ECIM: Exponent Computing in Memory for an Energy-Efficient Heterogeneous Floating-Point DNN Training Processor.
Juhyoung LeeJihoon KimWooyoung JoSangyeob KimSangjin KimHoi-Jun YooPublished in: IEEE Micro (2022)
Keyphrases
- floating point
- instruction set
- memory bandwidth
- training process
- floating point arithmetic
- memory subsystem
- fixed point
- level parallelism
- square root
- wireless sensor networks
- floating point unit
- memory requirements
- memory access
- memory management
- sensor networks
- multithreading
- graphics processing units
- energy efficient
- fast fourier transform
- parallel implementation
- interval arithmetic
- data access
- main memory
- fine grained
- computer systems
- intel xeon