earticle

영어

Binarized neural network (BNN) is one of the most efficient neural network for low-cost convolution operations. In BNN, binarized data is utilized to reduce memory size and complexity of convolution operations. A content addressable memory (CAM) based BNN accelerator can perform convolution operations efficiently by taking an advantage of fully parallel search operations in CAM. However, one of the critical issue of CAM based BNN hardware is that the operation reliability is severely degraded by the process variation during ML sensing operation. Therefore, we propose new CAM array design which can reduce hardware error probability. The proposed CAM based accelerator achieves 62% reduction in XNOR-popcount operations, and the classification accuracy drop of Fashion MNIST data set reduces from 2.33% to 1.26%.