Speeding up artificial intelligence

A group at Politecnico di Milano has developed an electronic circuit able to solve a system of linear equations in a single operation in the timescale of a few tens of nanoseconds. The performance of this new circuit is superior not only to classical digital computers, but also to quantum computers. It will be soon possible to develop a new generation of computing accelerators that will revolutionize the technology of artificial intelligence.

Solving a system of linear equations means finding the unknown vector X which satisfies the equation Ax = b, where A is a matrix of coefficients and b is a known vector. To solve this problem, a conventional digital computer executes an algorithm that takes several operations, thus translating into considerable time and .

The new circuit, which has been developed in the frame of the ERC European project Resistive Switch Computing Beyond CMOS (RESCUE), solves systems of linear equations (Ax=b) thanks to an innovative method of in-memory computing, where the coefficients of matrix A are stored in a special device called a memristor. The memristor is able to store analogue values, so a memristor matrix can physically map a coefficient matrix A within the circuit, thus strongly accelerating the computation.

The memristor circuit has been tested and validated on a wide set of algebraic problems, such as the ranking of internet websites and the solution of complicated differential equations including the Schrödinger for the computation of the quantum wavefunction for an electron. All these problems are solved in a single operation.

These results have been published in the Proceedings of the National Academy of Science.


Explore further

Memory-processing unit could bring memristors to the masses

More information: Zhong Sun et al. Solving matrix equations in one step with cross-point resistive arrays, Proceedings of the National Academy of Sciences (2019). DOI: 10.1073/pnas.1815682116
Provided by Politecnico di milano
Citation: Speeding up artificial intelligence (2019, March 15) retrieved 15 October 2019 from https://phys.org/news/2019-03-artificial-intelligence.html
This document is subject to copyright. Apart from any fair dealing for the purpose of private study or research, no part may be reproduced without the written permission. The content is provided for information purposes only.
40 shares

Feedback to editors

User comments

Mar 17, 2019
That is freaking awesome. If memory companies don't see the value in resistive memory tech at this point they are stoopid. Reading about normal cross point memory arrays (not this), I guess there's issues with parasitic leakage without another selector layer sandwiched in there, but the interesting thing about AI is there are situations where some noise is no problem, and in certain situations it could even help. As the power of analogue compute is released for AI, we may have to face that AI is intrinsically unpredictable if sufficiently powerful.

Please sign in to add a comment. Registration is free, and takes less than a minute. Read more