Hybrid analog-digital matrix processors
AssigneeLightmatter, Inc.
InventorsTyler J. Kenney, Martin B. Z. Forsythe, Tomo Lazovich, Darius Bunandar
Life of this patent
Estimating the term…
Abstract
Techniques for computing matrix operations for arbitrarily large matrices on a finite-sized hybrid analog-digital matrix processor are described. Techniques for gain adjustment in a finite-sized hybrid analog-digital matrix processor are described which enable the system to obtain higher energy efficiencies, greater physical density and improved numerical accuracy. In some embodiments, these techniques enable maximization of the predictive accuracy of a GEMM-based convolutional neural network using low-precision data representations.
Claims18
A hybrid analog-digital processor comprising:
- circuitry comprising an analog processor, wherein the circuitry is configured to perform a multi-layer neural network, the performing comprising:
- obtaining a first scaling factor for a first matrix associated with a first layer of the multi-layer neural network;
- scaling at least a portion of the first matrix using the first scaling factor to obtain a first scaled matrix;
- performing the first layer of the multi-layer neural network by controlling the analog processor to perform matrix multiplication using the first scaled matrix;
- obtaining a second scaling factor for a second matrix associated with a second layer of the multi-layer neural network, wherein the first scaling factor is different from the second scaling factor;
- scaling at least a portion of the second matrix using the second scaling factor to obtain a second scaled matrix;
- and
- performing the second layer of the multi-layer neural network by controlling the analog processor to perform matrix multiplication using the second scaled matrix.
A method of performing a multi-layer neural network, comprising:
- obtaining a first scaling factor for a first matrix associated with a first layer of the multi-layer neural network;
- scaling at least a portion of the first matrix using the first scaling factor to obtain a first scaled matrix;
- performing the first layer of the multi-layer neural network by controlling an analog processor to perform matrix multiplication using the first scaled matrix;
- obtaining a second scaling factor for a second matrix associated with a second layer of the multi-layer neural network, wherein the first scaling factor is different from the second scaling factor;
- scaling at least a portion of the second matrix using the second scaling factor to obtain a second scaled matrix;
- and
- performing the second layer of the multi-layer neural network by controlling the analog processor to perform matrix multiplication using the second scaled matrix.
Dependencies are inferred from the claim text.