On November 12, the 23rd China International Semiconductor Expo (IC China 2026), jointly hosted by the China Center for Information Industry Development and the China Semiconductor Industry Association, opened at the National Convention Center in Beijing. The 8th IC Entrepreneurs Conference (hereinafter referred to as "the Conference") was held concurrently. Zhao Yi, General Manager of Zhuhai Silicon Core Technology Co., Ltd., delivered a keynote speech at the Conference, stating that computational power demand is surging in the post-Moore era, heterogeneous integration is becoming a trend, and traditional EDA design paradigms are no longer suitable for 2.5D/3D stacked ChipsZ. A new 3D IC EDA toolchain and design methodology need to be established.
Heterogeneous Integration Is the Direction of Industry Development
In the era of AI computing power, the demand for AI-driven computational growth severely mismatches the rate of chip performance improvement. The path of enhancing single-chip performance through transistor scaling in the Moore's Law era is approaching physical limits. Huawei's "Tau Scaling Law" has pushed 3D IC stacking to the forefront of the industry, prompting exploration of multi-chiplet lateral and vertical stacking solutions.
"Essentially, companies pursuing advanced computing will transition to heterogeneous integration in the post-Moore era; advanced packaging is the process means to achieve this," said Zhao Yi. He pointed out that current processes and equipment can already manufacture 3D stacked ChipsZ, but the matching EDA design tools and design concepts have not kept up. There is an urgent need for a shift towards 3D IC-oriented EDA and design paradigms.
This poses challenges on two levels for EDA tools. On one hand, the two core goals of advanced packaging are improving interconnect density and bonding density; achieving underlying, board-level, and system-level interconnects is a critical issue. On the other hand, each type of stacking process requires re-adaptation, placing new demands on the standardization of design tools.
The transformation of design paradigms faces two categories of objective needs: first, scenario requirements, where different application scenarios have different goals. High-performance computing focuses on bandwidth after stacking and memory architecture, while RF systems prioritize gain metrics; second, physical implementation needs, where chip realization must address physical challenges such as signal integrity and power integrity brought about by stacking. Driven by these two major needs, the entire design paradigm is shifting from the DTCO (Design-Technology Co-Optimization) methodology for single 2D SoCs to the STCO (System-Technology Co-Optimization) methodology for 3D systems.
Demand-Oriented Approach Leads the STCO Collaborative Design Paradigm for Stacked ChipsZ
Zhao Yi stated that the biggest feature of the STCO methodology is being driven by scenario requirements. Multi-scheme exploration must be conducted during the system-level planning phase to determine process nodes, chiplet selection, and stacking modes based on specific scenarios. For example, digital and storage ChipsZ for high-performance computing typically involve various forms such as digital-RF-analog chip stacking or silicon photonics stacking. Different stacking scenarios impose new requirements on the 3D IC EDA toolchain.
At the chiplet and IP level, mature IP modules within a single chip face more complex situations when placed into multi-chiplet stacking scenarios. In addition to data flows within chiplets, there are cross-chiplet data flows, many of which require customized processing. Even if the same chiplet interconnect standard is adopted, significant custom adaptation work is still required if IPs come from two different companies. Therefore, custom adaptation and I/O allocation analysis must be clearly defined at the architecture stage.
Entering the physical design phase, 3D stacking introduces multi-level collaborative design challenges. The complexity of power networks and signal interconnects in stacked structures is far higher than in traditional single ChipsZ. In some scenarios, package design may even constrain internal chip design.
He also emphasized that simulation needs to be "shifted left." Various early simulations should be performed during the architecture design phase to assess failure risks such as cracking and delamination by analyzing current and power density. This requires a series of supporting design and simulation tools, as well as coordination across multiple links and parties to implement effectively.
Common Technical Challenges Across Multiple Industry Segments
"Testing is the primary challenge for 3D ICs," Zhao Yi stated, citing years of industry research conclusions. Beyond performance improvements, reliability and yield in 3D stacking are pressing issues for the industry. During the testing phase, test models and vectors need to be redeveloped; test circuits and self-repair redundant circuits are distributed across different dies and must interact, making it difficult to configure redundancy individually. There is still room for improvement in linking test standards with test tools. Having participated in the formulation of IEEE 1838 standards for 3D IC Design-for-Testability (DFT), he acknowledged that the 3D IC DFT field still requires a long period of exploration.
Besides testing, many areas of 2.5D and 3D multi-chiplet verification remain global "no-man's lands." Clear and mature standards for 3D DRC (Design Rule Check) have not yet been formed, unlike the well-established rules for 2D ChipsZ, and mature 3D PDKs (Process Design Kits) are missing. This requires EDA vendors, foundries, and design firms to explore and mine these rules together for different scenarios. Only when tools and processes are deeply bound can stable solutions emerge.
Under this entirely new 3D design rule system, design thinking must also evolve simultaneously. Regarding the "optimal path" proposed by the Tau Scaling Law, he cautioned that the optimal signal interconnect path is not necessarily the physically shortest path. In 3D stacking, signal, power, and test paths must be analyzed coupled together.
AI Agents Will Not Replace EDA at the Current Stage
Addressing the industry debate on the relationship between AI and EDA, Zhao Yi believes that AI Agents and 3D IC EDA should not be viewed as opposing forces. At the current stage, native AI Agents will not directly replace EDA software. A more reasonable model is for AI Agents to cooperate with the new 3D IC toolchain, creating a business closed loop that integrates "Design for AI" (chip design for AI computing power) and "AI for Design" (using AI to empower chip design).
According to him, Silicon Core Technology has developed a dedicated AI Agent (ChipsZ Z) for 2.5D/3D heterogeneous integration based on its self-developed "3Sheng Integration" full-process 3D IC toolchain. It covers high-level functional simulation, architecture exploration, early pre-analysis, and placement verification, with plans to deliver trials to customers in December. It supports various complex scenarios, including logic plus HBM stacking, large-scale homogeneous stacking, and mixed digital-analog-RF stacking.
Finally, Zhao Yi stated that in the 3D IC era of advanced packaging, EDA is undergoing significant changes and needs to deeply match processes and various stacking design scenarios. Once this path is fully navigated, standardized and reusable chiplet units will form. This process comes with huge challenges but also brings rare opportunities to the industry.
Author | Liu Lu
Editor | Wu Lilin
Art Editor | Ma Liya Producer | Lian Xiaodong