Paper title:
Eyeriss: A Spatial Architecture for Energy-Efficient Dataflow for Convolutional Neural Networks
Publication:
ISCA’16
Problem to solve:
Although highly-parallel compute paradigms, such as SIMD/SIMT, effectively address the computation requirement to achieve high throughput, energy consumption still remains high as data movement can be more expensive than computation. Accordingly, finding a dataflow that supports parallel processing with minimal data movement cost is crucial to achieving energy-efficient CNN processing without compromising accuracy.
Major contribution:
-
A taxonomy that classifies existing CNN dataflows from previous research.
-
A spatial architecture based on a new CNN dataflow, called row stationary, which is optimized for throughput and energy efficiency. This dataflow has been demonstrated on a fabricated chip.
-
An analysis framework that can quantify the energy efficiency of different CNN dataflows under the same hardware constraints. It can also search for the most energy efficient mapping for each dataflow.
Lessons learnt:
The paper for the first time clarifies the CNN dataflow, i.e., row stationary dataflow that optimizes for all types of data movement energy costs to achieve energy efficiency.
Experiments using the CNN configurations of AlexNet show that the proposed RS dataflow is more energy efficient than existing dataflows in both convolutional (1.4× to 2.5×) and fully-connected layers (at least 1.3× for batch size larger than 16). The RS dataflow has also been demonstrated on a fabricated chip, which verifies the energy analysis.