当前位置:首页 > 报告详情

通过先进的包装创意推动人工智能发展.pdf

上传人: 明**** 编号:1011545 2025-12-21 22页 2.35MB

1、Propelling AI forward through Advanced Packaging CreativityYin ChangASE,Inc.OCP APAC/August 5,2025Taipei,TaiwanAI IS HEREAI applications permeating global life,creating new efficiencies and new markets,from data center to edge devices.AI economy projected to soar from$189 billion in 2023 to$4.8 tril

2、lion by 2033 a 25-fold increase in just a decade AI adoption could boost global GDP by 15%by 2035,based on game-changing capabilities and safe deployment.Unprecedented data generation is driving global datasphere trajectory towards 200ZB by 2030HEALTHCARETELECOMMUNICATIONSRETAILFINANCIAL SERVICES2So

3、urce:UN Trade and Development(UNCTAD),PwC,IDC 2025Multimodal AISystems to mimick human ability to process diverse sensoryTrending AI models generating dataAgentic AISystems that exhibit autonomy,proactivity,learn and the ability to act independently.3ASE Confidential ProprietaryGrowing Demands and C

4、hallenges in AI EraPerformanceMemoryPowerThermalArea2x 7xincreased computeworkload2x-10 x network bandwidth&memory capacity3 kW thermal coolingsExascale 21 mW to zettascale 500mW power delivery2x10 x reticle 2x-6x Pkg size(20120)5High Performance Demands by AI IndustryAcross all domains of AI,datase

5、t size for train models doubles every 6 months with tens of trillions wordsHigh-performance compute to train models grew 4-5x yearly 2024 with 1.6x/year increase chip quantity&1.6x/year improvement in performance per chip3.7x/yearSize of datasets 6.7x/year4.2x/yearSuper compute performanceSource Epo

6、ch AI6Memory Bandwidth Trend for AIEpoch AIMemory&interconnect bandwidth(1.4x/2year)slower than compute capability(3.1x/2year)Memory access delays(Memory Wall)cannot meet demands of compute,creating performance&SI challengesMemory bandwidth bottleneck drives more HBM&DRAM,increasing power and module

word格式文档无特别注明外均可编辑修改,预览文件经过压缩,下载原文更清晰!
三个皮匠报告文库所有资源均是客户上传分享,仅供网友学习交流,未经上传用户书面授权,请勿作商用。
根据文章内容,以下是全文主要内容的简明概括: 1. **AI发展趋势**:AI应用广泛渗透,预计到2033年AI经济将增长25倍,达到4.8万亿美元。 2. **AI对经济的影响**:AI的采用预计到2035年能提升全球GDP 15%。 3. **AI挑战**:AI时代对性能、内存、功率和热管理提出了更高的要求。 4. **数据增长**:AI模型训练数据集每6个月翻一番,高性能计算需求每年增长4-5倍。 5. **内存带宽瓶颈**:内存带宽增长速度慢于计算能力,导致性能和功耗挑战。 6. **先进封装技术**:先进封装技术如Chiplets、异构集成、SiPh等是应对AI挑战的关键。 7. **市场需求**:AI性能需求增速超过摩尔定律,推动半导体市场对新型设备、材料和架构的需求。 8. **创新生态系统**:通过异构集成和封装创新,加速AI经济的发展。
"AI时代,先进封装如何突破性能瓶颈?" AI驱动下,半导体市场如何创新?" "芯片集成新纪元,先进封装如何引领未来?"
客服
商务合作
小程序
服务号
折叠