notesum.ai
Published at November 14PIMCOMP: An End-to-End DNN Compiler for Processing-In-Memory Accelerators
cs.AR
Released Date: November 14, 2024
Authors: Xiaotian Sun1, Xinyu Wang1, Wanqian Li1, Yinhe Han1, Xiaoming Chen1
Aff.: 1Institution unspecified

| Level | Parameter | Level | Parameter |
| Accelerator | #Chip in X | Core | VFU functionality |
| #Chip in Y | #VFU | ||
| Off-chip bandwidth | LocalMem bandwidth | ||
| Chip | #Core in X | LocalMem size | |
| #Core in Y | Instruction execution model | ||
| Connection type | PIMFU | #Crossbar in X | |
| Connection bandwidth | #Crossbar in Y | ||
| Comm. mechanism | Management granularity | ||
| GlobalMem bandwidth | Crossbar Array | Crossbar array size | |
| GlobalMem size | Cell precision |