A novel 3D 1T1R RRAM architecture for memory-centric Hyperdimensional Computing

3D Parallel ReRAM Computation-in-Memory for Hyperdimensional Computing

“…Furthermore, in 3D integration, the construction of SL and BL on BEOL metallic layers as shown in Fig. 5 provides significant benefit [9,15]. Implementation of the N-gram encoder by the ReRAM CiM is shown in Fig.…”

“…Typically, a HV consists of 1 bit × 10,000 dimensions. Almost all operations are bitwise operations, therefore HDC is suitable for parallel processing and Computation-in-Memory (CiM) [5][6][7][8][9][10][11]. CiM is one of the many parallel processing methods.…”

“…For large-scale algorithms such as Machine Learning, the cost of data transfer is a major barrier, known as the von Neumann bottleneck, and CiM is seen as a promising way to solve this problem. In addition, using ReRAM, one of the emerging Non-Volatile Memories (NVMs), grants some advantages such as energy efficiency [9][10][11][12][13].…”