MC-CIM: A Reconfigurable Computation-In-Memory For Efficient Stereo Matching Cost Computation
TimeWednesday, July 13th10:53am - 11:15am PDT
Location3002, Level 3
In-memory and Near-memory Computing
DescriptionThis paper proposes the design of a computation-in-memory for stereo matching cost computation. The matching algorithm incurs large energy and latency overhead because of frequent memory access. To overcome previous limitations, this work named MC-CIM introduces several key features. (1) A symmetric lightweight XOR computing circuit is realized in SRAM array to minimize area overhead and expand sensing margin. (2) the programmable digital design enable switching computing mode in different scenario. (3) This work introduces flexible data mapping method for different algorithm. The proposed design is implemented with 28nm technology and achieves an average performance of 277 TOPs/W.