LUT based synapse weight update scheme in STDP neuromorphic systems

1. A method for updating synapse weight values in a Spike Time Dependent Plasticity model neuromorphic system, comprising:

selectively performing, by a hardware-based synapse weight incrementer or decrementer, one of a synapse weight increment function or a synapse weight decrement function, the hardware-based synapse weight incrementer and decrementer being independent of each other and each using a respective lookup table, to generate updated synapse weight values responsive to spike timing data, the lookup tables including contents determined at least in part by applying the synapse weight increment function and the synapse weight decrement function, the synapse weight increment function being based on an axon timer value as a function of time and synapse weight as a function of time, and the synapse weight decrement function being based on a dendrite timer value as a function of time and the synapse weight as a function of time;

storing the updated synapse weight values in a memory;

implementing a multiplexer to select one of a result of the synapse weight increment function and a result of the synapse weight decrement function to update synapse weight values; and

performing, by a hardware-based processor, a learning process to integrate the updated synapse weight values stored in the memory into the Spike Time Dependent Plasticity model neuromorphic system for improved neuromorphic simulation.

2. The method of claim 1 , wherein the contents of the respective lookup tables are further based on contents of a simulation, or a hardware measurement, of synapse models.

3. The method of claim 2 , further comprising determining that missing data exists in the simulation or hardware measurement, and calculating the missing data by at least one of interpolation and extrapolation, each using existing data.

4. The method of claim 1 , further comprising supplying the contents of the respective lookup tables based on a Spike Time Dependent Plasticity theory.

5. The method of claim 1 , further comprising preloading the contents of the respective lookup tables into the respective lookup tables.

6. The method of claim 1 , wherein the contents of the respective lookup tables further include delta values of one of the synapse weight increment function and the synapse weight decrement function.

7. The method of claim 1 , further comprising determining increment amounts of the synapse weight values using current synapse weight values and current Axon Timer values when at least one dendrite output (neuron fire) spike arises.

8. The method of claim 1 , further comprising determining decrement amounts of the synapse weight values using current synapse weight values and current Dendrite Timer values when at least one axon input spike arises.

9. The method of claim 1 , wherein the contents for the respective lookup tables are determined by applying:

sw ( t+ 1)= sw ( t )+ ax _ tmr ( t ) 2 /p *(( sw ( t )− q )/ q ) 2 q ) 2

for the synapse weight increment function, and

sw ( t+ 1)= sw ( t )− dr _ tmr ( t ) 2 /p *( sw ( t )/ q ) 2

for the synapse weight decrement function,

wherein sw is the synapse weight, t is the time, ax_tmr(t) is the axon timer value for time t, dr_tmr(t) is the dendrite timer value for time t, and p and q are the optimization parameters.

10. The method of claim 1 , further comprising:

switching from a learning mode to a non-learning mode.

11. The method of claim 1 , further comprising:

determining an update for a neuron membrane potential while incorporating a leak effect to generate an updated neuron membrane potential; and

storing the updated neuron memory potential in the memory

wherein performing the learning process further includes simultaneously integrating the updated synapse weight values and the updated neuron membrane potential stored in the memory into the Spike Time Dependent Plasticity model neuromorphic system for improved neuromorphic simulation.

12. A computer-readable storage medium including a computer-readable program for updating synapse weight values in a Spike Time Dependent Plasticity model neuromorphic system, wherein the computer-readable program when executed on a computer causes the computer to perform the steps of:

selectively performing, by a hardware-based synapse weight incrementer or decrementer, one of a synapse weight increment function or a synapse weight decrement function, the hardware-based synapse weight incrementer and decrementer being independent of each other and each using a respective lookup table, to generate updated synapse weight values responsive to spike timing data, the lookup tables including contents determined at least in part by applying the synapse weight increment function and the synapse weight decrement function, the synapse weight increment function being based on an axon timer value as a function of time and synapse weight as a function of time, and the synapse weight decrement function being based on a dendrite timer value as a function of time and the synapse weight as a function of time;

implementing a multiplexer to select one of a result of the synapse weight increment function and a result of the synapse weight decrement function to update synapse weight values;

storing the updated synapse weight values in a memory; and

performing, by a hardware-based processor, a learning process to integrate the updated synapse weight values stored in the memory into the Spike Time Dependent Plasticity model neuromorphic system for improved neuromorphic simulation.

13. The computer readable storage medium of claim 12 , wherein the computer-readable program when executed on the computer causes the computer to further perform the step of preparing, by the hardware-based processor, the contents of the respective lookup tables, wherein the contents are further based on at least one of:

contents of a simulation;

a hardware measurement of synapse models; and

some equations to approximate the model performance.

14. A system for updating synapse weight values in a Spike Time Dependent Plasticity model neuromorphic system, comprising:

at least one of a hardware-based synapse weight incrementer and a hardware-based synapse weight decrementer, each respectively configured to perform either a synapse weight increment function or a synapse weight decrement function, the hardware-based synapse weight incrementer and decrementer being independent of each other and each using a respective lookup table, to generate updated synapse weight values in response to spike timing data, the lookup tables including contents determined at least in part by applying the synapse weight increment function and the synapse weight decrement function, the synapse weight increment function being based on an axon timer value as a function of time and synapse weight as a function of time, and the synapse weight decrement function being based on a dendrite timer value as a function of time and the synapse weight as a function of time;

a memory configured to store the updated synapse weight values;

a multiplexer configured to select one of a result of the synapse weight increment function and a result of the synapse weight decrement function to update synapse weight values; and

a hardware-based processor configured to perform a learning process to integrate the updated synapse weight values stored in the memory into the Spike Time Dependent Plasticity model neuromorphic system for improved neuromorphic simulation.

15. The system of claim 14 , wherein the hardware-based processor is further configured to prepare the contents of the respective lookup tables, wherein the contents are based on at least one of:

contents of a simulation;

a hardware measurement of synapse models; and

some equations to approximate the model performance.

16. The system of claim 15 , wherein, in response to determining any missing data in the simulation or hardware measurement, the hardware-based processor is further configured to calculate the missing data by at least one of interpolation and extrapolation, each using existing data.

17. The system of claim 14 , wherein the hardware-based processor is further configured to supply the contents of the respective lookup tables based on a Spike Time Dependent Plasticity theory.

18. The system of claim 14 , wherein the hardware-based processor is further configured to determine increment amounts of the synapse weight values using current synapse weight values and current Axon Timer values when at least one dendrite output (neuron fire) spikes arise.

19. The system of claim 14 , wherein the hardware-based processor is further configured to determine decrement amounts of the synapse weight values using current synapse weight values and current Dendrite Timer values when at least one axon input spikes arise.

20. The system of claim 14 , wherein the hardware-based processor is further configured to:

determine an update for a neuron membrane potential while incorporating a leak effect to generate an updated neuron membrane potential;

store the updated neuron memory potential in the memory; and

perform the learning process by simultaneously integrating the updated synapse weight values and the updated neuron membrane potential stored in the memory into the Spike Time Dependent Plasticity model neuromorphic system for improved neuromorphic simulation.