Quartet: A 22nm 0.09mJ/lnference Digital Compute-in-Memory Versatile AI Accelerator with Heterogeneous Tensor Engines and Off-Chip-Less Dataflow.
Yikan QiuYufei MaMeng WuYifan JiaXinyu QuZecheng ZhouJincheng LouTianyu JiaLe YeRu HuangPublished in: CICC (2024)
Keyphrases
- metal oxide semiconductor
- artificial intelligence
- circuit design
- cmos technology
- mixed signal
- memory subsystem
- high speed
- nm technology
- case based reasoning
- random access memory
- dynamic random access memory
- digital signal processors
- phase locked loop
- compute intensive
- memory requirements
- silicon on insulator
- cmos image sensor
- parallel computing
- low cost
- multithreading
- power dissipation
- single chip
- ai systems
- design methodology
- higher order
- power consumption
- high order
- expert systems
- random access
- high density
- knowledge representation
- memory bandwidth
- low power
- memory management
- analog to digital converter
- main memory
- dimensionality reduction
- ibm power processor
- data flow
- physical design