Location via proxy:   [ UP ]  
[Report a bug]   [Manage cookies]                

Paper 23 PDF

Download as pdf or txt
Download as pdf or txt
You are on page 1of 9

640 IEEE TRANSACTIONS ON COMPUTERS, VOL. C-17, NO.

7, JULY 1968

Computer Aided Memory Design Using


Transmission Line Models
KUO A. CHEN, MEMBER, IEEE

Abstract-This paper describes a computer program for analyzing Several such models are possible, one of which is for-
a class of transmission line configurations which arise in the design mulated in Weeks.'
of memory arrays. The systems analyzed by the program consist of
interconnected pairs of mutually coupled distributed parameter The computer program described below extends the
transmission lines driven and terminated by linear lumped element model formulated in Weeks by including several pairs
networks. A brief discussion of the mathematical model and its im- of cascaded coupled transmission lines instead of one
plementation is given and applications of the program to the design single pair of coupled transmission lines. The termina-
of 2ID and 3D memory arrays are described. The applications illu- tions of the cascaded coupled transmission lines now
strate how the program can be used to evaluate pulse propagation
characteristics and noise generation mechanisms within a memory can be any linear network instead of a predescribed net-
array. work. The cascaded coupled lines can be connected in
such a way that a 3D memory sense line coupled into
Index Terms-Distributed parameter, mathematical models, an x line can be described. This allows simulation of
memory analysis, memory array models, memory design, memory noise cancellation schemes for 3D sense windings.
simulation, transmission line models. As long as the models are linear, the mathematical
INTRODUCTION operations required to construct a solution are nearly
independent of the details of the model. The analysis
r HE PROBLEM of constructing a mathematical given below is therefore essentially independent of the
model for simulating the electrical properties of memory array model, the point of departure being the
memory arrays has been considered by Weeks' most general set of equations describing pulse propaga-
who found that a memory array can be approximated by tion on a pair of mutually coupled transmission lines.
a system of linear transmission lines. In Weeks,' equiv- Following the mathematical discussion, a computer
alent circuits are established for ferrite cores and for program implementing the mathematical analysis is
intersections of wires crossing each other at right angles. described. Applications of the program to the design of
These equivalent circuits are then distributed uniformly actual memory arrays are illustrated by two examples.
over the length of a line in the memory array, thus
arriving at a transmission line model for the line. If the
mutual coupling between lines is sufficiently weak, then BASIc EQUATIONS
it is possible to characterize pulse propagation on the Consider a pair of uniform, parallel, linear, mutually
line by a single pair of transmission equations. If the coupled transmission lines which share a common return
coupling is not weak, then it is necessary to characterize path (see Fig. 1).
the array by at least two mutually coupled transmission Let x represent the distance measured along the lines.
lines. In the latter case, it is necessary to solve at least Let Vl(x, t) be the voltage between line 1 and the return
two pairs of mutually coupled transmission line equa- path at position x and time t. Let I1(x, t) be the current
tions. Practical considerations have limited the models in line 1 at position x and time t. Similarly, let V2(x, t)
in use to either single transmission line models or to and I(x, t) be the corresponding voltage and current
models consisting of two mutually coupled transmission for line 2. Let 17,(x,co), 12 (X, O), I,(x, co), and h2(x, (D) be
lines. the frequency transforms of Vl(x, t), V2(x, t), Ii(x, t),
Whereas the mathematical analysis of single trans- and 12(x, t), respectively. The general equations, in the
mission line models has already been treated in detail,' frequency domain, for a pair of mutually coupled, uni-
the present paper is devoted to the analysis of models form, linear transmission lines can be written in the
based on a pair of mutually coupled transmission lines. following form:
Manuscript received September 2, 1967. dV1(x, w)
The author was with IBM Corporation, Systems Development
Div., Poughkeepsie, N. Y. He is now on educational leave of absence
-________ = Zl(W)I1(x, CO) +
dx
ZM(CW)12(X, C) (1)
from IBM and is working toward the doctoral degree in electrical en-
gineering at Syracuse University, Syracuse, N. Y.
1 W. T. Weeks, "Computer simulation of the electrical properties
of memory arrays," IEEE Trans. Electronic Computers, vol, EC-12,
pp. 874-886, December 1963.
- d =2(X
dx
' = ZM(W)I,1(X, C) + Z2(o)h2(X, W) (2)
CHEN: COMPUTER AIDED MEMORY DESIGN 641

x X+ A x

LINE 2 12(x,t)
x+
LINE I II( x,t)

RETURN PAT H

Fig. 1. A pair of uniform, parallel, linear, mutually Fig. 2. Equivalent circuit of an incremental section of a pair of
coupled transmission lines with a common return path. coupled transmission lines.

d- (x,CO b) For the case of periodic solutions with the period


dx
= (Y1(o) + YM(M))fl(x,W) Tp from the Fourier series
- YM(CW)V2(X, W) (3)
Ik(X, t) =
-E k con)eintI
Ik(XW k = 1,2
- =_____ - YM(W)Vl(X,C) + (Y2(W)
TP n-oo
dx (6)
+ YM(W))172(X, w), (4) 27+00 27rn
Vk(X, i) = - V k(x, wn)ejiwt, Con =

where Zj(co), Z2(CO), ZM(W), Yi(w), Y2(w), and YM(C)


p n-oo Tv
are complex functions of the angular frequency c, the The process of constructing a solution is best begun
function Z1 is the distributed self-impedance per unit by rewriting (1) through (4) in matrix form. Thus, if
length of line 1, Z2 is the self-impedance per unit length one defines the following matrices
of line 2, and Zm is the mutual impedance per unit
length between the lines. Similarly, Y1 is the self-
admittance per unit length of line 1, Y2 is the self- V= [V%(xs,c&)] (7)
2(x, co)
admittance per unit length of line 2, and YM is the
mutual admittance per unit length between the two r(XX,10
I2
lines. (8)
The interpretation of these functions is given in Fig. 2 L 2(X, CO)
which shows the equivalent circuit for an incremental
section of the two transmission lines. The explicit func- Z [Z (w), ZM (CO)]
(9)
tional form of the three impedances and three admit- ZM(C)), Z2 (CO)
tances depends on the details of the mathematical mod-
els used to approximate memory array lines. At this rYI(CW) + YM(@), YM(CW)
(10)
point, however, it is not necessary to specify the func- YM(c), Y2(W) + YM(o) -

tional forms of these quantities. Later, two possible


choices of this set of six frequency functions will be then (1) through (4) can be written as the two matrix
given, corresponding to two memory array models. equations
In order to calculate the currents and voltages on the dV
two transmission lines, it is necessary to solve (1) - = ZI (11)
through (4) for the current and voltage transforms. The dx
time domain response of the system can then be calcu- dI
lated as follows. - = YV. (12)
dx
a) For the case of aperiodic solutions from the Fourier
integrals (1 1) and (1 9) renresent twn nairs of counled differen-
tial equations. In general, the off-diagonal elements of
Ik(x, t) = f_k(X, w)ej"ldw, the Z and Y matrices are not zero. Consequently, the
k = 1, 2 (5) voltage on one line is a function of the currents on both
_00
lines, and the current on one line is a function of the
voltages of both lines. However, by performing a linear
64I+2 IEEE TRANSACTIONS ON COMPUTERS, JULY 1968

transformation on the column matrices I and V, it is and


possible to decouple the equations. To accomplish this dU2
transformation, let - -~=Z2J2 (24)
dx
I= TJ (13)
dJ2
(T')-'U, (14) =y2U2. (25)
-

V -
dx
where
These are the desired decoupled equations. It should
_ Til T12] be observed that each pair is in the form of the differ-
LT21 T2J' (15) ential equations for a single transmission line. The solu-
tions of equations of this form are well known and can
T' is the transpose of T, (T')-l is the inverse of T', and be written down at once. They are given by
J and U are the column matrices
J1 = F1e-'lx + F2eyIx (26)
J rJ (X, c)
(16) Ul = Zoi(Fe-lx - F2eylx) (27)
,72(Xi c)
and
Lo2(x, co) (17) J2 = F3e-Y2 + F4eY2x (28)
U2 = Zo2(F3eY2x - F4el2x) (29)
The elements of the T matrix are, in general, complex
functions of the angular frequency w, but are indepen- where
dent of the distance variable x. The problem is to choose V/Zi/yi,
T so that (11) and (12) transform into two pairs of
Zoi = 71i= Vzlyl,
uncoupled equations, i.e., T is to be chosen so that (11) Z02= \Z2/Yy2 72 = VZ2y22
and (12) yield a pair of equations involving J1 and U1,
but not J2 and U2, and a pair of equations involving J2 The coefficients F1, F2, F3, and F4, which may depend
and C2, but not L1 and U1. on frequency but not on the distance variable x, are
Inserting (13) and (14) into (11) and (12), one finds determined by the boundary conditions imposed at the
the transformed equations ends of the transmission line system.
The current J1 and voltage U1 make up one of the two
dU independent modes of propagation of the system; the
-
dx
= (T'ZT)J (18) current J2 and voltage 12 make up the other mode.
Note that, in general, each mode has its own char-
dJ acteristic impedance Zo, or Z02 and its own propagation
- - - Th'Y(T')-1U. (19)
dx constant y1 or 72. The actual current and voltage trans-
forms are linear combinations of the mode currents and
After a lengthy but straightforward calculation of mode voltages, respectively. Thus, from (13), (14), and
eigenvalues and eigenvectors one finds a transformation (26) through (29) one obtains
matrix T such that T'ZT and T-1 Y(T')-l are diagonal.
That is, 1 = T,l(Fier-y + F2eY1Y) + T12(F3eY2x + F4eY2X) (30)

T'ZT = [r ] (20) i2 = T21(Fle-,cx + F2e'yIx) + T22(F3e-Y2 + F4eY2x) (31)

V, -
T22ZO1(Fle1lx - F2eylx)
T-1 Y(T)-l = BY ]. (21) - T21ZO2(F3e-"2 - F4e62x) (32)

Putting (20) and (21) and (16) and (17) into (18)
V2 = T12Zo1(Fle-7x - F2e7lx)
and (19), one obtains + T11ZO2(F3 e-82x- F4e72x). (33)
dUl In order to calculate the time domain response of the
- ___= z,Jl (22)
dx transmission line system, it is necessary to compute the
frequency domain solutions [(30) through (33)] at all
- Y--Ul (23) required frequencies, and then make use of either (5)
dx or (6) as the case may be,
CHEN: COMPUTER AIDED MEMORY DESIGN 64i"W3
LINE 2

LINE 2,,/ , ,, ,,/


/ --

LINE

Fig. 3. Memory array configuration-Type 1. Fig. 4. Memory array configuration-Type 2.

MEMORY ARRAY MODELS Y,(w) = jwCll (43)


In the next section, a computer program which per- Y2(@) = jcC12 (44)
forms the calculations discussed above will be described.
The program offers a choice of two models for approxi- YM(w) =jdJCM. (45)
mating ferrite core memory arrays. The first model is in- In these equations Ri', Li', Ri2, Li2, C1, C12, CM, and
tended to describe the case where the two transmission LM have the same meaning as before. In this case, how-
lines pass through separate sets of ferrite cores as is ever, only a single set of core parameters R2 and L2 is
shown in Fig. 3. needed since both lines pass through the same set of
Mathematically, this model is defined by the follow- cores. The parameter X is a number much smaller than
ing equations: the one which can be used to account for imperfect
jcoL21R21 transformer coupling of the lines due to the cores. For
Z1(w) Ril + jwL,l + R L1R2'
= perfect coupling, X would be zero. It should be noted
R21 + jo.L21 (4
that, in this model, the mutual impedance contains a
=R,2+ frL, ± fL2 2R,22 term which accounts for the mutual coupling of the
Z2 (CO)= R12+ jwL12 + R2 + jcoL22 (35) two lines through the ferrite cores. For an account of
how one arrives at models such as the two described
ZM(w) = J6oLM (36) above and a more detailed discussion of the meaning of
Yl(W = jwCll (37) the parameters which described the models, see Weeks.'
Y2(cO) = jWC12 (38) The computer program described in the next section
is not restricted to the two models described above for
YM(w) - jCA{, (39) ferrite core arrays. It can handle any bit storage devices
where R,' is the s"eries resistance of line 1, Li' is the that can be approximated by a linear equivalent cir-
series inductance of line 1, and R2e and L21 are the cuit. It is only necessary to replace the ferrite core sub-
equivalent core parameters for the cores on line 1. routine by a subroutine describing the equivalent cir-
Similarly, R,2 is the series resistance of line 2, L12 is the cuit of the new device. The brief summary given above
series inductance of line 1, and R22 and L22 are the is intended, more than anything else, to indicate the
equivalent core parameters of line 2. Also, C,' is the functional forms of the impedance and admittance
shunt capacitance of line 1, and C12 is the shunt capaci- parameters commonly encountered in memory array
tance of line 2. Finally, LM and CM are the mutual in- simulation.
ductance and mutual capacitance, respectively, between
the two lines. THE COMPUTER PROGRAM
The second model available is intended to represent
the case where the two lines pass through a common set The IBM 7090 program described in this section,
of ferrite cores as shown in Fig. 4. This model is defined which solves (1) through (4) by using the procedures
by the following equations: described above, requires as input the distributed re-
sistances, inductances, and capacitances which describe
the pair of transmission lines. Also included as input is
Zi(o) = RI' + jcL,l + (1 + X) jwL22 (40) sufficient information to define the circuitry at the driv-
R2 + joL2
ing and load ends of the lines. The calculated output
jcoL,R2 consists of current and voltage waveforms given as
Z2(CO) = R12 + jwL12 + (1 + X) jcoL2 (41) functions of time. Waveforms can be obtained at any
R2f+ LR2
desired position along the transmission lines.
ZM(co) = jwLM + (1 -
X) jcoL2R2 (42) The program solves the transmission line configura-
- ~~R2 + jcoL2 tion shown in Fig. 5, consisting of N pairs of mutually
644 IEEE TRANSACTIONS ON COMPUTERS, JULY 1968
coupled transmission lines in cascade. The three net- A termination network and its equivalent circuit are
works, called generator, middle, and terminal networks, shown in Figs. 9 and 10.
consist of at most 24 elements each. Each element A sense amplifier is located across the transformer of
is either a resistor, a capacitor, a self- or mutual in- the terminal network. If a core is switched, then there
ductor, a voltage source, or a current source. All re- will be a difference voltage appearing across the trans-
sistors, capacitors, and inductors are linear. former which can be detected by the sense amplifier.
Each network must be such that it is possible to find During the reading operation at most 1 core will be
its Thevenin equivalent circuit from the transmission switched on a pair of sense-bit lines; the rest of the cores
line terminals. The middle network can be located be- are not switched. Therefore, the cores can be represented
tween any sections of the coupled transmission lines. by an RL parallel circuit.'
Voltage and current sources shown in Fig. 5 can also For a pair of bit sense lines, the equivalent circuit
be inserted between any sections of the transmission used to represent an infinitesimal section of the coupled
lines. transmission line is shown in Fig. 11. Z, Y parameters
Up to 30 cascaded coupled transmission line pairs for this equivalent circuit are given in (34) through (39).
are allowed. If for some reason, such as to simulate a z Once the parameters of the coupled lines and the
sense line coupled with an x line in a 3D memory, one equivalent circuit elements of both driving and termina-
wishes to connect points A to B and C to D rather than tion networks are known, the computer program can be
A to C and B to D, one can do so by specifying the used to simulate the memory array operation on the
discontinuity locations in the input data. computer. In a reading operation, if a core is switched,
The method of finding the Thevenin equivalent cir- one can represent the core switching signal by a voltage
cuit for the networks will not be given here. The basic source. This voltage waveform will be superimposed on
theory has been given already.2'3 Once the equivalent top of the driving waveform (the noise). The termina-
circuits and the ABCD parameters of the transmission tion-network (sense-amplifier) end will have the
line sections are found, one can solve the problem easily. switched core signal appear on top of the driving wave-
The overall procedure used in solving equations of a form. If the signal is large enough to meet the require-
pair of coupled transmission lines is similar to the pro- ments of the array specifications, then the array is de-
cedure described in Weeks.' The periodic input voltages signed properly. Otherwise, some modification would
and/or currents are transformed into the frequency have to be made to the array design.
domain; that is, the coefficients of the Fourier series are Note that one can easily simulate the worst possible
obtained. Then the diagonalization of (1) through (4) case on the computer by unbalancing the coupled trans-
proceeds as described in the section entitled Basic mission lines by, for example, assigning one line 10 per-
Equations. cent more resistance and inductance than the other line.
Boundary conditions are applied to (30) through (33). Or one can change the core equivalent circuit to simu-
The calculation of the coefficients Fl, F2, F3, and F4 can late "1" state on one line and "0" state on the other line
be bypassed. Instead, one obtains directly the voltages in order to see the delta noise at the sense-amplifier end
and currents at the generator-network end and load (see Fig. 12).
terminal-network end. Summing the Fourier series for
each voltage and current, the time domain solutions are Example 2: 3D Array
obtained. The second example is a 3D-array design problem.
Two examples are given below to illustrate the useful- The array has 16 x lines, 128 y lines, and 72 bit planes
ness of the program in aiding the design of memory (see Fig. 13).
arrays. As in Example 1, one can specify all the driving and
termination networks for x, y, and z sense lines. Since
EXAMPLES the y lines are perpendicular to both the x and the z
Example 1: 22D Array sense lines, the line inductive coupling between them
is negligible. Thus, one has only to consider the coupling
Assume a memory array engineer is planning to de- effect between the x and the z sense lines. Assuming that
sign a 2 D array with 2048 cores on a bit sense line the coupling effect between the x line and itself (and the
where each word line contains 72 bits (see Fig. 6). The z sense and itself) is negligible, one can expand the z
array is resting on top of a ground plane. sense and x lines as shown in Fig. 14. In Fig. 14, it is
A simplified driving network and its equivalent cir- assumed that the xi line is selected and that one is in-
cuit are shown in Figs. 7 and 8. terested in the detection problem of the first bit (top
plane). This assumption is made arbitrarily. One could
2 H. E. Koenig and W. A. Blackwell, Electromechanical System choose any other x line and plane. In Fig. 14, dashed
Theory. New York: McGraw-Hill, 1961. lines mean a short circuit connection with no R, L, or C
3S. Seshu and M. B. Reed, Linear Graphs and Electrical Networks.
Reading, Mass.: Addison-Wesley, 1961. whatsoever, with A being connected to B and C to D.
CHEN: COMPUTER AIDED MEMORY DESIGN 645;

N PAIRS OF COUPLED TRANSMISSION LINES

Fig. 5. Basic transmission line configuration for the computer program where EK1 and EK2 are voltage sources, and IK1 and IK2
are current sources.

WORD I WORD 2 WORD 2047 WORD 2048

Fig. 6. 2jD memory array.


646 IEEE TRANSACTIONS ON COMPUTERS, JULY 1968

ARRAY

Fig. 9. Termination network.

LI -I-

I+
Fig. 7. Driving network.

Fig. 10. Termination network equivalent circuit.

R 22 t/C2
R2 2
2I'
-IL
-Wv-fL2
+
T CM
LM
Fig. 8. Driving network equivalent circuit.
RI L
I

Cl

Fig. 11. Equivalent circuit for coupled transmission line.

4.4.
+
4.
4.
Un 4.
4.
0 4.
4. 4.
-J
J
+ 4.4.44.4.4 +++ Ee|++++++++
++++++++++ +l+
+++++++++ +++++++++ + ~~~~~~~~~~~~+......
+ +
.......................................
+ +
+@

1 I I 1 1 III I

NANOSECONDS
Fig. 12. Typical calculated sense-end output of 2' D array.
CHEN: COMPUTER AIDED MEMORY DESIGN 647

7 PLANE

PLANE 2

--.i 1.
-,
it 1/t f w- A ;- ,
-~~~~~~~~~~~~~~~~~~~I
7L/ - Ile I
~L
'/

/ YI /
0 TERMINAT ION/
* //

/ /LANE 72

XI / / / /I / /I
TERMINATION A _ -.I

Fig. 13. 3D memory array.

COUPLING BETWEEN COUPLING BETWEEN


X AND SENSE LINES X AND SENSE LINES

Fig. 14. Model for 3D memory x-to-sense coupling.


648 IEEE TRANSACTIONS ON COMPUTERS, JULY 1968

ao
0
a 0
O
a 00a
0 0 0 0 0
a0 0
U, O 0000 000 0 0
-J 0 00
a O 0000000000000000000000000000
0 0 0a

ji
4i 0
0
0
0
0
oo 0

lI I I I I I I I I I I I I I l lI_
NANOSECONDS
Fig. 15. Typical computed sense amplifier output of 3D array.

Coupling of the sense line to the x line through ferrite


z 1, one can unbalance the lines to simulate delta noise
cores is obtained by using the model described by (40) (see Fig. 15).
through (45). As a word of caution, the model used in the computer
As before, one can simulate the 3D memory array program for a 3D memory array is not an ideal model
operation on the computer by specifying line parame- because some of the second-or higher-order effect param-
ters, driving, and termination circuit elements. In a eters are not included, such as the mutual couplings
reading operation, noise is induced on the z sense line between the sense segments and the mutual couplings
by the x line driving pulse. (Noise induced on the z between the x line and itself. However. for a first-order
sense line by the y line is neglected here, but it can be approximation, the model is a fairly good representation
included if necessary by specifying more voltage gen- of the array. A modified version of the program will be
erators along the z sense line.) If a core is switched, a available shortly which will be able to take into account
signal will be induced on the z sense line riding on top of the mutual couplings between sections of the x line.
the noise. Depending on the signal level appearing These programs are the initial effort toward a more per-
across the sense amplifier, one can decide whether the fect mathematical model to simulate a 3D memory
design meets the required specifications. As in Example array.

You might also like