Compute-in-Memory Chiplet 2.5D Packaging for Warpage and Stress Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for Compute-in-Memory chiplet-based 2.5D Packaging face challenges in efficiently handling complex nonlinear relationships and multi-objective constraints, leading to structural warping and stress issues that affect product performance and reliability.
Innovation Solution
An intelligent optimization method combining machine learning with the MOPSO algorithm to co-optimize material and structural parameters, using Taguchi orthogonal experiments, finite element modeling, and MOPSO optimization to determine optimal bump parameters and materials, reducing warpage and stress through a multi-objective comprehensive evaluation and optimization.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If traditional optimization methods are used for packaging structure, then the design process is simpler, but the optimization efficiency and accuracy deteriorate due to inability to handle complex nonlinear relationships and multi-objective constraints
Solution Approach 1:
The patent combines machine learning algorithms with MOPSO (Multi-Objective Particle Swarm Optimization) to create a hybrid optimization system. The machine learning component predicts warpage and stress outcomes, while MOPSO searches for optimal parameter combinations, together resolving complex nonlinear relationships and multi-objective constraints that neither method could handle alone efficiently
Solution Approach 2:
The patent introduces an equivalent finite element model as an intermediary between the optimization algorithm and the actual packaging structure. This model serves as a computational surrogate that captures the complex physical behaviors (thermal expansion, stress distribution) without requiring full-scale physical experimentation, enabling efficient multi-objective optimization
2Ease of manufacture
If unreasonable material combination or structural design is used, then the manufacturing process is easier, but structural warping occurs affecting coplanarity and causing chip fracture or delamination
Solution Approach 1:
The patent systematically varies key parameters including C2 bump diameter (0.15-0.25mm), C2 bump height (0.05-0.15mm), C4 bump diameter (0.30-0.50mm), C4 bump height (0.20-0.40mm), and material properties to identify optimal combinations that minimize warpage and stress while maintaining manufacturing feasibility. The machine learning model captures the nonlinear relationships between these parameters and reliability outcomes
3Manufacturing precision
If simple optimization methods are used, then the computational resources required are fewer, but the accuracy in predicting and reducing warpage and stress deteriorates
Solution Approach 1:
The patent employs machine learning to pre-process and predict warpage and stress outcomes based on input parameters before the full MOPSO optimization runs. This preliminary prediction step filters the search space and guides the optimization algorithm toward promising regions, achieving high precision in warpage and stress prediction while reducing the computational energy required for exhaustive searches
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
The method significantly improves the efficiency and accuracy of optimizing structural and material parameters, reducing warpage and stress, thereby enhancing packaging reliability and product life.
Implementation Method 1
Electronic devices are subjected to multiple repeated thermal loadings during fabrication and service, and Coefficient of thermal expansion (CTE) of joint structures introduced the generation of internal stresses
Implementation Method 2
Although these stresses are much smaller than the fracture strength of the joint, stress can cause damage at critical locations with the repeated external thermal cycling loads
Data Source
AI summary
This invention presents an intelligent reliability optimization method for Compute-in-Memory chiplet 2.5D packaging using MOPSO. The approach integrates Taguchi orthogonal experiments to establish a finite element model of critical bump parameters, coupled with signal-to-noise ratio range analysis and weighted percentage evaluation for multi-objective prioritization. Stress and warpage relationships are mathematically modeled via least squares fitting and optimized through a multi-objective particle swarm algorithm. The optimized packaging achieves a 21.2% reduction in thermal stress and 61.7% lower warpage deformation compared to conventional designs. With balanced thermo-mechanical reliability and performance metrics, this method surpasses existing packaging solutions in proactive risk mitigation, supporting heterogeneous integration for next-generation 3D-IC architectures. The framework enables predictive optimization of structural parameters prior to fabrication, addressing reliability challenges in advanced chiplet-based systems.