DOSA: Differentiable Model-Based One-Loop Search for DNN Accelerators
This addresses the hardware design space exploration problem for DNN accelerator developers, offering a more efficient optimization approach, though it is incremental as it builds on existing differentiable models and gradient-based methods.
The paper tackles the problem of simultaneously optimizing hardware parameters and algorithm-to-hardware mappings for DNN accelerators, which is challenging due to combinatorial explosion from separate exploration. It introduces DOSA, a differentiable model-based one-loop search method that outperforms random search and Bayesian optimization by 2.80x and 12.59x in improving energy-delay product, and achieves a 1.82x improvement when applied to a real DNN accelerator.
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.