AdexSimMaxeler

Simulation of the Adaptive Exponent Integrate-and-Fire Neuron model with Spike-timing Dependent Plasticity accelerated on Maxeler Dataflow Engines

Getting Started

The following instructions help build and execute the simulation on a Maxeler DFE.

Prerequisites

The program has been compiled with MaxCompiler 2017.2.1 and targeted on a MAIA DFE Model.

Compilation

The following instructions are needed to compile the program.

For Simulation

cd /CPUCode
export PATH=$HOME/bin:/opt/altera/13.1/quartus/bin/:$PATH
source "/opt/maxcompiler2017.2.1/settings.sh"
make RUNRULES="Simulation"

For DFE

cd /CPUCode
export PATH=$HOME/bin:/opt/altera/13.1/quartus/bin/:$PATH
source "/opt/maxcompiler2017.2.1/settings.sh"
make RUNRULES="DFE"

Running

There are two options to run the program. The first one is running on the machine's CPU simulating the DFE operation and the second one is running on DFEs.

For Simulation

cd /CPUCode
source "/opt/maxcompiler2017.2.1/settings.sh"
make RUNRULES="Simulation" startsim // And then follow the instructions of the output of this instruction
./../RunRules/Simulation/Simulation Number_of_Timesteps N_S N_Group_S N_Group_T SpikeInt WriteEn AllConnected RunCPU

For DFE

./Simulation Number_of_Timesteps N_S N_Group_S N_Group_T SpikeInt WriteEn RunCPU AllConnected

Number of Timesteps: Number of Simulation Steps
N_S: Number of Input Neurons
N_Group_S: Number of AdEx Neurons as Source
N_Group_T: Number of AdEx Neurons as Target
SpikeInt: Input Neurons Spike Interval
WriteEn: 1: Neuron Values output for every timestep in "NeuronOut.csv", 0: Output only of the final step
AllConnected: 1: All connected, 0: Connections read from "connections.txt"
RunCPU: 1: Run CPU Code with doubles to check acceleration and errors, 0: Only DFE run

Example 1

To run the following Simulation:
Number of Input Neurons: 10
Number of AdEx Neurons: 10
NxM Simulation (Input Neurons connected only to AdEx)
Simulation Steps: 1000 (1s)
Input Spikes Frequency: 1kHz (Spike Interval = 1 step)
Writing output in file for every timestep
All Input Neurons connected with all the AdEx Neurons
No running on CPU

./Simulation 1000 10 0 10 1 1 0 1

Example 2

To run the following Simulation:
Number of Input Neurons: 0
Number of AdEx Neurons: 10
MxM Simulation (AdEx Neurons connected to each other)
Simulation Steps: 1000 (1s)
Input Spikes Frequency doesn't matter
Writing output in file for every timestep
Specific connections (Synapses) between AdEx Neurons specified in "connections.txt"
No running on CPU

./Simulation 1000 0 10 10 1 1 0 0

The file "connections.txt" that specifies the connections (Synapses) between Neurons must be formated as the example file in the repository, as an Adjacency Matrix.
The rows correspond to AdEx Neurons as Targets and the columns to Input Neurons or AdEx Neurons as Sources.