Impact of On-Chip Interconnect on In-Memory Acceleration of Deep Neural Networks