Behavioral Simulation of Biological Neuron Systems in SystemC

Electronic Systems Design Group

University of Southampton, UKSchool of Electronics and Computer Science

BehavioralBehavioral Simulation of Biological Neuron Simulation of Biological Neuron Systems in SystemCSystems in SystemC

S. S. Modi S. S. Modi 11, P.R. Wilson , P.R. Wilson 11, A.D. Brown , A.D. Brown 1 1 and J.E. Chad and J.E. Chad 22

School of Electronics and Computer ScienceSchool of Electronics and Computer Science11 &&School of NeuroscienceSchool of Neuroscience22

University of SouthamptonUniversity of SouthamptonSO17 1BJ, UKSO17 1BJ, UK

25/10/2004 2



There is a disconnect between….There is a disconnect between….

NeurobiologistsNeurobiologists• working in labs & using

software

EngineersEngineers• developing software

25/10/2004 3



Why is there is a disconnect ?Why is there is a disconnect ?

You cannot do full analogue simulation on systems You cannot do full analogue simulation on systems with > 1000 nodes practicallywith > 1000 nodes practically• Simulation Time is long• Accuracy is an issue

You cannot do meaningful system modelling on You cannot do meaningful system modelling on neural networks with < 10000 nodesneural networks with < 10000 nodes• What is the point of small systems – do experiments

instead

25/10/2004 4



Take a simple example: C. Take a simple example: C. eleganselegans

ShouldShould be be simulatablesimulatable• Has only 302 neurones • Topology known

But….But….• Everything is connected to

everything else• Connection 'strengths' are not

known

25/10/2004 5



Further MotivationFurther MotivationEmergence of new class of Neural Architectures based on Emergence of new class of Neural Architectures based on Neuroscience Neuroscience

• Pulsed Neural Network with Spatial-temporal coding and Spike Trains

• Novel Analogue and Digital Architectures with radically different type of information processing

Neural Core

Future sophisticated Future sophisticated SOCsSOCs will contain such neural coreswill contain such neural cores

25/10/2004 6



Bringing it all together…Bringing it all together…We need biologically realistic modelling and simulation of We need biologically realistic modelling and simulation of Biological Neuron Systems ( BNS ) : Biological Neuron Systems ( BNS ) :

• To gain insight of the Information Processing of BNS • To perform “Virtual experiments” on BNS• Verification of behaviour of neural SOC core

Desired characteristics for the modelling framework: Desired characteristics for the modelling framework: • Efficient, fast simulation of large neuron aggregates • Effective visualization of results for meaningful interpretation• Extensible and flexible framework, working on various level of

abstraction• Easy integration in SOC design environment

25/10/2004 7



Biological NeuronsBiological Neurons

Dendrite

Soma

Axon Hillock

Node of Ranvier

Schwam Cell

Axon

Presynaptic terminal

Synapse

0

40

-70

mV

Restig Potential( Leaking Iron Channels)

Depolarization due tosodium channels

Inactivation of sodiumchannels & activation of

pottasium channels

closing of sodium andpotassium channels

Spike

Action potential Spike

Generic neuron

Synapse

25/10/2004 8



Neuron ModellingNeuron ModellingBiophysical Abstract

Molecular Kinetic Models

Iron channel Models

Compartmental Models

Binary Perceptron

SigmoidalModels

Integrate and Fire Models

State Automata based models

25/10/2004 9



SimulatorsSimulators

The current simulators:The current simulators:• Bio-physical Models• Analogue Simulation • No standard, reusable and flexible platform• Do not use OOP and component library approach • Not very suitable for integrating in SOC design environment

Previous work developed a bespoke simulatorPrevious work developed a bespoke simulator• can we be more general/standard?

Low efficiency, restricted to small neural aggregates

25/10/2004 10



Previous important workPrevious important work

BehaviourBehaviour of neural aggregates of neural aggregates impliesimplies statistical statistical connectivityconnectivity• Not contentious

Initial results Initial results indicateindicate randomrandom networks (with the right networks (with the right statistics) exhibit the statistics) exhibit the samesame behaviourbehaviour• Health warning – single source so far

••E.T. E.T. ClaverolClaverol, A.D. Brown and J.E. Chad 2002 'Scalable cortical simulations o, A.D. Brown and J.E. Chad 2002 'Scalable cortical simulations on Beowulf architectures', n Beowulf architectures', NeurocomputingNeurocomputing, 43, pp 307, 43, pp 307--315, Mar 2002. 315, Mar 2002. ••E.T. E.T. ClaverolClaverol, A.D. Brown and J.E. Chad, 'Discrete simulation of large aggreg, A.D. Brown and J.E. Chad, 'Discrete simulation of large aggregates of neurons', ates of neurons', NeurocomputingNeurocomputing. 47, . 47, pp 277pp 277--297 Oct 2002 297 Oct 2002 ••E.T. E.T. ClaverolClaverol, A.D. Brown and J.E. Chad, 'A large scale simulation of the , A.D. Brown and J.E. Chad, 'A large scale simulation of the piriformpiriform cortex by a cell automaton based cortex by a cell automaton based network model', IEEE Transactions on Biomedical Engineering. 49,network model', IEEE Transactions on Biomedical Engineering. 49, no 9, Sept 2002 pp 921no 9, Sept 2002 pp 921--935935

25/10/2004 11



What would we like to do?What would we like to do?

The development of a computational model of:The development of a computational model of:• a single mammalian neurone• that concisely characterizes the information processing

capabilities.

A Btab

25/10/2004 12



How do we accomplish this?How do we accomplish this?

The development of a specialised (extremely fast) simulator The development of a specialised (extremely fast) simulator that canthat can• Handle the simulation of extremely large networks of the model. • The system will allow topological changes to be made to the

network• Predicated dynamically on the simulation results themselves (plasticity).

• Use a standard modeling and simulation platform for• Interoperability• Portability• Efficiency• Reliability• Development of libraries of standard elements

25/10/2004 13



What else would we like to do?What else would we like to do?

The development of a 'statistical assembler‘The development of a 'statistical assembler‘• Will automatically assemble large bodies of neurone models, such that the gross

connectivity properties of the collection can be manipulated by the researcher.

25/10/2004 14



Can we develop a design technique?Can we develop a design technique?

The development of an 'activity The development of an 'activity browser‘browser‘

• Allows the investigator to study the macroscopic behaviour of the assembled system in a manner sympathetic to the aims of the research.

• Simple user interface• Decouple main user from the

complex programming required

25/10/2004 15



Basic platform: Choice of SystemC Basic platform: Choice of SystemC

• A natural candidate: an HDL with inbuilt concurrency and time• Built on the general purpose C++ programming language• Designed to handle simulation of very large system • Efficient and fast event driven kernel• Open source and uses standard compilers• Object oriented, promotes libraries with Plug-and-play

components• Scalable designs, small executables specifications • C/C++ is pervasive in the scientific/ engineering community• Designed for SOC design and verification, enables seamless

integration of BNS modelling and simulation in SOC environment

25/10/2004 16



SystemC FrameworkSystemC FrameworkStructure Structure

N e u r o n 2

S y s te m C s im u la t io n s y s te m

m o n ito r _ d r iv e r ( to p le v e l m o d u le )

n _ s in g n a l

n e tw o r k

N e u r o n

n _ s in g n a lT h r e s h o ld p r o c e s s

B u r s t G e n e r a to r p r o c e s s

O s c i l la to rp r o c e s s

f u n c t io n u p d a te _ w _ s u m ( W s y n )

w _ s u m

e v e n t t r ig g e r _ g e n

r e m a n in g _ b u r s t s

S y n

S y n

S y n

S y n

S y n

S y n

N e u r o n 2

N e u r o n 2

S y n

S y n

n _ s in gn a l

S y n a p s ew a i t ( p r e s y n a p t ic _ n e u r o n _ to _ t r ig g e r ) ;w a i t ( s y n a p t ic _ d e la y ) ;in c r e a s e _ m e m b r a n e _ p o te n t ia l ( W s y n ) ;w a i t ( s y n a p t ic _ d u r a t io n ) ;d e c r e a s e _ m e m b r a n e _ p o te n t ia l ( W s y n ) ;

25/10/2004 17



Some Design and Performance IssuesSome Design and Performance Issues

• Modelling of Transport delay

• Pointer access instead of Signal communication

• Use of shared inter-process variable

• Run-time Re-configurable network

25/10/2004 18



Some Results: VisualizationSome Results: Visualization

Waveform viewerActivity Browser

25/10/2004 19



Animated VisualizationAnimated Visualization

25/10/2004 20



Modelling of Modelling of C.ElegansC.Elegans Nervous SystemNervous System

• Standard Animal for or study of Neuroscience ( 302 Neurons ) • Known Neuron topologies and properties• Simulation of its locomotory nervous system ( 85 Neurons )

Animated visualization Animal MovementSimulated activity Reference results

25/10/2004 21



Calibration of the model…Calibration of the model…

Wet experiments to ‘calibrate’ the Wet experiments to ‘calibrate’ the model/simulatormodel/simulator

25/10/2004 22



OutcomesOutcomesOutcomes: Outcomes:

• Use of SystemC for Biological Neuron System simulation• Effective• Fast• Efficient

• Flexible, modular and extensible framework in very short design time

• Tcl/ Tk activity browser • Results consistent with previous work

• NB 3 years modeling work condensed to 3 months using a standard platform

• Users can develop and use these models with a minimal learning curve

25/10/2004 23



Short Term GoalsShort Term Goals

Large scale statistical experiments on emergent Large scale statistical experiments on emergent behaviourbehaviourof artificial neural aggregatesof artificial neural aggregates

25/10/2004 24



Long Term GoalsLong Term Goals

Interaction with the chemical ambientInteraction with the chemical ambient• How the models behaviour changes depending on the ambient

chemistry

25/10/2004 25



The bottom line….The bottom line….

This approach allows Neuroscientists to leverage the power This approach allows Neuroscientists to leverage the power of advanced simulation techniques to analyse neuron of advanced simulation techniques to analyse neuron structuresstructures

Engineers delivering a practical usable platform, and also Engineers delivering a practical usable platform, and also learning from how nature deals with complexitylearning from how nature deals with complexity

Documents

Behavioral Simulation of Biological Neuron Systems in SystemC