Figure 6
From: A processing-in-pixel-in-memory paradigm for resource-constrained TinyML applications
Algorithm-circuit co-design framework to enable our proposed P2M approach optimize both the performance and energy-efficiency of vision workloads. We propose the use of ①large strides, ② large kernel sizes, ③ reduced number of channels, ④ P2M custom convolution, and ⑤ shifted ReLU operation to incorporate the shift term of the batch normalization layer, for emulating accurate P2M circuit behaviour.
