SSM-CIM: An Efficient CIM Macro Featuring Single-Step Multi-bit MAC Computation for CNN Edge Inference

Zhang, Heng; He, Shuhai; Xin, Lü; Guo, Xin; Wang, Shaodi; Du, Yuan; Du, Lin

doi:10.1109/tcsi.2023.3301814

Cited by 4 publications

(2 citation statements)

References 40 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The evaluation of different MAQE strategies focuses on the normalized throughput, demoted as 𝑇 . The calculation of 𝑇 is illustrated in (20), where 𝐶 , 𝐶 , 𝐶 , and 𝐶 , represent the cycle count of DDMA, WDMA, RDMA, and macro computing, respectively. 𝐾 is the normalization factor.…”

Section: ) Evaluation Matricsmentioning

confidence: 99%

See 1 more Smart Citation

A Compilation Framework for SRAM Computing-in-Memory Systems with Optimized Weight Mapping and Error Correction

Bai,

Li,

Zhang

et al. 2024

Preprint

View full text Add to dashboard Cite

Deploying convolution-based algorithms into SRAM computing-in-memory (CIM) systems faces various challenges, such as operator incompatibility and intrinsic non-ideal error. This paper proposes a compilation framework to address this issue. Efficient weight mapping strategies are introduced to improve the utilization of SRAM-CIM macro. The intrinsic non-ideal errors of SRAM-CIM macro are also taken into consideration, and two efficient error correction schemes are proposed, which include calibration of computation voltage linear error (CCVLE) and the mitigation of analog-to-digital quantization error (MAQE). In addition, bit-width flexibility and signed-unsigned reconfigurability are also supported to facilitate the deployment of various convolution-based algorithms. ResNet18, finite impulse response (FIR) filtering, and Gaussian image filtering are deployed into a multi-macro SRAM-CIM system. These algorithms serve as deployment representatives of convolutional neural network (CNN), digital signal processing (DSP), and digital image processing (DIP), respectively. The results show that the introduced weight mapping strategies improve the macro utilization by 63.29% and 21.10% for two types of frequently used convolution layers compared to the traditional strategy. Moreover, the proposed error correction schemes achieve similar algorithm accuracy to the floating-point results, and the deployment result of ResNet18 achieves 66.3%~70.1% top-1 classification accuracy evaluated on the ImageNet dataset with different throughput tradeoffs.

show abstract

Section: ) Evaluation Matricsmentioning

confidence: 99%

“…This error can be classified into computation voltage linear error (CVLE) and analog-to-digital quantization error (AQE). The CVLE encompasses slope skew and zero-point drift in the voltage-MAC result curves, both caused by analog computing nonidealities [17,18,20], as shown in Fig. 1(b).…”

Section: Introductionmentioning

confidence: 99%