SEADA: An efficient methodology for optimizing mixed-precision DNNs on multi-precision spatial architectures
This work provides DNN accelerator designers with a robust framework to optimize mixed-precision networks for improved latency, energy, and memory, addressing a key bottleneck in deploying efficient DNNs on specialized hardware.
SEADA introduces an efficient methodology for optimizing mixed-precision DNNs on multi-precision spatial architectures, addressing challenges in precision assignment, mapping, and cost estimation. The methodology enables effective design-space exploration, achieving near-optimal mappings and efficient precision selection.
Mixed-precision computation has been introduced in deep neural networks (DNNs) as an effective approach to reduce latency, energy consumption, and memory footprint. However, efficiently mapping mixed-precision networks onto multi-precision spatial architectures poses several challenges. These include determining the appropriate precision for each layer, balancing layer-wise accuracy sensitivity to quantization against architectural heterogeneity and system-level constraints, and accurately estimating the system-level cost of heterogeneous precision assignments. This work presents SEADA, an efficient methodology designed to address these challenges. SEADA comprises: (i) a configurable system-level analytical cost model of a multi-precision spatial accelerator architecture; (ii) a fast mapping tool that identifies near-optimal mappings of DNN workloads onto the target integer accelerator; (iii) analytical models for floating-point layers to estimate the overall benefits of mixed-precision execution; and (iv) a per-layer precision selection methodology based on bit-level entropy, enabling efficient assignment across multiple numerical precisions. SEADA's efficiency provides designers with a robust framework for the design-space exploration of multi-precision architectures.