Analog Alchemy: Neural Computation with In-Memory Inference, Learning and Routing

TL;DR

This paper explores using memristive devices for neural computation, enabling in-memory inference, learning, and routing by leveraging physical device dynamics, aiming to overcome energy and efficiency bottlenecks of traditional architectures.

Contribution

It introduces a memristive-based neural computation framework that integrates inference, learning, and routing, utilizing physical device properties for improved efficiency and scalability.

Findings

Hardware evidence of local learning adaptability to memristive devices

Development of new material stacks and circuit blocks for credit assignment

Efficient routing mechanisms for scalable analog crossbar architectures

Abstract

As neural computation is revolutionizing the field of Artificial Intelligence (AI), rethinking the ideal neural hardware is becoming the next frontier. Fast and reliable von Neumann architecture has been the hosting platform for neural computation. Although capable, its separation of memory and computation creates the bottleneck for the energy efficiency of neural computation, contrasting the biological brain. The question remains: how can we efficiently combine memory and computation, while exploiting the physics of the substrate, to build intelligent systems? In this thesis, I explore an alternative way with memristive devices for neural computation, where the unique physical dynamics of the devices are used for inference, learning and routing. Guided by the principles of gradient-based learning, we selected functions that need to be materialized, and analyzed connectomics principles for efficient wiring. Despite non-idealities and noise inherent in analog physics, I will provide hardware evidence of adaptability of local learning to memristive substrates, new material stacks and circuit blocks that aid in solving the credit assignment problem and efficient routing between analog crossbars for scalable architectures.