Recently, Prof. Chen Jiezhi’s team from the School of Information Science and Engineering, at Shandong University, published their research titled “An Efficient Flash-Based Computing-in-Memory (CIM) Demonstration of High-Precision (32-bit) Nonlinear Partial Differential Equation (PDE) Solver With Ultra-High Endurance and Reliability” in IEEE Transactions on Circuits and Systems I: Regular Papers (TCAS-I), a flagship journal in the integrated circuits field. The paper’s first author is Feng Yang, a 2021 Ph. D. candidate, and Prof. Chen is the corresponding author. Shandong University is credited as the primary affiliation.

As data-intensive applications like artificial intelligence demand greater computational power and efficiency, computing in memory (CIM) has emerged as a critical solution. By integrating computational units directly within memory arrays, CIM significantly enhances energy efficiency and computational density, making it a pivotal technology for high-performance computing. This study addresses the growing need for high-precision CIM units, particularly for solving nonlinear partial differential equations (PDEs).

Fig 1 Mapping Principle of the FNO Algorithm and High-Precision CIM System

The team developed a 32-bit flash-based high-precision CIM unit utilizing the Fourier Neural Operator (FNO) algorithm. They successfully implemented a hardware system to solve nonlinear PDEs, combining reliability analysis, circuit testing, and system-level simulation to validate performance. The system achieved a pulse output speed of 40ns and demonstrated an energy efficiency of 2 TOPS/W, completing complex computations with reduced cost and power consumption. This work offers a novel pathway for developing energy-efficient general-purpose CIM systems.

This research received funding from several prestigious programs, including the National Key R&D Program, the National Natural Science Foundation of China, the Major Research Plan for Post-Moore Computing, and CSC funding.