Semiconductor Chip Protection Area Designs with 3D Thermal-Mechanical Simulations

Harvard Dataverse hosts 65,000 engineering designs for semiconductor chip protection areas, generated by Zi Hen Lin. The dataset, updated in May 2026, contains 2D designs governed by manufacturing constraints and evaluated with 3D finite element method simulations for thermal-mechanical stress. It provides cross-dimensional benchmarks for machine learning tasks related to semiconductor manufacturing processes.

Use Cases

• Solve the design-to-stress forward problem based on simulated sequential thermal-mechanical stress fields.
• Solve the stress-to-design inverse problem for design optimization using the provided stress benchmarks.
• Train machine learning models to predict thermal-mechanical defects from 2D design parameters.
• Benchmark finite element method simulation accuracy against a large corpus of design-stress pairs.
• Analyze the impact of manufacturing constraints on stress fields across 12 metal stack configurations.

Strengths

• Contains 65,000 distinct engineering designs, providing a substantial corpus for analysis.
• Includes simulations across 12 metal stacks of a dummy technology node, offering configurational variety.
• Provides sequential thermal-mechanical stress fields corresponding to physical manufacturing states.
• Enables two defined cross-dimensional machine learning benchmarks: forward and inverse problems.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• The full dataset (1.2 TB, 110k designs) is hosted on a separate external platform, requiring additional access.
• Row count for the available 608 GB subset is unknown, which may limit suitability assessment.

Provenance

Source

Harvard Dataverse, author Zi Hen Lin.

Collection Method

2D designs generated under manufacturing constraints, evaluated with 3D finite element method quasi-static structural analysis.

Time Range

null

Freshness

Last updated 2026-05-04 23:17:45; freshness should be verified.

Geography

null