An Adaptive Window Function based Memristor

Model
2023 33rd International Conference Radioelektronika (RADIOELEKTRONIKA) | 979-8-3503-9834-2/23/$31.00 ©2023 IEEE | DOI: 10.1109/RADIOELEKTRONIKA57919.2023.10109029

Ananda Y. R. Arkadeep Hajra Gaurav Trivedi

Dept. of EEE Dept. of ETCE Dept. of EEE
IIT Guwahati Jadavpur University IIT Guwahati
Guwahati, India Kolkata, India Guwahati, India
yrananda@iitg.ac.in arkadeep548@gmail.com trivedi@iitg.ac.in

Abstract—In the proposed work, a new window function for boundary lock issue, but its limited nonlinearity leads to a
the memristor model is developed, incorporating an intelligent degraded pinched hysteresis loop (PHL). Prodromarkis et al.
adaptive system. For the first time, an adaptive constant param- [11] propose a new window function that offers substantial
eter for the window function is proposed in this paper, which
introduces higher nonlinearity at the boundaries as compared scalability and nonlinearity as compared to the above window
to classical window functions. This parameter depends on the functions for a certain voltage and frequency range. However,
input voltage and frequency to generate a pinched hysteresis loop this window function retains the boundary lock problem and
(PHL). It enhances the nonlinearity of the memristor model for produces distorted PHL when voltage and frequency exceed
a wide range of voltages and frequencies. The memristor model a permissible range. For extending the nonlinear behaviour,
using the proposed window function provides an acceptable PHL
from 0.01 Hz to 10 kHz for various voltage levels compared to Jinxiang et al. [12] showcase a novel window function, with
other available window functions based memristor models. no boundary lock problem and provides PHL for a more
Index Terms—Memristor, nonlinearity, adaptive parameter, extended range of voltage and frequency than the previous
memristor modeling, window function. ones. Nevertheless, increasing voltage and frequency further
limits the nonlinearity of this window function, which leads
I. I NTRODUCTION to distorted PHL.
The term memristor was initially introduced by Leon Chua This paper proposes a novel window function incorporating
[1], and it is considered as the fourth fundamental circuit an additional parameter, the adaptive constant. The features of
element. The memristor is a two-terminal passive nano ionic the proposed adaptive window function are higher nonlinearity,
device that exhibits a resistor with memory. The relationship scalability, and a more extended range of voltage and fre-
between flux (φ) and the charge (q) gives rise to nonlinear quency operations. The adaptive constant parameter introduces
characteristics due to the presence of dynamic negative differ- intelligence in the window function to estimate applied voltage
ential resistance (DNDR) [2]. and frequency for realizing the nonlinear behaviour of the
The versatile properties of memristor make it suitable for memristor correctly. It aids in producing the desired PHL
several applications, such as ultra-high density nonvolatile without any distortion. The key contributions of the proposed
memory [3], neuromorphic computing systems [4], convolu- window function are as follows.
tion neural networks [5], programmable analog ICs [6], and • The proposed window function incorporated with Adap-
several other bio-inspired applications [7] etc. The first physi- tive constant (AD ) works for a higher voltage and fre-
cal memristor was fabricated in 2008 by Strukov et al. from the quency range than the standard window functions re-
HP lab [8] using a TiO2 oxide layer sandwiched between two ported in the literature.
platinum electrodes. The mathematical model derived using • No need to change any static parameters for operating
this memristor is based on several assumptions considering voltage or frequency change.
a uniform electric field. As we know, a small change in the • Not require a high value of j for higher frequency due to
input voltage of a memristor introduces significant variations adaptive nonlinear behaviour.
in the electric field. Hence, a uniform electric field assumption The rest of the paper is organized as follows. Section II
does not hold, and the memristor inherently exhibits nonlinear describes the working principle of a memristor. In section III,
characteristics. Therefore, a window function is introduced in a proposed adaptive window function is presented. Section IV
the memristor device model to achieve nonlinear behaviour delineates the performance of the proposed memristor model,
and satisfy boundary conditions. and the paper is concluded in section V.
There are many window functions reported in the literature.
The window function proposed by Joglekar et al. [9] has II. T HE WORKING PRINCIPLE OF A MEMRISTOR
several limitations. It exhibits less scalability, limited nonlin- The structure of a memristor comprises an oxide layer
earity and a boundary lock problem. With improved scalability, sandwiched between two electrodes, as depicted in Fig. 1.
Biolek et al. [10] present a new window function with no Depending on the direction of current flowing through the

979-8-3503-9834-2/23/$31.00 © 2023 IEEE

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY CALICUT. Downloaded on June 23,2023 at 04:13:13 UTC from IEEE Xplore. Restrictions apply.
 device, the width (w) of the doped region (high conductive window functions offer limited nonlinearity and scalability and
region) varies, and it is considered as the ON state of the suffer from various issues, such as boundary lock problems
device. Similarly, the current flowing in the opposite direction etc. Thus, the need of the hour is to design a window function
reduces width (w) and drives the memristor to the OF F state. that can address the abovementioned issue. It motivates us to
Therefore, the memristor has two resistive states, Ron and explore the possibility of a novel window function that can
Rof f , as shown in Fig. 2. provide realistic memristor behaviour in various conditions.
Our proposed window function is described in the following
section in detail.
III. P ROPOSED ADAPTIVE WINDOW FUNCTION
As we know, every device model has advantages and limi-
tations. It is found in the literature that a memristor window
function can be improvised further to depict realistic behaviour
Fig. 1: Physical structure of the TiO2 based memristor at a wide range of voltages and frequencies. Among many
limitations imposed by the memristor model using various
window functions [8]–[14], producing a desired pinched hys-
teresis loop (PHL) by these window functions only in the
range of 0.5 Hz to 100 Hz. Therefore, there is a need to
extend the frequency and voltage range of the memristor. This
paper presents a memristor model using a proposed window
function to obtain desired PHL in a broader frequency and
voltage range.
Fig. 2: Equivalent circuit of the TiO2 based memristor
A. Mathematical model of adaptive window function
The memristor fabricated by Strukov et al. [8] satisfies An adaptive window function based memristor model is
above mentioned device characteristics. Therefore, the mem- proposed. Initially, it evaluates the frequency and amplitude
ristance of the device M (q) can be computed as mentioned estimation of the applied input. Our proposed window function
below. introduces a new parameter for estimating the frequency and
amplitude of the applied input signal. This new parameter is
w(t) w(t) called an Adaptive constant.
M (q) = Ron + Rof f (1 − ) (1)
D D Also, the window function [12] is modified to achieve high
Where w is the width of the doped layer, D is the total width nonlinear characteristics of a memristor and the mathematical
of the memristor, and Ron and Rof f are the ON and OF F expression for the new window function is formulated as
resistance of the memristor. follows.
Now normalizing the state variable w with respect to D,
 
1
f (x) = AD 1−
we get a new variable x, where x = w(t) D . As stated below, 2(e − 1)
the abovementioned equations can be reformulated using the 
( (x−1)2q −1)

+ e(x −1)
2q
e
variable x.
  p i
2
M (x) = Ron x + Rof f (1 − x) (2) 1 − (x − 0.5) + 0.75 (5)
Where q and p are control parameters and AD is an adaptive
dx(t) Ron constant.
= µv 2 i(t)f (x) (3)
dt D
Where µv is an average ion mobility. Now we have to analyze how the adaptive constant should
It helps to define a normalized window function f (x) in change to get a pinched hysteresis loop for the chosen voltage
numerous ways. One of them is a polynomial having zero and frequency range. According to an experiment, we have
values at x = 0 and x = 1 and is defined below for seen that if we use the expression in equation 6 for the
completeness. adaptive constant, it gives a pinched hysteresis loop for the
f (x) = xm (1 − x)n (4) amplitude range (VL , VH ) and the frequency range (fL , fH ).
The observed result of the experiment is presented in Table I.
Where m and n are integers. Similarly, another window As seen in Table I, the optimal value of the adaptive
function satisfying the same boundary conditions is f (x) = constant, in any case, is approximately equal to the “golden
1−(2x−1)2p , where p is an integer. Another window function ratio”, defined as the ratio of the frequency and the amplitude
is presented in [10], which considers current i flowing through of the applied sinusoidal voltage. The expression for Adaptive
the memristor and a factor involving a unit step function of Constant (AD ) is mentioned below.
current in the window function. In [10], f (x) is defined as estimated min . f requency of input
f (x) = [1 − (x − stp(−i))2p ]. It can be observed that these AD = (6)
estimated min . amplitude of input

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY CALICUT. Downloaded on June 23,2023 at 04:13:13 UTC from IEEE Xplore. Restrictions apply.
 TABLE I: Experimental analysis of adaptive constant T /2
Ron
Z
∆x = 2 µv i(t)f (x)dt
Voltage (V) Frequency (f) Golden ratio Optimal 0 D2
(in V) (in Hz) ( fv ) value of AD
T /2
Ron v(t)
Z
100 10 0.1 0.105
∆x = 2 µv f (x)dt
100 1 0.01 0.011 0 D2 M (x)
10 100 10 10.25
10 10 1 0.975 T /2
Ron v(t)
Z
10 0.1 0.01 0.01075 ∆x ≤ 2 µv f (x)dt
1 100 100 100 0 D2 Ron
1 10 10 10 Z T /2
0.1 100 1000 1000 2µv Am AD
0.1 10 100 100 ∆x < sin(2πf t)dt
D2 0

2µv Am AD
The detailed analysis and mathematical formulation of the ∆x <
πf D2
proposed Adaptive Constant (AD ) are described in the follow-
ing section. Therefore,
2µv Am AD
(∆x)max ≅ (10)
B. Formulation of adaptive constant πf D2
The I-V characteristic of the device being nonlinear is due From equation 10, we make Adaptive constant = Afm ; the
to different values of δv
δi at different points of the curve. In the memristor response will be appropriately scaled for sinusoid
case of a memristor, its pinched hysteresis loop is generated of any amplitude and frequency.
because of different values of δv δi at different points of the Also, by considering the physical realization of the mem-
curve, as stated in below mathematical model. ristor, it has a specified range of frequency and ampli-
δv tude, which needs to be incorporated into our proposed
= M (x) = [Rof f − (Rof f − Ron )x] (7) Adaptive constant to obtain its complete mathematical
δi
model as follows.
The derivative of state variable x is described by the
following mathematical expression.
min(fH , max(fL , estimated min . f req. of input))
dx(t) Ron AD =
= µv 2 i(t)f (x) (8) min(vH , max(vL , estimated min . ampl. of input))
dt D
C. Design methodology of memristor model using a proposed
To ensure the swing of the variable x is optimum, we have
window function
to ensure that the output of f (x) needs to be compensated to a
larger value when the pinched hysteresis loop starts to shrink To estimate the frequency and amplitude of an input signal,
to a straight line or it starts to lose its nonlinear behaviour we have designed a memristor model by using a diode and
(which happens when a low-amplitude and high-frequency capacitor in series with a copy of the input signal (a copy
input is applied). Alternatively, in order to compensate for of the input signal can easily be made using a voltage-
the output of f (x) to a smaller value, when the pinched dependent voltage source with gain=1) using LTSpice. Also,
hysteresis loop becomes too nonlinear and starts to become a capacitor of 1 F is connected in parallel with the input
distorted (which happens when a high-amplitude and low- signal, and the current flowing through the capacitor becomes
frequency input signal is applied). Therefore, encapsulation of dv/dt = 2πAm f cos(2πf t), as shown in Fig. 5.
the Adaptive Constant (AD ) in equation 6 with the proposed
window function for memristor modelling to generate the D
desired PHL.
The mathematical expression of Adaptive constant(AD )
is derived by considering an estimation of the maximum
possible swing of the state variable x when the frequency Vin C=1 F
f (Time-Period, T = 1/f ) and amplitude Am for an input
signal Am sin(2πf t). The range of window function and
memristance is as follows,
• The range of the proposed window function is
(0, Adaptive constant), i.e., 0 < f (x) < Fig. 5: Peak detector circuit
Adaptive constant(AD ).
• The range of memristance is Ron ≤ M (x) ≤ Ro f f .
Additionally, a CCVS with trans-resistance=1 Ω with cur-
Consider the swing of the state variable x is ∆x. Now, ∆x rent flowing through the capacitor as the controlling current.
can be approximated by the following expression. Then its output becomes 2πAm f cos(2πf t). The estimated
Z T /2 frequency value is obtained by dividing the peak amplitude of
dx
∆x = 2 dt (9) this voltage signal by 2πAm .
0 dt

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY CALICUT. Downloaded on June 23,2023 at 04:13:13 UTC from IEEE Xplore. Restrictions apply.
 (a) (b) (c)
Fig. 3: Numerical analysis of memristor model using different window functions with 1 V (a) PHL at 100 Hz (b) PHL at 1
kHz and (c) PHL at 10 kHz

(a) (b) (c)

(d) (e) (f)

Fig. 4: Numerical analysis of memristor model using different window functions (a) PHL using Biolek and Joglekar at 0.05
V and 0.01 Hz (b) PHL using Prodromakis, Jinxiang, and Proposed at 0.05 V and 0.01 Hz (c) PHL at 0.05 V and 1 Hz (d)
PHL at 0.01 V and 0.05 Hz (e) PHL at 10 V and 100 Hz and (f) PHL at 10 V and 1 kHz

TABLE II: Comparison of different window functions

Window functions Strukov [1] Joglekar [9] Biolek [10] Prodromakis [11] Jinxiang [12] Proposed
Scalability No Limited Limited Moderate Moderate High
Nonlinearity Limited Limited Limited Moderate Moderate High
Boundary lock Yes Yes No Yes No No
Operating voltage and frequency range Limited Limited Limited Moderate Moderate High

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY CALICUT. Downloaded on June 23,2023 at 04:13:13 UTC from IEEE Xplore. Restrictions apply.
 IV. P ERFORMANCE ANALYSIS OF THE PROPOSED WINDOW function. The advantages of the proposed window function are
FUNCTION BASED MEMRISTOR MODEL mentioned as follows.
• The adaptive constant (AD ) aids in enhancing the nonlin-
ear behaviour of the memristor model compared to other
The numerical analysis of various window functions is per-
window functions.
formed using the LTSpice framework. As can be seen in Fig.
• No need to adjust any static parameters for any variation
3, PHL (fingerprint of a memristor) is generated for various
in applied voltage or frequency.
window functions [9]–[12] for different frequencies in the
• Not require a high value of j for higher frequency due
range of 100 Hz to 10 kHz at 1 V. However, PHL is generated
to adaptive nonlinear behaviour.
only for memristor model using a proposed window function at
varied ranges of frequencies as depicted in Fig. 3 and the lobe Therefore, it is substantially flexible for obtaining the desired
area of PHL starts shrinking when the frequency increases characteristics of a memristor, which can be further employed
from 100 Hz to 10 kHz, which verifies the correctness of to design a wide range of potential applications.
memristor behaviour [15] using a proposed window function. ACKNOWLEDGMENTS
The functionality of different window functions is also verified
The work has been partially supported by the Ministry of
for low voltage and low frequency range, as shown in Fig.
Electronics and Information Technology’s (MeitY) sponsored
4. The distorted PHL is generated at 0.05 V and 0.01 Hz
project Electronics and ICT Academy at IIT Guwahati. The
using Biolek and Joglekar, as shown in Fig. 4a. No PHL
authors acknowledge the infrastructure support provided by
is generated for the Prodromakis window function. However,
the Academy, which helped in the successful and timely
PHL is generated for the Jinxiang and proposed window
conclusion of the proposed work.
functions, as shown in Fig. 4b. Increasing frequency to 1 Hz
at 0.05 V, PHL is generated only for the proposed window R EFERENCES
function, as illustrated in Fig. 4c. By reducing voltage to 0.01 [1] L. Chua, “Memristor-the missing circuit element,” IEEE Transactions
V at frequency 0.05 Hz, PHL is generated for Prodromakis and on circuit theory, vol. 18, no. 5, pp. 507–519, 1971.
proposed window functions, as depicted in Fig. 4d. Applying [2] T. Hickmott, “Low-frequency negative resistance in thin anodic oxide
films,” Journal of Applied Physics, vol. 33, no. 9, pp. 2669–2682, 1962.
a high voltage of 10 V at 100 Hz, PHL is generated slightly [3] R. Waser and M. Aono, “Nanoionics-based resistive switching memo-
for the Prodromakis window function, and PHL of high lobe ries,” Nanoscience And Technology: A Collection of Reviews from Nature
area is generated for the proposed window function, which Journals, pp. 158–165, 2010.
[4] T. D. Dongale et al., “TiO2 based nanostructured memristor for RRAM
is shown in Fig. 4e. A stable PHL is generated only for and neuromorphic applications: A simulation approach,” Nano conver-
the proposed window function at high voltage of 10 V with gence, vol. 3, no. 1, pp. 1–7, 2016.
increased frequency up to 1 kHz, as shown in Fig. 4f. The [5] E. Lehtonen and M. Laiho, “CNN using memristors for neighbor-
hood connections,” in 2010 12th International Workshop on Cellular
results of the numerical analysis show that the memristor Nanoscale Networks and their Applications (CNNA 2010). IEEE, 2010,
model using the proposed window function generates PHL pp. 1–4.
for different frequency ranges from 0.01 Hz to 10 kHz due [6] S. Shin, Kim et al., “Memristor applications for programmable analog
ICs,” IEEE Transactions on Nanotechnology, vol. 10, no. 2, pp. 266–
to the incorporation of the Adaptive constant (AD ), which 274, 2010.
further enhances the scalability and nonlinearity. Therefore, [7] S. Spiga et al., Memristive Devices for Brain-Inspired Computing.
our proposed window function based memristor model pro- Woodhead Publishing, 2020.
[8] D. B. Strukov et al., “The missing memristor found,” nature, vol. 453,
vides desirable I-V characteristics for a higher range of voltage no. 7191, pp. 80–83, 2008.
and frequency without any degradation compared to other [9] Y. N. Joglekar and S. J. Wolf, “The elusive memristor: properties of
approaches presented in Table II. basic electrical circuits,” European Journal of physics, vol. 30, no. 4,
pp. 661–675, 2009.
[10] Z. Biolek et al., “SPICE model of memristor with nonlinear dopant
V. C ONCLUSIONS drift,” Radioengineering, vol. 18, no. 2, 2009.
[11] T. Prodromakis et al., “A versatile memristor model with nonlinear
The main features of the memristor model using the pro- dopant kinetics,” IEEE transactions on electron devices, vol. 58, no. 9,
posed window function are higher nonlinearity, scalability, pp. 3099–3105, 2011.
and adaptability. The proposed model has four new user- [12] J. Zha et al., “A novel window function for memristor model with
application in programming analog circuits,” IEEE Transactions on
defined parameters: VL , VH , fL , and fH , respectively. Thus, it Circuits and Systems II: Express Briefs, vol. 63, no. 5, pp. 423–427,
provides a PHL for a much broader amplitude and frequency 2015.
range because the user can define the voltage and frequency [13] S. Benderli and T. Wey, “On SPICE macromodelling of TiO2 memris-
tors,” Electronics letters, vol. 45, no. 7, pp. 377–379, 2009.
range using those newly introduced parameters. For instance, [14] A. Rak and G. Cserey, “Macromodeling of the memristor in SPICE,”
our proposed window function based memristor model can IEEE Transactions on Computer-Aided Design of Integrated Circuits
provide a pinched hysteresis loop for 0.01 Hz to 10 k Hz fre- and Systems, vol. 29, no. 4, pp. 632–636, 2010.
[15] S. P. Adhikari, M. P. Sah, H. Kim, and L. O. Chua, “Three fingerprints
quency than other state-of-the-art window functions. Because of memristor,” IEEE Transactions on Circuits and Systems I: Regular
the memristor model using the proposed window function will Papers, vol. 60, no. 11, pp. 3008–3021, 2013.
adapt itself for any change in voltage and frequency due to
the presence of an adaptive constant in the voltage and the
frequency due to an adaptive constant in the proposed window

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY CALICUT. Downloaded on June 23,2023 at 04:13:13 UTC from IEEE Xplore. Restrictions apply.