Machine Learning Based Power Estimation for

CMOS VLSI Circuits

V. Govindaraj (  [email protected] )
Dr NGP Institute of Technology
B. Arunadevi
Dr NGP Institute of Technology

Research Article

Keywords: power estimation, Random Forest, BPNN, CMOS, VLSI

DOI: https://doi.org/10.21203/rs.3.rs-723965/v1

License:   This work is licensed under a Creative Commons Attribution 4.0 International License.
Read Full License
MACHINE LEARNING BASED POWER ESTIMATION FOR CMOS
VLSI CIRCUITS

Dr V.GOVINDARAJ* and Dr B.ARUNADEVI1

*Assistant Professor, Dr NGP Institute of Technology, Coimbatore, Tamilnadu, India

Author Email: [email protected]
1
Professor, Dr NGP Institute of Technology, Coimbatore, Tamilnadu, India.

single chip. Increase in number of transistor and

Abstract: operating speed in a chip and speed due which
ultimately increase the power consumption and has
Nowdays, machine learning (ML) algorithms are
become a crucial concern in submicron technology.
receiving massive attention in most of the engineering
application since it has capability in complex systems In VLSI circuits power estimation at an earlier
modelling using historical data. Estimation of power for stage is highly needed, because it has a major
CMOS VLSI circuit using various circuit attributes is impact on the reliability of VLSI circuits. Under
proposed using passive machine learning based this condition, at the higher levels of design
technique. The proposed method uses supervised abstraction average power estimation before a chip
learning method which provides a fast and accurate manufacturing is very much essential to calculate
estimation of power without affecting the accuracy of
power budget and to take necessary steps to reduce
the system. Power estimation using random forest
power consumption. Hence less complex and low
algorithm is relatively new. Accurate estimation of
cost power estimation techniques is needed.
power of CMOS VLSI circuits is estimated by using
random forest model which is optimized and tuned by
using multi-objective NSGA-II algorithm. It is inferred 1.1 Previous work in power
from the experimental results testing error varies from
estimation
1.4 percent to 6.8 percent and in terms of and Mean
Square Error is 1.46e-06 in random forest method when Power dissipation of the circuit depends on the
compared to BPNN. Statistical estimation like inputs, frequency and operating voltage.
coefficient of determination (𝑅) and Root Mean Square Simulation and non simulation based method
Error (RMSE) are done and it is proven that random are two main categories to estimate average
Forest is best choice for power estimation of CMOS power. Amuru et al (2020) proposed an
VLSI circuits with high coefficient of determination of
estimation technique for leakage power, delay
0.99938. and low RMSE of 0.000116.
in standard CMOS/FinFET digital cells.
Bhanja and Ranganathan (2003) explained about
Keywords: power estimation; Random Forest ;
switching activity estimation using Bayesian
BPNN; CMOS;VLSI
Networks considering both in internal nodes and
inputs. Burch et al. (1993) and saxena et al

1. Introduction (1997) proposed Simulation based method

which includes Monte Carlo approach.
Due to recent development in VLSI
Estimate the power for combinational and
technology million of transistors are fabricated in a
 sequential circuits was discussed by Buyukus Estimation of Cumulative Pollution Index of
et al (2006) and Kozhaya et al (2001). Insulator Strings Leakage using random forest
Adaptive neuro fuzzy application to power was proposed by de santos et al[2020] . Metal
estimation was discussed by Govindaraj et al Oxide Surge Arrester based surface condition
(2018). Kirei et al(2019) explained about identification using random forest is done by
estimation of power in CMOS integrated Das AK et al [2020]. Impedance estimation in
circuits with discrete time filters in which power system using machine learning is
filter of various complexities are considered proposed by Liang et al [2020]. Pilarski et al
and power and area estimates are obtained . 2012 , S. Seyedzadeh et al 2018 Peng, H., et
Ligang Hou et al. (2006) proposed a method al 2017 developed human and robot
using Levenberg-Marquardt function to integration, Monitoring of electric power grid
estimate power using Neural Network. and building energy management systems
Murugavel et al (2002) proposed a method using using machine learning. Air Pollution
Least square estimation to estimate average measurement using Machine Learning
power which minimize MSE through each Techniques was discussed by srivatsava et
iteration when compared to Monte Carlo al[2018] . A perspective on machine learning
approach. Under Probabilistic class, Najm et in turbulent flows was discussed by Sandeep
al. (1993) devised an algorithm to propagate Pandey et al (2020) similarly kanchan et
of transition density values from the primary al(2020) proposed Medical Internet of things
input to the primary output. BPNN and Radial using machine learning algorithms for lung
Basis Function Neural Network (RBFNN) based cancer detection.
power estimation of ISCAS’89 sequential
Benchmark circuits have been performed by Machine Learning models will discover
Ramanathan et al. (2013). Only two training the relationship between input variables and
functions Traingdm and Traingdx were used outputs of interest from the system being
for analysis out of eleven different training studied or learn from measured data or
algorithms available in MATLAB tool. Omnia simulated data that represents the physical
S et al (2016) discussed about estimation of problem. Nowadays, machine learning (ML)
dynamic power consumption using circuit models are receiving enormous attention in
nodes and logic picture for CMOS most of the fields due to their capability in
combinational logic circuits A. K. Shaw modeling complex systems using historical
Halwai(2018) proposed a method to estimate data. The use of data-driven models have been
power dissipation of phase detector circuit. successfully demonstrated in applications
demanding for real-time estimation of the

1.2 Related work using Random targets values. The proposed work employs an
Random forest algorithm which has the
Forest
ability to estimate the power of CMOS VLSI
Chao Chen et al (2020) discussed about
circuits, without the knowledge on actual
classification of neural activities , brain-
circuit structure and interconnections. The
Computer Interfaces, classification of finger
Random forest algorithm results are compared
gestures during motor execution and imagery.
 with BPNN results. Error percentage for from the literature (Ligang Hou 2006) which is
BPNN and Random Forest is calculated to shown in Tables 1 and 2. BPNN network is trained
find the deviation from actual power to by using database of 20 Benchmark ISCAS’89
predicted power in which Random Forest sequential circuits and 5 the same circuits are used
outperforms BPNN. for testing purpose. The training and testing
consists of attributes such as considered for

2. Training and testing data sequential circuits are number of inputs, outputs, D
flip-flops, inverters, total number of gates, AND

The database used for training and testing the gates, NAND gates, OR gates and NOR gates.

BPNN and random forest network is obtained

Table 1. ISCAS’89 benchmark circuit data set for training BPNN/RF (Ligang Hou et al.
2006)

BENCHMARK Monte Carlo Simulation

GATE AND INV NOR NAND OR DFF IN OUT
CIRCUIT power in mw.

S208 66 21 38 16 15 14 8 10 1 0.00698
S298 75 31 44 19 9 16 14 3 6 0.00912
S349 104 44 57 31 19 10 15 9 11 0.01856
S386 118 83 41 0 0 35 6 7 7 0.0162
S400 106 11 58 34 36 25 21 3 6 0.01065
S420 160 49 78 34 29 28 16 18 1 0.00903
S444 119 13 62 34 58 14 21 3 6 0.01172
S713 139 94 254 0 28 17 19 35 23 0.03743
S820 256 76 33 66 54 60 5 18 19 0.02831
S838 288 105 158 70 57 56 32 34 1 0.01292
S953 311 49 84 112 114 36 29 16 23 0.02458
S1238 428 134 80 57 125 112 18 14 14 0.06347
S1423 490 197 167 92 64 137 74 17 5 0.07181
S1494 558 354 89 0 0 204 6 8 19 0.06018
S5378 1004 0 1775 765 0 239 179 35 49 0.23357
S9234 2027 955 3570 113 528 431 228 19 22 0.28004
S15850 3448 1619 6324 151 968 710 597 14 87 0.51991
S35932 12204 4032 3861 0 7020 1152 1728 35 320 1.22048
S38417 8709 4154 13470 2279 2050 226 1636 28 106 1.14518
S38584 11448 5516 7805 12 278 1452 2621 1185 1.87987
Table 2. ISCAS’89 benchmark circuit data set for testing BPNN/RF (Ligang Hou et al.
2006)
BENCHMARK
GATE AND INV NOR NAND OR DFF IN OUT
CIRCUIT
S344 101 44 59 30 18 9 15 9 11
S382 99 11 59 34 30 24 21 3 6
S641 107 90 272 0 4 13 19 35 24
S1488 550 350 103 0 0 200 6 8 19
S13207 2573 1114 5378 98 849 512 669 31 121
3. Power estimation using BPNN 3.2 Training phase
The method consists of the following steps. 1: Two-third of the input vectors from ISCAS’89
benchmark circuits database are extracted and to
 Construction of neural network train the BPNN (Harris et al 1994).
 Training phase 2: Normalization is done for input vectors and their
 Testing phase corresponding target vectors . Since the the hidden
3.1 Construction of neural network layers has tan-sigmoidal activation function
normalization is done between the −1 to +1.
A back-propagation neural network is constructed 3: The BPNN network is trained with the
normalized input vectors and their corresponding
with four-layer feed-forward. First layer is set with normalized target vectors .
‘linear’ transfer function and ‘tansig’ function is
chosen for remaining layers. BPNN network 3.3 Testing phase
Parameters such as learning rate, epoch , hidden
1: One-third of the input vectors from ISCAS’89
layer and momentum constant are varied for benchmark circuits database are used for testing
different training algorithms such as Traingd, (Harris et al 1994).

Trainscg Traingda, Trainbfg,, Trainrp, Trainoss, 2: Before testing, the parameters used for input test
vectors are normalized.
Traingdx, Trainsgf, Traingdm, Traincgp and
3:Outputs vectors are generated for these
Traincgb. The data base of sequential circuits normalized test input vectors by the BPNN
consists of nine attributes, therefore number of network.

inputs for the network is considered is nine. The 4: Original value is obtained from Reverse

generalised architecture of BPNN is shown in normalization process.

Figure 1

f1(e)
x1
x2 y1
f1(e)
y2
x3
fX(e)

fn(e) yn
xr
Output
Input fH(e) signal
signal
Hidden layer f(e) - Activation function

Figure 1 Architecture of BPNN

 The work flow of RF Algorithm is shown
4 Proposed Random forest based
in Figure 2 in which training is done by using two
power estimation
third of system dataset and then training algorithm
In the proposed method, a Random Forest (RF) and training parameters are chosen. Maximum
model is preferred due to its ability to predict depth is varied from 10 to 15 ,number of trees from
multiple output values Simultaneously. Until one 150 to 750, neighbor order is 1.The parameter
record remain as subset RF divide the records in to neighbor order gives the relationship between
smaller and smaller subset since RF is an ensemble adjacent subcells. The neighbor condition can be
of randomized decision trees. The nodes and leaf measured by the Neighbor-order parameter.
nodes are called as inner and final sets. RF needs a
number of numbers of hyper-parameters to be set
3.1 Steps involved in RF algorithm
and to achieve maximum accuracy, these
1. From the training dataset k data points are
parameters are tuned as per simulated data to get pick in random.
accurate power estimation of CMOS VLSI Circuits
2. Using the k data points a decision tree
.. The constructed model is trained and tested using associated with data points is constructed.
10-fold cross-validation. Implementation is done 3. Step 1 and 2 are repeated to build N
number N of trees .
using Python programming language, and the
4. By using each one of tree in N –tree, Output
proposed work is carried out on a Computer with
value Y corresponding to a new data point is
Intel Core i7-6700 3.4GHz CPU, 16GB RAM. determined and average of all predicted Y
values is calculated and final Y value is
identified.

Figure 2 Work flow of RF Algorithm

5. Results and discussion Momentum constant, activation function and
training algorithm as four important parameter.
BPNN is a feed-forward back propagation type Momentum constant is varied between 0.1 to 0.9,
neural network consists of Learning rate, number of epochs is varied between 150 to 2700,
hidden layer neurons are selected between 10 to 17, gives Regression results comparison of ISCAS’89
0.3 to 0.8 is the range of variation for learning rate benchmark circuits using BPNN for 11 different
for 11 different training algorithm and activation training functions.
functions are chosen as tansig and logsig . Table 3
Table 3 Regression results comparison of ISCAS’89 benchmark circuits using BPNN
Number of Regression Deviation
Training Layer Y-
Epochs Input Slope MSE
Function Size intercept Value in %
Attributes
Trainscg 9:15:15:1 253 9 1 -0.0034 0.9999 0.01 6.254E-05
Traincgp 7:15:15:1 300 9 1.8 -0.025 0.9988 0.12 0.02672
Traincgb 9:16:15:1 100 9 1.7 -0.023 0.9993 0.07 0.02403
Trainoss 8:14:15:1 600 9 0.99 0.0017 0.9991 0.09 0.01924
Trainbfg 9:15:15:1 500 9 1.4 -0.016 0.9993 0.07 0.01793
Traingd 9:15:15:1 750 9 1.8 -0.027 0.9940 0.6 0.014096
Traingdm 8:13:14:1 400 9 0.015 0.0016 0.9920 0.8 0.02485
Traingdx 8:14:15:1 225 9 0.34 0.018 0.9979 0.21 0.01602
Traincgf 7:15:15:1 256 9 1.1 -0.0036 0.9990 0.1 0.00528
Traingda 9:15:14:1 500 9 0.58 0.011 0.9956 0.44 0.025339
-5
Trainrp 8:16:15:1 250 9 1 -1.6x10 0.9998 0.02 0.06346

Trainscg training function which is based other 11 training algorithms. Trainscg under
on conjugate gradient algorithm provides better conjugate gradient category is best suited for power
results for sequential circuits when compared to estimation for sequential circuits for a layer size of
other training algorithms. From the regression 9:15:15:1 with 9 inputs and 253 epochs. From the
analysis it is found that the results for ISCAS’89 Figure 3, it is inferred that trainscg predicts power
sequential benchmark circuits deviates from ideal very close to power estimator tool than remaining
power by 0.01% to 0.8%. The trainscg training training algorithm.
function gives the minimum regression value among

1,002
1
Regression Value

0,998
0,996
0,994
0,992
0,99
0,988

Training Algorithms

Figure 3 ISCAS’89 benchmark circuits regression result comparison using BPNN for 11
training algorithms
Table 4 Mean square error of ISCAS’89 Benchmark circuits using BPNN/RF
BPNN (MSE) Random Forest (MSE)
Benchmark
Actual Output Predicted Output Squared Error Predicted Output Squared Error
Circuits

7.056E-07 0.0182 6.76E-08

S344 0.01846 0.0193
5.098E-05 0.0103 2.56E-08
S382 0.01046 0.0176
3.686E-06 0.0153 9.604E-07
S386 0.01628 0.0182
4.84E-08 0.0104 6.25E-08
S400 0.01065 0.01087
0.0001367 0.0338 6.2E-06
S641 0.03629 0.0246
3.842 E-05 1.46E-06
Mean Square Error

Predicted power output and Mean square error of this shows ML models discover relationships
ISCAS’89 Bench mark circuit is shown in Table 4, between input variables and outputs of interest from
which is obtained by comparing actual circuit output the system being studied, learn from measured or
with neural network. and random forest predicted simulated data. Figure 4 Comparison of ISCAS’89
output. MSE is calculated for both BPNN and RF. benchmark circuits Actual output with RF and
MSE of RF is1.46E-06 whereas BPNN is 3.842E-05 BPNN predicted output.

0,04
0,035
0,03
Power in mw

0,025
0,02 Actual Output

0,015 BPNN output

0,01 RF OUTPUT

0,005
0
S344 S382 S386 S400 S641
ISCAS'89 Circuits

Figure 4 Comparison of ISCAS’89 benchmark circuits Actual output with RF and BPNN
predicted output.
5.1 Prediction Error
Training and testing prediction error is calculated using equation (1)

𝐴 − 𝐵 𝑋 100
𝐸𝑟𝑟𝑜𝑟 % = (1)
𝐴

where A is the actual value and B is the predicted output value by testing the network.

Prediction error comparison of BPNN with because it can deal with regression and classification

RF is reported in table 5. It is infer that RF problems of multiclass, small sample data , and

prediction error is varying from 1.4% to 6.8 % but without data pre processing procedures. Figure 5

BPNN prediction error is varying from 4.5% to gives Comparison of Prediction error of RF and

68.2% . RF prediction error is less than BPNN BPNN in which RF outperforms BPNN.

Table 5 Error calculation for BPNN and RF

Benchmark Circuit BPNN Error in (%) Random Forest Error in (%)

4.55038 1.408451
S344
68.26 1.529637
S382
11.7936 6.019656
S386
2.06573 2.347418
S400
32.21273 6.861394
S641

80
70
60
50
Error in %

40
BPNN (%)
30
Random Forest(%)
20
10
0
S344 S382 S386 S400 S641
ISCAS'89 Benchmark Circuits

Figure 5 Comparison of Prediction error of RF and BPNN

 5.2 Statistical Analysis of BPNN and RF Mean Square Error (RMSE) and Coefficient of
determination (R) which is given below in
Performance of RF and BPNN network can be
Equation (2) – (3).
determined by doing statistical analysis such Root

∑𝑁 𝑂
𝐼=1(𝑌𝐼 − 𝑌𝐼
𝐶 )2
𝑅𝑀𝑆𝐸 = √ (2)
𝑁

𝑂 𝐶
∑𝑁 𝑂 𝐶
𝐼=1(𝑌𝐼 − 𝑌 )(𝑌𝐼 − 𝑌 )
𝑅= (3)
√∑ 𝑁 𝑂 𝑁 𝑂 2 𝐶 𝐶 2
𝐼=1(𝑌𝐼 − 𝑌 ) ∑𝐼=1(𝑌𝐼 − 𝑌 )

coefficient of determination 𝑅 determines a linear

where YO is the mean of observed value , YIO is
correlation between measured values and values
the observed value YC is the mean of calculated
simulated by the model, optimal value is 1. RMSE
value. and YIC is the calculated value. RMSE is
and R for BPNN and ANFIS is shown in Table 6,
used as a measure to find the difference between
from the table we infer that RMSE of RF is less
predicted values and measured values. RMSE can
when compared to BPNN and R value of RF is very
be used as an ndicator of model accuracy or
close to 1. Considering these measures, the RF
precision. A good network /model should has
model achieves high accuracy for power estimation.
RMSE value low, or as close to zero. The

Table 6 Statistical analysis test on BPNN and RF

Parameter BPNN Random Forest
RMSE 0.0004499 0.000116

R 0.84696 0.99938

6. Conclusion
In this paper, a supervised learning method to interconnections are not required. Random Forest is
estimate the Power of CMOS VLSI circuits is less computationally than BPNN. A random forest
presented. The proposed RF model is an alternative for a decision tree will give a different
approach to the conventional techniques like SPICE interpretation but with better performance on the
circuit simulation which are based on the other hand BPNN requires much more data for the
assumption of predefined empirical equations and result to be effective. Hence, RF with exclusive
arbitrary parameters. The results of RF are highly characteristics appears as a better choice for
accurate and the details of circuit structure and estimation of power in CMOS VLSI circuits. It is
proven that by using statistical estimation like Conflict of Interest: Author don’t have
coefficient of determination and Root Mean Square
any conflict of interest in submitting the
Error (RMSE), RF performs well for power
paper to this journal.
estimation application with high coefficient of
determination of 0.99938 and low RMSE of
0.000116.

REFERENCES
[1] Amuru, M. S. Ahmed and Z. Abbas (2020). An Efficient Gradient Boosting Approach for PVT Aware Estimation of Leakage
Power and Propagation Delay in CMOS/FinFET Digital Cells. IEEE International Symposium on Circuits and Systems (ISCAS),
Sevilla, 2020, pp. 1-5, doi: 10.1109/ISCAS45731.2020.9180600
[2] Bai, Y. Li, X. Wang, J. Xie and C. Li (2016).Air pollutants concentrations forecasting using back propagation neural network based
on wavelet decomposition with meteorological conditions", Atmos. Pollut. Res., vol. 7, no. 3, pp. 557-566, 2016
[3] Bhanja, S., & Ranganathan, N. (2003). Switching activity estimation of VLSI circuits using Bayesian networks. IEEETransactions
on Very Large Scale Integration Systems, 11(4), 558–567. doi:10.1109/TVLSI.2003.816144
[4] Burch, R., Najm, F. N., Yang, P., & Trick, T. N. (1993). A Monte Carlo approach for power estimation. IEEE Transactions on Ve ry
Large Scale Integration Systems, 2(1), 63–71. doi:10.1109/92.219908
[5] Buyuks, M. K., & Najm, F. N. (2006). Early power estimation for VLSI circuits. IEEE Transactions on Computer Aided Design of
Integrated Circuits and Systems, 24(7), 1076–1088. doi:10.1109/TCAD.2005.850904.
[6] Chao Chen, Peiji Chen, Abdelkader Nasreddine Belkacem, Lin Lu, Rui Xu, Wenjun Tan, Penghai Li, Qiang Gao, Duk Shin,
Changming Wang & Dong Ming (2020) .Neural activities classification of left and right finger gestures during motor execution and
motor imagery, Brain-Computer Interfaces. DOI: 10.1080/2326263X.2020.1782124.
[7] de Santos H and M. Á. Sanz-Bobi(2020).A Cumulative Pollution Index for the Estimation of the Leakage Current on Insulator
Strings. IEEE Transactions on Power Delivery, vol. 35, no. 5, pp. 2438-2446, doi: 10.1109/TPWRD.2020.2968556.
[8] Das A K, S. Dalai and B. Chatterjee, "Cross Stockwell transform aided Random Forest based surface condition identification of
Metal Oxide Surge Arrester employning Leakage current signal," 2020 IEEE Region 10 Symposium (TENSYMP), 2020, pp. 1775-
1778, doi: 10.1109/TENSYMP50017.2020.9230802.
[9] Govindaraj, V & Ramesh, J (2018). Adaptive neuro fuzzy inference system-based power estimation method for CMOS VLSI
circuits. International Journal of Electronics, Taylor and Francis,, vol. 105, no. 3, pp. 398-411.
[10] Harris, C. J. (1994). Advances in intelligent control. Boca Raton, FL: CRC Press. ISBN 9780748400669.
[11] Hou, L., Zheng, L., & Wu, W. (2006). Neural Network Based VLSI Power Estimation. 2006 8th International Conference on Solid-
State and Integrated Circuit Technology Proceedings. doi:10.1109/icsict.2006.306506
[12] Kanchan Pradhan & Priyanka Chawla (2020) .Medical Internet of things using machine learning algorithms for lung cancer
detection. Journal of Management Analytics, 7:4, 591-623, DOI: 10.1080/23270012.2020.1811789
[13] Kirei, S C. Farcas and M. D. Topa (2019). Power and Area Estimation of Discrete Filters in CMOS Integrated Circuits. Signal
Processing: Algorithms, Architectures, Arrangements, and Applications (SPA), Poznan, Poland, 2019, pp. 67 -70, doi:
10.23919/SPA.2019.8936762.
[14] Klein, F., Leao, R., Araujo, G., Santos, L., & Azevedo, R. (2009). A multi-model engine for high-level power estimation accuracy
optimization. IEEE Transactions on Very Large Scale Integration Systems, 17(5), 660–673. doi:10.1109/ TVLSI.2009.2013627
[15] Kle K. Shaw Halwai (2020). Estimation of Power and Delay of CMOS Phase Detector and Phase-Frequency Detector Using Nano
Dimensional MOS Transistor. IEEE VLSI DEVICE CIRCUIT AND SYSTEM (VLSI DCS), Kolkata, India, 2020, pp. 294-298,
doi: 10.1109/VLSIDCS47293.2020.9179921.
[16] Kozhaya, J. N., & Najm, F. N. (2001). Power estimation for large sequential circuits. IEEE Transactions on Very Large Scale
Integration Systems, 9,no(2), 400–407. doi:10.1109/ICCAD.1997.643581.
[17] Liang, H. Guo and T. Zheng(2019).Real-Time Impedance Estimation for Power Line Communication. IEEE Access, vol. 7, pp.
88107-88115, 2019, doi: 10.1109/ACCESS.2019.2925464.
[18] Murugavel, A. K., Ranganathan, N., Chandramouli, R., & Chavali, S. (2002). Least-square estimation of average power in digital
CMOS circuits. IEEE Transactions on Very Large Scale Integration Systems, 10(1), 55–58.doi:10.1109/92.988730
[19] Peng, H., Lima, A.R., Teakles, A (2017). Evaluating hourly air quality forecasting in Canada with nonlinear updatable machine
learning methods. Air Qual. Atmos. Hlth. No. 10, Issue 2, pp 195–211, March 2017.
[20] Pilarski, M. R. Dawson, T. Degris, J. P. Carey, and R. S. Sutton (2012).Dynamic switching and real-time machine learning for
improved human control of assistive biomedical robots. 4th IEEE RAS & EMBS International Conference on Biomedical Robotics
and Biomechatronics (BioRob), Rome, Italy, June 2012, pp. 296-302.
[21] Ramanathan, P., Surendiran, B., & Vanathi, P. T. (2013). Power estimation of benchmark circuits using artificial neural networks.
Pensee Journal, 75(9), 427–433.
[22] Saxena, V., Najm, F. N., & Hajj, I. N. (1997,March 17–20). Monte-Carlo approach for power estimation in sequential circuits.
European Design and Test conference, (ED&TC 97), pp. 416–420. doi:10.1109/EDTC.1997.582393
[23] Sandeep Pandey, Jörg Schumacher & Katepalli R. Sreenivasan (2020) .A perspective on machine learning in turbulent
flows. Journal of Turbulence, 21:9-10, 567-584, DOI: 10.1080/14685248.2020.1757685.
[24] Seyedzadeh, F. Pour Rahimian, P. Rastogi, and I. Glesk (2020).Tuning machine learning models for prediction o f building energy
loads. Sustainable Cities and Society, 47, p. 101484, 2019/05/01/ 2019.
[25] Seyedzadeh, F. P. Rahimian, I. Glesk, and M. Roper, "Machine learning for estimation of building energy consumption and
performance: a review," Visualization in Engineering, 6, 5.
[26] Srivastava, S. Singh and A. P. Singh (2018).Estimation of Air Pollution in Delhi Using Machine Learning Techniques.
International Conference on Computing, Power and Communication Technologies (GUCON), 2018, pp. 304-309, doi: 10.1109