A DRAM-Based Process-in-Memory Using Data Redundancy and Differential Bit-Line Computation
1
조회
0
다운로드
0
인용
0
참고문헌
키워드
초록
This paper presents a novel Dynamic Random Access Memory (DRAM) - In Memory Computing (IMC) structure achieving high throughput without altering the existing cell configuration. Multiple Word Line (WL) activations are utilized to enhance the throughput of Multiply and Accumulate (MAC) operations. The issue of data destruction during simultaneous WL activations is addressed by employing the adjacent MAT and differential operation of the sense amplifier. Moreover, the problem arising from non-ideality in WL switches is alleviated through the differential bit-line computation operation. Consequently, an accuracy of 99.01% was achieved on the MNIST dataset.
Abstract
This paper presents a novel Dynamic Random Access Memory (DRAM) - In Memory Computing (IMC) structure achieving high throughput without altering the existing cell configuration. Multiple Word Line (WL) activations are utilized to enhance the throughput of Multiply and Accumulate (MAC) operations. The issue of data destruction during simultaneous WL activations is addressed by employing the adjacent MAT and differential operation of the sense amplifier. Moreover, the problem arising from non-ideality in WL switches is alleviated through the differential bit-line computation operation. Consequently, an accuracy of 99.01% was achieved on the MNIST dataset.
참고문헌 (7)
- S. Lee, "A 1ynm 1.25V 8Gb, 16Gb", s, 2022 IEEE International Solid- State Circuits Conference (ISSCC), pp.2022[DOI]
- Y. -C. Kwon, "25.4 A 20nm 6GB Function-In- Memory DRAM, Based on HBM2 with a 1.2TFLOPS Programmable Computing Unit Using Bank-Level Parallelism, for Machine Learning Applications", 2021 IEEE International Solid- State Circuits Conference (ISSCC), pp.350~335 (2021)
- S. Xie, "16.2 eDRAM-CIM: Compute-In-Memory Design with Reconfigurable Embedded-Dynamic- Memory Array Realizing Adaptive Data Converters and Charge-Domain Computing", 2021 IEEE International Solid- State Circuits Conference (ISSCC), pp.248~250 (2021)
- R. W. Erickson, "Fundamentals of Power Electronics" (2001), Springer
- M. He, "Newton : A DRAM-maker’s accelerator-inmemory(AiM)architecture for machine learning", 2020 53rd Annual IEEE, 2020, ~
- Y. Kim, "Input-splitting of large neural networks for power-efficient accelerator with resistive crossbar memory array", Proceedings of the International Symposium on Low Power Electronics and Design (2018)
- J. Kim, "Area-efficient and variation-tolerant in-memory BNN computing using 6T SRAM array", 2019 Symposium on VLSI Circuits (2019)