Neural cell and a neural network

1. A neural network, comprising multiple layers of neural cells;

wherein a certain neural cell of a certain layer of neural cells comprises:

a first plurality of one-bit inputs;

an adder and leaky integrator unit; and

an activation function circuit that has a one-bit output;

wherein the first plurality of one-bit inputs are coupled to a first plurality of one-bit outputs of neural cells of a layer that precedes the certain layer;

wherein the adder and leaky integration unit is configured to calculate a leaky integral of a weighted sum of a number of one-bit pulses that were received, during a time window, by the first plurality of one-bit inputs; and

wherein the activation function circuit is configured to apply an activation function on the leaky integral to provide a one-bit output of the certain neural cell.

2. The neural network according to claim 1 , wherein the activation function is a sigmoid and wherein the activation function unit comprises a random number generator for generating random numbers and a comparator for comparing the leaky integral to a sum of at least two of the random numbers.

3. The neural network according to claim 1 , wherein the activation function is a sigmoid and wherein the activation function unit comprises a pseudo-random number generator for generating pseudo-random numbers and a comparator for comparing the leaky integral to a sum of at least two of the pseudo-random numbers.

4. The neural network according to claim 1 , wherein the activation function is a rectified linear unit, and the activation function circuits is configured to calculate the rectified linear unit.

5. The neural network according to claim 1 , wherein the certain neural cell comprises multiple flip-flops that are configured to store the one-bit pulses, a memory for storing a bias value and for storing weights associated with the one-bit inputs, and an adder that is configured to calculate the leaky integral using the content of the multiple flip-flops, the bias value and the weights.

6. The neural network according to claim 1 , wherein the adder and leaky integrator unit comprises a shift register that is configured to apply a leaky function.

7. The neural network according to claim 1 , wherein the neural network is free of multipliers.

8. The neural network according to claim 1 , wherein the neural network consists of up to 200 neurons.

9. The neural network according to claim 1 , wherein neural cells of an input layer of the neural network have a single one-bit input.

10. The neural network according to claim 1 , wherein the neural network is a purely digital neural network.

11. A method for activating a certain neural cell of a neural network, wherein the method comprises:

calculating, by an adder and leaky integration unit of the certain neural cell, a leaky integral of a weighted sum of a number of one-bit pulses that were received, during a time window, by a first plurality of one-bit inputs of the certain neural cell; wherein the certain neural network belongs to a certain layer of neural cells of the neural network; wherein the first plurality of one-bit inputs are coupled to a first plurality of one-bit outputs of neural cells of a layer that precedes the certain layer; and

applying, by an activation function circuit of the certain neural cell, an activation function on the leaky integral to provide a one-bit output of the certain neural cell.

12. The method according to claim 11 , wherein the activation function is a sigmoid and wherein the applying of the activation function comprises generating random numbers by a random number generator of the activation function circuit and comparing, by a comparator of the activation function circuit, the leaky integral to a sum of at least two of the random numbers.

13. The method according to claim 11 , wherein the activation function is a sigmoid and wherein the applying of the activation function comprises comprises generating, by a pseudo-random number generator of the activation function circuit, pseudo-random numbers and comparing, by a comparator of the activation function circuit, the leaky integral to a sum of at least two of the pseudo-random numbers.

14. The method according to claim 11 , wherein the activation function is a rectified linear unit, and wherein the applying of the activation function comprises comprises calculating the rectified linear unit.

15. The method according to claim 11 , comprising storing by multiple flip-flops of the certain neural cell the one-bit pulses, storing by a memory of the certain neural cell a bias value and weights associated with the one-bit inputs, and calculating by an adder of the certain neural cell, the leaky integral using the content of the multiple flip-flops, the bias value and the weights.

16. The method according to claim 11 , comprising applying, by a shift register of the adder and leaky integrator unit, a leaky function.

17. The method according to claim 11 , wherein the neural network is free of multipliers.

18. The method according to claim 11 , wherein the neural network consists of up to 200 neurons.

19. The method according to claim 11 , wherein neural cells of an input layer of the neural network have a single one-bit input.

20. The method according to claim 11 , wherein the neural network is a purely digital neural network.