Who will break through Nvidia’s “iron wall”?

Processor chips are key components in computer systems and are used to perform various computing tasks. These chips can be divided into several categories based on their design and purpose, including central processing units (CPUs), graphics processing units (GPUs), field programmable gate arrays (FPGAs), etc. They have different architectures, performance characteristics, and application areas, such as CPUs for general computing tasks, GPUs focusing on graphics and parallel computing, and FPGAs that can be customized on demand for specific tasks. Therefore, choosing the appropriate processor chip is crucial in different application scenarios.

The full name of CPU is Central Processing Unit (Central Processing Unit), which is the core component of electronic computer system. The CPU is responsible for interpreting computer instructions and processing data operations in computer software. The CPU chip integrates arithmetic logic units, control units, registers and other components. It can read instructions, decode instructions, execute instructions, coordinate the work between various components, and is the "brain" of the entire computer system. Unlike other chips, the CPU can perform different tasks according to the order of instructions. The clock frequency of the CPU determines the running speed of the computer, and the instruction system determines the types of tasks that can be performed.

GPU stands for Graphics Processing Unit, which is a microprocessor that specializes in processing graphics and image tasks. GPU can perform floating-point calculations very efficiently through its massively parallel stream processor architecture, and is particularly suitable for matrix calculations. Compared with CPU, which pays more attention to serial calculation and control flow, GPU is better at data flow calculation. GPU chips have more computing cores and higher memory bandwidth, making them perform far better than CPUs in applications that require large-scale parallel computing, such as image processing and scientific computing. In recent years, GPUs have been widely used in emerging applications such as artificial intelligence, which can significantly improve computing performance.

FPGA stands for Field-Programmable Gate Array and is a programmable integrated circuit. The FPGA chip is composed of programmable logic blocks (CLB) and programmable interconnections. Users can determine the functions of each logic block in the chip and the connection methods between logic blocks through programming according to their needs. Compared with traditional application-specific integrated circuits ASIC, FPGA has the characteristics of strong programmability, short development cycle, and high design flexibility. It can quickly implement design ideas and perform functional verification, and is very suitable for product design of small and medium-sized batches and multiple variants.

Figure 1: Comparison of characteristics of CPU, GPU and FPGA

Source: Baidu Encyclopedia, CSDN

The instruction set is a collection of computer operation instructions, a "dictionary" for computer operation, and one of the core technologies at the bottom of the CPU. Domestic CPU manufacturers have chosen four instruction set routes out of their own considerations, each with its own advantages and disadvantages.

Figure 2: Overview of mainstream domestic CPU manufacturers

Data source: Compilation of public information, Xinshiye Research Institute

Whether in the server market or PC market, x86 processors are absolutely mainstream. Research data from Counterpoint Research shows that in the global server market in 2022, the x86 instruction set will account for as much as 91% of the server CPU market. Haiguang and Zhaoxin use x86 architecture, which is highly compatible with mainstream operating systems and has obvious ecological advantages. However, it is difficult to obtain the latest technology authorization and can only improve performance by increasing the number of physical CPU cores.

The ARM instruction set is open to CPU manufacturers for licensing by charging licensing fees. Feiteng and Huawei Kunpeng adopt the ARM v8 architecture, which has faster performance improvement and better ecological adaptability, and has released a series of PC and server products based on the ARM architecture. However, the ARM licensing model is also the sword of Damocles hanging over its head. Once supply is cut off, subsequent independent iterations will be difficult to sustain.

It is worth mentioning that on July 12, 2023, China Electronics and Huawei decided to merge the Kunpeng ecosystem and the PKS ecosystem to jointly create the "Pengteng" ecosystem that supports both Kunpeng and Feiteng processors, and work together with industry partners to develop together and create a universal A new landscape of computing power.

Different from the previous two routes, Loongson chose to independently develop the LoongArch instruction set after obtaining the MIPS instruction set authorization, while Sunway independently developed the SW-64 instruction set based on the Alpha instruction set. Although it has its own necessity and controllability, urgency, but its ecological maturity has yet to be improved.

Among them, Loongson's upper-layer application ecosystem has been gradually built and has certain applications in party and government and special markets. Sunway's products focus on the field of special supercomputing servers, with exclusive barriers to fully controllable technology and ecology.

In addition to the six major domestic CPUs, the RISC-V processor Xuantie 910 released by Pingtou Ge in 2019 officially brought the RISC-V architecture into the public eye. The biggest advantage of RISC-V is that it is open source and free. Its open attributes and rich instruction set library fundamentally eliminate the possibility of "stuck", and its development prospects are broad.

从全球市场来看，CPU行业龙头集中效应显著，国产CPU市场空间广阔。根据市场调研机构Counterpoint最新公布的报告显示 ^[1] ，2022年英特尔占据服务器CPU市场70.77%的份额，AMD则占19.84%；在2022年的数据中心CPU市场，英特尔有71%的市占率，与2021年相比下滑了16%，AMD则以20%的市占率位居第二。

图3：2021 & 2022全球服务器CPU市场格局

资料来源：Counterpoint

另据Mercury Research数据 ^[2] ，2022年Q4英特尔和AMD桌面CPU全球出货量份额分别约为68.1%和18.6%，预计2026年全球CPU出货量达到29亿颗，市场规模达到1336亿美元。

当前全球PC GPU芯片市场主要由英伟达、英特尔和AMD三家厂商垄断，国内厂商加速布局。

图4：2022年全球GPU市场格局

资料来源：Jon Peddie Research

国产GPU公司主要有两大方向，一类是面向图形处理的GPU芯片，包括景嘉微、芯动科技、摩尔线程、芯瞳半导体等；另一类是面向通用计算的GPGPU芯片，包括寒武纪、壁仞科技、海光信息、沐曦等。近年来，国产厂商在GPU赛道发力，推出了较为成熟的产品，在性能上不断追赶行业主流产品，在特定领域达到业界一流水平。

图5：主流国产GPU厂商梳理

资料来源：企业官网，芯师爷

不过需要重视的是，尽管国产GPU在价格方面有一定优势，但在生态建立方面，仍然不如英伟达完善。业内人士称“CUDA是英伟达最深的护城河”，英伟达通过建立CUDA生态聚拢了大批客户和开发者，得以称霸高性能计算市场。除了CUDA，目前主流的开发平台还包括AMD ROCm以及OpenCL。当前国产GPU大多兼容英伟达CUDA，融入大生态进而实现客户端导入，仍需逐步建设生态系统，提高用户体验，从而带来更大的市场份额。

据Global Market Insights数据 ^[3] ，全球GPU市场预计将以CAGR 25.9%持续增长，至2030年达到4000亿美元规模。其中AI领域大语言模型的持续推出以及参数量的不断增长有望驱动模型训练端、推理端GPU需求快速增长。我国GPU厂商正处于技术追赶环节，市场份额较小，随着国内数据中心、智能驾驶及终端侧GPU市场需求的提升，国产GPU市场份额有望实现渗透。

根据Frost&Sullivan的数据 ^[4] ，预计全球FPGA市场规模将从2021年的68.6亿美元增长至2025年的125.8亿美元，年均复合增长率约为16.4%；中国FPGA市场2020年的市场规模约150.3亿元，预计2025年中国FPGA市场规模将达到332.2亿元，复合增速为17.2%。

国产FPGA公司营收体量相差不大，竞争优势各不相同。 复旦微电 进入市场较早，推出亿门级FPGA主要应用于高可靠等特定领域，毛利率水平普遍较高 ^[5] ； 安路科技 的主要优势在于小型FPGA产品的市场基础扎实，凤凰系列28nm工艺打入全球FPGA中高端市场 ^[6] ；

智多晶 目前已实现55nm、28nm中低端FPGA的量产，并针对性推出了内嵌Flash、SDRAM、DDR等集成化的产品，每年发货量超过1千万片； 紫光同创 覆盖高、中、低端等多层次FPGA市场； 易灵思 在机器视觉、LED显示、工业控制等领域有明显优势并且获得头部客户认可，积极推进16nm工艺节点的新产品。

图6：主流国产FPGA厂商梳理

资料来源：企业官网、招股书、财报，芯师爷

工艺制程是评价FPGA首先考虑的指标。纵观海外FPGA市场，赛灵思Versal系列已经进入7nm制程工艺，英特尔Altera与之对标的Agilex系列也已采用Intel 7制程工艺。

而中国FPGA市场 仍以低端容量和中端制程为主 ，可触达的市场空间依然广阔。Frost&Sullivan数据显示，2019年28nm-90nm制程的FPGA占中国市场份额为63.3%，28nm以下制程的FPGA芯片占据20.9%的市场份额，大于90nm制程的FPGA占比15.8%。由于28nm-90nm制程的FPGA在性价比和良品率方面具有优势，28nm以上制程是FPGA国产替代的重要方向。

图7：2019年中国FPGA芯片格局（按芯片制程工艺）

资料来源：Frost&Sullivan、安路科技招股书，芯师爷

此外，FPGA需要由FPGA芯片、EDA软件及IP方案组成软硬件生态系统，在海外厂商建立成熟的EDA软件生态环境的同时，已有多家国产FPGA厂商推出自研EDA软件，如紫光同创的Pango Design Suite、复旦微的Procise、安路科技的TangDynasty和FutureDynasty以及易灵思的Efinity等。

图8：国内外FPGA领先厂商EDA软件工具一览

资料来源：企业官网，芯师爷

根据芯师爷统计，国产FPGA厂商高度重视产品技术研发，报告期内研发投入占营业收入的比例高达30%-52% ^[7] ，处于较高水平，且较上年同期均呈现明显增长。以加强其技术壁垒，保持其市场领先地位。

图9：国内FPGA领先厂商2023年半年度研发投入

资料来源：企业2023半年度财报，芯师爷

RISC-V架构凭借其开源、精简、可扩展性强的特性，极有希望成为半导体领域的第三大架构生态。2023 RISC-V中国峰会发布数据显示，经过约5年时间建设，中国的RISC-V生态已初具规模。2022年，全球采用RISC-V架构的处理器已出货100亿颗，中国企业RISC-V芯片出货量达50亿颗，占据半壁江山，预计2025年全球RISC-V处理器出货量将突破800亿颗。

图10：三大指令级架构对比

资料来源：公开资料整理，芯师爷

近年来，国内涌现出了众多RISC-V赛道的处理器企业，如 阿里平头哥 已推出3大系列8款RISC-V处理器，是中国RISC-V领域影响力和市占率最大的处理器企业； 华为海思 于2021年底发布了首款商用级别的RISC-V架构CPU，并基于此打造了高清电视芯片； 易灵思 是全球第一家商用RISC-V的FPGA厂家，在16nm、40nm制程有长期的产品规划； 安路科技 则将FPGA和RISC-V融合在一起，推出了FPSoC， 助力实现视频图像接口转换和工业控制交互。

从市场来看，据调研机构Frost&Sullivan报告显示，预计全球FPGA需求将从2021年68.6亿美元增长为2025年的125.8亿美元，年均复合增长率为16.4%。中国FPGA市场将从2020年的150.3亿元人民币增长至332.2亿元人民币，年均复合增长率达到23.1%。

由此可见，RISC-V有望助力中国在处理器芯片领域绕开ARM和X86的高昂授权费以及贸易战下的卡脖子风险，为芯片产业链自主可控提供可贵的历史机遇。

RISC-V作为一个开源指令集架构，其生态系统的丰富程度直接影响着未来发展，健全的RISC-V生态将促进其快速商业化，使其成为可竞争的通用芯片架构选择。为此， RISC-V基金会 应运而生，累计吸引了全球3000多家芯片设计公司加入，其中中国公司占比将近一半。

图11：RISC-V基金会成员（节选部分）

资料来源：Semiwiki

2023年6月，谷歌、英特尔、英伟达、高通以及平头哥等13家企业发起了 全球RISC-V软件生态计划“RISE” ，进一步加速RISC-V在移动通信、数据中心、边缘计算及自动驾驶等领域的技术和商业化进程。

中国的RISC-V生态建设亦进行得如火如荼：2018年， 中国RISC-V联盟（CRVA） 依托中科院计算所成立，倪光南担任联盟理事长，目前成员包含百度、华为、腾讯、兆易创新、国芯科技等。

2023年8月31日，中国电子工业标准化技术协会RISC-V工作委员会正式成立，是从事RISC-V产业领域相关单位及组织等自愿组成的全国性、行业性、非营利性社会团体。

此外，在第三届滴水湖中国RISC-V产业论坛上，芯原半导体、平头哥半导体、芯来科技、赛昉科技等九家企业携手成立了国内首个“ RISC-V专利联盟 ”，旨在打造一个RISC-V专利互不诉讼的生态系统，共同推动国产RISC-V技术的不断创新和快速发展。

GPU凭借其强大的并行运算能力、深度学习能力、极强的通用性和成熟的软件生态，成为数据中心加速的首选，90%左右的AI服务器采用GPU作为加速芯片。

Trendforce集邦咨询预计 ^[8] ，2023年AI服务器出货量约为120万台，同比增长38.4%，占整体服务器出货量9%，2022至2026年AI服务器出货量CAGR将达22%，而AI芯片2023年出货量将增长46%。其中，GPU作为数据并行处理的核心，是AI服务器的核心增量，用于加速图形处理、深度学习、计算机视觉和自然语言处理算法等领域。

可以确定的是，未来5至10年，在国产GPU已有自主产品发布并且大模型训练缺乏基础设施的情况下，市场增速相较过去5至10年一定会呈现更高速增长。

随着汽车逐渐步入智能化时代，传感器数量的增加及交互能力的提升，将带来数据的几何式增长，必然要求车端拥有强大的数据分析和处理能力。在自动驾驶领域，各类自动驾驶芯片得到广泛的应用。根据Yole数据 ^[9] ，全球自动驾驶市场2025年将达到780亿美金，其中用于自动驾驶的AI芯片超过100亿美元。

在这一领域，GPU主要应用于车端及其配套设施智能芯片，负责处理来自摄像头、普通雷达、激光雷达等传感器数据，实现智能驾驶。ICVTank的自动驾驶渗透数据显示 ^[10] ，假设GPU在L2中渗透率15%，在L3-L5中渗透率50%，估算得到GPU在自动驾驶领域的市场规模将从2020年的7.1亿美元上升至2025年的44亿美金，CAGR为44%。

FPGA芯片下游涉及通信、工业、军工、航天、汽车和数据中心等领域，其中通信市场占比最高，工业、汽车市场不断扩大。多家受访企业向芯师爷表示， 通信与工业在国内的合计市场份额可高达70% ，与Frost&Sullivan的调研数据基本吻合：该机构认为，2022年通信/工业/数据中心/汽车/消费电子/人工智能在国内FPGA下游应用中占比分别为41.52%/31.23%/10.54%/6.94%/5.89%/3.88%。

此外该机构预计，2025年FPGA在中国通信领域的市场规模可达140.4亿元，3年CAGR为17.43%；在中国工业领域将达到100.8亿元，CAGR为15.6%。

新兴应用市场需求的涌现，持续拓宽FPGA应用场景。FPGA确定性的低时延和低功耗使其在汽车电子上具有非常大的优势，可为汽车电子需求提供灵活的低成本高性能解决方案。据芯师爷不完全统计，当前已有高云半导体、易灵思等国内FPGA厂商积极布局汽车赛道。

此外，不少企业认为FPGA在光伏储能的工业控制领域，矩阵运算、图像处理、机器学习等人工智能领域以及可穿戴设备领域应用前景广阔。

Mordor Intelligence Research&Advisory预测 ^[11] ，2023年全球FPGA的市场规模为87.6194亿美元，2028年预计将达到131亿美元，年复合增长率可达8.32%。

另据福建电子信息集团统计，2021年我国FPGA国产化率已由2018年的3%增长至17%，预计2025年有望达到37%。随着我国集成电路设计产业在FPGA领域研发投入的不断增加和人才培养力度的不断加大，国产FPGA产品有望缩小与国际先进水平的差距，国产化率得到持续提升。

“信创”即信息技术应用创新，旨在针对硬件及云等基础设施、基础软件、应用软件、网络安全等IT产业链核心技术产品进行自主研发，为我国经济发展、社会运转构建安全可控的信息技术支撑，避免核心技术受制于人。

CPU和GPU作为信创基础硬件发展环节的重中之重，也是决定信创底层发展逻辑的关键所在。当前我国信创芯片面临制程工艺、半导体配套供应链和应用软件生态三大挑战。在信创生态建设上，国内已形成由 中国电子、中电科、华为系、中科系 组成的四大信创生态体系，可满足部分关键领域和重要信息系统最基本的应用需求。

目前，信创产业已经从 “试点实践期”进入到“规模化推广期”的关键阶段，产业需求正在全面打开，市场订单正在井喷式增长，产业红利预计将会持续到2027年前后。自2020年以来，信创产业由党政逐渐向其他行业覆盖，以金融、电信等为代表的行业信创也进入规模化应用阶段。以电信、移动、联通为代表的运营商2020年开始集采国产服务器，目标是5年内实现完全替换。

以飞腾为例，受益于国家信创大趋势，其营收自2019年的2.07亿元增长至2021年的22.18亿元，呈现出高速发展趋势。根据《从端到云基于飞腾平台的全栈解决方案白皮书》 ^[12] ，飞腾未来五年内将投入超过150亿元用于新品研发、生态建设和区域客户保障，以百万、千万供货为目标提升产品交付能力，目标到2024年实现年营收超过100亿元。

景嘉微的JM7200高性能GPU芯片则对标Nvidia GT640，参与国产生态构建，已完成民用信创市场批量落地。 据《2023年中国信创产业研究报告》显示 ^[13] ，预计到2026年，我国信创市场将超过2000亿规模，未来市场前景广阔。

Chiplet可以实现不同工艺制程、不同类型芯片间立体集成，实现更大芯片面积、更大存储容量和更快互连速度，有望助力国产半导体厂商突破海外科技领域制裁。

目前包括英特尔、AMD、英伟达、海思等国际厂商在内，已有超过25%的高性能CPU和GPU都采用了Chiplet技术设计，如英特尔的Sapphire Rapids和AMD的RDNA3。国产处理器方面，龙芯中科发布了龙芯3D5000处理器，首次使用Chiplet将2个龙芯3C5000封装在一起，做到了32核。

在暗硅趋势与成本压力下，“CPU+GPU”和“CPU+协处理器”的异构计算无疑是未来的发展趋势。CPU与GPU形成的异构计算系统，可以完成多种不同的计算任务，其中GPU承担并行加速运算，CPU负责系统控制。这种计算单元协同的异构模式，能够在复杂的计算场景下获得更高的性能。

“CPU+协处理器”以FPGA为例，复旦微电在国内发布的首款高性能PSoC产品采用异构计算，突破了现有FPGA产品的发展瓶颈，大幅提升了芯片的任务处理性能；安路科技推出的FPSoC产品，则以单芯片实现FPGA和MCU芯片优势的融合。

随着英伟达GPU H100、A100采用了HBM2e、HBM3技术，HBM应用逐步走向成熟，凭借超高带宽的特性成为AI服务器与高端GPU的标配，为处理器发展速度不均衡、数据搬运慢、搬运能耗大等问题提供了解决方案。

市场调研机构Omdia预测 ^[14] ，随着并行计算和AI技术的发展，适用于并行计算的HBM需求将成倍增长。 到2025年，HBM市场的总收入将达到25亿美元，其规模约为2020年水平的5倍 。对于处理器芯片而言，HBM的重要性在于提高传输速率和存储容量，实现大规模并行计算下算力、存力、运力三者同时匹配。

随着大模型时代的到来，未来的AI模型对算力的需求将大幅增加，预计将突破1000 TOPS，这意味着需要处理更大规模的数据。因此,“存算一体”芯片架构的优势将更加明显。这种架构将所有的计算操作集成在存储器内，从根本上消除了传统将存储和计算分离带来的性能瓶颈，极大地提升了数据的处理速度，同时大幅降低功耗。未来HBM、MRAM和RRAM等新型存储介质将日渐成熟并商业化,处理器芯片的发展前景将更加广阔。

上海安路信息科技股份有限公司（688107.SH），创立于2011年11月，是国内领先的集成电路设计企业。公司具备FPGA芯片硬件和FPGA编译软件的自主研发能力，专注于研发通用可编程逻辑芯片技术及系统解决方案。于2021年在上交所科创板成功上市，成为A股首家专注于FPGA业务的上市公司。公司根植本土，面向世界，矢志改变行业格局，以成为全球可编程逻辑器件一流的供应商为愿景。

PH1A90

PH1A90 is a "Phoenix" series product launched by Anlu Technology. It includes 11.5K LUTs, high-speed serial I/O, PCIE hard core, DDR3\DDR4 storage interface and rich IP resources, targeting the cost-effective programmable logic market; PH1A It provides best-in-class transceiver and signal processing capabilities while maintaining low power consumption. PH1A can meet the size, weight, power consumption and cost-sensitive markets such as communication infrastructure, medical, industrial control, consumer electronics and other markets, providing developers with better solutions.

Xi'an Zhiduojing Microelectronics Co., Ltd., established in 2012, focuses on the technology research and development of programmable logic circuit devices (FPGA, CPLD), and provides system manufacturers with highly integrated, cost-effective programmable logic devices, programmable Logic device IP cores, related software design tools and system solutions.

Zhiduojing has a registered capital of 35.267 million yuan. It is a national high-tech enterprise and has passed ISO9001-2015 quality management system certification. At present, mass production of 55nm/28nm process FPGA has been achieved, and 14nm process products are under development. And targeted products with embedded Flash, DDR, SDRAM, ARM, serdes and other integrated solutions have been launched. Products have been widely used in 5G communications (base stations, servers, antennas), industrial control (servo systems, industrial controllers), video processing (LED control boards, security monitoring) and other fields.

SA5T-366-D0-8H900C

SA5T-366-D0-8H900C product adopts 28nm process, has 366k logic units, adopts advanced 6-LUTs logic architecture, supports DDR4 PHY, and integrates DDR2/DDR3 hard core controller and high-speed serial interface (serdes) into the chip internally, enabling system designers to reduce costs while meeting growing high-performance application requirements. It is widely used in industries such as wireless and wired communications, industrial control, image processing, artificial intelligence, data processing centers and cloud information.

Shenzhen China Microelectronics Technology Co., Ltd. is a chip design company founded in April 2009 and is affiliated to China Electronics Information Industry Group Co., Ltd. Headquartered in Shenzhen, with wholly-owned subsidiaries in Shanghai, Nanjing, Chengdu, and Xi'an, it is the national team for GPU R&D in the Xinchuang industry chain.

The company has rich technology and intellectual property accumulation in chip R&D and design. The independently developed instruction set MVP ISA was rated as a "completely independent intellectual property instruction set" by the Ministry of Industry and Information Technology. It has successfully conducted four SoC chip tape-outs based on MVP cores and obtained processing 22 core invention patents in the field of medical devices. The company is one of the few companies in China that integrates processor core IP and solutions. Related chips have been used in fields such as smart homes, edge computing and artificial intelligence.

"Nanfeng No. 1"

"Nanfeng No. 1" uses the "Nanfeng No. 1" high-performance graphics processor chip independently developed and designed in China. It is a display card with completely independent intellectual property rights. Through the function of hardware acceleration, users can better realize the rendering effect of the application layer and process 3D complex applications more efficiently. The graphics card supports OpenGL3.1, the power consumption is <20W, and the video memory capacity is 4G. It adopts PCIe2.0x4 interface, equipped with 2 HDMI2.0 display interfaces, and the screen resolution is 1920x1080@60Hz. At present, the graphics card is fully compatible with Great Wall Xinchuang desktop computers, Kirin operating system, Feiteng processor, QiAnxin trusted browser, etc., runs stably, and its performance can meet the computer graphics card needs of Xinchuang industry chain such as party and government, finance and security.

Yilingsi is a domestic FPGA company that uses the XLR structure with interchangeable logic and routing and revolutionaryly invented the breakthrough Quantum® architecture. Its PPA advantage is 4 times that of the traditional world's leading FPGA companies. Recently, Yilinsi launched a titanium series FPGA product, which once again increased PPA by 8 times and is very suitable for applications in edge computing, ADAS and AIoT. The core team members of Yilinsi come from the early experts and management teams of technology companies such as Xilinx, Intel and Microsemi, with an average industry experience of 25 years. From architecture and IC design, process, packaging and testing, cost/quality/delivery control, to EDA tool design, IP and application solution design, marketing and technical support, the company adheres to both quality and quantity, and strives to win the industry crown.

Titanium product series

Yilinsi Titanium series FPGA has a logic unit capacity of 35K to 1000K and a performance of up to 300-500MHz.

1.5Gbps LVDS, 2.5Gbps MIPI DPHY

3733Mbps LPDDR4/4x, 16Gbps/25.8Gbps Serdes

Ultra-small package as low as 3.5mm*3.4mm@60K LE, power consumption as low as 1/4 of competing products

Can be widely used in multiple cameras, high-definition video, AI acceleration, wireless communications and other fields

source:

[1]Counterpoint, Data Center CPU Market: AMD Surpasses Intel in Share Growth

[2]Mercury Research, AMD and Intel CPU Market Share Report

[3] Research organization Global Market Insights

[4] Research organization Frost & Sullivan

[5] Fudan Microelectronics, corporate prospectus

[6] Anlu Technology, corporate prospectus

[7] Fudan Microelectronics & Anlu Technology, 2023 Semi-annual Report

[8]Trendforce, Market Status Update

[9] Research organization Yole Development

[10] Research organization ICVtank

[11]Mordor Intelligence, Field Programmable Gate Array (FPGA) Market Size

[12] Feiteng, a white paper on full-stack solutions based on Feiteng platform from end to cloud

[13] iResearch, 2023 China Information and Innovation Industry Research Report

[14]Omdia, DRAM Market Dynamics

If there are more companies and chip products not mentioned in this field that are available for market selection, please send your company profile and contact information to email: news@gsi24.com . After receiving the company information, we will display more "representative companies and product selection references" in this field in subsequent updated series of reports.

Report author: Valencia Wu