Exploring the Performance Improvement of Tensor Processing Engines through Transformation in the Bit-weight Dimension of MACs