A novel generic memristor, dubbed the 6-lobe Chua corsage memristor, is proposed with its nonlinear dynamical analysis and physical realization. The proposed corsage memristor contains four asymptotically stable equilibrium points on its complex and diversified dynamic routes which reveals a 4-state nonlinear memory device. The higher degree of versatility of its dynamic routes reveal that the proposed memristor has a variety of dynamic paths in response to different initial conditions and exhibits a highly nonlinear contiguous DC V-I curve. The DC V-I curve of the proposed memristor is endowed with an explicit analytical parametric representation. Moreover, the derived three formulas, exponential trajectories of state xnt, time period tfn, and minimum pulse amplitude VA, are required to analyze the movement of the state trajectories on the piecewise linear (PWL) dynamic route map (DRM) of the corsage memristor. These formulas are universal, that is, applicable to any PWL DRM curves for any DC or pulse input and with any number of segments. Nonlinear dynamics and circuit and system theoretic approach are employed to explain the asymptotic quad-stable behavior of the proposed corsage memristor and to design a novel real memristor emulator using off-the-shelf circuit components.
Ministry of Education2016R1A2B4015514Cooperative Research Program for Agriculture Science and Technology DevelopmentPJ0120642016Air Force Office of Scientific ResearchFA9550-18-1-0016Rural Development Administration1. Introduction
Memristor, the acronym of memory resistor, is one of the most propitious elements in the emerging memory sector due to its exclusive attributes under DC or AC excitations, as well as its miniature nanoscale physical dimension. Extensive research is ongoing on memristors and the memristive system after the seminal paper published by hp in 2008 [1]. Memristor, the fourth basic circuit element, was postulated by Chua [2] and later generalized to a broader class of dynamical devices which exhibit interesting and valuable circuit-theoretic properties [3].
Recently, several researchers investigated the multistate phenomena in generic and extended memristors [4–7]. This important research direction could lead to another stage of technical innovation in the memristor area. The principle of the multistate memristor can be explained using the nonlinear dynamics theory as well as circuit and system theoretic concepts [4–6]. For example, the locally active generic Chua corsage memristor exhibits an asymptotical stability via the supercritical Hopf bifurcation [8–9]. Once an initial state is set, the state alters following its nonlinear dynamic route. The state changing is repeated until the state reaches a particular state which is termed as an “attractor.” In this type of memristors, the state space contains various attractors and each attractor has its own basin of attraction [10]. When inputs or noises are applied at a stable equilibrium state of the generic corsage memristor, the equilibrium state is moved by the amount of time integral of the inputs. However, unless the state moves beyond the boundary of current basin of attraction, the state returns back to its original equilibrium state (attractor) [8, 9]. Therefore, it can become a robust memory device. Since a part of the previous programming history of the corsage memristor is lost in this procedure, the phenomenon is known as “local fading memory” in bistable and multistate memory devices [4].
Another feature of this type of multistate corsage memristor is the alteration of the stable equilibrium states. In this case, a sufficiently large amplitude with short pulse width or a minimum pulse amplitude with lengthy pulse width is applied across the memristor to switch the state from one stable equilibrium state to another stable state by converging into the basin of a new stable attractor. In this way, the equilibrium state of a multistate corsage memristor is changed to a new stable state where the resistances or conductances of each stable equilibrium state are distinguishably different from each other [7]. The alteration of the stable equilibrium states of corsage memristors is determined by the function of its input and initial condition, and henceforth, the corsage memristors exhibit multistability and eventually can be used as multistate memory devices.
In this paper, we demonstrate a novel quad-stable generic memristor, dubbed the 6-lobe Chua corsage memristor. The dynamic routes of the 6-lobe corsage memristor have four asymptotically stable equilibrium points and three unstable equilibrium points at the DC input voltage V = 0 V. The four asymptotically stable equilibrium points of the proposed memristor define the corresponding four distinct resistance levels and can be used to develop a multibit-per-cell memory device similar to the unidirectional spin Hall magnetoresistance [11]. The multistable memory states are distinguishable by resistance levels in accordance to stable equilibrium points where the memory states can be defined with a pair of bits. To ease the demonstration of the switching kinetics of multistable memory states of the proposed memristor, we derived three universal formulas regarding the exponential state xnt, the time period tfn, and the minimum pulse amplitude VA.
In addition to the theoretical insights, we have designed and built a real emulator circuit of the proposed corsage memristor. For the physical realization of the piecewise linear 6-lobe Chua corsage memristor, we use the Graetz bridge [12] circuit in parallel with an active and locally active resistor [5]. Concepts from circuit and system theory, and techniques from nonlinear dynamics theory, are employed in this paper to elucidate the key mechanisms underlying the emergence of switching strategies of quad-stable memory.
The rest of the paper is organized as follows: the 6-lobe corsage memristor is designed and introduced in Section 2. The parametric representation and DC V-I curve are analyzed in Section 3. The switching kinetics and the physical implementation of the proposed corsage memristor are described in Sections 4 and 5, respectively, followed by the concluding remarks in Section 6.
2. 6-Lobe Chua Corsage Memristor Model
The 6-lobe Chua corsage memristor is an extension of the 1st-order locally active Chua corsage memristor [8]. It is a piecewise linear (PWL) memristor whose state-dependent Ohm’s law and state equation are as follows:
(1)i=Gxv,where
(2)Gx=G0x2,and
(3)dxdt=fx+v,where
(4)fx=33−x+x−6−x−12+x−20−x−30+x−42−x−56,and x, i, and v denote the memristor state, current, and voltage, respectively. In practice, G0 is a scaling constant chosen to fit the intrinsic memductance scale of the memristor. In this paper, we choose G0=10−6 so that the current of the 6-lobe Chua corsage memristor can be measured in milliamperes (mA) [7].
2.1. Frequency-Dependent Pinched Hysteresis Loops
The frequency-dependent pinched hysteresis loops of a device, when driven by any periodic input current or voltage source with a zero DC component, are a signature of a memristor or memristive system [13]. The 6-lobe Chua corsage memristor defined in (1), (2), (3), and (4) exhibits frequency-dependent pinched hysteresis loops when it is driven by a sinusoidal input signal vt=Asinωt where A=5V, as shown in Figure 1. The input voltage vt and the corresponding memristor current it are shown in the upper-right side of Figure 1(a), and the memristor state xt and memductance Gt are shown in the lower-right side of Figure 1(a), whereas the left side of Figure 1(a) shows the memristive circuit diagram with AC excitation. The frequency-dependent pinched hysteresis loops are shown in Figure 1(b) for frequencies ω=1rad/s, 10rad/s, 20rad/s, and 100rad/s. The lobe area of the pinched hysteresis loops shrinks as the frequency increases and tends to a straight line for ω≥100rad/s as shown in Figure 1(b) [14]. It follows that the proposed corsage memristor is a generic memristor [15].
Frequency-dependent pinched hysteresis loop fingerprint of the 6-lobe Chua corsage memristor, calculated with initial state x(0) = 5: (a) the input voltage v(t) (in red), corresponding memristor current i(t) (in blue), memristor state x(t) (in magenta), and memconductance G(t) (in green) of the 6-lobe Chua corsage memristor; (b) frequency-dependent pinched hysteresis loops for zero-mean periodic input v(t) = Asin(ωt) where A = 5 V and frequency ω = 1 rad/s, 10 rad/s, 20 rad/s, and 100 rad/s.
2.2. Dynamic Routes with Their Phase Portrait
The dynamic route of a nonlinear system prescribes the dynamics of nonlinear differential equations [14]. The dynamic route of the short-circuited (v = 0 V), namely, the power-off-plot (POP), 6-lobe Chua corsage memristor is shown in Figure 2 where (5) is used to plot the loci of dx/dt∣v=0versusx.
(5)dxdtv=0=f^x=33−x+x−6−x−12+x−20−x−30+x−42−x−56.
Dynamic route map of the 6-lobe Chua corsage memristor at V = 0 V is called the power-off-plot (POP).
The arrowheads in Figure 2 indicate the direction of motion of the state variable x from any initial state x0.
Figure 2 shows that for any initial state x0 on the upper half of the POP, where dx/dt>0, the state variable xt must move to the right as xt increases with time, depicted by the purple arrowheads pointing to the right in Figure 2. On the contrary, for any initial state x0 on the lower half of the POP, where dx/dt<0, the state variable xt decreases with time and must move to the left, depicted by the black arrowheads pointing to the left in Figure 2. In the theory of nonlinear dynamics [16], the stationary points where dx/dt=0 or f^x intersects the x-axis; are known as equilibrium points. Figure 2 shows that f^x intersects the x-axis at seven points, namely, x=XQ1=3 (Q1), x=XQ2=9 (Q2), x=XQ3=15 (Q3), x=XQ4=25 (Q4), x=XQ5=35 (Q5), x=XQ6=49 (Q6), and x=XQ7=63 (Q7). The equilibrium points Q1, Q3, Q5, and Q7 are stable whereas Q2, Q4, and Q6 are unstable equilibrium points because the state variable xt diverges away from Q2, Q4, and Q6. Moreover, the equilibrium points Q1, Q3, Q5, and Q7 in Figure 2 are stable as the corresponding eigenvalues of those equilibrium points are negative real numbers whereas Q2, Q4, and Q6 are unstable as the eigenvalues are positive [17].
Figure 2 shows that for any initial state x0>XQ+δx, where XQ∈XQ2,XQ4,XQ6, the unstable equilibrium points Q2, Q4, and Q6 converge to stable equilibrium points Q3, Q5, and Q7, respectively, to their right as shown with purple arrowheads. In contrast, for any initial state x0<XQ−δx, Q2, Q4, and Q6 converge to stable equilibrium points Q1, Q3, and Q5, respectively, to their left as shown with black arrowheads.
The phase portrait of stable equilibrium states Q1, Q3, Q5, and Q7 is shown in Figure 3 where the dotted straight lines represent the separatrices between two stable equilibrium states and pass through the unstable equilibrium points Q2, Q4, and Q6, respectively. Similar to Figure 2, Figure 3 also shows that for any x0>XQ2orXQ4orXQ6, the trajectories of xt converge to Q3, Q5, and Q7, respectively, as shown with purple arrowheads. Conversely, for x0<XQ2orXQ4orXQ6, xt converges to Q1, Q3, and Q5, respectively, as shown with black arrowheads.
Phase portrait of the stable equilibrium states Q1, Q3, Q5, and Q7. The horizontal dotted lines are the separatrices and pass through the unstable equilibrium states Q2, Q4, and Q6, respectively.
The dynamic routes in Figure 2 and the phase portrait in Figure 3 illustrate that the proposed memristor can be used as a 4-state or multibit-per-cell (2-bit) memory device at v = 0 V.
The more stable equilibrium states of the 6-lobe corsage memristor increases the memory efficiency per device 50% and 25% compared to 2-lobe and 4-lobe corsage memristors, respectively, and eventually enhance the capability to represent a desired function more closely than 2-lobe or 4-lobe corsage memristors.
3. Parametric Representation and the DC V-I Curve
In mathematics, parametric representation of an object is a collection of parametric equations which are used to express the coordinates of the points that make up a geometric object [18] where those parametric equations are defined by a group of quantities based on a function of one or more independent variables [19].
3.1. Parametric Representation
The parametric representation of the 6-lobe Chua corsage memristor can be derived by equating state (3) to zero (dx/dt=0) and solving for the following equilibrium points (6) for each DC input voltage v=V, at the DC equilibrium state x=XV:
(6)dxdt=33−x+x−6−x−12+x−20−x−30+x−42−x−56+v=0.
The DC voltage of the proposed corsage memristor is given explicitly by
(7)V=−33−x+x−6−x−12+x−20−x−30+x−42−x−56≜v^X.
The parametric representation of the DC current of the 6-lobe corsage memristor can be derived by substituting V given by (7) for v in (1) with G0 = 10−6, namely,
(8)I=−G0X233−x+x−6−x−12+x−20−x−30+x−42−x−56≜i^X.
The parametric representations of the proposed corsage memristor are shown in Figure 4 where Figures 4(a) and 4(b) show the loci of the parametric representation of V=v^xversusX and I=i^xversusX, respectively. The loci of the parametrically represented V=v^xversusI=i^x are shown in Figure 4(c).
Parametric representations of the 6-lobe Chua corsage memristor (a) voltage V=v^X versus state variable X and (b) current I=i^X versus state variable X, for each value of χ=X:−12≤X≤78. (c) DC V-I plot, where the coordinates (V, I) of each point are extracted from (a) and (b), for each value of χ=X:−7≤X≤73 for better illustration.
For convenience of readers, several points of the parametric representation of V=v^x and I=i^x of the 6-lobe Chua corsage memristor over the range χ=X:−12≤X≤78 are listed in Table 1.
Numerical values of the 6-lobe corsage memristor obtained from parametric representation over −12 ≤ X ≤ 78.
X
V=v^X (V)
I=i^X (mA)
−12
−15
−2.16
0
−3
0
3
0
0
6
3
0.11
9
0
0
10
−1
−0.1
12
−3
−0.43
15
0
0
20
5
2
25
0
0
30
−5
−4.5
35
0
0
40
5
8
42
7
12.35
49
0
0
50
−1
−2.5
56
−7
−21.95
60
−3
−10.8
63
0
0
70
7
34.3
78
15
91.26
3.2. DC V-I Plot
A circuit-theoretic approach is used to derive the DC V-I loci of the voltage-controlled 6-lobe Chua corsage memristor. Each DC value of voltage V and current I is computed using the following steps:
For each value of V listed in Table 1, we calculate all equilibria x=Xk, 1≤k≤7, of the proposed memristor using state (3) where dx/dt=0(9)V=−33−x+x−6−x−12+x−20−x−30+x−42−x−56≜v^X.
Then we determine the DC current i=I of the memristor corresponding to each equilibrium point X=X1,X2,…,XN, 1≤N≤7:
(10)I=−G0X233−x+x−6−x−12+x−20−x−30+x−42−x−56≜i^X.
Finally, we draw the DC V-I curve by plotting the coordinates (V, I) on the V-I plane for each value of X.
The DC V-I loci of the 6-lobe corsage memristor is shown in Figure 5 over the input voltage range −10 V ≤ V ≤ 10 V where the solid curves correspond to stable equilibrium states and the dash curves correspond to unstable equilibrium states. Since the DC V-I curve contains six contiguous lobes, henceforth call it the “six lobe corsage V-I curve.” The seven different colored DC V-I branches in Figure 5 represent the equilibrium points of the corresponding colors in Figure 3. At v=0V, the state variables are x=3 (red DC V-I curve Q1), x=9 (fluorescent green DC V-I curve Q2), x=15 (blue DC V-I curve Q3), x=25 (magenta DC V-I curve Q4), x=35 (cyan DC V-I curve Q5), x=49 (brown DC V-I curve Q6), and x=63 (green DC V-I curve Q7). As the values of the state variable of each DC V-I branch at the origin are different, their slopes (i.e., conductances Gx) at the origin are also different according to (2), as tabulated in the upper-left inset of Figure 5. The tabulated upper-left inset shows that the red DC V-I curve represents the lower conductance state (higher resistance state) whereas the green DC V-I curve represents the higher conductance state (lower resistance state). Moreover, the lower-right inset of Figure 5 shows a zoomed portion of the red DC V-I curve over the range −5 V ≤ V ≤ 2 V. The zoomed red DC V-I curve contains a negative-slope region over the voltage range −3 V < V < −1 V which affirms that the proposed corsage memristor is locally active over the −3 V < V < −1 V range as ReZiω<0 for DC input voltage (ω=0) [8]. The locally active negative slope region of the 6-lobe Chua corsage memristor is significant in circuit theory as it might give rise to complexity through which complex phenomenon and information processing might emerge [20, 21].
DC V-I plot of the 6-lobe corsage memristor over input voltage −10 V ≤ V ≤ 10 V. The left inset shows the conductance values at V = 0 V. The right inset shows the zoomed portion of the red DC V-I curve over the voltage range −5 V ≤ V ≤ 2 V.
One of the most important features of the 6-lobe corsage memristor is the contiguousness of its DC V-I curve which is different from many other published nonlinear DC V-I curves which exhibit several disconnected branches [14].
Another impressive feature is that the parametric representation and the DC V-I curve of the proposed corsage memristor has an explicit analytical equation, which rarely happens.
4. Switching Strategies of Memory States
The power-off-plot in Figure 2 shows that the 6-lobe Chua corsage memristor can be used as a 4-state or 2-bit memory device at v=0V. Conceptually, the simplest way to switch the memory states of the 6-lobe corsage memristor is to apply a square pulse with an appropriate pulse amplitude VA and pulse width Δw. For a successful switching between the memory states of the proposed corsage memristor, the square pulse should have a minimum pulse width for an appropriate pulse amplitude, VA. Any square pulse with less than the minimum pulse width results in switching failure.
The switching kinetics of the 6-lobe Chua corsage memristor can be represented through its dynamic route map (DRM). The solution of each straight-line segment of the dynamic route map of our corsage memristor is an exponential function where the complete solution x(t) is made of a sequence of the exponential waveforms, joined at the various breakpoints in the dynamic routes. In this paper, we derived the following universal exponential state variable xnt formula related to a straight-line segment around an equilibrium point Qn of the piecewise linear DRM (detailed derivation of xnt, tfn, and VA are provided in the supplementary document (available here).):
(11)xnt=Qn−mvt1−emt−t0n−Qn−xt0nemt−t0n,where m represents the sign value of the straight-line slope,
(12)m=sgndx/dtstart−dx/dtendxstart−xend,and t0n is the initial time of the segment whereas xt0n represents the initial state at t0n. The universal formula of the time, tfn, required for the trajectory of xnt to move from any initial point xt0n to the end of the straight-line segment is also derived as follows:
(13)tfn=t0n+1mlnQn−mvt−xtfnQn−mvt−xt0n.⏟Δtfn
The appropriate pulse amplitude VA is computed by replacing t=tfn in (11) and substituting the value of tfn from (13) to (11) where the resultant equation is shown as follows:
(14)VA>Qn−1–xt0n−1,where Qn−1 and xt0n−1 represent, respectively, the immediate before equilibrium point and the initial state of the resultant memory state Qn.
The derived universal formulas in (11), (12), (13), and (14) are applicable for any piecewise linear DRM curve of any number of segments and any DC or pulse input vt. Such exponential analytical solutions can be derived from no nonlinear functions other than the PWL functions.
The dynamic route map (DRM) in Figure 6(a) shows an application of successful switching for an appropriate pulse amplitude VA and pulse width Δw where the 6-lobe corsage memristor switches from high-resistance (low conductance) state Q1 to low-resistance (high conductance) state Q5. To switch from Q1 to Q5, we choose the pulse amplitude VA=5.5V which satisfies the condition VA>Q4−xt04=5. To compute the appropriate pulse width Δw, we choose the final state (with input VA) xΔwt=26.5 for a square pulse vt by satisfying the condition xΔwt>Q4, as shown in Figure 6(a). For xΔwt≤Q4, the proposed corsage memristor fails to switch from memory state Q1 to Q5 and converges to memory state Q3. However, pulse width Δw is equal to the time required for the trajectories to move from x1(t01) to xΔwt and can be computed by summing the time needed for each individual straight-line segment to reach the terminal points and express as
(15)Δw=Δtf1+Δtf2+Δtf3+Δtf4.
Memory state switching kinetics of the 6-lobe Chua corsage memristor from memory state Q1 to Q5 for an input square pulse VA=5.5V and Δw=7.48s. (a) Dynamic routes of the switching kinetics of the 6-lobe Chua corsage memristor. The two magenta-color vertical line segments indicate an instantaneous jump between the red and the blue piecewise-linear plots in the dynamic route map. (b) Movement of the exponential trajectories of xt with respect to time t.
The total time period tf required to move from memory state Q1 to Q5 is expressed as
(16)tf=Δw+Δtf4wv+Δtf5,as shown in Figure 6. The sequence of exponential x(t) obtained from (11) is shown as follows:
(17)xt=x1t=Q1+vt1−e−t−t01−Q1−x1t01e−t−t01,t01≤t<tf1,x2t=Q2−vt1−et−t02−Q2−x2t02et−t02,t02≤t<tf2,x3t=Q3+vt1−e−t−t03−Q3−x3t03e−t−t03,t03≤t<tf3,x4t=Q4−vt1−et−t04−Q4−x4t04et−t04,t04≤t<tf4,x4wvt=Q4−vt1−et−t04wv−Q4−x4wvt04wvet−t04wv,t04wv≤t<tf4wv,x5t=Q5+vt1−e−t−t05−Q5−x5t05e−t−t05,t05≤t,and by inserting the initial states, equilibrium points, and time period for the trajectories to move from initial states to final states, the x(t) can be expressed as follows:
(18)xt=x1t=3+vt1−e−t−t01,t01=0≤t<tf1=0.778,x2t=9−vt1−et−t02−3et−t02,t02=tf1=0.778≤t<tf2=2.008,x3t=15+vt1−e−t−t03−3e−t−t03,t03=tf2=2.008≤t<tf3=4.841,x4t=25−vt1−et−t04−5et−t04,t04=tf3=4.841≤t<tf4=7.48,x4wvt=25−vt1−et−t04wv+1.5et−t04wv,t04wv=tf4=7.48≤t<tf4wv=8.684,x5t=35+vt1−e−t−t05−5e−t−t05,t05=tf4wv=8.684≤t.
The memory state switching from Q1 to Q5 in Figure 6(a) shows that the applied square pulse with VA = 5.5 V is equivalent to translating the red curve fx upwards by 5.5 units, as shown by the blue curve. The dynamic route starting from Q1 (x=3), at t=0−, would jump abruptly from Q1 on the red curve to a point directly above Q1 on the blue curve (yellow circle) at t=0+ (shown with the upward green arrow) as the pulse input increases from 0 V to 5.5 V. Since the blue curve is located above the x-axis (where dx/dt>0) over the range of interest, its motion can only move to the right, until time t=Δw. When the square pulse returns to zero at t=Δw, the point (shown with the green circle) on the blue curve reverts back abruptly to the point xΔwt=Δw=26.5 on the red curve (shown with the light cyan circle followed by the downward green arrowhead), whereupon the dynamics must continue to move along the dynamic route indicated by the black arrowheads, until it converges to the low-resistance memory state Q5 (x=35).
The exponential trajectories of the xt related with the individual piecewise linear segments are shown in Figure 6(b). Observe from Figure 6(b) that the total time period (tf) needed for the xt trajectories to reach Q5 from Q1 is the summation of all the time periods needed for an individual trajectory to propagate through the piecewise linear segments which is tf=Δw+Δtf4wv+Δtf5=17.2s.
To switch back from the low-resistance (high conductance) state Q5 to the high-resistance (low conductance) state Q1 of our corsage memristor, we simply applied a negative voltage pulse with amplitude VA = −5.5 V and pulse width Δwb=7.684s where Δwb is computed using (15). The dynamic route and the state trajectories xt of switching back kinetics from memory states Q5 to Q1 is shown in Figures 7(a) and 7(b), respectively. In Figure 7(a), at tb=Δwb, the state variable xΔwbtb=Δwb=8.9 and the slope dx/dt<0 at that linear segment for which the state variable xt must move to the right and eventually converge to the equilibrium memory state Q1 (x=3). The similar phenomenon with exponential trajectories of xt is shown in Figure 7(b) where the xt decreases as the time increases and converges to xtf1b=3 where xtf1b=3 is regarded as the Q1 memory state. To switch back from Q5 to Q1, the total time tfb=20.701s is needed as shown in Figure 7(b).
Switching kinetics from low-resistance state Q5 to high-resistance state Q1 for an input square pulse VA=−5.5V and Δwb=7.684s. (a) Dynamic routes of the switching-back kinetics of the proposed corsage memristor. (b) Movement of the exponential trajectories of xt with respect to time t.
The pulse amplitude VA and the pulse width Δw play a crucial role in the switching kinetics of the memory states of the 6-lobe Chua corsage memristor. An inappropriate pulse amplitude or pulse width may result in switching failures. To choose the appropriate pulse amplitude VA, we already provided (14) whereas we illustrate the inappropriate pulse width scenario in Figure 8. In Figure 8, we provide the same pulse amplitude VA=5.5V to switch from memory state Q1 to Q5 with a different pulse width Δw=7s. Observe from Figure 8(b) that the exponential trajectories are converging to memory state Q3 (x=15) rather than converging to memory state Q5 (x=35). The reason behind such switching failure is the pulse width as at t=Δw and the state variable xΔwt=Δw=23.812 which lies in the left-hand side of Q4 (x=25), as shown in Figure 8(a). According to Section 2.2, any point that lies in the left side of Q4 (x = 25) follows the dynamic route dx/dt<0 (as shown with the black arrowhead in Figure 2) and converges to equilibrium state Q3, and in this case, the state variable xt follows the same route dx/dt<0 and converges to Q3 (x = 15) as xΔw(t = Δw) < Q4.
Switching failures of the 6-lobe corsage memristor from Q1 to Q5 for a pulse amplitude VA=5.5V and pulse width Δw=7s. (a) Dynamic routes of the switching kinetics and (b) movement of the exponential trajectories of xt with respect to time t. The switching failure happens due to insufficient pulse width.
For convenience of the readers, we plotted the hyperbolic relationship between the pulse amplitude VA versus the pulse width Δw of the switching memory states between Q1 and Q5 of our 6-lobe corsage memristor as shown in Figure 9.
Relationship between the pulse amplitude VA and pulse width Δw for the switching kinetics of the 6-lobe corsage memristor.
5. Physical Realization of the 6-Lobe Chua Corsage Memristor
For physical realization of the 6-lobe Chua corsage memristor, we modified the circuit in Figure 1(a) with the switching kinetics closer to the behavioral attributes of our 7-segment PWL hypothetical memristor which is shown in Figure 10. The novel circuit consists of the cascade between a passive nonlinear-resistive two-port and a dynamic first-order one-port [5]. The passive nonlinear-resistive two-port is composed of parallel connected Graetz bridges [12] with opposite diode directions whereas the dynamic first-order one-port is made up of a C-R parallel circuit.
Circuit diagram of the real 6-lobe Chua corsage memristor emulator with quad-stable input dynamics.
The active and locally active resistor R0 in Figure 10 should exhibit the contiguous six breakpoints on its DC V-I curve similar to the 6-lobe Chua corsage memristor. To design the nonlinear resistor R0, we applied the circuit-theoretic analysis on an op-amp circuit [22] to obtain the desired DC V-I breakpoints at specific voltages. According to [22], the driving-point characteristic of a single positive and negative feedback op-amp circuit provides two breakpoints on its piecewise linear DC V-I curve. To obtain a six-breakpoint piecewise linear DC V-I curve, we combine three op-amp circuits in parallel as shown in Figure 11. The current i coming out from the parallel op-amp circuit in Figure 11 provides a piecewise linear DC V-I curve with six breakpoints when plotted in the V-I plane.
Circuit diagram of nonlinear resistor R0 of the real 6-lobe corsage memristor using off-the-shelf components.
The active and locally active two-port (marked with the black box) in Figure 11 consists of three op-amp circuits (marked with red, blue, and magenta boxes) in parallel. The circuit parameters and the components of the three op-amp circuit in Figure 11 are similar for each box except the negative feedback resistances (R41, R42, and R43) which determine the effective saturation voltage Esat of an individual op-amp circuit. The saturation voltage Esat along with the negative feedback resistance play a key role to achieve the V-I breakpoints at specified voltages such as V=3V, V=5V, and V=7V. The op-amp circuits in Figure 11 also contain a positive feedback path where the difficulty arises with the driving-point and transfer function. To resolve this problem, we replace the op-amp circuit in the red box (in Figure 11) by its three ideal models, such as the “Linear region,” “+Saturation region,” and “−Saturation region” as shown in Figures 12(a)–12(c), respectively.
Region-based ideal models of the op-amp circuit: (a) Linear region, (b) +Saturation region, and (c) −Saturation region.
The Linear region of the op-amp circuit in Figure 12(a) shows that the potential difference between the noninverting terminal (v+) and the inverting terminal (v−) is zero, so the differential voltage vd=v+−v−=0 and eventually inverting terminal voltage,
(19)v−=v+=vx=v.
The following relation between output voltage v03 and inverting terminal voltage v− can be computed by the voltage divider rule
(20)v−=vx=R33R33+R43v03=βv03,where β=R33/R33+R43 and henceforth
(21)v=βv03.
Pedagogically, in the linear region, the relation between the saturation voltage (±Esat) and output voltage v03 is as follows:
(22)−Esat<v03<Esat,for which, in the Linear region, the relation between the input voltage v and saturation voltage (±Esat) is
(23)−βESat<v<βEsat.
The loop
(24)provides linear region current ilin as
(25)ilin=1R73v1+R73R63−v03.
For the +Saturation region shown in Figure 12(b), the relation between the output voltage v03 and the saturation voltage Esat is as follows:
(26)v03=Esat,and the differential voltage vd>0, so that v+−v−>0, and eventually the relationship between input voltage and saturation voltage is
(27)v≥βEsat.
The current for the +Saturation region i+sat is computed as follows:
(28)i+sat=1R73v1+R73R63−Esat.
For the −Saturation region shown in Figure 12(c), the relation between the output voltage v03 and the saturation voltage Esat is
(29)v03=−Esat,and vd<0, so that v+−v−<0, and henceforth, the relationship between the input voltage and saturation voltage is
(30)v≤−βEsat,and the −Saturation region current is computed as
(31)i−sat=1R73v1+R73R63+Esat.
The current i3 flowing out of the op-amp circuit in Figure 11 is computed by adding all currents (ilin, i+sat, and i−sat):
(32)i3=1R73v1+R73R63−v03⏞−βEsat≤v≤βEsat⏟Linearregion+1R73v1+R73R63−Esat⏞v≥βEsat⏟+Saturationregion+1R73v1+R73R63+Esat⏞v≤−βEsat⏟−Saturationregion.
Plotting the output voltage v03 and input current i3 with respect to the input voltage v of the op-amp circuit with positive and negative feedback paths over the specified ranges of all the three regions provides a piecewise linear curve as shown in Figure 13(a). The op-amp circuit enclosed in the red box in Figure 11 exhibits two breakpoints at V = ±7 V with circuit parameters R33 = 1K, R43 = 1K, R63 = 100K, R73 = 1K, and Esat = 14 V over the input voltage range −14 V ≤ V ≤ 14 V. The mathematical simulation results shown in Figure 13 shows that the linear region of the op-amp circuit lies in the range −7 V ≤ V ≤ 7 V (as βEsat=7V) whereas the positive and negative saturation regions lie at V>7V and V<−7V, respectively. Moreover, the output voltage v03 at the positive and negative saturation regions are v03=+Esat=+14V and v03=−Esat=−14V, respectively, and in the linear region, v03 increases proportionately to the input voltage v. However, the current i3 of the op-amp circuit increases linearly in the saturation region and decreases in an inversely proportional manner to the input voltage v in the linear region.
Mathematical simulation of input currents (i3, i2, and i1) and output voltages (v03, v02, v01) with respect to input voltage v. Loci of (a) v03 versus v and i3 versus v, (b) v02 versus v and i2 versus v, and (c) v01 versus v and i1 versus v of the individual op-amp circuits.
Similarly, following the circuit-theoretic concepts mentioned above, the input currents i2 and i1 of the op-amp circuits in Figure 11 are computed as
(33)i2=1R72v1+R72R62−v02⏞−βEsat≤v≤βEsat⏟Linearregion+1R72v1+R72R62−Esat⏞v≥βEsat⏟+Saturationregion+1R72v1+R72R62+Esat⏞v≤−βEsat⏟−Saturationregion,and
(34)i1=1R71v1+R71R61−v01⏞−βEsat≤v≤βEsat⏟Linearregion+1R71v1+R71R61−Esat⏞v≥βEsat⏟+Saturationregion+1R71v1+R71R61+Esat⏞v≤−βEsat⏟−Saturationregion.
The loci of v02 versus v and i2 versus v and v01 versus v and i1 versus v are shown in Figures 13(b) and 13(c), respectively.
The total current i flowing out of the 3 parallel connected op-amp circuits in Figure 11 is equal to the summation of the three currents (i1, i2, and i3) of the individual op-amp circuits and computed as
(35)i=i1+i2+i3.
The six-breakpoint V-I curve of nonlinear resistor R0 is shown in Figure 14. Mathematical simulation results of current i and the equivalent nonlinear resistance R0 are shown in Figure 14(a) with the parameters (the measured resistor values of circuit implementation are used in mathematical and SPICE simulations) R31=R32=R33=0.985K, R61=R62=R63=100.5K, R71=R72=R73=1.001K, R41=3.888K, R42=1.797K, R43=0.987K, and Esat=14V. The plots of the current i and the active and locally active resistance R0 obtained by SPICE simulation and the actual circuit implementation are shown in Figures 14(b) and 14(c), respectively.
DC current I versus DC voltage V and DC resistance R0=V/I derived and measured from the 6-lobe corsage memristor emulator. (a) I versus V and R0 versus V plot of mathematical model, (b) I versus V and R0 versus V plot of spice circuit simulation, and (c) I versus V and R0 versus V plot of the actual circuit implementation.
The mathematical simulation presented in Figure 14(a) shows that the V-I curve and the resistance value of the nonlinear resistor R0 has six breakpoints at V = ±6.99 V, ±4.95 V, and ±2.82 V. The slope of R0 is different between these breakpoints and hence defines different memory states. Similar to the mathematical model, the plots of the V-I curve and the nonlinear resistance R0 obtained by SPICE simulation and the circuit implementation also have six breakpoints at V = ±6.9 V, ±4.89 V, and ±2.77 V and V = ±5.74 V, ±4.05 V, and ±2.15 V, respectively. The insets in Figures 14(a)–14(c) show the zoomed figure of R0 near the origin and show that R0=0 at input voltage V=0V.
The nonlinear resistance waveform obtained from the SPICE modelling in Figure 14(b) shows that the R0 is constant at resistance 147.47 Ω for an input voltage range −2.77 V < V < 2.77 V, except for a tiny interval at the origin (V = 0 V). However, for 2.77 V < V < 4.89 V and 4.89 V < V < 6.9 V, the R0 increases linearly from 147.47 Ω to 216.07 Ω and from 216.07 Ω to 336.10 Ω, respectively, whereas for V > 6.9 V, R0 increases almost linearly. Due to the linear increment of R0 with a constant slope over the abovementioned voltage range, it can be acclaimed that the real 6-lobe corsage memristor emulator contains four different memory states, namely, R01, R02, R03, and R04 where R01 = (R0 = 147.47 Ω), R02 = (147.47 Ω < R0 ≤ 216.07 Ω), R03 = (216.07 Ω < R0 ≤ 336.10 Ω), and R04 = (R0> 336.10 Ω).
Similar to the SPICE model, the mathematical and the circuit implementation plots of R0 also contains the four different memory states as shown in Figures 14(a) and 14(b), respectively. The fluctuation of R0 at the −2.16 V < V < 2.16 V voltage range in Figure 14(c) is negligibly small and can be regarded as R0 = 137 Ω. The fluctuation was induced due to computational difficulties at V = 0 V in the oscilloscope. Although the resistance R0 of the mathematical model of the 6-lobe corsage memristor and the real emulator are quantitatively different, they are also qualitatively identical.
The breakpoints of the V-I curve of the nonlinear resistor R0 in mathematical simulation, shown in Figure 14(a) and the SPICE simulation shown in Figure 14(b), are slightly different. This deviation happens due to the nonideal circuit components of the SPICE module. Moreover, the V-I curve breakpoints of R0 measured from the circuit implementation is further deviated from the mathematical and SPICE simulation. The reason behind such deviation is the nonideal characteristic of the op-amp circuit as well as the noise induced from the DC power supply and the oscilloscope probe. Another reason for such deviation is the used op-amp’s rated saturation voltage (Esat=13.7V) which is slightly less than the mathematical and SPICE saturation voltage Esat=14V.
Although the breakpoints of the DC V-I curve in Figure 5 and the breakpoints of Figure 14 are quantitatively different, they are qualitatively similar. In this artifact, one of our primary motives is to show that the basic method explained in [4, 5] and [14] can be used to convert the DC V-I curve of any real nonlinear resistor into a memristor. We prove this analogy by analyzing the parallel connected op-amp circuit in Figure 11 which has the capabilities to emulate the attributes of the 6-lobe Chua corsage memristor as it exhibits a 7-segment PWL DC V-I curve as shown in Figure 14.
The SPICE simulation of switching of memory states of the real 6-lobe Chua corsage memristor emulator (in Figure 10) is shown in Figure 15. Figure 15(a) shows the example of successful switching between the memory states R01 and R03. To switch from R01 to R03, a pulse input VA=5.5V with a pulse width Δw=7.5s is applied across our real emulator circuit in Figure 10. Observe from Figure 15(a) that the resistance R0=147.47Ω at t=0+, and it gradually increases during the pulse period Δw and saturated at R0=216.5Ω and remained there although the input voltage become zero for t≥Δw. The saturated resistance R0=216.5Ω lies over the memory state R03 = (216.07 Ω < R0 ≤ 336.10 Ω) which confirms the successful switching from memory state R01 to R03 for an input pulse VA=5.5V and Δw=7.5s. However, to fit the resistance scale, we truncated the t≤0 part of the R0 in Figure 15(a) as that part is insignificant because at t=0+, R0 immediately rises from R0=0Ω to R0=147.47Ω.
Switching kinetics of the memory states of real 6-lobe Chua corsage memristor emulator. (a) Successful switching: the corsage memristor emulator switches from memory state R01 to R03 with a resultant resistance R0=216.5Ω for an input pulse VA=5.5V and pulse width Δw=7.5s. (b) Switching failure: the proposed corsage memristor emulator fails to switch from R01 to R03 for an input pulse VA=5.5V and pulse width Δw=7s as the resultant resistance R0=215.6Ω lies in the memory state range R02.
In this paper, we also demonstrate the switching failure scenario of the real 6-lobe Chua corsage memristor as shown in Figure 15(b). In Figure 15(b), a pulse input VA=5.5V with a pulse width Δw=7s is applied across our real emulator. The nonlinear resistance R0 is saturated at R0=215.6Ω and remained there as the time increases although the voltage becomes vt=0 for t≥Δw. The resultant resistance R0=215.6Ω lies over the memory state R02 = (147.47 Ω < R0 ≤ 216.07 Ω) which confirms the failure of switching as our intention is to switch from memory state R01 to R03 for an input pulse VA=5.5V and Δw=7s, but we converge on memory state R02.
The switching failure scenario gives us the insights that the emulator circuit of our proposed corsage memristor is also dependent on the appropriate pulse amplitude and the pulse width like its mathematical model. To illustrate the relationship of pulse amplitude and the pulse width in our real emulator circuit, we plot the pulse amplitude VA versus pulse width |Δw| curve as shown in Figure 16. The hyperbolic relationship in Figure 16 shows that the maximum pulse amplitude VA requires less pulse width Δw to switch from one memory state to another state whereas the minimum pulse amplitude requires maximum pulse width.
Relation between amplitude VA versus pulse width Δw of the real 6-lobe Chua corsage memristor emulator.
6. Conclusion
The recent interest in inherently nonlinear memristor devices is bringing to a new life to the theory of nonlinear circuits and systems. In this paper, we design and build a highly nonlinear novel device, namely, the 6-lobe Chua corsage memristor, and its real emulator circuit using the nonlinear circuit theory. The proposed generic memristor can be used as a multistate, specifically 4-state, memory device with an increased efficiency of 50% compared to the 2-lobe and bistable extended memristor whereas the efficiency of the proposed memristor increased by 25% compared to the 4-lobe corsage memristor. Moreover, due to the presence of more equilibrium points compared to the 2-lobe or 4-lobe corsage memristors, the proposed corsage memristor exhibits a higher variety of dynamic routes in response to different initial conditions x0 which enhance the capability to represent a desired function more closely than 2-lobe or 4-lobe corsage memristors. Due to the diversified dynamic routes and the enhancement in stable memory states, the proposed corsage memristor is more versatile and effective than its predecessor 2-lobe and 4-lobe corsage memristors. Moreover, the diversified dynamic routes reveal a contiguous highly nonlinear DC V-I curve with six distinct contiguous hysteresis lobes, unlike the most published highly nonlinear disconnected DC V-I curves. Furthermore, the universal formulas, derived in Section 4, ease the demonstration of the switching kinetics of the 6-lobe corsage memristor and assist to switch the memory states preciously with an appropriate pulse amplitude and pulse width in accordance to an initial condition x0. The universal formulas are applicable to any device which exhibits a PWL dynamic route map with any number of segments for any DC or pulse input. Following the introduction of a purely mathematical memristor model with quad stability (2-bit memory system) at DC and pulse input, this paper elucidates the mechanisms behind the emergence of the first real emulated 6-lobe Chua corsage memristor using off-the-shelf elements. Nonlinear system theoretic concepts were applied to the model of the two-port memristive element to gain a deep insight into the quad-stable characteristic of its dynamics where the quantitative attributes of the real emulator might not be similar to the mathematical model but they are qualitatively same.
Data Availability
The data used to support the findings of this study are available from the first author or corresponding author upon request.
Conflicts of Interest
The authors declare that they have no conflicts of interest.
Acknowledgments
The authors would like to thank Professor Leon Chua for his wonderful assistance for revising the manuscript and technical advice. This work was supported in part by the National Research Foundation of Korea (NRF) grant funded by the Korean Government (2016R1A2B4015514), the “Cooperative Research Program for Agriculture Science and Technology Development (Project no. PJ0120642016),” the Rural Development Administration, Republic of Korea, and the US Air Force Office of Scientific Research under Grant no. FA9550-18-1-0016.
Supplementary Materials
Finding of universal formulas. Figure 1: dynamic routes of the switching kinetics of the 6-lobe Chua corsage memristor. The two magenta-color vertical line segments indicate an instantaneous jump between the red and the blue piecewise-linear plots in the dynamic route map. Figure 2: movement of the exponential trajectories of x(t) with respect to time t.
StrukovD. B.SniderG. S.StewartD. R.WilliamsR. S.The missing memristor found20084537191808310.1038/nature069322-s2.0-4304912683318451858ChuaL.Memristor-the missing circuit element197118550751910.1109/TCT.1971.10833372-s2.0-0015127532ChuaL. O.KangS. M.Memristive devices and systems197664220922310.1109/PROC.1976.100922-s2.0-0016918810AscoliA.TetzlaffR.ChuaL. O.The first ever real bistable memristors—part I: theoretical insights on local fading memory201663121091109510.1109/TCSII.2016.26045672-s2.0-85027034178AscoliA.TetzlaffR.ChuaL. O.The first ever real bistable memristors—part II: design and analysis of a local fading memory system201663121096110010.1109/TCSII.2016.26135602-s2.0-85002498878StathopoulosS.KhiatA.TrapatseliM.CorteseS.SerbA.ValovI.ProdromakisT.Multibit memory operation of metal-oxide bi-layer memristors201771, article 1753210.1038/s41598-017-17785-12-s2.0-8503844258029235524MannanZ. I.YangC.KimH.Oscillation with 4-lobe Chua corsage memristor2018182142710.1109/MCAS.2018.28217242-s2.0-85047949061MannanZ. I.ChoiH.KimH.Chua corsage memristor oscillator via Hopf bifurcation201626410.1142/S02181274163000932-s2.0-84966659153MannanZ. I.ChoiH.RajamaniV.KimH.ChuaL.Chua corsage memristor: phase portraits, basin of attraction, and coexisting pinched hysteresis loops201727310.1142/S02181274173001172-s2.0-85017325090StrogatzS. H.Phase plane19942ndMA, USAAddison-Wesley Publishing Co.145150ch. 6, sec 6.1–6.3AvciC. O.MannM.TanA. J.GambardellaP.BeachG. S. D.A multi-state memory device based on the unidirectional spin Hall magnetoresistance20171102010.1063/1.49837842-s2.0-85029907968StrzeleckiR.ZinovievG. S.Overview of power electronics converters and controls2008LondonSpringer-Verlag Limited55105Power Systems10.1007/978-1-84800-318-7_3AdhikariS. P.SahM. P.KimH.ChuaL. O.Three fingerprints of memristor201360113008302110.1109/TCSI.2013.22561712-s2.0-84887136051ChuaL.If it’s pinched it’s a memristor2014291010.1088/0268-1242/29/10/1040012-s2.0-84907222758ChuaL.Everything you wish to know about memristors but are afraid to ask201524231936810.13164/re.2015.03192-s2.0-84932605254ChuaL. O.1969New York, USAMcGraw-HillSlotineJ. J. E.LiW.Advance stability theory1991New JerseyPrentice-Hall, Inc.ThomasG. B.FinneyR. L.19795thMA, USAAddison-Wesley Publishing Co.StoverWeissteinE. W.Parametric equationshttp://mathworld.wolfram.com/ParametricEquations.htmlMainzerK.ChuaL.2013UKImperial College Press10.1142/p8822-s2.0-84967495602SahM. P.MannanZ. I.KimH.ChuaL.Oscillator made of only one memristor and one battery201525310.1142/s02181274153001042-s2.0-84928550667ChuaL. O.DesoerC. A.KuhE. A.Operational-amplifier circuits1985New York, USA187212ch. 4, sec. 3