Deep In-Memory Architectures for Machine Learning

by Mingu Kang

1 Introduction1.1 The Energy Problem in Machine Learning1.2 Digital ML Architectures1.3 In-memory ML Architectures1.4 Book Organization2 The Deep In-memory Architecture (DIMA)2.1 Data-flow of Machine Learning Algorithms2.2 DIMA Overview2.3 Inference Architectures: A Shannon-inspired Perspective2.4 DIMA Design Guidelines and Techniques2.5 DIMA Models of Energy, Delay, and Accuracy2.6 ConclusionAppendices3 DIMA Prototype Integrated Circuits3.1 The Multi-Functional DIMA IC3.2 Measured Results3.3 Random Forest (RF) DIMA IC3.4 Random Forest IC Prototype3.5 Measured Results3.6 Conclusion4 A Variation-Tolerant DIMA via On-Chip Training4.1 Background and Rationale4.2 Architecture and Circuit Implementation4.3 Experimental Results4.4 Conclusion5 Mapping Inference Algorithms to DIMA5.1 Convolutional Neural Network (CNN)5.2 Mapping CNN on DIMA (DIMA-CNN)5.3 Energy, Delay, and Functional Models of DIMA-CNN5.4 Simulation and Results5.5 Sparse Distributed Memory (SDM)5.6 DIMA-based SDM Architecture (DIMA-SDM)5.7 Energy, Delay, and Functional Models of DIMA-SDM5.8 Simulation Results5.9 Conclusions6 PROMISE: A DIMA-based Accelerator6.1 Background6.2 DIMA Instruction Set Architecture6.3 Compiler6.4 Validation Methodology6.5 Evaluation6.6 Conclusion7 Future ProspectsIndex

• Format: Paperback
• Language: English
• ISBN: 3030359735
• ISBN13: 9783030359737
• Release Date: January 2021
• Publisher: Springer
• Length: 174 Pages
• Weight: 0.60 lbs.
• Dimensions: 0.4" x 6.1" x 9.2"