Beruflich Dokumente
Kultur Dokumente
1, JANUARY 2010 93
Abstract—In this paper, a method for the statistical design of variations, which, in turn, are related to the bit-cell transistor di-
the static-random-access-memory bit cell is proposed to ensure a mensions. Therefore, to meet the specifications for all the bit-
high memory yield while meeting design specifications for perfor- cell design metrics (some of which are conflicting) in the min-
mance, stability, area, and leakage. The method generates the nom-
inal design parameters, i.e., the widths and lengths of the bit-cell
imum possible area, the widths and lengths of the bit-cell tran-
transistors, which provide maximum immunity to the variations sistors must be chosen optimally [12]. The impact of variability
in a transistor’s dimensions and intrinsic threshold-voltage fluc- on the design metrics should be considered up front during the
tuations. Moreover, the need to deviate from the conventional bit- design phase to maximize yield.
cell sizing strategy to obtain a high-yield low-leakage design in the However, the current industrial practice is to first develop a
nanometer regime is demonstrated. database, by simulations, which characterizes the design met-
Index Terms—Circuit optimization, design methodology, static- rics for various transistor sizes. This is used to carefully choose
random-access-memory (SRAM) chips. the sizes of the bit-cell transistors. Monte Carlo (MC) simula-
tions are run for the chosen design to verify if variations in the
design metrics such as the SNM are within the desirable bounds.
I. INTRODUCTION
The chosen design is updated, and MC simulations are run it-
URRENTLY, more than 50% of the area of system-on-
C chip designs is occupied by embedded memory [1]. This
is due to the increasing integration of functional blocks that re-
eratively, until all the design metrics meet the specifications.
Another approach is to develop analytical functions for the de-
sign metrics, as a function of the device sizes. These are then
quire large memories for data manipulation and storage. The use deployed to choose nominal transistor sizes. These procedures
of minimum-size transistors in static random access memories have many drawbacks. The characterization, through simula-
(SRAMs), along with technology scaling, increases the intrinsic tions, is quite time consuming. These methods are iterative and
variability, causing a large deviation in the transistor proper- rely on manual transistor-size selection. Moreover, the chosen
ties such as the threshold voltage . This and the varying design may not be optimal, it may be an overdesign with larger
transistor dimensions adversely impact the yield of SRAMs. area. In this paper, a systematic statistical design method to de-
Memory yield is the percentage of the number of bit cells that termine the optimal size of the bit-cell transistors is proposed.
meet all the functionality and performance requirements, even The proposed framework is capable of eliminating the charac-
under variability. For an SRAM, the bit cell is functional if it al- terization step completely. Currently, the electrical bit-cell de-
lows for a nondestructive read and a successful write. Tradition- sign in the industry takes about two to three weeks or even more.
ally, these have been evaluated as static noise margin (SNM) and However, with the proposed method, the optimal design can be
write trip voltage (Vtrip), respectively. The performance con- obtained in a day or two.
straints are the desired memory speed and leakage. The various causes of variability in transistor dimensions are
It should be noted that the SRAM design involves the evalu- subwavelength lithography, proximity effects, etc. [2], [3]. The
ation of several bit-cell architectures and layout topologies for intrinsic variability is caused by the atomistic-level differences
process–layout interactions. It also requires the choice of SRAM between the devices, even though they might have identical lay-
specific physical design rules and the assessment of their robust- outs and environments. This manifests primarily as variations
ness. At this level, process developers are involved. However, in the device . The main reasons for variations are fluc-
this phase of the bit-cell development is not analyzed in this tuations in the number and location of the dopant atoms in the
paper. This paper is of significance to the circuit designer, who channel [random dopant fluctuation (RDF)], line edge rough-
is involved in the ‘electrical’ design phase. This entails the op- ness, and oxide-thickness variations [4]. Of these, the most sig-
timal selection of the transistor sizes to avoid parametric failures nificant factor is observed to be the RDF [4], [5], [6], [8]. In addi-
such as destructive read, write, and access failures and excessive tion, the distribution due to RDF is normal, and the variance
leakage, which can occur due to variations in the transistor pa- is inversely proportional to the channel area. In [7], the
rameters. From now on, ‘bit-cell design’ refers to this electrical channel-area component of the for fabricated MOSFETs
design phase. is observed to be dominant.
In large measure, the variability in the bit-cell metrics, such Because of the intrinsic variations, the SNM, read current,
as the SNM, Vtrip, speed, and leakage, is caused by intrinsic and Vtrip are observed to have normal distributions [9]–[11].
Technology scaling has a twofold impact on the SRAM. First,
Manuscript received June 15, 2008; revised November 15, 2008. First pub- an increase in the due to scaling SRAM transistors causes
lished March 06, 2009; current version published January 22, 2010. . This paper the distributions of SNM, read current, and Vtrip to take a larger
was recommended by Associate Editor J. Pineda de Gyvez. sigma value. Second, the increase of the memory density at each
The authors are with the Department of Electrical and Computer Engineering,
University of Waterloo, Waterloo, ON N2L3G1, Canada (e-mail: vgupta@vlsi.
successive technology node requires the bit cell to tolerate a
uwaterloo.ca; manis@vlsi.uwaterloo.ca). larger number of sigma variations (e.g., – ) in the design
Digital Object Identifier 10.1109/TCSI.2009.2016633 characteristics [13] to ensure a satisfactory memory yield.
1549-8328/$26.00 © 2010 IEEE
Authorized licensed use limited to: K.S. Institute of Technology. Downloaded on January 25, 2010 at 15:06 from IEEE Xplore. Restrictions apply.
94 IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS—I: REGULAR PAPERS, VOL. 57, NO. 1, JANUARY 2010
Authorized licensed use limited to: K.S. Institute of Technology. Downloaded on January 25, 2010 at 15:06 from IEEE Xplore. Restrictions apply.
GUPTA AND ANIS: GUPTA AND ANIS: STATISTICAL DESIGN OF THE 6T SRAM BIT CELL 95
Authorized licensed use limited to: K.S. Institute of Technology. Downloaded on January 25, 2010 at 15:06 from IEEE Xplore. Restrictions apply.
96 IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS—I: REGULAR PAPERS, VOL. 57, NO. 1, JANUARY 2010
Fig. 2. SNM variation with voltage and temperature. (a) = 1. (b) = 1:3.
channel length and width is negligible [12]. Therefore, in this
paper, the intrinsic variation due to RDF is considered as
the main source of intradie variation. (3)
The variations of six transistors are considered to be six
independent and uncorrelated Gaussian random variables [10],
[12]. Moreover, —the standard deviation of the dis-
tribution of a minimum-sized transistor—is an input parameter.
is usually available in the process development kits of
vendors. Then, the for a transistor of width and length
is related to the transistor size as follows [4], [12]:
(2)
At the circuit-design level, the designer can specify only the (4)
nominal values of the geometrical transistor dimensions (de-
sign parameters) and has little control over statistical parameters
such as the variations due to mismatch. However, as shown Equations (3) and (4), contain the of the driver, access,
by (2), the choice of design parameters is used to control the ex- and load transistors, which can be calculated by (2). is
tent of device mismatch. the simulated SNM at mean values. The partial derivative
terms are also computed at mean values numerically by sim-
III. PROBLEM FORMULATION ulation. In the aforementioned two equations, the term ‘SNM’
can be replaced by Vtrip, Iread, or leakage.
A. Intradie Variation The results of modeling have been verified by MC simula-
To consider the impact of intrinsic fluctuation due to RDF, tions. For example, the SNM average and standard deviation
a conservative method is to calculate the worst case change from MC simulations are 131.8 and 22.1 mV, respectively
for each of the six transistors (e.g., sigma) and evaluate (Fig. 3). The corresponding estimated values from modeling
the design constraint for this case [34]. However, this approach are 132.1 and 21.6 mV, respectively. Similarly, Fig. 4(a) shows
overestimates variability and gives poor results in terms of bit- that the average leakage and standard deviation, from the 45-nm
cell area. Therefore, the statistical approach is applied in this bit-cell leakage MC results, are 77.2 and 8.9 nA, respectively.
paper to model the intradie variations. The estimated values from modeling are 77.7 and 8.57 nA, re-
Because of the intrinsic variations, the design met- spectively. For our purpose of modeling, this level of accuracy
rics—SNM, Vtrip, and read current—exhibit Gaussian distri- is sufficient.
butions [13]. To statistically model the impact of the intrinsic The coefficient of on the right-hand side (RHS) of (4)
variations due to RDF, on the design metrics, the statistical contains the slopes of SNM versus and . These will be
average and variance of the Gaussian distributions of the design interchanged if the opposite data value (1 instead of 0) is stored
Authorized licensed use limited to: K.S. Institute of Technology. Downloaded on January 25, 2010 at 15:06 from IEEE Xplore. Restrictions apply.
GUPTA AND ANIS: GUPTA AND ANIS: STATISTICAL DESIGN OF THE 6T SRAM BIT CELL 97
Authorized licensed use limited to: K.S. Institute of Technology. Downloaded on January 25, 2010 at 15:06 from IEEE Xplore. Restrictions apply.
98 IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS—I: REGULAR PAPERS, VOL. 57, NO. 1, JANUARY 2010
(7)
rectangular overlap that can be attained between the feasible re- In (8), is the normalized value of , and and are the
gion and the tolerance box and can be used to estimate the yield lower and upper bounds, respectively, of the DB random vari-
directly [37]. In six dimensions (which is the case of bit-cell able . and are the shape parameters, and distributions such
design problem), the 6-D volume of the inner box, henceforth as uniform, triangular, and Gaussian can be obtained by using
called the yield box, defines the yield. different values of and . A truncated Gaussian shape is used
In this paper, the feasible region is determined implicitly, i.e., in this paper with variation as the design spread around the
whether a design point lies in the feasible region is determined nominal design . Therefore
by simulations and by checking if all the design constraints are
met. The typical design centering approach needs an explicit
feasible region, i.e., the design constraints should be expressed (9)
as analytical expressions. This is not always possible and re- In (9), represents the maximum spread on the design param-
quires either a polytope approximation or simulations for curve eter . The closed-form DB-CDF can be obtained by integrating
fitting the data in the region of interest, which becomes difficult and is given by
with the increase in the number of dimensions [25].
Qualitatively, the problem is reduced to finding and , the (10)
coordinates of the yield box in Fig. 6, such that the following
conditions are satisfied: 1) the maximum difference between Due to the symmetrical nature of the distribution of the design
and is not more than the maximum spread in design parame- parameters, the final optimized design solution is the center of
ters (the yield box should lie within the tolerance box) and 2) the the yield box and is computed as
yield box lies in the feasible region (all the points lying within
the box satisfy all the design constraints that cover within-die (11)
variations).
If the aforementioned conditions are met, then for the nom- Therefore, by using DB-CDF and (7), (10) and (11)
inal design placed at the center of the yield box, the probability
(yield) that the design constraints are satisfied in the presence of
parameter variations is estimated in two dimensions (for illus-
trative purposes) as follows:
- and
Authorized licensed use limited to: K.S. Institute of Technology. Downloaded on January 25, 2010 at 15:06 from IEEE Xplore. Restrictions apply.
GUPTA AND ANIS: GUPTA AND ANIS: STATISTICAL DESIGN OF THE 6T SRAM BIT CELL 99
Fig. 8. CDF plot of deviation in across dies. is a measure of the
within-die variation. It is observed that nearly 99.3% of the dies have their
6
within 1 mV of the nominal value.
Authorized licensed use limited to: K.S. Institute of Technology. Downloaded on January 25, 2010 at 15:06 from IEEE Xplore. Restrictions apply.
100 IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS—I: REGULAR PAPERS, VOL. 57, NO. 1, JANUARY 2010
TABLE II
GENERAL-PURPOSE HIGH-PERFORMANCE DESIGN. OPTIMIZATION RESULTS FOR VARYING READ-CURRENT BOUNDS WITH Ileak = 60 nA
cells per 1000 fall outside the range [13]. The chosen
mathematically corresponds to a single bit-
cell failure in an array of . Typically, the largest
size of an embedded SRAM block is 256 kbits [33]; for
larger blocks, performance begins to degrade. Therefore,
the chosen value of covers a wide range of embedded
memory sizes. With redundancy, the designer can reduce
the value of .
3) The transistor dimension is altered in steps of 1 nm. This is
the step size available for altering the layout geometries in
the 45-nm technology. The ranges for transistor width and
length are set to 90–500 and 45–80 nm, respectively.
4) Usually, the required residuals for the SNM and Vtrip are
set to zero, and the designers attempt to obtain a value of
4–5 for [11]. However, in this paper, the SNM and Vtrip
residuals are selected as 15 and 25 mV, respectively, to set
a higher reliability margin. Note that these are the desired
bounds for the worst case voltage-temperature corners, as
in Table I.
Authorized licensed use limited to: K.S. Institute of Technology. Downloaded on January 25, 2010 at 15:06 from IEEE Xplore. Restrictions apply.
GUPTA AND ANIS: GUPTA AND ANIS: STATISTICAL DESIGN OF THE 6T SRAM BIT CELL 101
Fig. 10. For varying read-current residuals: (a) Nominal read current (in microamperes) and bit-cell leakage (in nanoamperes), (b) cell ratio and W =L , and
(c) nominal SNM (in millivolts).
Authorized licensed use limited to: K.S. Institute of Technology. Downloaded on January 25, 2010 at 15:06 from IEEE Xplore. Restrictions apply.
102 IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS—I: REGULAR PAPERS, VOL. 57, NO. 1, JANUARY 2010
Fig. 12. Yield and average values of SNM, Vtrip, read current, and bit-cell leakage obtained by MC simulations for the low-leakage bit cell. In every figure, only
one of the bit-cell dimensions is varied while the other dimensions are kept equal to the obtained optimal value.
TABLE III
LOW-LEAKAGE BIT-CELL DESIGN RESULTS. Iread = 10 A. average values of the design constraints when is varied
Ileak = 25 nA from 143 to 155 nm, while , ,
, , and . It is evident
that the yield degrades as the design point is moved away from
the obtained optimum.
V. CONCLUSION
In this paper, a statistical method to design the SRAM bit cell
has been proposed. The method accounts for the manufacturing
TABLE IV variability in the transistor dimensions and the intrinsic vari-
LOW-LEAKAGE BIT-CELL DESIGN RESULTS. Iread = 10 A.
Ileak = 25 nA ations due to RDF. In addition, the widths and lengths of the
transistors are chosen so as to satisfy the constraints of SNM,
Vtrip, read current, leakage, and area. The developed method
is flexible, involves a small infrastructure in terms of mathemat-
ical computations, and uses readily available models and tools in
the industry, so that the extent of approximation in the proposed
method is small. Robust bit-cell designs for high-performance
moderate-leakage and moderate-performance low-leakage con-
ditions have been developed and analyzed to demonstrate the
working of the proposed method. The method can be upgraded
Fig. 12 shows the variation of the MC yield in the neigh- to include the impact of other sources of variability such as the
borhood of the obtained low-leakage-design solution. oxide thickness fluctuation and line-edge roughness. It is nec-
and refer to SNM at low-voltage high-temperature and essary to evaluate how much these other sources impact the
high-voltage high-temperature conditions, respectively, as per threshold voltage of the transistor and how they can be mod-
Table I. Running the MC simulations to explore the entire eled mathematically.
search space in order to find the globally maximum yield point
is infeasible. Therefore, in Fig. 12, it is verified that the proposed ACKNOWLEDGMENT
optimization approach leads to, at least, a locally maximum The authors would like to thank the Associate Editor and the
yield. In each figure in Fig. 12, one design parameter is varied, reviewers who helped improve the quality of this paper. The au-
whereas the others are kept constant at the values mentioned in thors would also like to thank Prof. P. Kumaraswamy for pro-
Table III. For example, Fig. 12(a) shows the MC yield and the viding the preliminary optimization code.
Authorized licensed use limited to: K.S. Institute of Technology. Downloaded on January 25, 2010 at 15:06 from IEEE Xplore. Restrictions apply.
GUPTA AND ANIS: GUPTA AND ANIS: STATISTICAL DESIGN OF THE 6T SRAM BIT CELL 103
Authorized licensed use limited to: K.S. Institute of Technology. Downloaded on January 25, 2010 at 15:06 from IEEE Xplore. Restrictions apply.
104 IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS—I: REGULAR PAPERS, VOL. 57, NO. 1, JANUARY 2010
Mohab Anis (S’98–M’03) received the B.Sc. degree Journal of Circuits, Systems and Computers, the ASP Journal of Low Power
(with honors) in electronics and communication engi- Electronics, and VLSI Design. His research interests include integrated-circuit
neering from Cairo University, Cairo, Egypt, in 1997 design and design automation for VLSI systems in the deep-submicrometer
and the M.A.Sc. and Ph.D. degrees in electrical en- regime.
gineering from the University of Waterloo, Waterloo, Dr. Anis is an Associate Editor of the IEEE TRANSACTIONS ON CIRCUITS
ON, Canada, in 1999 and 2003, respectively. AND SYSTEMS—II: EXPRESS BRIEFS. He is also a member of the program com-
He is currently an Associate Professor with the mittee for several IEEE conferences. He was the recipient of the 2009 Early
Department of Electrical and Computer Engineering, Research Award from Ontario’s Ministry of Research and Innovation. He was
University of Waterloo. He has authored/coauthored the recipient of the 2004 Douglas R. Colton Medal for Research Excellence
over 80 papers in international journals and confer- in recognition of his excellence in research, leading to new understanding and
ences and is the author of the book Multi-Threshold novel developments in microsystems in Canada. He won the 2002 International
CMOS Digital Circuits—Managing Leakage Power (Kluwer, 2003). He is Low-Power Design Contest.
the Cofounder of Spry Design Automation. He is an Associate Editor of the
Authorized licensed use limited to: K.S. Institute of Technology. Downloaded on January 25, 2010 at 15:06 from IEEE Xplore. Restrictions apply.