Random matrix hardware for machine learning
View Patent ↗A computer-implemented method for training a random matrix network is presented. The method includes initializing a random matrix, inputting a plurality of first vectors into the random matrix, and outputting a plurality of second vectors from the random matrix to be fed back into the random matrix for training. The random matrix can include a plurality of two-terminal devices or a plurality of three-terminal devices or a film-based device.
1. A computer-implemented method executed by a processor for training a random matrix network, the method comprising:
initializing values of a random matrix by a random initializer;
inputting, by an input generator, a plurality of first vectors into the random matrix; and
outputting, by an output vector reader, a plurality of second vectors from the random matrix to be fed back into the random matrix for training.
2. The method of claim 1 , wherein the random matrix includes a plurality of two-terminal devices.
3. The method of claim 2 , wherein the two-terminal devices are resistive devices formed in a cross-bar array.
4. The method of claim 1 , wherein the random matrix includes a plurality of three-terminal devices.
5. The method of claim 4 , wherein the three-terminal devices are field effect transistors (FETs) formed in a cross-bar array.
6. The method of claim 1 , wherein the random matrix includes a film-based device.
7. The method of claim 6 , wherein the film-based device is formed that such top and bottom electrodes are aligned in a cross-bar array and a film is formed between the top and bottom electrodes.
8. The method of claim 1 , wherein the initializing occurs by electrical pulses inputted to the random matrix.
9. The method of claim 1 , wherein the initializing occurs by employing radiation used to adjust device parameters of devices forming the random matrix.
10. The method of claim 1 , wherein the initializing occurs by adjusting fabrication parameters of devices forming the random matrix to obtain specific random distributions.
11. A random matrix network, comprising:
a random initializer for initializing values of a random matrix;
an input generator for inputting a plurality of first vectors into the random matrix; and
an output vector reader for reading a plurality of second vectors outputted from the random matrix to be fed back into the random matrix for training.
12. The random matrix network of claim 11 , wherein the random matrix includes a plurality of two-terminal devices.
13. The random matrix network of claim 12 , wherein the two-terminal devices are resistive devices formed in a cross-bar array.
14. The random matrix network of claim 11 , wherein the random matrix includes a plurality of three-terminal devices.
15. The random matrix network of claim 14 , wherein the three-terminal devices are field effect transistors (FETs) formed in a cross-bar array.
16. The random matrix network of claim 11 , wherein the random matrix includes a film-based device.
17. The random matrix network of claim 16 , wherein the film-based device is formed that such top and bottom electrodes are aligned in a cross-bar array and a film is formed between the top and bottom electrodes.
18. The random matrix network of claim 11 , wherein the initializing occurs by electrical pulses inputted to the random matrix.
19. The random matrix network of claim 11 , wherein the initializing occurs by adjusting fabrication parameters of devices forming the random matrix to obtain specific random distributions.
20. A digital processing architecture for training and/or inferencing a random matrix, comprising:
a random matrix network including at least the random matrix having an array of elements connected between row and column lines;
at least one first single instruction, multiple data processing (SIMD) unit connected to the row lines;
at least one second SIMD unit connected to the column lines; and
a shared memory shared between the at least one first and second SIMDs.