chip market segment of domestic GPUs competition and development.

The AspenCore analyst team has compiled 9 domestic GPU chip manufacturers, including:

Among the 9 domestic GPU manufacturers, only Jingjiawei is a listed company, Innosilicon has many years of experience in custom chip design, and the other companies are start-ups, but they have obtained considerable venture capital financing (among them Moore Thread, The cumulative financing amount of Biren Technology and Hanbo Semiconductor is more than 2 billion yuan).

The "Fenghua No. 1" GPU released by Innosilicon at the end of 2021 adopts the 12nm process, supports GDDR6 / GDDR6X (maximum rate of 19Gbps), the capacity is optional 4GB / 8GB / 16GB, and supports HDMI2.1 / DP1.4 /VGA multi-channel independent Output, support X86, ARM, Loongson and other instruction sets; support Linux, Android, Kirin, Tongxin UOS and other operating systems; support Kunpeng/Ampere and other server platforms. "Fenghua No. 1" is divided into two types: A type and B type. The specific performance indicators are as follows.

Tianshu Zhixin's 7nm general-purpose parallel (GPGPU) cloud training chip BI was successfully "lighted up" in December 2020. Based on this fully self-developed general-purpose computing GPGPU chip, Tianshu Zhixin's hardware products focus on cloud training and reasoning, release powerful programmability and application versatility through self-developed instruction sets, and provide industry-leading AI computing power density and energy efficiency Compare. It has a software and hardware architecture designed for cloud AI training and HPC general computing; supports floating-point, fixed-point, and multiple precision data types; provides ultra-high bandwidth local storage and inter-chip interconnect expansion. Tianshu Zhixin can support general computer vision related network models such as ResNet, Vgg, Inception, Alexnet, SSD, Mask R-CNN, etc.; provide ecologically compatible software suites; support multi-precision data type standard & hybrid training, and support model in-depth optimization.

Domestic FPGA

The recent successful acquisition of Xilinx by AMD means that it is difficult for FPGA to become a large-scale independent market and can only be used as a dedicated processing method in the field of high-performance computing. However, in the fields of 5G, industrial control and dedicated subdivision applications, FPGA still has the advantages that CPU/GPU/AI chips cannot replace. The overall technical strength of domestic FPGA manufacturers still lags far behind international manufacturers such as Intel and Xilinx, but several domestic manufacturers have been seen in the low- and medium-performance FPGA market.

The AspenCore analyst team compiled 11 domestic FPGA manufacturers, of which Anlu Technology, Ziguang Guowei and Fudan Microelectronics are listed companies, although the FPGA business does not account for a large proportion of the total revenue of Ziguang Guowei and Fudan Microelectronics. In addition to traditional FPGAs, some manufacturers have developed application-specific software and hardware processing solutions based on FPGAs. For example, the FPGA based on Quantum technology of Yilingsi can optimize the "power-performance-area" (PPA) by up to 4 times. Class 16nm FPGAs target autonomous driving, smart cockpit and electrification applications in new energy vehicles.

Lianjie Technology develops FPGA-based multimedia heterogeneous computing solutions for data center image and video, which can improve performance and efficiency by an order of magnitude, and has obtained US and Chinese patents. Lianjie's high-throughput, low-latency FPGA image processing acceleration technology solutions have been widely used in smartphone cloud applications, cloud storage and online video websites.

In terms of FPGA chip architecture, Anlu Technology, which has recently been listed on the Science and Technology Innovation Board, has developed the first-generation PHOENIX FPGA architecture that supports up to 600K logic array capacity. Second-generation FPGA architecture. In terms of system integration, based on the first generation of small-capacity FPSoC chips, the company will lay out the next-generation FPSoC chips from two directions of low power consumption and high performance, integrating CPU, FPGA and dedicated data processing modules to meet future application markets. trend.

In terms of dedicated EDA software, Anlu Technology's TangDynasty (TD) software is a self-developed FPGA integrated development environment that supports industry-standard design input, includes a complete circuit optimization process and rich analysis and debugging tools, and provides good The third-party design verification tool interface provides strong support for all application designs based on Anlu Technology's FPGA products. In addition, Anlu Technology will upgrade the core software algorithm for the PHOENIX2 architecture, develop system-level software compilation tools for FPSoC chips, and effectively support a rich product line of hardware products.

Details of 35 domestic processor chip manufacturers

Below we will show these 35 companies one by one in terms of core technology, main products, target market and competitiveness.

Electrotech Shentai

Suzhou Guoxin

Hexin Technology

Flathead Semiconductor

Biren Technology

35家国产处理器芯片(CPU/GPU/FPGA)厂商调研报告-电子工程专辑

Hubble Semiconductor

Moore thread

Core Pupil Semiconductor

Login Technology

Lianjie Technology

Zhongke Yihai

Yi Lingsi

Fudan Microelectronics

Zhongke Godson

Tianjin Feiteng

Haiguang Information

Zhaoxin

Huawei Hisilicon

Ziguang Zhanrui

Rockchip

Beijing Junzheng

Allwinner Technology

Jing Jiawei

Days Smart Core

Innosilicon

GOWIN Semiconductor

Shanghai Anlu

Ziguang Guowei

Jingwei Qili

Wisdom Polycrystalline

Chengdu Huawei Electronics

GLOBALFOUNDRIES

Amlogic

Guokewei

Zhongxingwei

35 domestic processor chip manufacturers research and analysis report.pdf
1562a786dd8a77bb0ffb718e2876af05.pdf (827.53 KB)
Editor in charge: Steve