DOSA: Differentiable Model-Based One-Loop Search for DNN Accelerators

Fuente: arXiv

Saved in:

Bibliographic Details
Main Authors:	Hong, Charles, Huang, Qijing, Dinh, Grace, Subedar, Mahesh, Shao, Yakun Sophia
Format:	Preprint
Published:	2025
Subjects:	Hardware Architecture Machine Learning
Online Access:	Acceder al recurso
Tags:	Add Tag No Tags, Be the first to tag this record!

_version_	1866915493451923456
author	Hong, Charles Huang, Qijing Dinh, Grace Subedar, Mahesh Shao, Yakun Sophia
author_facet	Hong, Charles Huang, Qijing Dinh, Grace Subedar, Mahesh Shao, Yakun Sophia
contents	In the hardware design space exploration process, it is critical to optimize both hardware parameters and algorithm-to-hardware mappings. Previous work has largely approached this simultaneous optimization problem by separately exploring the hardware design space and the mapspace - both individually large and highly nonconvex spaces - independently. The resulting combinatorial explosion has created significant difficulties for optimizers. In this paper, we introduce DOSA, which consists of differentiable performance models and a gradient descent-based optimization technique to simultaneously explore both spaces and identify high-performing design points. Experimental results demonstrate that DOSA outperforms random search and Bayesian optimization by 2.80x and 12.59x, respectively, in improving DNN model energy-delay product, given a similar number of samples. We also demonstrate the modularity and flexibility of DOSA by augmenting our analytical model with a learned model, allowing us to optimize buffer sizes and mappings of a real DNN accelerator and attain a 1.82x improvement in energy-delay product.
format	Preprint
id	arxiv_https___arxiv_org_abs_2509_10702
institution	arXiv
publishDate	2025
record_format	arxiv
spellingShingle	DOSA: Differentiable Model-Based One-Loop Search for DNN Accelerators Hong, Charles Huang, Qijing Dinh, Grace Subedar, Mahesh Shao, Yakun Sophia Hardware Architecture Machine Learning In the hardware design space exploration process, it is critical to optimize both hardware parameters and algorithm-to-hardware mappings. Previous work has largely approached this simultaneous optimization problem by separately exploring the hardware design space and the mapspace - both individually large and highly nonconvex spaces - independently. The resulting combinatorial explosion has created significant difficulties for optimizers. In this paper, we introduce DOSA, which consists of differentiable performance models and a gradient descent-based optimization technique to simultaneously explore both spaces and identify high-performing design points. Experimental results demonstrate that DOSA outperforms random search and Bayesian optimization by 2.80x and 12.59x, respectively, in improving DNN model energy-delay product, given a similar number of samples. We also demonstrate the modularity and flexibility of DOSA by augmenting our analytical model with a learned model, allowing us to optimize buffer sizes and mappings of a real DNN accelerator and attain a 1.82x improvement in energy-delay product.
title	DOSA: Differentiable Model-Based One-Loop Search for DNN Accelerators
topic	Hardware Architecture Machine Learning
url	https://arxiv.org/abs/2509.10702