Chapter 15

Analysis of Crank-Nicolson - Successive Over Relaxation Semiconductor

Device Model

Elayaraja Aruchunan, Zailan Siri, Muhamad Hifzhudin Bin Noor Aziz, Muhammad Hakimi Bin
Ab Wahab, Mohana Sundaram Muthuvalu and Jumat Sulaiman

The junction temperatures (JT) of many microelectronic devices have been reduced in order to
improve device reliability and extend operating life. When the JT exceeds a certain threshold,
electronic devices can suffer irreversible damage. As a result, a thermal control system can be
used to achieve high performance in electronic systems. For predicting peak junction
temperatures (PJT) of semiconductor devices, the heat conduction equation will be discretized
using a numerical approach. In the first step, the heat conduction equation will be discretized
using the Crank-Nicolson finite difference method. The Successive Over Relaxation (SOR)
and Gauss-Seidel (GS) methods will be used to solve the generated system of linear equations
iteratively. The SOR method uses fewer iterations and takes less time to compute than the
traditional GS iterative method, based on the results. In terms of PTJ accuracy, however, both
methods are comparable.

Keywords Peak Junction Temperature; Crank-Nicholson; Successive Over Relaxation;

Semiconductor; Thermal Conductivity.

15.1 Introduction

High power is a common occurrence in power device applications, and it is one of the most
important factors in ensuring that power devices are reliable in their demanding applications.
In order to achieve this goal, reliability and performance must be taken into account [1]. It is a
designer's job to build the best shape of semiconductor devices in order to extend the equipment
life expectancy. They will leave no room for failure, ensuring that all semiconductor-based
devices function properly. Every device must adhere to its own design by undergoing careful
measurement and processing. As a result, JT is a critical parameter to measure when
determining the reliability and performance of semiconductor devices and power devices [2].
Because many companies want to build a small device with increasing system reliability
and performance, thermal consideration has become one of the most important things in
electronic packaging design in this era of globalisation [3]. Most researchers believe that heat
management is the most serious issue because it can cause devices to malfunction or break
immediately. As a result, careful consideration must be given to ensuring that semiconductor
devices can be used until the maximum operating junction temperature is reached in order to
determine semiconductor reliability. This will help to extend the life of electronic packages
and prevent thermal failure. In this study, mathematical modelling is used to predict IC JT in
order to monitor the temperature of the semiconductor [3].
The performance of the Crank-Nicolson - Successive Over Relaxation (CN-SOR) iterative
method in determining the PJT of semiconductor devices is examined in this paper. Previously,
the SOR method had been widely used to solve a variety of matrix equation problems. On
linear and nonlinear systems, the effectiveness of the SOR method and its variants has been
studied and tested; see [4–7] for recent papers. The remainder of this paper is organised as
follows. In Sections 2 and 3, the mathematical modelling and numerical approach for
determining the PTJ of semiconductor devices will be discussed. Some simulation results are
included in Section 4. Section 5 contains the discussion and concluding remarks.

15.2 Model of Heat Equation

Fig. 1. FDM thermal circuit networks. (Source [8])

The following one-dimensional heat conduction equation will be used to predict IC JT:
𝜕 2 𝑇(𝑥, 𝑡) 𝜕𝑇(𝑥, 𝑡)
𝐾 = 𝜌𝑐 (15.1)
𝜕𝑥 2 𝜕𝑡

where T, K, 𝜌, c represents the absolute temperature, semiconductor device (silicon) thermal

conductivity, silicon mass density, and specific heat of silicon, respectively.
The following boundary conditions are satisfied by equation (15.1).
𝜕𝑇
𝑆𝐾 | = −𝑃𝑖𝑛 (15.2)
𝜕𝑥 𝑥=0
𝑇(𝐿, 𝑡) = 𝑇𝑖𝑛 (15.3)

where S, 𝑃𝑖𝑛 , L and 𝑇𝑖𝑛 are, respectively, silicon surface, input power, vertical power device
thickness, and input temperature.
15.3 Discretization of Crank-Nicolson method

CN will be used to discretize the equation. The CN is as shown below [9]

𝜕𝑇 𝑇𝑖,𝑗+1 − 𝑇𝑖,𝑗
= (15.4)
𝜕𝑡 ∆𝑡

𝜕 2𝑇 1
2
= [ ∅] (15.5)
𝜕𝑥 2(∆𝑥)2
Where ∅ = (𝑇𝑖+1,𝑗 − 2𝑇𝑖,𝑗 + 𝑇𝑖−1,𝑗 ) + (𝑇𝑖+1,𝑗+1 − 2𝑇𝑖,𝑗+1 + 𝑇𝑖−1,𝑗+1 )
By applying formulas (15.4) and (15.5) to equation (15.1), it will result to
𝑇𝑖,𝑗+1 − 𝑇𝑖,𝑗 𝐾 1
= ( [∅]) (15.6)
∆𝑡 𝜌𝑐 2(∆𝑥)2

𝐾∆𝑡
Let 𝛼 = 2𝜌𝑐(∆𝑥)2 , equation (15.6) can be rewrite as

−𝛼𝑇𝑖+1,𝑗+1 + (1 + 2𝛼)𝑇𝑖,𝑗+1 − 𝛼𝑇𝑖−1,𝑗+1 = 𝛼𝑇𝑖+1,𝑗 + (1 − 2𝛼)𝑇𝑖,𝑗 + 𝛼𝑇𝑖−1,𝑗 (15.7)

Boundary condition gives,

𝑇𝑖,𝑗+1 − 𝑇0,𝑗+1
𝑆𝐾 = −𝑃𝑖𝑛 (15.8)
∆𝑥
𝑇𝑛,𝑗+1 = 𝑇𝑖𝑛 (15.9)

Generally, by following boundary conditions from equation (15.8) and (15.9), equation (15.7)
can be represented in matrix form as below [10]
𝐴𝑇 = 𝐵 (15.10)

Where
1+𝛼 −𝛼 0 0 0 0
−𝛼 1 + 2𝛼 −𝛼 0 0 0
0 −𝛼 1 + 2𝛼 −𝛼 0 0
𝐴= ⋮ ⋱ ⋱ ⋱ ⋱ ⋮
0 0 −𝛼 1 + 2𝛼 −𝛼 0
0 0 0 −𝛼 1 + 2𝛼 −𝛼
[ 0 0 0 0 −𝛼 1 + 2𝛼 ]
 𝑇1,𝑗+1 𝐵1,𝑗
𝑇2,𝑗+1 𝐵2,𝑗
𝑇3,𝑗+1 𝐵3,𝑗
𝑇= ⋮ and 𝐵 = ⋮
𝑇𝑛−3,𝑗+1 𝐵𝑛−3,𝑗
𝑇𝑛−2,𝑗+1 𝐵𝑛−2,𝑗
[𝑇𝑛−1,𝑗+1 ] [𝐵𝑛−1,𝑗 ]
With
∆𝑥
𝛼𝑇𝑖+1,𝑗 + (1 − 2𝛼)𝑇𝑖,𝑗 + 𝛼𝑇𝑖−1,𝑗 + 𝛼 ( ) 𝑃𝑖𝑛 , 𝑖 = 1
𝑆𝐾
𝐵𝑖,𝑗 𝛼𝑇
𝑖+1,𝑗 + (1 − 2𝛼)𝑇𝑖,𝑗 + 𝛼𝑇𝑖−1,𝑗 , 𝑖 = 2, 3, … , 𝑛 − 3, 𝑛 − 2
{ 𝛼𝑇𝑖+1,𝑗 + (1 − 2𝛼)𝑇𝑖,𝑗 + 𝛼𝑇𝑖−1,𝑗 + 𝛼𝑇𝑖𝑛 , 𝑖 = 𝑛 − 1

15.4 Formulation Gauss-Seidel (GS) method

The generated linear system (15.10) will be solved by using CN-GS and CN-SOR methods.
The formula of CN-GS is [12]
𝑇 (𝑛+1) = (𝐷 − 𝐿)−1 𝑈𝑇 (𝑛) + (𝐷 − 𝐿)−1 𝐵 (15.11)

The algorithm of CN-GS method is shown below,

Step 1 Set κ = 1

Step 2 While (κ ≤ N) do steps iii-vi

Step 3 For i = 1, … , n − 1

(κ+1) 1 (κ+1) (κ)

Set, 𝑇i = a [− ∑i−1
j=1 a ij 𝑇j − ∑n−1
j=i+1 a ij 𝑇j + B]
ii
(κ+1)
Step 4 If ‖T (κ+1) − T (κ) ‖ < ε (STOP) OUTPUT: Xi

Step 5 Set κ = κ + 1

(κ)
Step 6 For i = 1, … , n − 1 set Ti = T.

Step 7 STOP when the maximum number of iterations surpassed.

15.5 Formulation of Successive Over Relaxation (SOR) method

Meanwhile, the formula of SOR is [12]

𝑇 (𝑛+1) = (𝐷 − 𝜔𝐿)−1 ((1 − 𝜔)𝐷 + 𝜔𝑈)𝑇 (𝑛) ((𝐷 − 𝜔𝐿)−1 )𝜔𝐵 (15.12)
Where D, -L and -U represent the diagonal, strictly lower triangular, strictly upper triangular
of matrix A, respectively.
The algorithm of SOR method is shown below [12]
Step 1 setκ = 1.
Step 2 While (κ ≤ 𝑁) do steps 3-6
Step 3 For 𝑖 = 1, … , 𝑛
1
Set 𝑥𝑖 = (1 − 𝜔)𝑋0𝑖 + [𝜔(− ∑𝑖−1 𝑛
𝑗=1 𝑎𝑖𝑗 𝑥𝑗 − ∑𝑗=𝑖+1 𝑎𝑖𝑗 𝑋0𝑗 + 𝑏𝑖 )]
𝑎𝑖𝑖

Step 4 If ‖𝑥 − 𝑋0‖ < 𝑇0𝐿 then OUTPUT 𝑥1 , 𝑥2 , … , 𝑥𝑛

(The procedure was successful.) STOP.
Step 5 Set κ = κ + 1,
Step 6 For 𝑖 = 1, … , 𝑛 set 𝑋0𝑖 ; = 𝑥𝑖 ;
Step STOP when the maximum number of iterations surpassed.

15.6 Numerical Treatments

The simulation results that obtained from CN-GS and CN-SOR iteration methods are
discussed. The different grid mesh sizes which are 30, 60, 90, 120 and 150 have been used to
establish the PJT. Itis critical to define the initial and boundary conditions correctly because
they can have an impact on the final result. For initial condition, at time 𝑡 = 0, which is the
room temperature, it is assumed to be 294 Kelvin. Meanwhile, for boundary conditions (refer
equation (15.3)), it defines the value at 𝑥 = 0 and 𝑥 = 𝐿. Assumed the temperature to be
constant, 𝑇𝑖𝑛 = 300.15 Kelvin. The thickness, 𝐿 is set to 0.055 cm, the surface of silicon, 𝑆 is
set to 0.1 𝑐𝑚2 and power input, 𝑃𝑖𝑛 = 200 𝑊 [1]. The properties of silicon used in this
experiment are shown in Table 1.
Table 1. Properties of Silicon
Properties SI Units Values
Thermal Conductivity, K W/cm/K 1.54
Mass Density, 𝜌 and
J/K/cm3 1.63
specific heat, c
 The number of iterations (k), computational time in seconds (CPU), and maximum
temperature (T max) are all measured as part of the numerical simulations. The values of 𝜔
used for CN-SOR method are 1.6, 1.7, 1.8 and 1.9. All the simulations are performed on a
computer model type Intel® Core™ i3-2310M (2.1GHz, 3MB L3 cache) and 4.0 GB RAM. All
the programs are compiled by using Borland C++. The convergence criterion for CN-GS and
CN-SOR methods is 𝜀 = 10-10 is used.
Based on the result obtained from Table 2, 3 and 4, it is shown that CN-SOR method is
more efficient than the CN-GS method.
Table 2. Comparison of numerical results at 𝑡 = 0.002
N Methods Number of Iterations CPU time 𝑇𝑚𝑎𝑥
CN-GS 12155 0.88 355.7851
30
CN-SOR 8056 (𝜔 = 1.6) 0.67 355.7851
CN-GS 45725 1.85 356.5180
60
CN-SOR 22963 (𝜔 = 1.8) 1.71 356.5180
CN-GS 99891 4.68 356.7612
90
CN-SOR 44286 (𝜔 = 1.8) 3.77 356.7612
CN-GS 173960 10.14 356.8826
120
CN-SOR 60852 (𝜔 = 1.9) 5.72 356.8826
CN-GS 211122 12.38 356.9554
150
CN-SOR 93698 (𝜔 = 1.9) 8.36 356.9554

Table 3. Comparison of numerical results at 𝑡 = 0.006

N Methods Number of Iterations CPU time 𝑇𝑚𝑎𝑥

CN-GS 31506 1.11 368.5538
30
CN-SOR 17892 (𝜔 = 1.7) 0.96 368.5538
CN-GS 120789 4.65 369.6896
60
CN-SOR 62369 (𝜔 = 1.9) 3.12 369.6896
CN-GS 264487 12.88 370.0674
90
CN-SOR 44286 (𝜔 = 1.9) 8.22 370.0674
CN-GS 388332 21.25 370.2562
120
CN-SOR 92442 (𝜔 = 1.9) 11.72 370.2562
CN-GS 573252 27.36 370.3694
150
CN-SOR 166122 (𝜔 = 1.9) 18.42 370.3694

N Methods Number of Iterations CPU time 𝑇𝑚𝑎𝑥

CN-GS 46424 1.55 369.1622
30
CN-SOR 25603 (𝜔 = 1.8) 1.36 369.1622
60 CN-GS 178336 6.95 370.3479
 CN-SOR 78988 (𝜔 = 1.9) 4.76 370.3479
CN-GS 329889 18.32 370.7430
90
CN-SOR 88504 (𝜔 = 1.9) 9.93 370.7430
CN-GS 450078 22.35 370.9406
120
CN-SOR 122598 (𝜔 = 1.9) 14.02 370.9406
CN-GS 506660 29.02 371.0591
150
CN-SOR 152454 (𝜔 = 1.9) 17.56 371.0591
Table 4. Comparison of numerical results at 𝑡 = 0.010

Graph 1. Number of iterations when 𝑡 = 0.002

Number of Iterations when t = 0.002

250000

200000

150000

100000

50000

0
30 60 90 120 150

GS SOR

Graph 2. Computational time when 𝑡 = 0.002

Computational time when t = 0.002

14 Number of Iterations when t = 0.006
12
1000000
10
8 500000
6
4 0
2 30 60 90 120 150
0 GS SOR
30 60 90 120 150

GS SOR
 Graph 3. Number of iterations when 𝑡 = 0.006

Number of Iterations when t = 0.006

700000
600000
500000
400000
300000
200000
100000
0
30 60 90 120 150

GS SOR

Graph 4. Computational time when 𝑡 = 0.006

Computational time when t = 0.006

30
25
20
15
10
5
0
30 60 90 120 150

GS SOR

Graph 5. Number of iterations when 𝑡 = 0.010

Number of Iterations when t = 0.010

600000
500000
400000
300000
200000
100000
0
30 60 90 120 150

GS SOR
 Graph 6. Computational time when 𝑡 = 0.010

Computational time when t = 0.010

40

30

20

10

0
30 60 90 120 150

GS SOR

According to the Table 2-4, at each of the mesh size, CN-GS and CN-SOR methods can give
the same maximum temperature which means that both methods deliver the good and same
performances to determine the peak junction temperature.
Refer to the Table 2-4, CN-SOR method can produce less number of iterations and
computational time compared to the CN-GS method.

15.7 Percentage Reduction Analysis

Refer to Table 5, for the percentage decrease calculation, it can be shown that, for each of the
elapsed time, the number of iterations can give high value of percentage means that the change
between two methods are large. Meanwhile, the percentage of computational time is at average
level mean that it has a slightly change between the two methods.
Table 5. Percentage decrease calculation
Elapsed time (s) Number of iteration (%) Computational time (%)
T = 0.002 33.72-65.02 7.57-43.59
T = 0.006 43.21-83.26 13.51-44.85
T = 0.010 44.85-73.17 12.26-45.80

15.8 Summary

In the study context, the Finite Difference Method (FDM) was applied to the one-dimensional
heat conduction equation. The Crank-Nicolson method was used to discretize the governing
heat conduction equation. This discretization will lead to the system of linear equation. The
GS and SOR methods were used to determine the peak junction temperature. Following the
results in Tables 2, 3 and 4, the implementation of SOR method improved the result obtained.
Moreover, the implementation of SOR iterative method significantly reduced the number of
iterations and computational time compared to the GS iterative method. Therefore, SOR
method proved to be better than GS method in solving heat conduction equation to determine
the peak junction temperature of semiconductor device.
References
[1] Muthuvalu, M. S., Asirvadam, V. S., & Mashadov, G. (2015, June 12). Performance
analysis of arithmetic mean method in determining peak junction temperature of
semiconductor device. Ain Shams Engineering Journal. Vol 6(4), pp. 1203-1210.
[2] Yangjun Z., Qinghai M., Xinghua Z. and Zhengsheng H., “A novel electrical
measurement method of peak junction temperature based on the excessive thermotaxis
effect of low current,” Chinese Journal of Semiconductors, vol. 30, (9), pp. 094005-1 –
094005-4, 2009.
[3] White M., Cooper M., Chen Y., and Bernstein J., “Impact of Junction Temperature on
Microelectronic Device Reliability and Considerations for Space Applications,” in 2003
IEEE International on Integrated Reliability Workshop Final Report, 2003, pp. 133-136.
[4] D. Noutsos Optimal stretched parameters for the SOR iterative method J. Comput. Appl.
Math., 48 (1993), pp. 293-308.
[5] D.J. Pierce, A. Hadjidimos, R.J. Plemmons Optimality relationships for p-cyclic SOR
Numer. Math., 56 (1990), pp. 635-643.
[6] R.S. Varga, W. Niethammer, D.Y. Cai p-cyclic matrices and the symmetric successive
overrelaxation method Linear Algebra Appl., 58 (1984), pp. 425-439.
[7] D.M. Young, Iterative Solution of Large Linear Systems, Academic Press, New York
(1971)
[8] Ammous A, Ghedira S, Allard B, Morel H, Renault D. Choosing a thermal model for
electrothermal simulation of power semiconductor devices. IEEE Trans Power Electron
1999;14(2): 300–7.
[9] Babajide Johnson, O., & Idowu Oluwaseun, L. (2020). Crank-Nicolson and modified
Crank-Nicolson scheme for one dimensional parabolic equation. International Journal of
Applied Mathematics and Theoretical Physics, 6(3), 35.
[10] Hailu, F., Gofe Gonfa, G., & Muleta Chemeda, H. (2020). Second degree generalized
successive over relaxation method for solving system of linear equations. Momona
Ethiopian Journal of Science, 12(1), 60–71.
[11] Youssef, I. K. 2012. On the successive overrelaxation method. Journal of Mathematics
and Statistics. Vol. 8(2), 176-184.
[12] Mayooran, T., & Light, E. 2016. Applying the successive over-relaxation method to a real
world problems. American Journal of Applied Mathematics and Statistics. Vol. 4, pp 113-
117.