A High-Precision Hybrid Floating-Point Compute-in-Memory Architecture for Complex Deep Learning

Guardado en:

Detalles Bibliográficos
Publicado en:	Electronics vol. 14, no. 22 (2025), p. 4414-4436
Autor principal:	Ma Zizhao
Otros Autores:	Wang, Chunshan, Chen, Qi, Wang, Yifan, Xie Yufeng
Publicado:	MDPI AG
Materias:	Accuracy Computer memory Deep learning Computer architecture Artificial intelligence Lookup tables Multiplication Design Architecture Energy efficiency Arrays Algorithms Machine learning Workloads Floating point arithmetic
Acceso en línea:	Citation/Abstract Full Text + Graphics Full Text - PDF
Etiquetas:	Agregar Etiqueta Sin Etiquetas, Sea el primero en etiquetar este registro!

Descripción
Resumen:	As artificial intelligence (AI) advances, deep learning models are shifting from convolutional architectures to transformer-based structures, highlighting the importance of accurate floating-point (FP) calculations. Compute-in-memory (CIM) enhances matrix multiplication performance by breaking down the von Neumann architecture. However, many FPCIMs struggle to maintain high precision while achieving efficiency. This work proposes a high-precision hybrid floating-point compute-in-memory (Hy-FPCIM) architecture for Vision Transformer (ViT) through post-alignment with two different CIM macros: Bit-wise Exponent Macro (BEM) and Booth Mantissa Macro (BMM). The high-parallelism BEM efficiently implements exponent calculations in-memory with the Bit-Separated Exponent Summation Unit (BSESU) and the routing-efficient Bit-wise Max Finder (BMF). The high-precision BMM achieves nearly lossless mantissa computation in-memory with efficient Booth 4 encoding and the sensitivity-amplifier-free Flying Mantissa Lookup Table based on 12T Triple Port SRAM. The proposed Hy-FPCIM architecture achieves 23.7 TFLOPS/W energy efficiency and 0.754 TFLOPS/mm2 area efficiency, with 617 Kb/mm2 memory density in 28 nm technology. With almost lossless architectures, the proposed Hy-FPCIM achieves an accuracy of 81.04% in recognition tasks on the ImageNet dataset using ViT, representing a 0.03% decrease compared to the software baseline. This research presents significant advantages in both accuracy and energy efficiency, providing critical technology for complex deep learning applications.
ISSN:	2079-9292
DOI:	10.3390/electronics14224414
Fuente:	Advanced Technologies & Aerospace Database