A new technical paper entitled “Cool-3D: an end-to-end thermo frame for exploring microfluidically cooled 3DICs” was published by researchers at the University of Michigan, the Shanghai Jiao Tong University and the University of Virginia.
Abstract
“The rapid further development of three-dimensional integrated circuits (3DICS) has increased the need for early phase design exploration (DSE) to minimize design editions and unexpected challenges. The emphasis on the design phase (Pre Register Transfer Level) (Pre-RTL) is crucial for the reduction of test and error costs. However, the 3DIC design introduces additional complexity due to thermal restrictions and an extended construction space as a result of vertical stacking and various cold strategies. Despite this need, existing DSE tools for 3DICS remain existing DSE tools for the front solutions, whereby available solutions are often lacking comprehensive design options and complete adaptation support. In order to close this gap, we present Cool-3D, an end-to-end frame for 3DIC design, which integrates simulators at architectural level, including GEM5, MCPAT and HOTSPOT 7.0, with advanced cooling models. COOL-3D enables wide and fine-grained exploration of constructions, built-in microfluidic cooling support for thermal analysis and an extension interface for non-parameterized adjustments, with which designer 3DIC architectures can model and optimize with greater flexibility and accuracy. In order to validate the Cool 3D framework, we carry out three case studies in which the ability to model various hardware design options and to record thermal behaviors. COOL-3D serves as a basic framework that not only facilitates comprehensive exploration of 3DIC Design Space, but also enables future innovations in 3DIC architecture, cooling strategies and optimization techniques. The entire frame together with the experimental data is currently released on Github. “
Here you will find the technical paper. March 2025.
Arxiv: 2503.07297.
https://doi.org/10.48550/arxiv.2503.07297
Authors: Runxi Wang, Zengeng Wang, Ting Lin, Jacob M. Raby, Mircea R. Stan, Xinfei Guo.
This work was partially carried out by the National Science Foundation China and the Semiconductor Research Corporation (SRC) Jump Center for research on intelligent storage and processing in memory (Crisp).