CoDR: Computation and Data Reuse Aware CNN Accelerator
This work addresses resource limitations in CNN accelerators for embedded or low-power applications, presenting an incremental improvement over prior compressed accelerators.
The paper tackled the problem of improving efficiency in CNN accelerators by proposing CoDR, which uses universal computation reuse and a customized encoding scheme to reduce memory access and energy consumption, achieving reductions of up to 7.99x in SRAM access and 6.84x in energy compared to existing accelerators.
Computation and Data Reuse is critical for the resource-limited Convolutional Neural Network (CNN) accelerators. This paper presents Universal Computation Reuse to exploit weight sparsity, repetition, and similarity simultaneously in a convolutional layer. Moreover, CoDR decreases the cost of weight memory access by proposing a customized Run-Length Encoding scheme and the number of memory accesses to the intermediate results by introducing an input and output stationary dataflow. Compared to two recent compressed CNN accelerators with the same area of 2.85 mm^2, CoDR decreases SRAM access by 5.08x and 7.99x, and consumes 3.76x and 6.84x less energy.