Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

MemIntelli: A Generic End-to-End Simulation Framework for Memristive Intelligent Computing

Houji Zhou, Ling Yang, Zhiwei Zhou, Yi Li, Xiangshui Miao

TL;DR

MemIntelli tackles the reliability and precision challenges of memristive in-memory computing by delivering an end-to-end co-simulation framework that spans device, circuit, and application levels. It introduces a variable-precision DPE built on a bit-slicing approach, coupled with PyTorch-compatible hardware layers to enable layer-wise mixed-precision training and inference. The framework models crossbar non-idealities, supports INT and FP data, and provides a computing-graph interface for seamless integration with NumPy and PyTorch. Through diverse experiments (linear solvers, wavelet transforms, clustering, and NN training/inference), MemIntelli demonstrates robust processing capabilities and serves as a practical tool for IMC pre-verification and co-design.

Abstract

Memristive in-memory computing (IMC) has emerged as a promising solution for addressing the bottleneck in the Von Neumann architecture. However, the couplingbetweenthecircuitandalgorithm in IMC makes computing reliability susceptible to non-ideal effects in devices and peripheral circuits. In this respect, efficient softwarehardwareco-simulationtoolsarehighlydesiredtoembedthedevice and circuit models into the algorithms. In this paper, for the first time, we proposed an end-to-end simulation framework supporting flexible variable-precision computing, named MemIntelli, to realize the pre-verification of diverse intelligent applications on memristive devices. At the device and circuit level, mathematical functions are employed to abstract the devices and circuits through meticulous equivalent circuit modeling. On the architecture level, MemIntelli achieves flexible variable-precision IMC supporting integer and floating data representation with bit-slicing. Moreover, MemIntelli is compatible with NumPy and PyTorch for seamless integration with applications. To demonstrate its capabilities, diverse intelligent algorithms, such as equation solving, data clustering, wavelet transformation, and neural network training and inference, were employed to showcase the robust processing ability of MemIntelli. This research presents a comprehensive simulation tool that facilitates the co-design of the IMC system, spanning from device to application.

MemIntelli: A Generic End-to-End Simulation Framework for Memristive Intelligent Computing

TL;DR

MemIntelli tackles the reliability and precision challenges of memristive in-memory computing by delivering an end-to-end co-simulation framework that spans device, circuit, and application levels. It introduces a variable-precision DPE built on a bit-slicing approach, coupled with PyTorch-compatible hardware layers to enable layer-wise mixed-precision training and inference. The framework models crossbar non-idealities, supports INT and FP data, and provides a computing-graph interface for seamless integration with NumPy and PyTorch. Through diverse experiments (linear solvers, wavelet transforms, clustering, and NN training/inference), MemIntelli demonstrates robust processing capabilities and serves as a practical tool for IMC pre-verification and co-design.

Abstract

Memristive in-memory computing (IMC) has emerged as a promising solution for addressing the bottleneck in the Von Neumann architecture. However, the couplingbetweenthecircuitandalgorithm in IMC makes computing reliability susceptible to non-ideal effects in devices and peripheral circuits. In this respect, efficient softwarehardwareco-simulationtoolsarehighlydesiredtoembedthedevice and circuit models into the algorithms. In this paper, for the first time, we proposed an end-to-end simulation framework supporting flexible variable-precision computing, named MemIntelli, to realize the pre-verification of diverse intelligent applications on memristive devices. At the device and circuit level, mathematical functions are employed to abstract the devices and circuits through meticulous equivalent circuit modeling. On the architecture level, MemIntelli achieves flexible variable-precision IMC supporting integer and floating data representation with bit-slicing. Moreover, MemIntelli is compatible with NumPy and PyTorch for seamless integration with applications. To demonstrate its capabilities, diverse intelligent algorithms, such as equation solving, data clustering, wavelet transformation, and neural network training and inference, were employed to showcase the robust processing ability of MemIntelli. This research presents a comprehensive simulation tool that facilitates the co-design of the IMC system, spanning from device to application.

Paper Structure

This paper contains 13 sections, 1 equation, 17 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Background and Motivation
  3. Memristive simulation frameworks
  4. Bit-slicing method
  5. Motivation
  6. MemIntelli tool design
  7. MemInteli overview
  8. Crossbar module
  9. Variable-precision hardware DPE module
  10. Hardware neural network module with computing graph
  11. DPE core Verification
  12. EXPERIMENTAL RESULTS
  13. Conclusion

Figures (17)

  • Figure 1: Mechanism of bit-slicing method. (a) Fully binary mapping. (b) Asymmetric mapping. (c) Bit-slicing matrix multiplication. (d) Shared exponent strategy for FP matrix
  • Figure 2: The overview of the MemIntelli framework
  • Figure 3: Conductance data generated by device model.
  • Figure 4: Simulation model of the core. (a) Circuit model of the crossbar. (b) Mathematic model.
  • Figure 5: The matrix multiplication flow of INT data and FP data.
  • ...and 12 more figures