.png){kind=icon}
返回
ISSA: Architecting CNN Accelerators Using Input-Skippable, Set-Associative Computing-in-Memory
.png){kind=icon}
PRE
AI
DOI：10.1109/TC.2024.3404060.png)
.png){kind=icon}
原文求助.png){kind=icon}
分享.png){kind=icon}
收藏
摘要
En 中文
Among several emerging architectures, computing in memory (CIM), which features in-situ analog computation, is a potential solution to the data movement bottleneck of the Von Neumann architecture for artificial intelligence (AI). Interestingly, more strengths of CIM significantly different from in-situ analog computation are not widely known yet. In this work, we point out that mutually stationary vectors (MSVs), which can be maximized by introducing associativity to CIM, are another inherent power unique to CIM. By MSVs, CIM exhibits significant freedom to dynamically vectorize the stored data (e.g., weights) to perform agile computation using the dynamically formed vectors. We have designed and realized an SA-CIM silicon prototype and corresponding architecture and acceleration schemes in the TSMC 28 nm process. More specifically, the contributions of this paper are fivefold: 1) We identify MSVs as new features that can be exploited to improve the current performance and energy challenges of the CIM-based hardware. 2) We propose SA-CIM to enhance MSVs (input-reordering flexibility) for skipping the zeros, small values, and sparse vectors. 3) We propose channel swapping to enhance the zero-skipping technique. 4) We propose a transposed systolic dataflow to efficiently conduct conv3 x 3 while being capable of exploiting input-skipping schemes. 5) We propose a design flow to search for optimal aggressive skipping scheme setups while satisfying the accuracy loss constraint. The proposed ISSA architecture improves the throughput by 1.91x to 2.97x speedup and the energy efficiency by 2.5x to 4.2x .
Keyword:
Vectors
Decoding
Common Information Model (computing)
Schedules
Registers
Computers
In-memory computing
Set associativity
convolutional neural networks
computing in memory
期刊
IF:
3.8
论文数:
5.4K
被引数:
9.8K
机构
引用论文
.png){kind=icon}
分享收藏分享收藏Challenges and Trends of SRAM-Based Computing-In-Memory for AI Edge DevicesAI边缘设备基于SRAM的内存计算的挑战和趋势
PREAI
分享收藏分享收藏Effects of canopy-mediated abrasion and water flow on the early colonisation of turf-forming algae
OAAI
OAAI分享收藏
查看更多内容