Available online at www.sciencedirect.

com

ScienceDirect
IFAC PapersOnLine 53-2 (2020) 5683–5688
A
modified hybrid Izhikevich neuron:
A
modified hybrid Izhikevich neuron:
A
modified
modeling, hybrid Izhikevich neuron:
synchronization, and
A
modified
modeling, hybrid Izhikevich
synchronization,
 neuron:
and
modeling,experiments
synchronization,
 and
modeling,experiments
synchronization,
 and
experiments 
experiments
Guillermo Ortega Ricardo Nunez Jonatan Pena Ramirez
Guillermo
Guillermo Ortega
Ortega Ricardo Ricardo Nunez Nunez Jonatan Jonatan Pena Pena RamirezRamirez
Guillermo
Guillermo Ortega Ortega Ricardo Ricardo Nunez Nunez Jonatan Jonatan Pena Pena RamirezRamirez
Center forOrtega
Guillermo ScientificRicardo ResearchNunez and Higher JonatanEducation Penaat Ramirez Ensenada
Center for
Center
(CICESE), forDepartment
Scientific Research
Scientific Research
of and Higher
and
Electronics Higher
and Education at
Education
Telecommunications at Ensenada
Ensenada Carr.
Center
Center for
(CICESE), Scientific
forDepartment
Scientific Research
Research
of and
and Higher
Electronics Higher
and Education
Education
Telecommunications at
atB.C.,Ensenada
Ensenada Carr.
(CICESE),
Tijuana-Ensenada Department
Center forDepartment
(CICESE), 3918,
Scientific of Electronics
Zona
Research
of Playitas, and
22860
and Higher
Electronics and Telecommunications
Ensenada,
Education
Telecommunications atB.C.,Ensenada Carr.
Mexico
Carr.
(CICESE),
Tijuana-Ensenada
Tijuana-Ensenada
(email:Department
3918,
3918, of Electronics
Zona
Zona
ortegamg@cicese.edu.mx; Playitas,
Playitas, and
22860
22860 Telecommunications
Ensenada,
Ensenada,
rnunez@cicese.mx) B.C., Carr.
Mexico
Mexico
(CICESE),
Tijuana-Ensenada
Tijuana-Ensenada
(email:Department
3918,
3918, of Electronics
Zona
Zona
ortegamg@cicese.edu.mx; Playitas,
Playitas, and
22860
22860 Telecommunications
Ensenada,
Ensenada,
rnunez@cicese.mx) B.C.,
B.C., Carr.
Mexico
Mexico
(email:
CONACYT-CICESE.
Tijuana-Ensenada
(email: ortegamg@cicese.edu.mx;
3918, Applied
Zona
ortegamg@cicese.edu.mx; Physics
Playitas, rnunez@cicese.mx)
22860 Division.
Ensenada,
rnunez@cicese.mx) DepartmentB.C., of
Mexico
(email:
CONACYT-CICESE.
CONACYT-CICESE.
Electronics and ortegamg@cicese.edu.mx;
Telecommunications,Applied
Applied Physics
Physics
Carr. rnunez@cicese.mx)
Division.
Division. Department
Department
Ensenada-Tijuana of
of
3918,
(email:
CONACYT-CICESE.
CONACYT-CICESE.
Electronics and ortegamg@cicese.edu.mx;
Telecommunications,Applied
Applied Physics
Physics
Carr. rnunez@cicese.mx)
Division.
Division. Department
Department
Ensenada-Tijuana of
of
3918,
Electronics Zona and
CONACYT-CICESE.
Electronics and Telecommunications,
Playitas, C.P.
Telecommunications, 22860,
Applied Carr.
Ensenada,
Physics
Carr. Ensenada-Tijuana
Division.B.C. Mexico
Department
Ensenada-Tijuana 3918,
of
3918,
Electronics Zona
Zona and Telecommunications,
Playitas,
Playitas, C.P.
C.P.
(email: 22860,
22860, Carr.
Ensenada,
Ensenada,
jpena@cicese.mx) Ensenada-Tijuana
B.C.
B.C. Mexico
Mexico 3918,
Electronics Zona and Telecommunications,
Playitas,
Zona Playitas, C.P. 22860,
C.P. 22860,
(email: Carr.
Ensenada, Ensenada-Tijuana
Ensenada, B.C. Mexico
jpena@cicese.mx) B.C. Mexico 3918,
Zona Playitas, (email: jpena@cicese.mx)
C.P. 22860,
(email: Ensenada, B.C. Mexico
jpena@cicese.mx)
(email: jpena@cicese.mx)
(email: jpena@cicese.mx)
Abstract: The original Izhikevich neuronal model is described by a nonlinear mathematical
Abstract:
Abstract:
model with aThe The original
original Izhikevich
Izhikevich neuronal
neuronal model
the factmodel is described
is described by a nonlinear
by nonlinear mathematical
Abstract:
Abstract:
model with The
aThe
static
static
reset map.
original
original
reset map.
Due to
Izhikevich
Izhikevich
Due analog neuronal
neuronal
to the fact
fact
that the
model
model is reset
is described
described
thatConsequently,
the reset by aaa instantaneously,
is applied
by
is applied
applied nonlinear mathematical
nonlinear
instantaneously,
it is not
mathematical
mathematical
it is
is not
not
model
easy
model to
Abstract: with
with a
implement
aThestatic reset
this
original
static reset map.
model Due
with
Izhikevich
map. Due to the
neuronal
to the fact that
circuits.
model
that the
theis reset
described
reset is
is this
by
applied a instantaneously,
paper
nonlinear presents
instantaneously, it
a modified
mathematical
it
model
easy
easy towith
to
Izhikevich a staticthis
implement
implement
neuronal reset
this
model, map.
model
modelin Due analog
with
with
which to
thethestatic
analog fact thatConsequently,
circuits.
circuits.
and the reset is applied
Consequently,
instantaneous this paper
this
reset instantaneously,
paper
is presents
presents
replaced by it is
is not
a modified
a modified
dynamic not
model
easy
easy towith a staticthis
implement reset map.
model Due analog
in with to thestatic
fact thatConsequently,
circuits. the reset is applied this
this paper instantaneously,
presents it is not
by aaa modified
reset,to
Izhikevich
Izhikevich
easy to
Izhikevich
implement
which neuronal
neuronal
implement
this
model,
model,
is physically
neuronal this
model,
modelin with
which
which
implementable.
modelin with
which
analog
the
the
analog
the
circuits.
static and
and Consequently,
instantaneous
instantaneous
Furthermore,
circuits.
static and Consequently,
instantaneousthe resultingreset
reset
this
reset
paper
is
is
system
paper
is
presents
replaced
replaced by
is modeled
presents
replaced by a
modified
dynamic
dynamicas a
modified
dynamic
Izhikevich
reset,
reset,
hybrid which
which neuronal
system is with
is model,
physically
physically
two in which
implementable.
implementable.
discrete the
modes. static and instantaneous
Furthermore,the
Furthermore,
Additionally, theoccurrence
the reset
resulting system
resulting is replaced
system
of by
is modeled
is a dynamic
modeled as
synchronization as
in aaa
Izhikevich
reset,
reset,
hybrid which
which neuronal
systemis is with model,
physically
physically
with in which
implementable.
implementable.
two discrete
discrete the
modes. static and instantaneous
Furthermore,
Furthermore,
Additionally, thethe
the reset
resulting
resulting
occurrence is replaced
system
system is by a
modeled dynamic
modeledofas
isstability
of synchronization
synchronization as
intheaa
hybrid
pair
reset,
hybrid of system
modified
which
system is with two
Izhikevich
physically
two neurons
implementable.
discrete modes.
is
modes. Additionally,
investigated and
Furthermore,
Additionally, the
a
thethe occurrence
comment
resulting
occurrence on the of
system
of local isstability
modeledofas
synchronization in
in aa
hybrid
pair
pair of
of
synchronous system
modified
modified with two
Izhikevich
Izhikevich
solution discrete
neurons
neurons
is given. modes.
Ultimately, is
is Additionally,
investigated
investigated and
and
the performance the
a occurrence
comment
aa comment
of on
on
the modified the
the of synchronization
local
local stability
Izhikevich of
model inthe
the
is
hybrid
pair
pair of system
modified
of modified
synchronous with two
Izhikevich
Izhikevich
solution discrete
neurons
neurons
is given.
given. modes.
Ultimately, is Additionally,
investigated
is investigated and the
and a comment
the performance
performance occurrence
comment
of the on the of
on the local
the modified
modified synchronization
local stability
stability
Izhikevich of
model inthe
of the isa
synchronous
experimentally
pair of
synchronous modified solution
validated
solution is
Izhikevich
is using
given. Ultimately,
electronic
neurons
Ultimately, is the
circuits.
investigated
the and
performance a of
comment
of the on
modified the Izhikevich
local stability
Izhikevich model
of
model is
the
is
synchronous solution
experimentally validated
experimentally is given.
validated using Ultimately,
using electronic
electronic circuits. the performance
circuits. of the modified Izhikevich model is
Copyright © 2020
synchronous
experimentally
experimentally The Authors.
solution
validated
validatedis given.
using
usingThis
Ultimately,
electronic
electronic access
is an opencircuits.
the article underofthe
performance
circuits. the BY-NC-ND
CCmodified license model is
Izhikevich
(http://creativecommons.org/
Keywords:
experimentally Izhikevich
validatedmodel, licenses/by-nc-nd/4.0)
usinghybrid
electronic system, dynamic reset, neuronal dynamics, electronic
circuits.
Keywords:
Keywords:
neuron Izhikevich model,
Izhikevich model, hybrid
hybrid system, system, dynamic
dynamic reset, reset, neuronal
neuronal dynamics,dynamics, electronicelectronic
Keywords:
Keywords:
neuron Izhikevich
Izhikevich model,
model, hybrid
hybrid system, system, dynamic
dynamic reset, reset, neuronal
neuronal dynamics,dynamics, electronicelectronic
neuron
Keywords:
neuron Izhikevich model, hybrid system, dynamic reset, neuronal dynamics, electronic
neuron
neuron1. INTRODUCTION Note that a physical implementation with electronic cir-
1. INTRODUCTION
1. INTRODUCTION Note
Note
cuits that
may aaabephysical
that physical
desirable, implementation
specially for with
implementation with electronic
electronic
the cases of largecir-
cir-
1. INTRODUCTION
1. INTRODUCTION Note
Note
cuits that
that
may a
be physical
physical
desirable, implementation
implementation
specially for with
with
the electronic
electronic
cases of cir-
cir-
large
There exist several 1. mathematical
INTRODUCTION models aiming to de- Note cuits
arrays
cuits may
of
that
may be
coupled
a
be desirable,
physicalneurons,
desirable, specially
where
implementation
specially a for
for the
physical
with
the cases
electronic
cases of
implementa-
of large
cir-
large
There existdynamic
several mathematical
mathematical models aiming to cell. cuits
arrays
arrays
tion
de- cuits may
ismay of
of
much be
coupled
coupled desirable,
faster neurons,
neurons,
than the specially
where
where
computer a for
aafor the
physical
physical cases
simulations, of
implementa-
implementa- large
seelarge
e.g.
There
scribe exist
the several behavior observedmodels
in aaiming
neuron to de- arrays
arrays
tion is of
of
is much be
coupled
coupled
much faster desirable,
faster neurons,
neurons,
than the specially
where
where
the computer
computer a the
physical
physical cases of
implementa-
implementa-
simulations, see see e.g.
e.g.
There
There
scribe existdynamic
exist
the several mathematical
several mathematical
behavior observedmodels
models
in aiming
aiming to cell.
to de- (Wagemakers
de- tion than
and neurons,
Sanjuán, 2013). a physicalsimulations,
scribe
There
scribe the
existdynamic
These models
the several
dynamic behavior
classified observed
may bemathematical
behavior according
observedmodelsto aaatheir
in
in
neuron
neuron
aiming
neuron de- arrays
to cell.
complex-
cell. tion
tion is
(Wagemakers of coupled
is much
much faster
faster
and than
than the
Sanjuán, the where
computer
computer
2013). simulations,
simulations, implementa-
see
see e.g.
e.g.
scribe
These
These the
models
models
ity or accordingdynamic
may
may be
to be
the behavior
classified
classified observed
according
according
type of behavior(s) in
to
to atheir
their
they neuron cell.
complex-
complex-
can repro- (Wagemakers
tion ispaper
muchpresents and
faster Sanjuán,
than 2013).
the computer simulations, see e.g.
scribe
These the
models dynamic
may be behavior
classified observed
according in
to atheirneuron cell.
complex- (Wagemakers
This
(Wagemakers and
and Sanjuán,
a modified
Sanjuán, 2013).
Izhikevich
2013). model, in which
These
ity
ity
duce.or models
or On
according
according
the mayto be
to
other the
the
hand,classified
type
typesomeof according
of behavior(s)
behavior(s)
of these tothey
models their
they complex-
can
can
are repro- (Wagemakers
repro-
biologi- Thisoriginal
paper presents
presents a modified
and Sanjuán, modified Izhikevich
2013). model, in which
which
These
ity
ity models
or On
or according
according may be
to the
the classified
type of according
of behavior(s) to their
they complex-
can repro- This
the paper static a
reset map Izhikevich
is replaced model,
by a in
dynamic
duce.
duce.
cally
ity
duce. On the
orinspired,
according
On asto
the other
the other to
other
hand,
hand,
thethe
hand,
type
type
some
some
Hodking-Huxley
some
behavior(s)
of thesemodel
of
of these
of behavior(s)
they
models
models
thesemodel they
models
canbiologi-
are
are repro-
and This
canbiologi-
(Hodgkin
are repro-
biologi-
This
the
the
reset.
paper
paper
original
presents
This presents
original static
static
modification
aa modified
modified
reset
reset map
map
allows
Izhikevich
Izhikevich
is
isa replaced
replaced
relatively
model,
model,
by
by
easy a
a
in
in which
which
dynamic
dynamic
physical
duce.
cally
cally
Huxley, On the
inspired,
inspired,
1990), other
as
as the
thehand,
whereas some
Hodking-Huxley
Hodking-Huxley
others of
are these models
model
computational are
(Hodgkin
(Hodgkin and This
biologi-
and
efficient, the
the
reset. paper
original
original
This presents
static
static
modification a modified
reset
reset map
map
allows Izhikevich
is
isa replaced
replaced
relatively model,
by
by
easy a
a in which
dynamic
dynamic
physical
duce.
cally
cally On1990),
the other
inspired,
inspired, as
as thehand,
the some are
Hodking-Huxley
Hodking-Huxley of these models
model
model areefficient,
(Hodgkin
(Hodgkin biologi-
and
and reset.
the
reset. This
implementation
original
This modification
of the
static
modification reset allows
Izhikevich
map
allows isaa model.
relatively
replaced
relatively easy
by
easy a physical
dynamic
physical
Huxley, whereas others computational reset. This modification allows a model. relatively easy physical
Huxley,
Huxley,
Huxley,
1990),
as theinspired,
cally 1990),as
1990),
whereas
integrate-and-fire others
models
the Hodking-Huxley
whereas
whereas others
others
are
are
computational
are(Jolivet et al.,
model
computational
computational
2004).
(Hodgkin and implementation
efficient,
efficient,
efficient, implementation
reset. This modification
implementation
of
of
of
the
the Izhikevich
Izhikevich
allows aismodel.relatively easymodeling
physical
as the
as the integrate-and-fire
Huxley,
integrate-and-fire
1990), whereas
models
models
others
(Jolivet
(Jolivet
are
et al.,
et
computational
al., 2004).
2004).
efficient,
The theory
implementation of the
of hybrid the Izhikevich
systems
Izhikevich model.
used in the
model.
as the integrate-and-fire
Among
as the integrate-and-fire
the existing models, models
models (Jolivet
the(Jolivet et al.,
Hodking-Huxley
et al., 2004).
2004). The
model implementation theory
The atheory
and designof of hybrid
of the
hybrid
procedure systems
Izhikevich
systems
for tuning is
is used
model.
used in
in
the dynamic the modeling
the modeling
coupling
Among
as
Among
Among
the existing
the
is considered
the
existing
the integrate-and-fire
as one of
existing
models,
models
models,
the
models,
the(Jolivet
the
most
the
Hodking-Huxley
et al.,
Hodking-Huxley
complete
Hodking-Huxley models 2004).in the The
model
model
model
The
and
and
is aatheory
theory
provided. design
design
of hybrid
hybrid systems
ofprocedure
procedure
Additionally, systems
for
forthetuning
tuning
is
is used
modified used
the
the
in
in the
dynamic
dynamicthe modeling
Izhikevich modeling
coupling
coupling
neuron
Among
is theit existing
considered as one models,
oneexplaining,
of the most the
most Hodking-Huxley
complete models model
in the and The
and
is theory
aa design
provided. of hybrid
designAdditionally,
procedure for
procedure systems
forthetuning
tuning is
modified used in
the dynamic
the dynamicthe
Izhikevich modeling
coupling
coupling
neuron
is considered
sense
Among
is that
the
considered as
allows
existing
as one of the
models,
of the in
the
most complete
a qualitative
Hodking-Huxley
complete models
models and in
model
in the
quan-
the is provided.
implemented Additionally,
with electronicthe modified
circuits Izhikevich
and a neuron
controller
is considered
sense that
that as
it allowsone
allows of the
explaining, most complete
in ofa qualitative
qualitative models and in the
quan- is and
is a
provided.
provided. design
implemented procedure
Additionally,
Additionally,
with for
electronicthetuning
the modified
modified the
circuits dynamic
Izhikevich
Izhikevich
and aa in coupling
neuron
neuron
controller
sense
titative
is considered
sense that it
manner, as
it allows the
one
allows explaining,
ofgeneration
the
explaining, mostin a the
complete
in of action
models and quan-
potential
in the is implemented
designed to with
induce electronic
synchronous circuits and
behavior controller
a pair
sense
titative
titative
in the that it
manner,
manner,
squid gianttheexplaining,
the generation
generation
axon. in
However, aa qualitative
of
qualitative
the
the
due action
action
the
and
and quan- is provided.
quan-
potential
potential
complexity is implemented
implemented
designed Additionally,
to with
with
induce electronic
electronicthe modified
synchronous circuits
circuits Izhikevich
and
and
behavior aa in neuron
controller
controller
aa pair
sense
titative
titativethat it
manner,
manner, allowsthe explaining,
theaxon.generation
generation in a
of
of qualitative
the
the action
action and quan-
potential
potential is
of
is designed
neurons.
implemented
designed to
to induce
Ultimately,
with electronic
induce synchronous
the proposed
synchronous behavior
circuits modified
and a in
behavior in aa pair
model
controller
pairis
in
in
of the squid
the
this squid
model, giant
giant axon.
simplified However,
However,
models whichduemainly
due the complexity
the complexity
focus on is
of designed
neurons. to induce
Ultimately, synchronous
the proposed behavior
modified inmodel pairis
titative
in
in the
the manner,
squid
squid giantthe generation
axon. However, of the
due action
the potential
complexity of
is neurons.
validated
designed Ultimately,
numerically
to induce and the proposed
experimentally.
synchronous modified
behavior inmodel
model is
a pair
of
of
the this
this model,giant
model,
membrane axon.have
simplified
simplified
potential, However,
models
modelsbeen which
whichduemainly
the complexity
mainly
developed. focus
focus
This is on of
on
the of neurons.
neurons.
validated Ultimately,
Ultimately,
numerically and the
the proposed modified
proposed
experimentally. modified model is
is
in
of the
this squid giant
model, potential, axon.
simplified have However,
models whichdue the
mainlyThis complexity
focus validated
on validated
of neurons. numerically
Ultimately, and the experimentally.
proposed
of
thethis
the
case model,
membrane
membrane
of the simplified
potential,
Izhikevich models
have
model, been
been which mainly
developed.
developed.
(Izhikevich, 2003), focus
This is
whichon
is the
the The
validated outline numerically
of this paper
numerically and
and is as follows. modified
experimentally.
experimentally. First, themodel original is
of
the
the
casethis model,
membrane
membrane
of the simplified
potential,
potential,
Izhikevich models
have
have
model, been
been which mainly
developed.
developed.
(Izhikevich, 2003), focus
This
This is
whichon validated
is the
the The outline of this
numerically paper
and is as follows.
experimentally. First, the original
case of The
Izhikevichoutline
outlinemodel of this
thisis paper is as
brieflyisdescribed follows.
follows.in First,
Section the
the2.original
Then,
the
case
case
is of the
of the Izhikevich
is biologically
membrane
the
biologically
inspired
potential,
Izhikevich
Izhikevich
inspired
model,
and have
model,
model,
and
(Izhikevich,
computationally
been developed.
(Izhikevich,
(Izhikevich,
computationally
2003),
2003),
2003),This which
efficient.
efficient. isThe
which
The
The outline
the Izhikevich
which The
Izhikevich
the problem
of this
of
model
model is
is
statement
paper
paper
briefly
briefly is is
as follows.
as
described
described
given in
in
in Section
First,
First,
Section
Section 3.the
original
original
2.
2.
Next, Then,
Then,
the
is biologically
Izhikevich
case
is of the
biologicallymodelinspired
is
Izhikevich
inspired a and computationally
nonlinear
model,
and model with
(Izhikevich,
computationally aefficient.
static
2003),
efficient. The
reset
which
The The
Izhikevich
Izhikevichoutline of
model
model this
is
is paper
briefly
briefly is as follows.
described
described in
in First,
Section
Section the original
2.
2. Then,
Then,
is biologically
Izhikevich modelinspired
is a and computationally
nonlinear model with aefficient.
static The
reset the
the
design problem
problemof the statement
statement
proposed is
is given
given
dynamic in
in Section
Section
reset and 3.
3.
the Next,
Next,
resultingthe
the
Izhikevich
map.
is Whenmodel
biologically
Izhikevich the
model is
inspired
is aa nonlinear
variable and describingmodel thewith
computationally
nonlinear model membrane
with aaefficient.
static
static reset
poten-
The
reset Izhikevich
the
the problem
problem model is briefly
statement
statement is
is described
given
given in
in inand
Section
Section
Section 3.
3. 2. Then,
Next,
Next, the
the
Izhikevich
map.reaches model
Whenathe the is a
variable nonlinear
describingmodel the with
membrane a static reset
poten- design
design
Izhikevich of
of the
the
model proposed
proposed
are dynamic
dynamic
presented in reset
reset
Section and4, the
the
and resulting
resulting
Section 5
map.
tial When
Izhikevich
map. When model
the variable
certain
is a
variableupper describing
threshold
nonlinear
describingmodel the
the membrane
value
with the
membrane a state
static poten-
of the
reset
poten- the
design
design problemof
of the
the statement
proposed
proposed is given
dynamic
dynamic in Section
reset
reset and
and 3.
the
the Next,
resulting
resultingthe
map.
tial When
reaches a the variable
certain upper describing
threshold the membrane
value the state poten-
of the Izhikevich
Izhikevich model
model are
are presented
presented in
in Section
Section 4,
4, and
and Section
Section 5
tial
map.
tial
tial
reaches
system is
When
reaches
reaches
a
reset certain
athe
a variable
certain
certain
upper
instantaneously
upper
upper
threshold
to
describing
threshold
threshold
a value
fixed
the the
value.
membrane
value
value the state
the
state
state of
Although the
poten-
of the
of
presents
design
Izhikevich
Izhikevich
the presents ofthe
thethe design
model
model proposed
design are
of a
areofpresented simple
dynamic
presented
aa simple
controller
reset
in Section
in Section to
controller andto the Section 555aa
synchronize
4, synchronize
4, resulting
and Section
and
system
system is
is reset
reset instantaneously
instantaneously to
to a fixed value. Although presents
pair of the
modified design of
Izhikevich simple
models. controller
After to
that, synchronize
the physical 5aaa
the reset
tial
system
the resetis map
reaches
system is reset
map
can
a certain
reset beupper
instantaneously
instantaneously
can be easily to a
easily threshold
to a fixed
implementable
a valuevalue.
fixed
fixed
implementable
in software,
the
value.
value.
in
Although
state
Although
software,
its presents
of the
Although its
Izhikevich
presents
pair
pair of
of
implementation
the
themodel
modified
modified
design
design areof ofpresented
Izhikevich
Izhikevich
with
aa simple
simple
electronic models.
models.
in Section
controller
controller
After
After
circuits
4, synchronize
to
to
that,
that,
of
and
the the
the
Section
synchronizephysical
physical
system and
the reset
physical
system
the map can
resetisimplementation
resetcan
map be easily
instantaneously
can be easily
easily implementable
with standard
to a fixedanalog
implementable in
value. software,
circuits
in software,Although
software, its
is presents
its pair
pair of the design
modified
of experiments
modified ofelectronic
Izhikevich
Izhikevich a simple
models.
models. controller
After
Afterof tothe
that,
that, synchronize
the
the physical
physical a
the reset
physical
physical
not map
implementation
implementation
an easy task, asbe mentioned implementable
with
with standard
standard
in (Kormaz in
analog
analog circuits
circuits
etsoftware,
al., 2016). is implementation
its
is implementation
some
pair of modified
with
with onelectronic
Izhikevich models.
circuits
circuits
synchronization After ofarethe
that,
system
system
provided
the
and
and
physical in
the reset
physical map can
implementation be easily implementable
with standard in
analog circuits its
is implementation
implementation
some experiments with
with electronic
electronic circuits
circuits of
of the
the system
system and
and
physical
not an easy
not implementation
easy task, as mentioned
mentionedwith standard
in (Kormaz analog et al., circuits
2016).is Section
some
some experiments6. Finally, aonon synchronization
synchronization
discussion and some are
ofare
provided
provided
conclusions in
in
are
not an
physical
not an easy task,
task, as
implementation
an easy task, as mentioned
as mentioned in
in (Kormaz
with standard
in (Kormaz
(Kormaz analog et
et al.,
et al.,
al.,
2016).
2016).is implementation
circuits
2016). some
Section
Section
provided
experiments
experiments
6.
6.in Finally,
Finally,
with
Section
on
aaon
7.
electronic
discussion
discussion
circuits
synchronization
synchronizationand
and some
some
are
arethe system and
provided
provided
conclusions
conclusions
in
in
are
are
not an easy task, as mentioned in (Kormaz et al., 2016). some
Section
Section
provided experiments
6.
6.in Finally,
Finally,
Section aaon
7. synchronization
discussion
discussion and
and some
some are provided are
conclusions
conclusions in
are
 Partially supported by CONACYT Mexico provided
Section
provided 6.in
in Section
Finally,
Section a7.
7. discussion and some conclusions are
 provided in Section 7.
 Partially supported by CONACYT Mexico provided in Section 7.
 Partially supported by CONACYT Mexico
 Partially
Partially supported
supported by
by CONACYT
CONACYT MexicoMexico
 Partially supported by CONACYT Mexico
2405-8963 Copyright © 2020 The Authors. This is an open access article under the CC BY-NC-ND license.
Peer review under responsibility of International Federation of Automatic Control.
10.1016/j.ifacol.2020.12.1592
5684 Guillermo Ortega et al. / IFAC PapersOnLine 53-2 (2020) 5683–5688
2. IZHIKEVICH MODEL
The Izhikevich model is described by the following set of
equations (Izhikevich, 2003)
ẋ1 = 0.04x21 + 5x1 + 140 − x2 + I,
ẋ2 = a(bx1 − x2 ),
with the auxiliary after-spike resetting
x1 c potential x1 reaches the upper-threshold value r1 , the
if x1 r1 , then
x2 x2 + d,
where x1 R and x2 R represent the membrane poten-
tial of the neuron and the membrane recovery variable,
respectively. The parameter a describes the time scale
of x2 and parameter b describes the sensitivity of x2 to
the subthreshold fluctuations of x1 . On the other hand, I
denotes the synaptic currents or injected dc-currents.
Note that when the membrane potential x1 reaches the
upper threshold value r1 , the state of the system is reset
according to (3), where r1 = 30mV .
Depending on the value of the parameter values a, b,
c and d, different types of neuronal behavior can be
reproduced by model (1)-(3). For example, regular spiking,
bursting, fast spiking, low-threshold spiking, among others
(Izhikevich, 2004; Nobukawa et al., 2015)
3. PROBLEM STATEMENT
The reset function (3) of the Izhikevich model updates
instantaneously the value of the variables x1 and x2 in Eqs.
(1)-(2). Although this reset can be easily implemented in
numerical simulations, its physical implementation with
analog circuits turns out to be complicated, see e.g. (Kor-
maz et al., 2016) (?) (?).
Therefore, the problem addressed in this paper consists in
finding a suitable replacement for the reset function (3),
such that it can be implemented with analog circuits and
without affecting the dynamic behavior observed in the
original Izhikevich system (1)-(3).
To achieve this, the theory of modeling hybrid systems is
used.
4. MODIFIED IZHIKEVICH MODEL
In this section, a modified Izhikevich system is presented.
As a first step, the reset function (3) is replaced by the
following dynamic reset

ẋ1 = −γ(x1 − c),
if x1 r1 , then ẋ2 = β, (4)
ż = 1.
Note that in the dynamic reset we have included a time
variable z which play the key role of determining the
amount of time that the system trajectories spend in
the dynamic reset mode (4). Specifically, while z < r2
for certain suitably chosen r2 > 0, the system evolves
according to (4) . After that, at z = r2 , the system ‘jumps
back’ to Eqs (1)-(2).
Consequently, the modified Izhikevich system can be mod-
eled by the hybrid automaton given in Figure 1. From
the automaton it is clear that the system operates in
two discrete modes: the mode q1 , in which the dynamics
are exactly the dynamics (1)-(2) of the original Izhikevich
(1) model and mode q2 , which is the proposed dynamic reset
(4).
(2)
Initially, the system starts in mode q1 and evolves in this
mode according to (1)-(2). Then, when the membrane
(3) system jumps to mode q2 , i.e. to the dynamic reset.
In mode q2 besides the equations describing the time
evolution of x1 and x2 , there is an additional variable,
namely z, which is a time variable. Then, the values of γ
and β in the dynamic equations describing x1 and x2 in
mode q2 are chosen such that when z = r2 , it holds that
x1 c + δ, δ << 0 and x2 = x2 (0) + d.
Note that, we are considering the practical case where
x1 c + δ and not the ideal case where x1 = c + δ. This is
due to the fact that the equality only holds when t .
However, the system remains in the reset mode q2 for a
finite and short period of time.
Fig. 1. Hybrid automaton for the modified Izhikevich
neuron
4.1 How to determine the values of γ and β in the dynamic
reset (4)
This part provides an algorithm for computing the values
of γ, β, and r2 in the dynamic reset (4).
First, from the first equation in (4) it follows that
x1 (t) = (x1 (0) − c)e−γt + c, (5)
where x1 (0) = r1 . Ideally, it is desired that x1 (t) = c but
this only occurs in the limit when t . Instead, the
practical value of interest of x1 (t) is x1 (t) = δ + c. In
other words, we are interested in the value of x1 when the
transient part of (5) satisfies
(x1 (0) − c)e−γt = δ, (6)
for certain δ << 1.
Then, the time at which (6) is satisfied is given by
 
1 δ
tδ := − ln . (7)
γ x1 (0) − c
Alternatively, the value of γ can be obtained from (6), i.e.
 Guillermo Ortega et al. / IFAC PapersOnLine 53-2 (2020) 5683–5688 5685

  Algorithm for tunning the parameters in the dynamic re-

1 δ
γ= ln . (8) set:
tδ x1 (0) c
Finally, note that the argument of the logarithmic function (1) Choose a transient time tδ << 1 and a δ << 1
in (7) is very small and consequently tδ > 0. (2) Find the value of γ from Eq. (8)
On the other hand, from the second equation in (4) it (3) Find the value of β from Eq. (13)
follows that
 t
x2 (t) = β dt + x2 (t0 ), (9) 4.2 Numerical results
0
= βt + x2 (0). (10) In this part, a numerical comparison between the original
Izhikevich model (1)-(3) and the modified model derived
From (3) it follows that the desired value of x2 (t) at time here described by Eqs. (1),(2) with dynamic reset (4),
tδ is is provided. Consequently, Eqs. (1)-(3) and Eqs. (1)-(2)
x2 (tδ ) = x2 (0) + d. (11) with dynamic reset (4) are numerically integrated by using
the following parameter values a = 0.002, b = 0.2, and
Hence, from Eq. (10) it follows that (11) is satisfied if I = 15. Two cases are considered, namely, the case of
regular spiking (c = 65 and d = 6) and bursting (c = 50
d and d = 2). For the former case the values of the dynamic
β= . (12) reset are tδ = 50µs and δ = 0.0043 and consequently
tδ
γ = 200000 and β = 12000. For the case of bursting the
By replacing tδ , given in (7) into (12) yields values of the dynamic reset are tδ = 50µs, δ = 0.0043,
γ = 1.9662e5 and β = 40000.
dγ The obtained results are shown in Figure 3, where the plots
β=  . (13) on the left correspond to the original Izhikevich model,
δ
ln x1 (0) c whereas the plots on the right correspond to the modified
Izhikevich model. Clearly, the obtained results are in good
Finally, it should be noted that the time the system spends agreement.
in mode q2 is tδ . Hence, the value of the threshold value
r2 , see Figure 1 is
 
1 δ
r2 = tδ = ln . (14)
γ x1 (0) c
This procedure is schematically depicted in Figure 2,
where the evolution of the system dynamics in mode q2 , i.e.
during the dynamic reset, is explained. The left plot shows
the time evolution of x1 , whereas the top and bottom
plots on the right depict the time evolution of x2 and z,
respectively.

x1 ≥ r1 x2(t2)=x2(t1)+d
Upper threshold value

d
x2(t1)

Fig. 3. Panels a) and c): time series obtained with the

original Izhikevich model. Panels b) and d): time
z≥r
2
series for the modified Izhikevich model. x-axis: time.
y-axis: x1 .

r2 5. SYNCHRONIZATION OF A PAIR OF MODIFIED

z(t1)=0
c+δ Lower
IZHIKEVICH MODELS
threshold value

mode q1 mode q
2
mode q
1
mode q1 mode q2 mode q1 This section presents the design of a controller for inducing
t0 t1 tδ t2 t0 t1 t t2
δ synchronization in a pair of modified Izhikevich models
coupled in a master-slave configuration.
Fig. 2. Time evolution of the system during the dynamic The master system is described by

reset (4). The time that the system spends in this ẋm = F (xm ),
mode is denoted by tδ , see Eq. (7). ΣM : (15)
ym = Cxm ,
5686 Guillermo Ortega et al. / IFAC PapersOnLine 53-2 (2020) 5683–5688

and the slave system Then, the value of k in the control input (19) should be

ẋs = F (xs ) + Bus , chosen as follows to ensure that A is Hurwitz
ΣS : (16)
ys = Cxs , k max 5 a, 5 b . (24)
where xi R2 and yi R, i = m, s are the state
and output, of master and slave systems, respectively, 6. EXPERIMENTAL RESULTS
and F (xi ) is a nonlinear function containing the intrinsic
dynamics of Izhikevich model, and is given by In this section a pair of modified Izhikevich models (1),(2)
  with dynamic reset (4) is implemented with analog cir-
0.04x2i1 + 5xi1 + 140 + I xi2
F (xi ) = , (17) cuits. Furthermore, the circuits are synchronized in a
a (bxi1 xi2 )
master-slave configuration, as given in (15)-(16), with in-
where a, b, and I are positive parameters and i = m, s. put and output vectors as given in (18) and using the
Furthermore, each system has a dynamic reset (4) and control input (19) .
 
1
B= , C = [1 0] . (18) The system (1)-(2) with dynamic reset (4) is scaled in
0
amplitude and time, introducing the following change of
variables t = τ η, x1 = αx̄1 , x2 = θx̄2 , z = Γz̄. Then, the
The static feedback control input us applied to the slave
scaled equations for the system Σi in (15)-(16) are given
system is
by
us = k(ym ys ) = kC(xm xs ), (19)  
1
where k R+ is the coupling strength. x̄i1 = 0.04αx̄2i1 + 5x̄i1 + (140 + I θx̄i2 + ui ) η,
α  α
The synchronization problem reduces to find a suitable x̄i2 = a bx̄i1 x̄i2 η,
value of coupling strength k, such that the master and θ
slave system synchronize, at least locally, such that the (25)
following is satisfied where denotes differentiation with respect to τ , and the
dynamic reset is
lim (xm (t) xs (t)) = 0. (20)  c
x 

 x̄˙ i1 = γ(x̄i1 )η,
r1 α
5.1 A comment on the local stability of the synchronous if x̄i1 , then x̄˙ = βη,1 (26)
solution α 
 i2
θ

z̄˙ = η/Γ,
The (global) stability analysis of the synchronous solution where i = m, s.
(20) in the coupled systems (15)-(16) is not a trivial
Likewise, the static feedback control input us , see (19), is
task, in part because of the hybrid nature of the system.
scaled as follows
However, under some assumptions it is possible to get some
insight into this issue. us = kα(xm1 xs1 ), (27)
Therefore, let assume that the master and slave modified and the parameter values are a = 0.002, b = 0.2, k = 5,
Izhikevich neurons start in mode q1 , see Figure 1, and and the scaling parameters are defined as α = 10, θ =
furthermore, let assume that the systems synchronize 100, Γ = 1, η = 1000.
before the reset takes place. The electronic implementation was constructed using the
Then, by defining the synchronization error as e = xm xs , scaled equations (25)(26) (27) and the corresponding
yields the following error dynamics for the coupled systems schematic diagram is given in Figure A.1. The constructed
(15)-(16), with function F defined in (17) experimental setup used to validate the numerical simula-
      2  tions is shown in Figure 4.
ė1 5 k 1 e1 x x21s
= + 0.04 1m (21)
ė2 ab a e2 0
     
A g(t,e)
Due to the reset, the trajectories of the master and slave
systems are always bounded and therefore, the term g(t, e)
can be considered as a perturbation ‘vanishing on e’,
satisfying

g(e) 2 γ e 2 , (22)
where γ is a nonnegative constant, and 2 represents the
Euclidean norm. Therefore, from the results presented in
(Khalil, 2002), it follows that a sufficient condition for the
local stability of the error dynamics (21), is that matrix
A should be Hurwitz. This condition can be easily verified
by looking at the characteristic polynomial of matrix A,
which is given by
det(λI A) = λ2 + (a + k 5)λ + a(b + k 5) = 0. (23) Fig. 4. Experimental setup at CICESE.
 Guillermo Ortega et al. / IFAC PapersOnLine 53-2 (2020) 5683–5688
The obtained experimental results are shown in Figures 5
to 7. The experimental data have been obtained by using
the TDS3012C Digital Oscilloscope from Tektronix.
Figure 5 shows the numerical simulations and the mea-
surements from the circuit, using different parameters.
Fig. 5. Comparison between numerical and experimental
results with parameters a = 0.02, b = 0.2, I = 15;
a) Numerical results for c = 50, d = 6; b) Circuit
measurements for c = 50, d = 6; c) Numerical
results for c = 50, d = 2; d) Circuit measurements
for c = 50, d = 2. x-axis: time. y-axis: x1
Figure 6 shows the synchronization of the master and slave
Izhikevich electronic neurons.
Fig. 6. Synchronized bursting behavior in the coupled
Izhikevich electronics neurons
The phase plane projection of the outputs of the systems
is shown in Figure 7. These experimental results confirm
that the modified Izhikevich model has been successfully
implemented with analog circuits.
5687
Fig. 7. Projection of the experimental solution of the
system onto the (x1 ,x2 ) plane
7. DISCUSSION AND CONCLUSIONS
A modified Izhikevich model has been presented, in which
the static reset function of the original Izhikevich model
has been replaced by an ad hoc designed dynamic reset.
The resulting system has been modelled as a hybrid
system with two discrete modes. It has been demonstrated
that the modified Izhikevich model presented here can be
physically implemented with analog circuits.
It is our hope that the derived model and circuit will
be useful in the study of neural dynamics, together with
other analog implementations of neuronal models, see e.g.
(Vromen et al., 2016; Savino and Formigli, 2009)
REFERENCES
Hodgkin, A.L. and Huxley, A.F. (1990). A quantitative
description of membrane current and its application
to conduction and excitation in nerve. Bulletin of
Mathematical Biology, 52(1), 25–71.
Izhikevich, E.M. (2003). Simple model of spiking neurons.
IEEE Transactions on Neural Networks, 14(6), 1569–
1572.
Izhikevich, E.M. (2004). Which model to use for cortical
spiking neurons? IEEE Transactions on Neural Net-
works, 15(5), 1063–1070.
Jolivet, R., Lewis, T.J., and Gerstner, W. (2004). Gen-
eralized integrate-and-fire models of neuronal activity
approximate spike trains of a detailed model to a high
degree of accuracy. Journal of Neurophysiology, 92(2),
959–976.
Khalil, H.K. (2002). Nonlinear systems; 3rd ed. Prentice-
Hall, Upper Saddle River, NJ.
Kormaz, N., Ozturk, I., and Kilic, R. (2016). Multiple
perspectives on the hardware implementations of biolog-
ical neuron models and programmable design aspects.
Turkish Journal of Electrical Engineering and Computer
Sciences, 24, 1729–1746.
Nobukawa, S., Nishimura, H., Yamanishi, T., and Liu,
J.Q. (2015). Analysis of chaotic resonance in Izhikevich
neuron model. PLOS ONE, 10(9), 1–22.
5688 Guillermo Ortega et al. / IFAC PapersOnLine 53-2 (2020) 5683–5688

Savino, G.V. and Formigli, C.M. (2009). Nonlinear elec- Appendix A. ELECTRONIC CIRCUIT
tronic circuit with neuron like bursting and spiking
dynamics. Biosystems, 97(1), 9 – 14.
Vromen, T.G.M., Steur, E., and Nijmeijer, H. (2016).
Training a network of electronic neurons for control of a
mobile robot. International Journal of Bifurcation and
Chaos, 26(12), 1650196.
Wagemakers, A. and Sanjuán, M.A. (2013). Electronic
circuit implementation of the chaotic Rulkov neuron
model. Journal of the Franklin Institute, 350(10), 2901
– 2910.

Fig. A.1. Schematic implementation of the modified Izhike-

vich model with analog circuits.