Skip to content

WaytoICT

通往ICT之路 · 光通信与AI网络技术解读

中文English
登录注册
Menu
Technology Optical comms · silicon photonics · CPO → Market Insights Financials · trends · industry → In-Depth Analysis Optical chips, devices & modules → Video Understand a topic in minutes → Documents Latest docs · summit materials →
Menu

The Disruptive Word on NVIDIA’s Roadmap: CoWoP

Posted on 2025-08-26 by

Recently, CoWoP, which has appeared in NVIDIA’s technology roadmap, has become a hot topic in the semiconductor industry. So how exactly does it compare with the previously much-discussed CoWoS and CoPoS technologies?

As we know, traditional 2.5D/3D packaging technologies usually use a “silicon interposer,” and achieve high-density interconnection between chips through the “through-silicon vias” (TSV) and “redistribution layer” (RDL) on it.

The Disruptive Word on NVIDIA's Roadmap: CoWoP

To continuously improve computing power and integration, three more advanced technology paths have been developed: CoWoS, CoPoS, and CoWoP. Today let’s talk about these three types of technology.

█ CoWoS

CoWoS (Chip-on-Wafer-on-Substrate): the current mainstream technology, like a “sandwich,” from top to bottom: silicon interposer → ABF substrate → PCB.

The Disruptive Word on NVIDIA's Roadmap: CoWoP

It has evolved into several types according to application needs:

  • CoWoS-S: uses a silicon interposer to integrate HBM and GPU, such as NVIDIA H100;
  • CoWoS-R: uses a redistribution layer, aimed at cost-sensitive designs;
  • CoWoS-L: combines silicon and redistribution layer, supporting a hybrid scheme of up to 12 HBM stacks.
The Disruptive Word on NVIDIA's Roadmap: CoWoP

To support this structure, the underlying PCB mainboard must use ultra-low-loss materials (such as Megtron 8) and complex stack-up designs with more than 28 layers, to ensure the integrity of high-speed signal transmission.

█ CoPoS

CoPoS replaces the silicon interposer with a panel-level redistribution layer (usually over 600×600 mm in size), forming a new “RDL → ABF substrate → PCB” structure.

The Disruptive Word on NVIDIA's Roadmap: CoWoP

The benefits of this approach are very obvious:

  • Panel utilization rises from 85% to over 95%, more economical;
  • Cost is reduced by 20-30%;
  • It can support the 10-12 HBM4 memory stacks required by future AI servers, providing up to 19 TB/s of bandwidth.

█ CoWoP

CoWoP (Chip-on-Wafer-on-PCB) directly bonds the silicon interposer to the PCB, eliminating the ABF substrate.

The Disruptive Word on NVIDIA's Roadmap: CoWoP

This scheme is said to reduce cost by 40-50%, with a shorter signal path, support for high-speed interfaces such as PCIe 6.0, and improved thermal management efficiency through direct contact with the heat spreader. But its implementation relies on the modified semi-additive process (mSAP), requiring the PCB to achieve line width/spacing ≤25 μm, void-free via filling, and the use of ultra-low coefficient of thermal expansion materials.

█ Comparison

From a technical comparison, the three schemes each have their own advantages.

The Disruptive Word on NVIDIA's Roadmap: CoWoP

Although CoWoP stands out in terms of cost, it still faces multiple challenges: PCB line width precision (25 μm) lags behind that of the ABF substrate (≤10 μm), the coefficient of thermal expansion mismatch between the chip and PCB materials, and an ecosystem that is not yet mature (currently only Apple’s SLP PCB is close to its requirements).

This also brings opportunities for PCB material innovation: new materials such as low-thermal-expansion glass cloth (such as Nittobo NE-glass), peelable copper foil (such as Fangbang Electronics), and cationic resins, as well as new processes such as mSAP fine-line processing and via filling electroplating (POFV), will become key breakthrough points.

CoPoS technology requires the PCB to be compatible with large panels from 310×310 mm to 700×700 mm, and to support redistribution layer integration through glass via (TGV) substrates.

According to market planning, this technology will establish pilot production lines in 2026-2027 (led by TSMC) and achieve mass production in 2028. The currently mainstream CoWoS technology still requires the PCB industry to focus on solving the impedance control of 56+ Gbps signals (deviation controlled within ±3%), as well as stack-up designs of hybrid FR-4/high-speed materials with more than 40 layers.

For PCB manufacturers, it may be necessary to advance strategic layout in stages:

  • Short-term (2025-2026): focus on meeting the 18-46 layer HDI PCB required by CoWoS-L;
  • Mid-term (2027-2028): cooperate with packaging and testing factories to develop CoPoS panel-level substrates;
  • Long-term (after 2029): deploy CoWoP production lines according to technology maturity, serving edge computing and AI accelerators.

It is worth noting that although CoPoS and CoWoP have significant cost and performance potential, CoWoS is still the mainstream technology supporting current AI chips (such as NVIDIA’s Rubin platform).

Thanks for reading!

Recommended reading:

DIP, SOP, QFP, and LCC Packaging of Chips

ZTE: A Liquid-Cooled Optical Transceiver Structure Supporting CPO

A CPO Packaging Structure Based on Glass Carrier

An Innovative CPO Fiber Vertical Coupling Scheme

SMIC’s Innovative Silicon-Germanium Planarization Process

Enjoy this article?

Subscribe to get new optical communication insights delivered to your inbox. Free, unsubscribe anytime.

Category: Technology

Leave a Reply

You must be logged in to post a comment.

订阅更新

订阅后,新文章发布时第一时间收到邮件通知

李东霏
光互连技术研究者

香农信息技术研究院,专注光模块、CPO、硅光与 AI 光互连的技术解读。

行业报告

光芯片·光器件·光模块 2026 报告汇编(含 OFC 2026)

光芯片 · 光器件 · 光模块
2026 报告汇编(含 OFC 2026)

¥1,900

上/下两册,收录 OFC 2026 一线报告
添加微信 WaytoICT 订购

广告

📢 广告位招商:面向全球光通信客户 侧边栏广告位 Email:Dongfei.li@shannoninstitute.cn

最新文档

  • 互连拐点:11 页解析机架内到多机架 Scale-Up 的 XPO/CPX/OCI 蓝图(OCP 2026 全球峰会)
  • AI 时代的相干光技术:17 页解析 Scale-Across、液冷 XPO 与全谱转发器(OCP 2026 全球峰会)
  • XPO 高密度光互连的系统级评估:20 页解析 212G SerDes 平台与 FRO/LRO/LPO 实测(OCP 2026 全球峰会)
  • CPO 与 OCI 标准及未来技术:17 页 Broadcom 精简版演讲提纲(OCP 2026 全球峰会)
  • VCSEL 高密度互连:7 页解析 50G NRZ「慢而宽」阵列与光 PCIe 演示(OCP 2026 全球峰会)

最新评论

暂无评论,欢迎登录后发表第一条评论。

最近访客

👥 今日访客 -
加载中…
© 2026 WaytoICT | 蜀ICP备2020035321号-2 | 隐私政策 | 免责声明
Wechat Pay
Alipay
请使用 支付宝 或 微信 扫码支付