Electronic-photonic Architectures for Brain-inspired Computing

feed news

HYBRAIN Paper: Using the IBM analog in-memory hardware acceleration kit for neural network training and inference

HYBRAIN Paper Using the IBM analog in-memory hardware acceleration kit for neural network training and inference

In a groundbreaking development funded by the European Union’s Horizon Europe research and innovation programme under grant agreement No 101046878 HYBRAIN project, researchers have delved into the intricacies of Analog In-Memory Computing (AIMC), a promising approach to reduce the latency and energy consumption of Deep Neural Network (DNN) inference and training.

However, the inherent noise and non-linear characteristics of AIMC chips, coupled with non-ideal peripheral circuitry require adapting DNNs to be deployed on such hardware to achieve accuracy equivalent to digital computing.

The recently released IBM Analog Hardware Acceleration Kit (AIHWKit), freely available at https://github.com/IBM/aihwkit, emerges as a pivotal tool in achieving these adaptations.

The AIHWKit is a Python library designed to simulate the inference and training of DNNs using AIMC. This tutorial offers an in-depth overview of the AIHWKit’s design, functionality and best practices for effective inference and training. Additionally, it introduces the Analog AI Cloud Composer, a platform providing the advantages of AIHWKit simulation in a fully managed cloud setting, combined with access to physical AIMC hardware, freely available at https://aihw-composer.draco.res.ibm.com.

A key aspect of the tutorial lies in the practicality of AIHWKit, demonstrated through Jupyter Notebook code examples, freely available at https://github.com/IBM/aihwkit/tree/master/notebooks/tutorial. These examples showcase how users can expand and customise the AIHWKit to meet their specific requirements.

This tutorial outlines several potential research directions enabled by AIHWKit. These include exploring device-level parameter specifications, implementing novel Analog optimizers for on-chip training, and studying the impact of low-precision arithmetic on digital operations for inference.

As technology continues to evolve, this tutorial serves as a beacon, guiding researchers and enthusiasts toward the transformative possibilities of Analog In-Memory Computing. With AIMC poised to reshape the landscape of computing, AIHWKit stands as a key enabler, fostering collaboration and innovation in this dynamic field.