Professional Documents
Culture Documents
Thewes - Explanation and Quantitative Model For The Matching Behaviour of Poly Silcon - Resistors - 1998
Thewes - Explanation and Quantitative Model For The Matching Behaviour of Poly Silcon - Resistors - 1998
of Poly-Silicon Resistors
analog CMOS process with three poly-Si options. The 1100 113 17.9
characteristic parameters of these poly-Si variants are 2,4 x lo2" = solubility limit
(= 4 x 1021)
summarized in Table 1. We consider the standard deviation
of the difference of the resistances of a pair normalized to the t [nml 200
mean value o(AR)/R. In Fig. 2a), this parameter is shown for application high ohmic
the high ohmic polySi2 variant as a function of in analog resistor
(device area).'" for different characterization conditions. circuits poly-Si 1 together with
Obviously, the relation o(AR)/R = A,,,i3(Lw)-"2 holds, and
the matching behavior is found independent of the
characterization conditions. The results of the other *uolv-Si
-
variants show a complete qualitative agreement with these Table I: Process parameters and field of application for the different poly-Si
findings. For direct comparison of the matching constants variants.
28.7.1
0-7803-4774-9/98/$10.00 0 1998 IEEE IEDM 98-771
0.8 I
‘in
Area-” [p”]
-5 12- 0
0.
E 10. 0 0 /
c poly-Si1
!4
L poly-Si2,lo
w 6 . \
....
....
I I1
3 $ 4 -
-6 2 -
0 .
/
I I I “char poly-Si2,hi
X- DECODER
I 0
Fig 1: Test structure used. Geometries UW [ p d p m ] investigated are: Fig. 2: a) Standard deviation o(AR/R) of poly-Si2,hi resistors as
poly-Si2,hi: 24/0.75, 4W0.75, 2411.5, 4 8 / 1 5 12/3.0, 2413.0, IU6.0, and 24/6.0; a function of (.LW)-’” (LW = device area) characterized in
poly-Si2,lo and poly-Sil: 24/0.75, 4W0.75, 12/6.0, and 24/6.0. As depicted, different operating points (constant voltage drops and constant
the devices under test (“R”) are arranged in pairs and surrounded by current per width). b) Matching constants Apoir =
dummies (“D). O(AR/R)X(LW)”~ for the different types of poly-Si resistors
investigated.
NG = W L / ~ =
1
-- -0.325x~=0.325-~-- 1
6 WL
N,= W L l n , (6)
n~(poly-Si2,lo)-n,(poly-Si2,lo)= n~(poly-Si2,hi)-n,(poly-Si2,hi) (7) =fixax/- x 1 = 0.84xax 1 (14)
28.7.2
772-IEDM 98
grains
k-- -w--- t- - -w---- 4 ,grains
[% pml
I
Apoir
measured
I Ro,r [QI
Apoir
a
[% pml
I
I
178
924
0.83
5.08
I
I
20.1
92.9
0.82
5.03
I
I
3.3
14.6
0.82
14.8
calculated
Fig 3: Schematic diagrams demonstrating the impact of the grain boundary-induced
resistances and the resistance of the inner grain on the total resistance of a poly-silicon layer Table 3: Experimentally obtained matching constant Aprrrr,
of length L and width W.a) Resistor with regular rectangle grains. b) Equivalent circuit for experimentally determined values of the most important
the resistor in a). The resistances inside the grains have values 1/2 RIJ,~, the resistances model parameters, and calculated values of Ap,,il for the
crossing the grain boundaries Rn,b. c) Resistor with arbitrarily distributed grain boundaries. d) different poly-Si variants. *) calculated values as
Equivalent circuit for the resistor in c) only depicting the boundary related contributions. The described in the text; **) assumption used: n, = solubility
values of resistors crossing the boundaries are inversely proportional to the length of the limit.
related boundary.
brackets in eq. (3)) based on the equivalent circuit in (lowest possible values for 1 and ne),o(n,)/n, is below 0.05 %
Fig. 3 b). Since in real poly-Si layers the grains are not for a device area of I pm2 so that the standard deviation of
arranged regularly (c.f. Figs. 3 c) and d)), a disorder this parameter can also be neglected.
parameter 6 must be introduced in eq. (3) to model RO,h A similar consideration can be performed for all
correctly. The magnitude of 6 is irrelevant in the following parameters in eq. (4), whose standard deviation is related to
discussion. We use only its variation which will be discussed the number of mobile and fixed charges (i. e., for the
in detail below. mobility, which is related to nD via eqs. (8) and (9), but also
Using eqs. (1) - (3), o(R)/R can be calculated as given by for the parameter r',,, whose analytical expressions are too
eqs. (4) and (5). (Note, that for the measured parameter complex to be discussed here in detail). In both cases, the
o(AR)/R = 2'" x o(R)/R holds.) In the following, the impact of these parameters in eq. (4) is negligibly small.
contribution of each term in eq. (4) will be discussed. The dominant term in eq. (4) is the standard deviation of
Since the lithography dependent parameters o(W)/W and the grain size, o(1)A. It can be expressed via the standard
o(L)/L do not lead to the area dependence which is deviation of the total number of grains NG in a given resistor
experimentally found they can be neglected. Due to the (eqs. (11) and (12)) leading to a direct relation between
excellent control mechanisms of layer thicknesses during mismatch and grain size. SEM photographs (Fig. 4) of the
processing and due to the relatively large values of poly-Si three poly-Si variants were taken to determine the grain size 1.
thickness t, variations of this parameter are also assumed to The results are given in Tab. 3.
be negligible.
In order to estimate o(n,)/n<,we calculate the total number
Ne of free electrons in a given resistor (eq. (6)), whose
standard deviation is Ne1". For quantitative evaluation of the
resulting relation o(n,)/n, = (W L I the values of ne are
determined. In case of poly-Sil, ne solubility limit can be
assumed due to the extremely high doping level. In case of
the poly-Si2 variants with lower doping, trapping of carriers
in traps located at the grain boundary, segregation effects, and
incomplete activation of dopants lead to a decrease of the
number o f free carriers inside the grain [l-41. These
mechanisms are considered in the evaluation. As a first order
approach we assume that the loss of free carriers is equal in
poly-Si2,hi and poly-Si2,lo (eq. (7)), since both poly-Si types
are processed identically. This holds despite the fact that
different doping levels are used. Using eqs. (1)-(3) and (6),
and the known values of the total square resistances, ne can be
calculated for both poly-Si variants. The resulting values are Fig. 4: SEM microphotographs of a poly-Si,l layer (left) and a poly-Si2,hi
layer (right) for determination of the grain size 1. The areas shown are
shown in Tab. 3. (Note, that the calculations related to ne also approximately 4.1 pm x 2.7 pm (left) and 1.2 pm x 1.8 pm (right). In case
lead to determination of the parameters Ro,i, RO,b, and a, of poly-Si2,lo (not shown in the figure) very similar results are obtained as
which are also shown in Table 3.) Even for the worst case for poly-Si2,hi.
28.7.3
IEDM 98-773
Step 1: Define dimensions x and y of an array of hexagonal grids and a
number No ( NGc< xy !) of grains.
Step 2: Select NGgrain centers in the array by a random process.
Step 3: Let grains grow from the grain centers in all six directions by
occupying the next grids in the neighborhood.
Step 4: As soon as two occupied areas meet, the respective grid points are
defined as grain boundary and the growth is stopped.
Step 5: Repeat steps 3 and 4 until the whole array is covered.
Step 6: Calculate the length of all grain boundaries of neighboring grains.
Step 6: Calculate resistances between neighboring grains which are
inversely proportional to the respective grain boundary lengths
(c.f. Fig. 3 d)) and short all resistors inside of each grain.
Step 7: Calculate the total resistance of the network achiwed.
Step 8: Repeat steps 2 to 7 many times and calculate the standard deviation Fig. 6: Schematic demonstration of the algonthms described
of the resistance of the network. in Fig. 5 for three grains with grain centers C1, C2, and C3. Left: Randomly
placing of the grain centers in an array with hexagonal grids. Middle: First
Fig. 5: Description of the Monte-Carlo algorithms used to determine the step of grain growth. Right: Situation after further steps of growth. The dark
standard deviation 0(6)/8 in eq. (4). fields represent the grain boundaries.
0 30 60 90 120 150 180 210 0.00 0.02 0.04 0.06 0.08 0.10 0.12 poly-Si2,hi poly-Si2,lo poly-Si1
Grainsize / 1000 [nm'] (Total number of grains No)-"'
Fig. 7: Grain size distribution for poly-Si,l Fig. 8: Result of the simulations as described in Fig. 9: Comparison of measured data and model
layers. The bars represent classes of grains with Figs. 5 and 6 used to determine the standard for the three poly-Si variants investigated.
an area of (n X 10000 5 5000) nm2 with n = 0, 1, deviation o(6)h in eq. (4). At least 500
2, 3, ... . This information is only used to check simulations per data point are performed. The
the plausibility of the Monte-Carlo simulation result is o(6)h= 0.325 NG-''' independent of the
(c.f. text and Figs. 5, 6, and 8). ratio x/y (c.f. Fig. 5) of the simulated array.
28.7.4
774-IEDM 98