Abstract:Analog computing using non-volatile memristors has emerged as a promising solution for energy-efficient deep learning. New materials, like perovskites-based memristors are recently attractive due to their cost-effectiveness, energy efficiency and flexibility. Yet, challenges in material diversity and immature fabrications require extensive experimentation for device development. Moreover, significant non-idealities in these memristors often impede them for computing. Here, we propose a synergistic methodology to concurrently optimize perovskite memristor fabrication and develop robust analog DNNs that effectively address the inherent non-idealities of these memristors. Employing Bayesian optimization (BO) with a focus on usability, we efficiently identify optimal materials and fabrication conditions for perovskite memristors. Meanwhile, we developed "BayesMulti", a DNN training strategy utilizing BO-guided noise injection to improve the resistance of analog DNNs to memristor imperfections. Our approach theoretically ensures that within a certain range of parameter perturbations due to memristor non-idealities, the prediction outcomes remain consistent. Our integrated approach enables use of analog computing in much deeper and wider networks, which significantly outperforms existing methods in diverse tasks like image classification, autonomous driving, species identification, and large vision-language models, achieving up to 100-fold improvements. We further validate our methodology on a 10$\times$10 optimized perovskite memristor crossbar, demonstrating high accuracy in a classification task and low energy consumption. This study offers a versatile solution for efficient optimization of various analog computing systems, encompassing both devices and algorithms.

Mixed-Precision Partial Differential Equation Solver Design Based on Nonvolatile Memory

A Low-Power In-Memory Multiplication and Accumulation Array with Modified Radix-4 Input and Canonical Signed Digit Weights

An Efficient and Robust Partial Differential Equation Solver by Flash-Based Computing in Memory

A Fast-Convergence Near-Memory-Computing Accelerator for Solving Partial Differential Equations

Mixed-Precision In-Memory Computing

A general memristor-based partial differential equation solver

Design-Technology Co-Optimizations (DTCO) for General-Purpose Computing In-Memory Based on 55nm NOR Flash Technology

A Scalable and Reconfigurable Bit-Serial Compute-Near-Memory Hardware Accelerator for Solving 2-D/3-D Partial Differential Equations

A fine-grained mixed precision DNN accelerator using a two-stage big-little core RISC-V MCU.

Precision-Aware Iterative Algorithms Based on Group-Shared Exponents of Floating-Point Numbers

Reconfigurable application-specific photonic integrated circuit for solving partial differential equations

Efficient Approximate Floating-Point Multiplier With Runtime Reconfigurable Frequency and Precision

In-memory photonic dot-product engine with electrically programmable weight banks

Improving Dense Linear Equation Solver on Hybrid CPU-GPU System.

Implementation of digital MemComputing using standard electronic components

Synergistic Development of Perovskite Memristors and Algorithms for Robust Analog Computing

Design-Technology Co-Optimization for NVM-based Neuromorphic Processing Elements

Reliability-Aware Training and Performance Modeling for Processing-In-Memory Systems

Memristor Based Mixed-Accuracy Computation-in-Memory System.

Programming Memristor Arrays with Arbitrarily High Precision for Analog Computing.

Compute-in-Memory for Numerical Computations