Method and device for binary coding of signals in order to implement digital MAC operations with dynamic precision

Patent metadata

Field	Value
Publication number	US-12003255-B2
Application number	US-202017787300-A
Country	US
Kind code	B2
Filing date	Dec 10, 2020
Priority date	Dec 18, 2019
Publication date	Jun 4, 2024
Grant date	Jun 4, 2024

Abstract

A computer-implemented method for coding a digital signal intended to be processed by a digital computing system includes the steps of: receiving a sample of the digital signal quantized on a number Nd of bits, decomposing the sample into a plurality of binary words of parameterizable bit size Np, coding the sample through a plurality of pairs of values, each pair comprising one of the binary words and an address corresponding to the position of the binary word in the sample, transmitting the pairs of values to an integration unit in order to carry out a MAC operation between the sample and a weighting coefficient.

First claim

The invention claimed is:

1. A computer-implemented method for coding a digital signal intended to be processed by a digital computing system, the method comprising the steps of:
receiving a sample of the digital signal quantized on a number N d of bits,
decomposing the sample into a plurality of binary words of parameterizable bit size N p ,
coding the sample through a plurality of pairs of values, each pair comprising one of the binary words and an address corresponding to the position of the binary word in the sample,
transmitting the pairs of values to an integration unit in order to carry out a MAC operation between the sample and a weighting coefficient.

2. The coding method as claimed in claim 1 , wherein only pairs of values comprising non-zero binary words are transmitted.

3. The coding method as claimed in claim 1 , comprising a step of deleting zero binary words before carrying out the MAC operation.

4. The coding method as claimed in claim 1 , wherein one and the same address corresponding to binary words coded from various samples is transmitted only once.

5. The coding method as claimed in claim 1 , comprising a step of determining the size N p of a binary word based on the statistical distribution of the values at 0 of the bits of the digital signal.

6. The coding method as claimed in claim 5 , wherein the size N p of a binary word is parameterized so as to minimize the energy consumption of a digital computing system wherein the processed signals are coded by way of said coding method.

7. The coding method as claimed in claim 6 , wherein the energy consumption is estimated by simulation or on the basis of an empirical model.

8. The coding method as claimed in claim 1 , wherein the digital computing system implements an artificial neural network.

9. The coding method as claimed in claim 8 , wherein the size N p of the binary words is parameterized independently for each layer of the artificial neural network.

10. A coding device comprising a coder configured to execute the coding method as claimed in claim 1 .

11. An integration device configured to carry out a multiply-accumulate (MAC) operation between a first number coded by way of the coding method as claimed in claim 1 and a weighting coefficient, the device comprising a multiplier (MUL) for multiplying the weighting coefficient by the binary word of the first number, a shift register (REG) configured to shift the output signal from the multiplier (MUL) by the value of the address of said binary word, an adder (ADD) and an accumulation register (RAC) for accumulating the output signal from the shift register (REG).

12. An artificial neuron (N), implemented by a digital computing system, comprising an integration device as claimed in claim 11 for carrying out a multiply-accumulate (MAC) operation between a received signal and a synaptic coefficient, and a coding device comprising a coder configured to execute the coding method as claimed in claim 11 for coding the output signal from the integration device the artificial neuron (N) being configured to propagate the coded signal to another artificial neuron.

13. An artificial neuron (N), implemented by a computer, comprising an integration device as claimed in claim 11 for carrying out a multiply-accumulate (MAC) operation between an error signal received from another artificial neuron and a synaptic coefficient, a local error computing module configured to compute a local error signal on the basis of the output signal from the integration device and a coding device comprising a coder configured to execute the coding method as claimed in claim 11 for coding the local error signal, the artificial neuron (N) being configured to back-propagate the local error signal to another artificial neuron.

14. An artificial neural network comprising a plurality of artificial neurons as claimed in claim 12 .

Assignees

• Commissariat Energie Atomique

Inventors

• Thiele Johannes Christian
• Bichler Olivier
• Duranton Marc
• Lorrain Vincent

Classifications

• G06F17/18

  for evaluating statistical data {, e.g. average values, frequency distributions, probability functions, regression analysis (forecasting specially adapted for a specific administrative, business or logistic context G06Q10/04)} · CPC title

• G06N3/063

  using electronic means · CPC title

• H03M7/28 Primary

  Programmable structures, i.e. where the code converter contains apparatus which is operator-changeable to modify the conversion process · CPC title

• G06F7/5443

  Sum of products (for applications thereof, see the relevant places, e.g. G06F17/10, H03H17/00) · CPC title

• G06N3/02

  Neural networks · CPC title