Sandler, Steven M, "A Comparison of Tolerance Analysis Methods," AEi Systems LLC, 1998

A Comparison of Tolerance Analysis Methods
by Steven M. Sandler
AEi Systems, LLC.
We have seen many methods of calculating the worst case tolerance limits for electronic
circuits. The intent of this paper is to demonstrate several different methods, and
determine the results, and the corresponding confidence factors for each method.
The calculation methods addressed, and a brief description of each method is shown
below:
1. Extreme Value Analysis - Each component is varied in the direction of the
sensitivities to obtain the absolute worst case values of the circuit performance.
2. EVA Sensitivity Analysis - The parameter sensitivities are computed by evaluating
the derivative of the output with respect to each component. The algebraic sum of
the individual component tolerances (ie temperature, radiation, initial..) is multiplied
by the sensitivity to determine the voltage variance for the component. The voltage
variances of each part are summed algebraically to obtain the worst case circuit
performance.
3. RSS Sensitivity 1 - The parameter sensitivities are computed at the nominal values.
sum of each individual component tolerance (ie temperature, radiation, initial..) is
multiplied by the sensitivity to determine the voltage variance for the component.
The square root of the sum of the squares of each voltage variance is defined as the
worst case circuit performance
4. RSS Sensitivity 2 - The parameter sensitivities are computed at the nominal values.
The square root of the sum of the squares of each individual component tolerance (ie
temperature, radiation, initial..) is multiplied by the sensitivity to determine the
voltage variance for the component. The square root of the sum of the squares of
each voltage variance is defined as the worst case circuit performance.
5. Monte Carlo - The component tolerances are algebraically added and entered into a
SPICE simulator. The simulator randomly selects component values within the
specified tolerance range, following a 12 point gaussian distribution. The results of
the simulation include the population standard deviation, the population mean and
normally, the 3 sigma limits for the worst case circuit performance.
Copyright © 1998 – AEi Systems, LLC
All Rights Reserved
Proprietary and Confidential
Page 1
A simple circuit was selected to apply each of these methods to The confidence level of
each approach is defined later in this article in order to compare the results from each
method.
The circuit selected for this example is an LM117 linear regulator circuit. The schematic
of the circuit is shown in figure 1.
LM117
OUT
IN
ADJUST
LOAD
INPUT
124
374
Figure 1 - Simple Evaluation Circuit
For this simple case the following symmetrical tolerances are defined for each part:
Component Tolerances
Part
R1
R2
LM117 Rout
LM117 Ref
LM117 Iadj
Load Current
124
374
.00625
1.25
55 uA
0.75A
Copyright © 1998 – AEi Systems, LLC
All Rights Reserved
initial
1.00%
1.00%
100.00%
4.00%
7.00E-05
100.00%
temp
0.19
0.19
0
2
0
0
Proprietary and Confidential
age
0.05
0.05
0
2.64
0
0
Total
1.24%
1.24%
100.00%
8.64%
7.00E-05
100.00%
Page 2
Extreme Value Analysis
nominal voltage and sensitivity calculations
R2
374
V out
dIadj
dR1
dR2
R1
Vref
124
I o. R o . 1
1.25
R2
R1
R2
R2
Vref I o. R o .
2
R1
Vref I o. R o
R1
I o. 1
R2
dIo
R o. 1
R2
1
Iadj
Iadj. R2
6
55. 10
Ro
.00625
Io
0.75
V out = 5.022
dIadj = 374
dRo
dVref
Vref
Iadj
R1
R1
R2
R1
Copyright © 1998 – AEi Systems, LLC
All Rights Reserved
dR1 = 0.03
dR2 = 0.01
dRo = 3.012
dIo = 0.025
dVref = 4.016
Proprietary and Confidential
Page 3
Extreme Value Worst Case Maximum and Minimum Voltages
Maximum Voltage
R2
R1
V out
374. 1
124. 1
1.24
100
1.24
100
1.25. 1
Vref
Ro
.00625. 1
I o. R o . 1
Vref
R2
R1
8.64
100
Iadj
100
100
Io
( 55
70 ) . 10
0.75. 1
6
100
100
Iadj. R2
Extreme Value Minimum output voltage
V out = 5.604
Minimum Voltage
R2
R1
Vout
374. 1
124. 1
1.24
100
1.24
Vref
100
Vref
Ro
I o. R o . 1
1.25. 1
.00625. 1
R2
R1
8.64
100
Iadj
100
100
Io
( 55
70 ) . 10
0.75. 1
6
100
100
Iadj. R2
Extreme Value Minimum output voltage Vout = 4.423
Copyright © 1998 – AEi Systems, LLC
All Rights Reserved
Proprietary and Confidential
Page 4
EVA Sensitivity Analysis
Part
R1
R2
ROUT
VREF
IADJ
ILOAD
Vnominal
Value
Sensitivity
1.24E+02
-3.03E-02
3.74E+02
1.01E-02
6.25E-03
-3.05E+00
1.25E+00
4.02E+00
5.50E-05
3.74E+02
7.50E-01
-2.51E-02
Relative
-3.76E-02
3.78E-02
-1.91E-04
5.02E-02
2.06E-04
-1.88E-04
initial
1.00%
1.00%
100.00%
4.00%
7.00E-05
100.00%
5.023
temp
0.19
0.19
0
2
0
0
EVA Tol
age
0.05
0.05
0
2.64
0
0
Tol
Abs Value
1.24%
0.047
1.24%
0.047
100.00%
0.019
8.64%
0.434
7.00E-05
0.026
100.00%
0.019
0.591 Volts
RSS1 Sensitivity Analysis
Part
R1
R2
ROUT
VREF
IADJ
ILOAD
Vnominal
Value
1.24E+02
3.74E+02
6.25E-03
1.25E+00
5.50E-05
7.50E-01
Sensitivity
-3.03E-02
1.01E-02
-3.05E+00
4.02E+00
3.74E+02
-2.51E-02
Relative
-3.76E-02
3.78E-02
-1.91E-04
5.02E-02
2.06E-04
-1.88E-04
initial
1.00%
1.00%
100.00%
4.00%
7.00E-05
100.00%
temp
0.19
0.19
0
2
0
0
age
0.05
0.05
0
2.64
0
0
Tol
Abs Value
1.24%
0.047
1.24%
0.047
100.00%
0.019
8.64%
0.434
7.00E-05
0.026
100.00%
0.019
5.023
RSS Tol
0.440 Volts
RSS2 Sensitivity Analysis
Part
R1
R2
ROUT
VREF
IADJ
ILOAD
Vnominal
Value
1.24E+02
3.74E+02
6.25E-03
1.25E+00
5.50E-05
7.50E-01
Sensitivity
-3.03E-02
1.01E-02
-3.05E+00
4.02E+00
3.74E+02
-2.51E-02
Relative
-3.76E-02
3.78E-02
-1.91E-04
5.02E-02
2.06E-04
-1.88E-04
5.023
Copyright © 1998 – AEi Systems, LLC
All Rights Reserved
initial
1.00%
1.00%
100.00%
4.00%
7.00E-05
100.00%
temp
0.19
0.19
0
2
0
0
RSS2 Tol
Proprietary and Confidential
age
RSS Tol RSS Row
0.05
1.02% 3.83E-02
0.05
1.02% 3.85E-02
0 100.00% 1.91E-02
2.64
5.19% 2.61E-01
0 7.00E-05 2.62E-02
0 100.00% 1.88E-02
0.268973 Volts
Page 5
SPICE Monte Carlo Analysis
The SPICE circuit for this circuit is shown in figure 2.
1
ROUT 6.2M
V(5)
VOUT
5.03
VREF
1.25
5
5.02
R1
124
6
3
3.78
3.78
IADJ
55U
ILOAD
.75
R2
374
E1 10000
Figure 2 Spice Schematic
Spice Monte Carlo Netlist
F:\TEMP\lm117
.OP
.TRAN 1U 100U
.PRINT TRAN V(5)
*ALIAS V(5)=VOUT
VREF 1 3 1.25 TOL=8.64%
RTOP 5 6 124 TOL=1.24%
RBOT 6 0 374 TOL=1.24%
IADJ 0 6 55U TOL=70U
ROUT 1 5 6.25M TOL=6.25M
ILOAD 5 0 .75 TOL=.75
E1 3 0 3 6 10000
.END
Copyright © 1998 – AEi Systems, LLC
All Rights Reserved
Proprietary and Confidential
Page 6
Monte Carlo Results
The results of 100 cases, performed in a Monte Carlo SPICE simulation, are shown
below.
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
5.1924
5.0099
5.107
5.1488
5.0898
4.9969
5.0128
5.3198
4.8813
5.099
5.1541
5.3301
4.915
5.1745
5.1916
4.9378
4.9538
4.7793
5.0798
5.2465
4.7666
5.056
5.056
5.2808
4.9699
4.9986
4.9667
4.9621
5.1424
5.0145
4.9698
5.0635
4.7392
4.6917
5.0959
5.0351
4.8466
4.987
5.207
Copyright © 1998 – AEi Systems, LLC
All Rights Reserved
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
5.1717
5.1344
4.9889
5.0089
4.8358
4.8783
4.9972
5.0934
4.8432
5.1633
5.0917
4.7834
4.9136
4.8822
4.9711
5.1044
5.0693
5.0622
5.3137
4.8554
5.1819
4.9533
5.1175
4.9699
5.0142
5.2482
5.2886
5.0802
4.9757
5.0127
4.9546
5.0153
5.1374
5.1161
5.0907
4.9393
5.1218
4.9501
5.1876
Proprietary and Confidential
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
Mean
Pop Stdev
5.1839
5.0727
5.1752
4.8133
4.9281
5.0251
4.8478
4.8168
4.9089
5.1033
5.0903
4.9508
5.1078
4.7521
5.1305
4.8451
5.2143
4.9566
5.3247
4.758
5.1846
4.6512
5.032088
0.14441
Page 7
Monte Carlo Histogram Results
40.00
Number of Number per Cell
30.00
20.00
10.000
x 4.579
< 1.000
x 5.397
< 0
>
>
1
0
4.434
4.582
4.731
4.880
5.028
5.177
5.325
5.474
5.623
.5 Sigma Cells
∆ x = 817.3M ∆ y = -1.000
Copyright © 1998 – AEi Systems, LLC
All Rights Reserved
Proprietary and Confidential
Page 8
Statistical Evaluation
The results of the Monte Carlo Analysis yield a population mean and a population
standard deviation. In order to determine the Extreme Value Worst Case circuit
performance we need to select a confidence level. Selecting a confidence level of
0.99998, meaning that we have 99.998 percent confidence that any device will remain
within these limits, we can define the number of standard deviations from the mean.
EXCEL was used to compute the confidence results.
The resulting number of standard deviations for a confidence level of 0.99998 is 4.265.
The resulting minimum and maximum values can be computed as
V min = mean − (4.265 * σ )
V max = mean + (4.265 * σ )
The results of the Monte Carlo analysis provided a population mean of 5.032 volts with a
population standard deviation of 0.14441 volts.
This results in a maximum of 5.648 volts and a minimum of 4.416 volts.
Comparative Results
The table below shows the mean, minimum, and maximum output voltages, the effective
number of standard deviations from the mean and the respective confidence level.
Method
Extreme Value
EVA Sensitivity
RSS1
RSS2
Monte Carlo
Mean
5.013
5.023
5.023
5.023
5.032
Minimum
4.423
4.432
4.583
4.754
4.416
Maximum
5.604
5.614
5.463
5.292
5.648
# of STDEV
Confidence
4.414*
4.092
3.048
1.863
4.265
100%
99.996
99.770%
93.750%
99.998
* 4.414 yields a confidence of 99.999%
Conclusions
Different circuits will result in different tolerances. This paper merely demonstrates the
relative performance of each of the methods. It does show that Monte Carlo may be a
Copyright © 1998 – AEi Systems, LLC
All Rights Reserved
Proprietary and Confidential
Page 9
reasonable method of determining the Extreme Value performance, if it is combined with
a confidence level for the result.
Copyright © 1998 – AEi Systems, LLC
All Rights Reserved
Proprietary and Confidential
Page 10