# Optimizing Predictive AI in Physical Design Flows with Mini Pixel Batch Gradient Descent

arXiv: https://arxiv.org/abs/2402.06034 · HTML: https://mlsanity.com/paper/2402.06034/

## Key facts

- Cited by 1 papers in the map (data through 14 Oct 2026).
- Builds on 10 of its 33 references that are in the map.
- First posted on arXiv 8 Feb 2024.

## Builds on (most cited first, 10 of 10)

- [Deep Residual Learning for Image Recognition](https://mlsanity.com/paper/1512.03385/) (2015)
- [U-Net: Convolutional Networks for Biomedical Image Segmentation](https://mlsanity.com/paper/1505.04597/) (2015)
- [Semi-Supervised Classification with Graph Convolutional Networks](https://mlsanity.com/paper/1609.02907/) (2016)
- [Inductive Representation Learning on Large Graphs](https://mlsanity.com/paper/1706.02216/) (2017)
- [DAMO: Deep Agile Mask Optimization for Full Chip Scale](https://mlsanity.com/paper/2008.00806/) (2020)
- [Differentiable Top-k Classification Learning](https://mlsanity.com/paper/2206.07290/) (2022)
- [Generic Lithography Modeling with Dual-band Optics-Inspired Neural Networks](https://mlsanity.com/paper/2203.08616/) (2022)
- [Painting on Placement: Forecasting Routing Congestion using Conditional Generative Adversarial Nets](https://mlsanity.com/paper/1904.07077/) (2019)
- [Physics-Informed Optical Kernel Regression Using Complex-valued Neural Fields](https://mlsanity.com/paper/2303.08435/) (2023)
- [A Machine Learning Approach to Improving Timing Consistency between Global Route and Detailed Route](https://mlsanity.com/paper/2305.06917/) (2023)

## Built on by (most cited first, 1 of 1)

- [Multimodal Chip Physical Design Engineer Assistant](https://mlsanity.com/paper/2510.15872/) (2025)

## Abstract (arXiv)

Exploding predictive AI has enabled fast yet effective evaluation and decision-making in modern chip physical design flows. State-of-the-art frameworks typically include the objective of minimizing the mean square error (MSE) between the prediction and the ground truth. We argue the averaging effect of MSE induces limitations in both model training and deployment, and good MSE behavior does not guarantee the capability of these models to assist physical design flows which are likely sabotaged due to a small portion of prediction error. To address this, we propose mini-pixel batch gradient descent (MPGD), a plug-and-play optimization algorithm that takes the most informative entries into consideration, offering probably faster and better convergence. Experiments on representative benchmark suits show the significant benefits of MPGD on various physical design prediction tasks using CNN or Graph-based models.