Numerical data and reproduction code for Residual-Bounded Transient Thermal Design Framework for Semiconductor Packages: A VCSEL Package Design-Space Study

Description

Recalibrated model-generated numerical data and reproduction scripts supporting "Residual-Bounded Transient Thermal Design Framework for Semiconductor Packages: A VCSEL Package Design-Space Study". Version 2 replaces the pre-recalibration VCSEL snapshot in Version 1; use Version 2 for the current manuscript. The effective bottom coefficient is calibrated at the nominal design on the same 27-node graph used for the package design-space search, then frozen across designs and selected spatial checks: h_eff = 66574.05281009956 W m^-2 K^-1. The archive records three searches, each with 27,291 center evaluations. Nominal/incumbent peak rises are 4.364751036688206/3.91450898023714 K; the remaining-opportunity computed gap is 0.07828988977246876 K. The prescribed heat pulse is 36.6 W for 1 ms every 10 ms. Version 1's coefficient, search count and gap must not be used for this manuscript. Contents include 32 CSV/JSON data files, 15 numerical/checking/figure scripts, 18 figure files, data schemas, SHA256 hashes, reproduction instructions and saved verification records. Unchanged two-node/rod benchmarks are retained; all VCSEL-dependent outputs reflect the same recalibration. The archived files match the current manuscript's numerical snapshot byte-for-byte. These are computational outputs, not measured transient device data. Exact-arithmetic bounds apply to the declared finite-graph design family; float64 evaluations do not enclose roundoff, continuum discretization or physical-model uncertainty. The 0.1374 K incumbent model-family sensitivity exceeds the 0.0783 K remaining-opportunity gap. A DC calibration anchor is not independent transient validation.

Steps to reproduce

Extract TCPMT_recalibrated_data_and_code_v2.zip and work from a copy: scripts overwrite generated results. README.md gives the ordered commands, DATA_FILES.csv identifies data schemas, and MANIFEST_SHA256.csv verifies the original snapshot. Use Python 3.9.6 with NumPy 2.0.2, SciPy 1.13.1 and Matplotlib 3.9.4 (requirements.txt). Set OPENBLAS_NUM_THREADS=1, OMP_NUM_THREADS=1, VECLIB_MAXIMUM_THREADS=1 and MPLBACKEND=Agg; use a writable Matplotlib cache. Run code/run_experiments.py, code/measure_costs.py, code/run_vcsel.py --repeats 3 and code/run_frozen_checks.py. Then run code/review_checks.py, code/review_vcsel.py, code/review_frozen.py and code/check_calibration_consistency.py. Generate figures with code/make_figures.py, code/make_revision_figures.py, code/make_frozen_figure.py and finally code/make_condensed_figure.py. Check h_eff = 66574.05281009956, 27,291 evaluations in each of three searches, and gap = 0.07828988977246876 K. Use tolerances for cross-platform numerical comparison; execution timings depend on the host/BLAS. Version 2 packaging preserves the previously recalculated files without a new scientific run. Method: periodic passive thermal-network response, contracting residual bounds, signed interface treatment and branch-and-bound screening.

Institutions

• Inner Mongolia University

  Inner Mongolia, Hohhot

Categories

Licence