The above is an axiomatic, or black box, definition
because the internal composition of the device is irrele-
vant. Indeed, not only can memristors be made from differ-
ent materials, they have even been found in amoeba, squids
and plants, and numerous other living beings [5].
Once a device is identified from experimental
measurements to be a memristor, it is natural to develop
a mathematical model which can mimic approximately the
measured pinched hysteresis loops.
For pedagogical reasons, it is convenient to classify
a memristor according to the complexity of its mathemati-
cal representation into the 4 classes listed in Tab. 1, in the
order of decreasing complexity.
The Venn diagram in Fig. 3 shows the memristor
universe and the relationship among the 4 classes of mem-
ristors listed in Tab. 1.
The simplest class of memristors defined in the lowest
part of Tab. 1 is called an ideal memristor, which coincides
with the original definition postulated in [2]. Indeed, we
can recover its constitutive relation, within an arbitrary
constant (0),via
0 (0) ( ) ( ) ˆ q
R d qq q . (1)
Observe that differentiating both sides of (1) with
respect to time t gives
d d R
dt t q d
q (2)
or
v Rq i (3)
upon identifying d
v
dt
and dq
dt i , respectively.
Equation (1), which is equivalent to (2), is called the
constitutive relation of a charge-controlled memristor in
[2], [6], [7]. The dual constitutive relation of a flux-con-
trolled memristor
q q ˆ (4)
is equivalent to the voltage-controlled memristor
i G( ) v , (5)
d
v
dt
(6)
defined in the lower right corner of Tab. 1. The uninitiated
readers are referred to several primers on the ideal mem-
ristor in [8], [9], [10].
3. Ideal Memristors
To avoid clutter, we will present this section only for
the class of Voltage-Controlled Memristors. The dual case
for Current-Controlled Memristors follows the same de-
velopments, mutatis mutandis.
The ideal voltage-controlled memristor is defined by
a single scalar function G(), called the memductance.
Since the input in this case is a voltage source, the inde-
pendent variable is the flux , defined by
0 () ( ) 0 ( ) t t
t d v d v . (7)
Once the voltage source function v = v(t), and the ini-
tial flux φ(0) are given, the flux waveform φ(t) is calculated
via (7), which when substituted into the memductance G(φ)
gives the time-varying conductance G(φ(t)), and the corre-
sponding current waveform i(t) via the state-dependent
Ohm’s law i(t) = G(φ(t)) v(t). Plotting the loci of (v(t), i(t))
in the i vs. v plane with the time t as parameter, the result-
ing graph is in general multi-valued, and always passes
through the origin, hence the name pinched hysteresis loop.
As an example, consider an ideal flux-controlled memristor
described by the odd-symmetric piecewise-linear (PWL)
constitutive relation (shown in the left column of Fig. 4).
For future reference, the equation of the identical PWL
curve in Figs. 4(a) and 4(c) is given in (8).
q q ˆ 0.01 0.04 0.25 0.04 0.25 . (8)
Let us connect a voltage source v(t) = 1.2 sin t across
this memristor, and choose 2 different initial states
identified as point on the q vs. φ curve in Figs. 4(a),
(c), respectively. The corresponding pinched hysteresis
loop is shown in Figs. 4(b), (d), respectively2
.
Pinched Hysteresis Loops Depend on Initial States !
Two pinched hysteresis loops measured from an
ideal memristor driven by the same periodic voltage
source (resp., current source) but different initial states
can look very different from each other.
1 Our axiomatic definition requires a pinched hysteresis loop to be measured not
only for one input signal, but for all possible periodic input signals
with zero mean. In practice, only a finite number of measurements could be made.
Our definition did not require the same pinched hysteresis loop to be measured
whenever the same input signal is applied, because for non-volatile
memristors, the pinched hysteresis loop depends not only on the input waveform i(t)
or v(t), but also on the initial conditions of the relevant state variables,
such as Fig. 42 of [5], where two different pinched hysteresis loops are measured
for the same input current source i(t) = 10 sin (ωt), but different initial
states x(0) = –6.3 and x(0) = 6.3, respectively.
2
Readers puzzled by the dramatic difference between the 2 pinched hysteresis loops
in Fig. 4(b) and 4(d) calculated from the same memristor
constitutive relation q q ˆ( ) , and the same input v(t) = 1.2 sin t are
referred to the exact calculations and graphical illustrations in Fig. 28 and Fig.
27