System and method for modeling stochastic behavior of a system of
View Patent ↗A system and method for modeling stochastic behavior of a system of N similar statistical variables using N uncorrelated/independent random model parameters. More particularly, a system and method of modeling device across chip variations and device mismatch. The method includes modeling stochastic behavior of a system of N similar statistical variables using N uncorrelated/independent random model parameters. The method includes providing a system of N similar statistical variables, wherein each stochastic variable has a same standard deviation. The method further includes partially correlating each and every pair of stochastic variables among N variables, wherein a degree of partial correlation is a same for all pairs of variables. A statistical model is constructed to represent a system of N stochastic variables in which only N independent stochastic model parameters are used. A one-to-one mapping relation exists between N model parameters and the N variables. The method further includes finding unique values of the N model parameters given a set of values of the N variables. Reversely, the method also includes finding the values of the N variables given a set of values of the N model parameters.
1. A method of modeling stochastic behavior of a system of N similar statistical variables using N uncorrelated/independent random model parameters, comprising:
providing a system of N similar statistical variables, wherein each stochastic variable has a same standard deviation, and N is equal to or greater than two;
partially correlating each and every pair of stochastic variables among N variables using a computing processor, wherein a degree of partial correlation is a same for all pairs of variables and there are N(N−1)/2 pairs;
constructing a statistical model to represent a system of N stochastic variables in which only N independent stochastic model parameters are used using the computing processor;
establishing a one to one mapping relation between N model parameters and the N variables;
finding unique values of the N model parameters given a set of values of the N variables; and
finding values of the N variables given a set of values of the N model parameters.
2. The method of claim 1 , wherein the system of N similar statistical variables include various VLSI circuit examples, semiconductor device examples and semiconductor device model examples.
3. The method of claim 1 , further-comprising partially correlating each and every pair of stochastic variables among N variables, where a degree of partial correlation is a same for all pairs of variables, where there are N(N−1)/2 pairs.
4. The method of claim 1 , further comprising providing an irreducible representation of N-instance statistical model, wherein the irreducible representation provides a one-to-one mapping relation between instance values and their representation in statistical simulation space.
5. The method of claim 1 , wherein a mismatch amount is related to the linear correlation coefficient r ij through a relation
( F i −F j ) 2 =2σ F 2 (1 −r ij ), i≠j, i,j= 1, 2, 3, . . . .
wherein F 1 , F 2 , . . . , F N are the N statistical variables, and σ F is a standard variation of each of N stochastic variable.
6. The method of claim 1 , wherein the statistical model reproduces required at least one-body statistics,
( F i −f 0 ) 2 1/2 =σ F , i= 1, 2 , . . . , N,
and two-body statistics,
( F i −F j ) 2 1/2 =√{square root over (2)}σ r , i≠j, i,j= 1, 2, 3 , . . . , N.
7. The method of claim 1 , wherein the mapping is given by:
(
F
1
F
2
F
3
⋮
F
N
)
=
f
0
(
1
1
1
⋮
1
)
+
U
N
×
N
(
N
σ
sym
g
1
σ
r
g
2
σ
r
g
3
⋮
σ
r
g
N
)
,
σ
sym
=
σ
F
2
-
(
1
-
1
N
)
σ
r
2
,
U
N
×
N
=
[
U
ij
]
N
×
N
=
(
1
N
1
2
·
1
1
3
·
2
⋯
1
N
(
N
-
1
)
1
N
-
2
-
1
2
·
1
1
3
·
2
⋯
1
N
(
N
-
1
)
1
N
0
-
3
-
1
3
·
2
⋯
1
N
(
N
-
1
)
⋮
⋮
⋮
⋱
⋮
1
N
0
0
⋯
-
N
-
1
N
(
N
-
1
)
)
,
(
1
)
wherein matrix elements are
U i1 =1 /√{square root over (N)}, i= 1, 2 , . . . , N ;
U
jj
=
-
j
-
1
j
(
j
-
1
)
,
j
=
2
,
3
,
…
,
N
;
U
ij
=
0
,
j
=
2
,
3
,
…
,
N
,
i
>
j
;
and
U
ij
=
1
j
(
j
-
1
)
,
j
=
2
,
3
,
…
,
N
,
i
<
j
;
wherein F 1 , F 2 , . . . , F N are the N statistical variables, and, f 0 is the average value of variable F i ,i=1, 2, . . . , N, g 1 , g 2 , . . . , g N are the N stochastic model parameters which provide the mapping from N model parameters to N variables; and
the matrix U N×N is an N-by-N normal-orthogonal matrix; and
any variable is similar to any other variables, an exchange of any two rows of the U N×N matrix; and further comprising:
multiplying all elements of i th column of the U N×N matrix by −1 which corresponds to replacing g i by −g i equation (1).
8. The method of claim 1 , wherein the modeling method uniquely determines the values of N model parameters given the values of the N stochastic variables,
(
g
1
g
2
g
3
⋮
g
N
)
=
1
σ
r
(
σ
r
/
(
N
σ
sym
)
0
0
⋯
0
0
1
0
⋯
0
0
0
1
0
⋮
⋮
⋮
⋱
⋮
0
0
0
⋯
1
)
U
N
×
N
T
(
F
1
-
f
0
F
2
-
f
0
F
3
-
f
0
⋮
F
N
-
f
0
)
,
where U N×N T is the transpose of the matrix U N×N .
9. The method of claim 1 , wherein each of the N stochastic variable is an independent stochastic variable, and the N independent stochastic variables form an N-dimensional space.
10. The method of claim 1 , further comprising providing a mismatch problem for at least one of VLSI design, macro models for VLSI logic delay circuits, semiconductor device design and semiconductor device modeling, which includes contributions from both correlated variations and uncorrelated variations.
11. The method of claim 10 , further comprising treating N instances of a device as N similar statistical variables, treating N stages of a VLSI delay circuit as N similar statistical variables, and treating N instances of a device model parameter as N similar statistical variables, and providing the one-to-one mapping relation between instance values and their representation, wherein the modeling method includes N independent stochastic variables, and is a set of stochastic functions defined in an N-dimensional space.
12. The method of claim 10 , wherein each and every stochastic variable of all examples has a same standard deviation, which includes contributions from both correlated variations and uncorrelated variations.
13. The method of claim 1 , wherein the modeling is at least a statistical model providing the one-to-one mapping relation and the one-to-one mapping relation is between instance values and their representation.
14. The method of claim 13 , wherein at least one of: (i) the statistical model in VLSI circuit simulation is a Monte Carlo model for Monte Carlo simulations of VLSI circuits and the Monte Carlo model represents g 1 , g 2 , . . . , g N by N independent Gaussian distributions; and (ii) the statistical model in VLSI circuit simulation is a corner model in which each of g 1 , g 2 , . . . , g N is an independent skewing parameter for corner simulation of VLSI circuits.
15. A computer program product comprising a computer non-transitory usable storage medium having readable program code embodied in the medium, the computer program product includes at least one component with the readable program code executable to:
provide a system of N similar statistical variables, wherein each stochastic variable has a same standard deviation, and N is equal to or greater than two;
partially correlating each and every pair of stochastic variables among N variables, wherein a degree of partial correlation is a same for all pairs of variables and there are N(N−1)/2 pairs;
construct a statistical model to represent a system of N stochastic variables in which only N independent stochastic model parameters are used;
establish a one to one mapping relation between N model parameters and the N variables;
find unique values of the N model parameters given a set of values of the N variables; and
find values of the N variables given a set of values of the N model parameters.