Linear symmetric self-selecting 14-bit kinetic molecular memristors

[Aimingatremission]

https://www.nature.com/articles/s41586-024-07902-2

Abstract

Artificial Intelligence (AI) is the domain of large resource-intensive data centres that limit access to a small community of developers1,2. Neuromorphic hardware promises greatly improved space and energy efficiency for AI but is presently only capable of low-accuracy operations, such as inferencing in neural networks3,4,5. Core computing tasks of signal processing, neural network training and natural language processing demand far higher computing resolution, beyond that of individual neuromorphic circuit elements6,7,8. Here we introduce an analog molecular memristor based on a Ru-complex of an azo-aromatic ligand with 14-bit resolution. Precise kinetic control over a transition between two thermodynamically stable molecular electronic states facilitates 16,520 distinct analog conductance levels, which can be linearly and symmetrically updated or written individually in one time step, substantially simplifying the weight update procedure over existing neuromorphic platforms3. The circuit elements are unidirectional, facilitating a selector-less 64 × 64 crossbar-based dot-product engine that enables vector–matrix multiplication, including Fourier transform, in a single time step. We achieved more than 73 dB signal-to-noise-ratio, four orders of magnitude improvement over the state-of-the-art methods9,10,11, while consuming 460× less energy than digital computers12,13. Accelerators leveraging these molecular crossbars could transform neuromorphic computing, extending it beyond niche applications and augmenting the core of digital electronics from the cloud to the edge12,13.

Layman’s terms

1. Current AI Limitations: Right now, AI relies on large, powerful data centers that use a lot of resources. This makes it hard for many developers to access and use AI.
2. Neuromorphic Hardware: This is a new type of hardware that mimics the human brain and promises to be much more efficient in terms of space and energy. However, it’s currently not very accurate and is mainly used for simple tasks like recognizing patterns in data.
3. High-Resolution Computing Needs: Tasks like processing signals, training neural networks, and understanding natural language require very high computing precision, which current neuromorphic hardware can’t provide.
4. New Development - Analog Molecular Memristor: Researchers have developed a new component called an analog molecular memristor. It’s based on a specific chemical compound and can achieve very high precision (14-bit resolution).
5. How It Works: This memristor can switch between two stable electronic states, allowing it to have 16,520 different levels of conductance. This means it can store and process information very precisely and efficiently.
6. Simplified Weight Updates: The memristor can update its conductance levels in one step, making it easier to adjust the weights in neural networks compared to current neuromorphic platforms.
7. Efficient Circuit Design: The circuit elements are designed to work in one direction, allowing for a simple and efficient layout that can perform complex mathematical operations like vector-matrix multiplication in one step.
8. Performance and Energy Efficiency: This new technology achieves a very high signal-to-noise ratio (over 73 dB) and is much more energy-efficient (460 times less energy consumption) compared to traditional digital computers.
9. Potential Impact: These molecular crossbars could significantly enhance neuromorphic computing, making it more useful for a wide range of applications and improving the performance of digital electronics from large data centers to small devices.

In essence, this new technology could make AI more accessible, efficient, and powerful by overcoming some of the current limitations of neuromorphic hardware.