Memory device for matrix-vector multiplications

• US 10,678,885 B2
• Filed: 02/25/2019
• Issued: 06/09/2020
• Est. Priority Date: 08/12/2016
• Status: Active Grant

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Le Gallo, Manuel, Sebastian, Abu
• Primary Examiner(s): Sandifer, Matthew D
• Application Number:
  US16/284,702
• Publication Number:
  US 20190188242A1
• Time in Patent Office: 470 Days
• CPC Class Codes: G06F 17/16 Matrix or vector computation , e.g. matrix-matrix or matrix-vector multiplication, matrix factorization matrix transposition G06F7/78
• CPC Class Codes: G11C 11/16 using elements in which the storage effect is based on magnetic spin effect
• CPC Class Codes: G11C 11/56 using storage elements with more than two stable states represented by steps, e.g. of voltage, current, phase, frequency
• CPC Class Codes: G11C 11/5607 using magnetic storage elements
• CPC Class Codes: G11C 11/5614 using conductive bridging RAM [CBRAM] or programming metallization cells [PMC]
• CPC Class Codes: G11C 11/5678 using amorphous/crystalline phase transition storage elements
• CPC Class Codes: G11C 11/5685 using storage elements comprising metal oxide memory material, e.g. perovskites
• CPC Class Codes: G11C 13/0002 using resistive RAM [RRAM] elements
• CPC Class Codes: G11C 13/0004 comprising amorphous/crystalline phase transition cells
• CPC Class Codes: G11C 13/0007 comprising metal oxide memory material, e.g. perovskites
• CPC Class Codes: G11C 13/0011 comprising conductive bridging RAM [CBRAM] or programming metallization cells [PMCs]
• CPC Class Codes: G11C 13/004 Reading or sensing circuits or methods
• CPC Class Codes: G11C 13/0069 Writing or programming circuits or methods
• CPC Class Codes: G11C 13/04 using optical elements ; using other beam accessed elements, e.g. electron or ion beam
• CPC Class Codes: G11C 13/047 using electro-optical elements G11C13/042 takes precedence

Abstract

A device for performing a multiplication of a matrix with a vector. The device comprises a plurality of memory elements, a signal generator and a readout circuit. The signal generator is configured to apply programming signals to the memory elements. The signal generator is further configured to control a first signal parameter of the programming signals in dependence on matrix elements of the matrix and to control a second signal parameter of the programming signals in dependence on vector elements of the vector. The readout circuit is configured to read out memory values of the memory elements. The memory values represent result values of vector elements of a product vector of the multiplication. The memory elements may be in particular resistive memory elements or photonic memory elements. Additionally there is provided a related method and design structure for performing the multiplication of a matrix with a vector.

Claims

1. A device for performing a multiplication of a matrix with a vector, the device comprising:

a plurality of memory elements;

a signal generator configured to

apply programming signals to the memory elements;

control a first signal parameter of the programming signals in dependence on matrix elements of the matrix; and

control a second signal parameter of the programming signals in dependence on vector elements of the vector; and

a readout circuit configured to read out memory values of the memory elements, the memory values representing result values of vector elements of a product vector of the multiplication, wherein the plurality of memory elements is configured as an array of memory elements, a memory element provided for each vector element of the product vector, wherein

to perform the multiplication the signal generator and the plurality of memory elements are configured to;

perform the multiplication by a plurality of successive steps in an accumulative manner; and

perform at each of the successive steps simultaneously a multiplication of column elements of one of the columns of the matrix with a corresponding vector element of the vector.

2. The device according to claim 1, wherein the signal generator is a pulse generator configured to apply pulses to the memory elements and wherein the first signal parameter is a first pulse parameter and the second signal parameter is a second pulse parameter.

3. The device according to claim 2, wherein the first pulse parameter is a duration of the pulses.

4. The device according to claim 2, wherein the second pulse parameter is an amplitude of the pulses.

5. The device according to claim 2, wherein the second pulse parameter is a number of the pulses.

6. The device as claimed in claim 1, wherein the device comprises

an array of input terminals configured to convey an input signal to the memory elements;

and

an array of transistors arranged in series to the memory elements;

wherein the signal generator is configured to control a control signal of a control terminal of the transistors.

7. The device as claimed in claim 1, wherein

the memory elements are resistive memory elements;

the programming signals are electrical programming signals; and

the memory values are resistance values.

8. The device as claimed in claim 7, wherein the resistive memory elements are one of Phase change memory (PCM) elements, Conductive bridge resistive memory elements, Metal-oxide resistive random access memory (RRAM) elements and Magneto-resistive random access memory (MRAM) elements.

9. The device as claimed in claim 7, wherein the resistive memory elements are Phase change memory elements and wherein the device is configured to:

apply a Reset-pulse to the PCM elements in order to bring the PCM-elements in the amorphous state;

apply current pulses to the PCM elements that heat the PCM elements above the crystallization temperature, thereby lowering the resistance of the PCM elements.

10. The device according to claim 7, wherein

the resistive memory elements are phase change memory elements;

the device comprises an array of counters; and

the device is configured to;

detect a predefined low resistance state of the resistive memory elements;

send an intermediate result value of a pending multiplication to a counter allocated to the respective resistive memory element having the low resistance state; and

apply a reset pulse to the respective resistive memory element having the low resistance state.

11. The device according to claim 10, wherein the result values of the vector elements of the product vector are represented by counter values of the counter.

12. The device as claimed in claim 1, wherein

the memory elements are photonic memory elements;

the programming signals are optical programming signals; and

the memory values are optical absorption values.

13. A method for performing a matrix multiplication of a matrix with a vector, the method comprising:

applying, by a signal generator, programming signals to a plurality of memory elements;

controlling a first signal parameter of the programming signals in dependence on matrix elements of the matrix;

controlling a second signal parameter of the programming signals in dependence on vector elements of the vector; and

reading out, by a readout circuit, memory values of the memory elements, the memory values representing result values of vector elements of a product vector of the multiplication, wherein the plurality of memory elements is configured as an array of memory elements, a memory element provided for each vector element of the product vector, wherein

to perform the matrix multiplication the signal generator and the plurality of memory elements are configured to;

perform the multiplication by a plurality of successive steps in an accumulative manner; and

perform at each of the successive steps simultaneously a multiplication of column elements of one of the columns of the matrix with a corresponding vector element of the vector.

14. The method according to claim 13, wherein the memory elements are resistive memory elements and the resistive memory elements are phase change memory elements and wherein the method further comprises:

detecting a predefined low resistance state of the resistive memory elements;

sending an intermediate result value of a pending multiplication to a counter allocated to the respective resistive memory element having the low resistance state; and

applying a reset pulse to the respective resistive memory element having the low resistance state.

15. A design structure tangibly embodied in a non-transitory machine readable medium for designing, manufacturing, or testing an integrated circuit, the design structure comprising:

a device for performing a multiplication of a matrix with a vector comprising

a plurality of memory elements;

a signal generator configured to

apply programming signals to the memory elements;

control a first signal parameter of the programming signals in dependence on matrix elements of the matrix;

control a second signal parameter of the programming signals in dependence on vector elements of the vector; and

a readout circuit configured to read out memory values of the memory elements, the memory values representing result values of vector elements of a product vector of the multiplication, wherein the plurality of memory elements is configured as an array of memory elements, a memory element provided for each vector element of the product vector, wherein

to perform the matrix multiplication the signal generator and the plurality of memory elements are configured to;

perform the multiplication by a plurality of successive steps in an accumulative manner; and

perform at each of the successive steps simultaneously a multiplication of column elements of one of the columns of the matrix with a corresponding vector element of the vector.