arXiv cs.AI· Varun Darshana Parekh, Anusha Devulapally, Sivani Devarapalli, Cassius Henderson, Shimeng Yu, Vijaykrishnan Narayanan·· 4 小时前AI 评分24
ThermLeT:面向 2.5D Chiplet 架构的 Transformer 温度预测模型
ThermLeT: Transformer-based Temperature Prediction for 2.5D Chiplet Architecture
AI 导读
ThermLeT 是一个基于多模态 Transformer 的尺度不变代理模型,可在 13 ms 内预测 2.5D chiplet 稳态温度,比 ANSYS 等有限元求解器快约四个数量级。
正文 · 原文
Abstract:The drive toward heterogeneous 2.5D chiplet packages promises unmatched performance and customization, yet the resulting density amplifies thermal challenges. To facilitate rapid design space exploration, we present ThermLeT, a multimodal transformer-based, scale invariant surrogate that integrates power density maps, spatial coordinate embeddings, material properties and heat transfer rate to predict steady state temperature in 13 ms, about four orders of magnitude faster than conventional finite element solvers such as ANSYS. To train the transformer model, we also create a dataset of 6,221 chiplet layouts perturbed from three representative architectures. On a 150 sample test set, ThermLeT achieves an RMSE of 3.16°C and an MAE of 2.24°C, improving accuracy by 35% over ThermEDGe and by 71% over HotSpot, while running approximately 6x and 11x lower inference latencies respectively. Ablation studies confirm robust generalization to unseen floorplans. Code and dataset available at this https URL
| Comments: | Published at VLSID 2026. Code and dataset: this https URL |
| Subjects: | Hardware Architecture (cs.AR); Artificial Intelligence (cs.AI); Machine Learning (cs.LG) |
| Cite as: | arXiv:2611.21954 [cs.AR] |
| (or arXiv:2611.21954v1 [cs.AR] for this version) | |
| https://doi.org/10.48550/arXiv.2611.21954 arXiv-issued DOI via DataCite (pending registration) |
|
| Journal reference: | VLSID 2026, pp. 512-517 |
| Related DOI: | https://doi.org/10.1109/VLSID68508.2026.00101
DOI(s) linking to related resources |
Submission history
From: Varun Parekh [view email]
[v1]
Mon, 16 Nov 2026 22:36:06 UTC (6,116 KB)
来源:arXiv cs.AI · arxiv.org