Cmpe 110 Class Notes
Cmpe 110 Class Notes
Cmpe 110 Class Notes
Digital Electronics
Class Notes
By
Dr. Ahmet Bindal
Revised August 2008
1
CHAPTER I. RLC NETWORKS
A. BASIC CONCEPTS ON PASSIVE ELEMENTS
Resistance:
vR(t)
R
iR(t)
vR(t) = R iR(t)
Capacitance:
vC(t)
C
iC(t)
dvc(t )
iC(t) = C
dt
dvc(t )
£ [iC(t)] = IC(s) = C £ [ ]
dt
∞
Laplace transform: F ( s ) = ∫ f (t )e − st dt
0
Thus:
∞ ∞
dvc(t ) dvc(t ) − st
£[ ] =∫ e dt = ∫ e− st dvc(t )
dt 0
dt 0
But,
∞ ∞ ∞
∫ e dvc(t ) = e vc(t )⏐0 - ∫ vc(t )(− s)e dt = [0- vc(0) ] + s ∫ vc(t )e dt = sVC(s) - vC(0)
− st − st ∞ − st − st
0 0 0
Then:
£ [ic(t )] = C[ sVc ( s ) − vc (0)] = sCVc ( s )
1
Thus, ZC(s) =
sC
2
Inductor:
vL (t)
L
iL(t)
diL(t)
vL(t) =
dt
diL(t)
£ [ vL(t)] = VL(s) = L £[ ] = L [sIL(s) – iL(0)] = sL IL(s)
dt
ZL(s) = sL
Integrator:
R
vout
+ i
vin C
-
Vin = Ri + vout
dvout
i=C
dt
dVin di i
=R +
dt dt C
di i 1 dVin
+ =
dt RC R dt
General Solution:
di 1
+ i=0
dt RC
3
Characteristic equation:
1 1
s+ =0⇒ s =−
RC RC
t
−
ig = Ae st = Ae RC
ip = 0
t
−
i (t ) = ig (t ) + i p (t ) = Ae RC
1 ⎛ Vin ⎞ − RCt
t t
1
vout = ∫ idt = ⎜ ⎟ ∫ e dt
C0 C⎝ R ⎠0
Vin − RCt
e ( − RC )⏐t0 = −Vin ⎢⎢e− RC −1⎥⎥
⎡ t ⎤
=
RC ⎣ ⎦
⎡ −
t
⎤
vout = Vin ⎢1 − e RC ⎥
⎣ ⎦
vout
Vin
0 t
Differentiator
C
vout
+ i
vin R
-
Vin = vc + vout
dvc di
0= +R
dt dt
4
i di
0= +R
C dt
di 1
+ i=0
dt RC
1
But, the characteristic equation: s + =0
RC
i (t ) = ig (t ) + i p (t ) = Ae st + 0
t
−
i (t ) = Ae = Ae
st RC
t
−
vout = Ri (t ) = ARe RC
vout
Vin
0 t
5
(ii) RL Circuits
V = v R + v out
v R = Ri Ohms law
di
v L (t ) = L Lorentz Law
dt
di
V = Ri + L
dt
di R V
+ i=
dt L L
To solve the general solution (the homogeneous part) use Laplace Transform:
£ ⎡⎢ dtdi + RL i⎤⎥ = 0
⎣ ⎦
R
I (s)S + I (s) = 0
L
⎡ R⎤
I (s)⎢S + ⎥ = 0
⎣ L⎦
I (s) ≠ 0
⎡ R⎤
∴ ⎢S + ⎥ = 0
⎣ L⎦
−R
S=
L
−R
t
i (t ) = Ae L
6
di p R V R V
+ ip = Æ 0+ K=
dt L L L L
V
K=
R
Now put the general and particular solutions together:
−R
t V
i (t ) = Ae L +
R
Now solve for A by using the initial condition i (0) = 0 :
V −V
i ( 0) = A + =0 Æ A=
R R
V ⎛⎜ ⎞
−R −R
V V t t
⎟
∴ i (t ) = − e L = 1− e L
R R R ⎜⎝ ⎟
⎠
Use Lorentz Law to solve for v out
di
v out = L
dt
⎛V ⎛ −R
t ⎞⎞
d ⎜ ⎜ 1 − e L ⎟⎟
⎜R⎜ ⎟⎟
V ⎛⎜ t ⎛ − R ⎞⎞
−R −R
⎝ ⎝ ⎠⎠ ⎟
t
v out =L =L 0−e L ⋅⎜ ⎟ = Ve L
dt R⎝ ⎜ ⎝ L ⎠⎠ ⎟
RESULTS:
V ⎛⎜ ⎞
−R
⎟
t
i (t ) = 1− e L
R ⎜⎝ ⎟
⎠
−R
t
v out = Ve L
7
(iii) RLC Circuits
vL (t)
L
R
vout (t)
i (t)
+
vin
C
-
Vin = Ri (t ) + vL (t ) + vout (t )
di (t ) dv (t )
vL (t ) = L and i (t ) = C out
dt dt
di
Vin = Ri (t ) + L + vout (t )
dt
Differentiating both sides yields:
di d 2i dv di d 2i i
0 = R + L 2 + out = R + L 2 +
dt dt dt dt dt C
R 1
Characteristic equation: s 2 + s+ =0
L LC
R R2 4
− ± − 2
L L 2
LC R ⎛ R ⎞ 1
s1,2 = = − ± ⎜ ⎟ −
2 2L ⎝ 2L ⎠ LC
R 1
Let α = and ω0 =
2L LC
s1, 2 = −α ± α 2 − ω0
2
Since the current through the inductor cannot change suddenly, iL(0+) = i(0+) = 0
Thus, i(0) = 0 = A + B
A=-B
8
di (t )
Evaluate :
dt
di (t )
= A s1 e s1t + A s2 e s 2t = A ( s1 e s1t − s2 e s 2t )
dt
When the switch is closed the voltage across the capacitor and the current through the
inductor cannot change instantaneously
di
Vin = vL ( 0+ ) = L
dt t =0+
Then:
Vin
= A( s1 − s2 )
L
Vin Vin
A= B= −
L ( s1 − s2 ) L ( s1 − s2 )
i(t) =
Vin
L ( s1 − s2 )
(e s 1t
− e s 2t )
di
vout = Vin − i R −L
dt
di Vin
where, = ( s 1 e s 1t − s 2 e s 2 t )
dt L ( s1 − s 2 )
vout = Vin −
Vin R
L ( s1 − s 2 )
(e s 1t
− e s 2t ) −
Vin
( s1 − s2 )
( s 1 e s 1t − s 2 e s 2 t )
⎡ e s 1t R e s 2t R ⎤
vout = Vin ⎢1 − ( + s 1) + ( + s 2) ⎥
⎣ ( s1 − s2 ) L ( s1 − s2 ) L ⎦
where, s1 = −α − α 2 − ω0 and s2 = −α + α 2 − ω0
2 2
s1 − s2 = −2 α 2 − ω0
2
9
vout
Vin
α - effect due to R
0 t
Case 1 is undershoot.
Integrator
R
Vout(s)
+ I(s)
Vin C
-
Since there is a switch in the circuit, Vin is considered a step function with respect to time
and it is valid for t > 0.
Thus, taking the Laplace transform of Vin yields:
∞ ∞
Vin
Vin ( s ) = ∫ Vin e dt = Vin ∫ e − st dt =
− st
0 0
s
10
Then, KVL dictates:
Vin ⎡
= I ( s) ⎢ R +
1 ⎤
= I ( s)
( sRC + 1)
s ⎣ ⎥
sC ⎦ sC
1 1 Vin sC
Vout ( s ) = I ( s ) =
sC sC s ( sRC + 1)
Vin 1
Vout ( s ) =
RC ⎛ 1 ⎞
s⎜s + ⎟
⎝ RC ⎠
1 A B ⎛ 1 ⎞ A
But, ≡ + A⎜ s + ⎟ + Bs ≡ 1 s ( A + B) + ≡1
⎛ 1 ⎞ s s+ 1 ⎝ RC ⎠ RC
s⎜s + ⎟
⎝ RC ⎠ RC
Thus:
A
A + B = 0 and = 1 yields
RC
A = RC and B = − RC
⎡ ⎤ ⎡ ⎤
⎢
Vin RC RC ⎥ ⎢ 1 1 ⎥
Vout ( s) = ⎢ − ⎥ = Vin ⎢ − ⎥
RC ⎢ s ⎛ 1 ⎞⎥ ⎢ s ⎛ 1 ⎞⎥
⎢⎣ ⎜s+ ⎟ ⎜s+ ⎟
⎝ RC ⎠ ⎥⎦ ⎢⎣ ⎝ RC ⎠ ⎥⎦
⎡ ⎤
⎢1 1 ⎥ ⎛ −
t
⎞
vout (t ) = Vin £ -1 ⎢ − ⎥ = Vin ⎜1 − e RC ⎟
⎢s ⎛s + 1 ⎞⎥ ⎝ ⎠
⎢⎣ ⎜ ⎟⎥
⎝ RC ⎠⎦
Differentiator
C
Vout (s)
I (s) R
+
Vin / s
-
Vin ⎡
= I ( s) ⎢ R +
1 ⎤
= I ( s )
( sRC + 1)
s ⎣ sC ⎥⎦ sC
11
RC Vin Vin
Vout ( s ) = I ( s ) R = =
RC ⎛ 1 ⎞ ⎛ 1 ⎞
⎜s+ ⎟ ⎜s+ ⎟
⎝ RC ⎠ ⎝ RC ⎠
⎡ ⎤
⎢ 1 ⎥ −
t
vout (t ) = Vin £ -1 ⎢ =
1 ⎥
RC
V in e
⎢s + ⎥
⎣ RC ⎦
Homework:
Find vout(t) by performing frequency-domain analysis
C
Vout (s)
+
Vin / s
-
I (s)
+
Vin / s
C
-
12
Vin
= Vout ( RCs + s 2 LC + 1)
s
1
Vout LC
=
R
Vin s ( s 2 + s + 1 )
L LC
R 1
Solving ( s 2 + s+ ) yields:
L LC
2
R ⎛ R ⎞ 1
s1,2 = − ± ⎜ ⎟ − or s1,2 = −α ± α 2 − ω0 2
2L ⎝ 2 L ⎠ LC
R 1
Where, α = and ω0 =
2L LC
1
Vout LC K M N
= ≡ + +
Vin s ( s − s1 )( s − s2 ) s ( s − s1 ) ( s − s2 )
( )
vout (t ) = K + Me s1t + Ne s2t Vin for t ≥ 0 .
s2 s − 2 s2
Where, K = 1 , M = and N = 1
s2 − s1 s2 − s1
13
(iii) STABILITY
Transfer Function
Poles
When solving for D(s) = 0, the solution should result in the following form:
s = α + ωj
where α is the real component and ω is the imaginary component. These solutions (there
could be more than one solution for s) can then be plotted real vs. imaginary.
14
Poles to the right of the imaginary axis shown as (0) and (1) are unstable causing the
resulting waveforms similar to the following:
Poles that lay on the imaginary axis shown as (2) are oscillatory causing waveforms
similar to the following:
15
Poles to the left of the imaginary axis shown as (3) and (4) are stable causing the
resulting waveforms similar to the following:
16
Example 1
1
Vout (s) 1 1
H(s) = = sC = =
Vin (s) R+
1 sRC + 1 ⎛ 1 ⎞
RC ⎜ s + ⎟
sC ⎝ RC ⎠
Now, to find the pole(s).
⎛ 1 ⎞
N ( s ) = RC ⎜ s + ⎟=0
⎝ RC ⎠
−1
s=
RC
17
From the position of the pole, we know that the circuit is stable and the waveform would
look as follows:
vout
vin
0 t
Example 2
18
1
Vout (s) sC 1
H(s) = = = 2
Vin (s) R + sL +
1 s CL + sRC + 1
sC
Now, to find the pole(s).
N ( s ) = s 2 CL + sRC + 1 = 0
2
− RC ± (RC )2 − 4CL ⎛ R ⎞
2 ⎛ 1 ⎞
− ⎜⎜ ⎟
R
s1,2 = =− ± ⎜⎜ ⎟⎟ ⎟⎟
2CL 2L ⎝ 2L ⎠ ⎜ CL
⎝ ⎠
s1 = −α 0 + α 0 − ω0
2 2
s2 = −α 0 − α 0 − ω0
2 2
where
R
α0 =
2L
1
ω0 =
LC
Assuming that L, R, and C are not zero and not negative there are three possible
solutions.
α 0 < ω0 Æ There is not imaginary part.
Plot these poles.
From the position of the pole, we know that the circuit is stable and the waveform would
look as follows:
vout
vin
0 t
19
α 0 = ω0 Æ There is not imaginary part.
Plot this pole.
From the position of the pole, we know that the circuit is stable and the waveform would
look as follows:
vout
vin
0 t
α 0 > ω0 Æ There is an imaginary part.
Plot these poles.
2 2
⎛ 1 ⎞ ⎛ R ⎞
⎜⎜ ⎟⎟ − ⎜ ⎟
⎝ CL ⎠ ⎝ 2 L ⎠
2 2
⎛ 1 ⎞ ⎛ R ⎞
− ⎜⎜ ⎟⎟ − ⎜ ⎟
⎝ CL ⎠ ⎝ 2L ⎠
20
From the position of the pole, we know that the circuit is stable and the waveform would
look as follows:
21
CHAPTER II. FUNDAMENTALS OF CMOS CIRCUITS
A. BRIEF THEORY OF SEMICONDUCTORS
INTRINSIC SEMICONDUCTOR
Si Si Si
electron
Si Si Si
Si Si Si
22
N-TYPE EXTRINSIC SEMICONDUCTOR
Si Si Si
extra
electron
Si As Si
Si Si Si
23
Si Si Si
free
positively charged electron
site (DONOR)
Si As Si
Si Si Si
24
P-TYPE EXTRINSIC SEMICONDUCTOR
Si Si Si
hole
Si B Si
Si Si Si
25
Si Si Si
Si B Si
Si Si Si
26
N-type Metal Oxide Semiconductor Field Effect Transistor (NMOSFET)
VDS
VGS
metal gate
gate oxide
drain source
N+ N+
electron path
e e
e
bulk
27
P-type Metal Oxide Semiconductor Field Effect Transistor (PMOSFET)
VSD
VSG
metal gate
P+ P+
hole path
h h
h
bulk
VBS
28
B. MOSFET CHARACTERISTICS
(i) NMOSFET
ID Saturation
Quadratic
VGS
Linear
D
G ID
VDS
VDsat=VGS-VT
S
μ nCoxWn ⎡ VDS 2 ⎤
ID = ⎢(VGS − VTn ) VDS − ⎥
L ⎣ 2 ⎦
μ nCoxWn
ID ≅ ⎡⎣(VGS − VTn ) VDS ⎤⎦ linear region for small VDS
L
μ nCoxWn
ID = (VGS − VTn ) for VDS ≥ VDsat = VGS - VTn saturation region for large VDS
2
2L
(ii) PMOSFET
ID
VSG
D
G ID
VSD
S VSDsat=VSG - VT
μ pCoxWp ⎡ VSD 2 ⎤
ID = ⎢(V SG − V Tp ) V SD −
2 ⎥⎦
L ⎣
μ pCoxWp
ID ≅ ⎡⎣(VSG − VTp ) VSD ⎤⎦ linear region for small VSD
L
μ pCoxWp
ID = (VSG − VTp ) for VSD ≥ VSDsat = VGS -VTp saturation region for large VSD
2
2L
29
C. LARGE SIGNAL EQUIVALENT CIRCUIT OF MOSFETS
(i) NMOSFET LARGE SIGNAL EQUIVALENT CIRCUIT
Rn
ID
VDS 1
Rn = =
μ nCoxWn
ID
(VGS − VTn )
L
1 μ nCoxWn
Rn = where Kn ≅
Kn (VGS − VTn ) L
Kn
I Dsat = (VGS − VTn )
2
1
If VGS =VDD & VTn ≅ 0.2VDD then: Rn = & I DSAT = 0.32KnVDD 2
0.8 KnVDD
ID Rp
1 μ pCoxWp
Rp = where Kp =
Kp (VSG − VTp ) L
Kp
(VSG − VTp )
2
I DSAT =
2
30
D. Complementary MOS (CMOS) Inverter Static Characteristics
VDD
VDD - vin vin(t)
S ID
G
D
vout VDD
+ D
vin G
- S
0 t
VSGp = VDD-vin
VGSn = vin
vout = VDSn VDD - vout = VSDn
Overlapping these two curves on top of each other yields:
vout = VDSn = VDD when VSDp = 0
vout = 0 when VSDp = VDD
ID
1
VGSn = vin
0 vout
VDD
1
At pfet is in linear region
nfet is in saturation region
31
Rp
vout
IDSAT
1 1
v out = VDD - RpI DSAT where Rp = =
Kp ( VSGp -VTp ) Κp ( VDD -v in -VTp )
Kn Kn
I DSAT = (VGSn − VTn ) = ( vin − VTn )
2 2
2 2
Kn ( vin − VTn )
2
vout = VDD −
2 Kp (VDD − vin − VTp )
ID
vout
ISDSAT
vout
IDSSAT
32
Kp
(VSGp − VTp )
2
I SDSAT =
2
Kn
= (VGSn − VTn )
2
I DSSAT
2
Kp (VGSn − VTn )
2
μ pCoxWp μ nCoxWn
We had Kp = and Kn =
L L
μ pWp vin − VTn
=
μ nWn VDD − vin − VTp
μ p Wp v −V Wp
= = in
=R T
2 μ pWn V − v − VDD
2Wn in T
R (VDD − VT ) + VT
vin =
1+ R
VDD
if Wp = 2Wn ⇒ R=1 ⇒ vin =
2
if Wp = Wn ⇒ R=0.71 ⇒ vin ≅ 0.41VDD
VGSn = vin
3 VSGp = VDD-vin
0 vout
33
At 3 nfet is in linear region
pfet is in saturation region
ISDSAT
vout
Rn
1 1
Where, Rn= =
Kn (VGSn − VTn ) Kn ( vin − VT )
Kp
(VSGp − VTp ) =
Kp
(VDD − vin − VTp )
2 2
I SDSAT =
2 2
Kp (VDD − vin − VTp )
2
v out =
2Kn ( vin − VTn )
Plotting vout in terms of vin yields:
vout
“ BELL CURVE ”
Kn ( vin − VTn )
2
v out = VDD −
2 Kp (VDD − vin − VTp )
34
ID
VSGp VGSn
2
VSGp VGSn
3
VSGp 1 VGSn
VDD vout
As vin approaches from 0 to VDD then the quiescent point (operating point) of the the
inverter transverses the DASHED ARC from 1 to 3 through 2 in ID vs. vout curve.
As vin approaches from VDD to 0, the quiescent point of the inverter follows the
DASHED ARC from 3 to 1 .
Wp
Now, let’s plot the BELL CURVE for different values of :
Wn
Wp
Wn vin at 2
1 0.41 VDD
2 0.5 VDD
3 0.55 VDD
4 0.58 VDD
6 0.63 VDD
0.41VDD
0.63VDD
0.5VDD
35
Wp
• When = 6 (strong pfet weak nfet) large increases in vin cannot turn on the nfet,
Wn
vout delays to decrease towards 0.
Wp/Wn=2
d v ou t
= −1
dt
v out dvout
= −1
dt
Wp/W n=1
d v ou t
= −1
dt
t t
36
When nfet is stronger with respect to pfet (Wp/Wn=1) in an inverter, then small input
voltages at vin can easily turn on the nfet. Therefore, the inverter with Wp/Wn=1 has low
noise margin or noise immunity for positive glitches at its input. But, the same inverter
exhibits high noise immunity for negative glitches at its input.
dvout
v out = −1
dt
Wp/W n=6
d v ou t
= −1
dt
For an inverter with Wp/Wn=6, the inverter exhibits high noise immunity for positive
glitches & low noise immunity for negative glitches.
If vin is changed from 0 to VDD without any transition time (rise time = 0 sec), the pfet is
turned OFF, and nfet is put in the linear region and shows resistive characteristics.
pfet is OFF
vout
Rn
1
When vin = VDD ⇒ Rn ≅
Kn (VDD − VTn )
37
1
Assume VTn ≅ 0.2VDD Rn ≅
0.8KnVDD
Now, if vin is switched back from VDD to 0 without any transition time, this time nfet is
turned off and pfet is put in the linear region and shows resistive characteristics.
Rp
vout
nfet is OFF
1
when vin=0 ⇒ Rp ≅
Kp (VDD − VTp )
1
Assume VTp ≅ 0.2VDD and Rp ≅
0.8KpVDD
This analysis aids to derive the propagation delay of the inverter with a load capacitor,
CL, at its output.
VDD
VDD - vin
S ID
G
D
vout
+ D
CL
vin G
- S
When vin changes from 0 to VDD, then nfet turns ON and pfet turns OFF.
38
Vout(0) = VDD
Rn CL
t
−
vout = VDD e RnCL
vin,vout
vout(t)
VDD/2 TpL
0 t
TpL
TpL
−
VDD
vout = = VDD e nCL
R
2
When vin goes from VDD to 0, then nfet is turned OFF and pfet is turned ON.
Rp
Vout (0) = 0
CL
t
−
vout = VDD (1 − e RpCL
)
39
vin,vout
vout(t)
vin(t) VDD/2
0 t
TpH
Let;
0.86CL M
M= L ⇒ TPL =
VDD μ pCox 2Wn
M
TPH =
Wp
For
Wp M M T
= 1 (strong nfet, weak pfet) ⇒ TPL = , TPH = ⇒ TPL = PH
Wn 2Wn Wn 2
Wp M M
= 2 (equal nfet and pfet) ⇒ TPL = , TPH = ⇒ TPL = TPH
Wn 2Wn 2Wn
40
Wp M M
= 4 (weak nfet, strong pfet) ⇒ TPL = , TPH = ⇒ TPL = 2TPH
Wn 2Wn 4Wn
FANOUT
Cin
Fanout of a gate (inverter in this case) is the number of identical gates at the output of this
gate.
Therefore,
TPL = 0.69 RnCL = 0.69 N RnCin
TPH = 0.69 RpCL = 0.69 N RpCin
Tp
TPH = 0.69 RpCL
CL
IMPORTANT OBSERVATION
Tp changes linearly with CL. That means more output capacitance (fan-out) linearly
produces propagation delay.
Note that, Tp= 0 when CL =0 F. However, this is impossible because every inverter has an
intrinsic load capacitor due to source/drain contact capacitance even though the external
load capacitor, CL, may not exist. Therefore, if one assumes CL as a physical capacitance
due to gate fanout or wiring, the practical Tp vs CL will be as follows.
41
Tp
TpL
TpH
TpL0
TpH0
CL
vin,vout
V DD
0.9V
DD
v (t)
in
v o ut(t)=VDD (1-e-t/Rp CL)
0.1VDD
0 t
t
1
t
2
TRise
⎛ −
t1 ⎞
⎜ RpCL ⎟
vout t =t 1
= 0.1 VDD = VDD ⎜1 − e ⎟
⎜ ⎟
⎝ ⎠
t1
−
0.1 − 1 = −e RpCL
42
⎛ −
t2 ⎞
⎜ ⎟
vout t =t 2
= 0.9 VDD = VDD ⎜ 1 − e pCL
R
⎟
⎜ ⎟
⎝ ⎠
t2
−
0.9 − 1 = −e pCL
R
v ,v
in ou t
v (t)
in
V DD
0.9VDD
v ou t(t)=VDD (e -t/R n C L)
0.1VDD
0
t 1 t 2
t
T FALL
⎛ − t1 ⎞
⎜ ⎟
vout t =t 1 = 0.9 VDD = VDD ⎜ e RnCL ⎟
⎜ ⎟
⎝ ⎠
t1 = − RnCL ln 0.9= 0.1 RnCL
⎛ − t2 ⎞
⎜ ⎟
vout t =t 2 = 0.1VDD = VDD ⎜ e RnCL ⎟
⎜ ⎟
⎝ ⎠
t2 = − RnCL ln 0.1= 2.3 RnCL
Wp
Example: Compare TRISE, TFALL for = 1 , 2 and 4. Assume VT ≈ 0.2VDD
Wn
43
1 1
Solution: Rn = & Rp =
0.8VDD Kn 0.8VDD Kp
μ nC Wn 2μ pCoxWn
Kn = ox
=
L L
μ pC Wp
Kp = ox
1 1 1 1
Rn = Rp =
0.8VDD μ nCox 2Wn 0.8VDD μ pCox Wp
L P
P
Let P then Rn = & Rp =
0.8VDD μ pCox 2Wn
Wp
Wp 1 1
For = 1 then TRISE =(2.2 C L P) and TFALL =(2.2 CL P)
Wn Wn 2Wn
T
therefore TRISE = FALL
2
Wp 1 1
For = 2 then TRISE =(2.2 C L P) and TFALL =(2.2 CL P)
Wn 2Wn 2Wn
therefore TRISE = TFALL
Wp 1 1
For = 4 then TRISE =(2.2 C L P) and TFALL =(2.2 CL P)
Wn 4Wn 2Wn
therefore 2TRISE = TFALL
H. Power Consumption
vout
i (t)
CL
T
1
T ∫0
P= i (t )vout (t )dt power spent to charge CL.
dvout
i = CL
dt
Then,
44
T
1 dv
P= ∫ CL out vout (t ) dt
T0 dt
CL vout 2 CL VDD 2
= =
VDD
0
T 2 T 2
Therefore,
vin
VDD
0 Tperiod t
Tperiod
2
1
⎛ Tperiod ⎞ ∫0
P1 = CL vout dvout
⎜ ⎟
⎝ 2 ⎠
Tperiod
1
P2 = ∫ CL vout dvout
⎛ Tperiod ⎞ Tperiod
⎜ ⎟ 2
⎝ 2 ⎠
CL VDD 2 1
P = P 1 + P2 = = CL VDD 2 f
Tperiod 2 2
vout
VDD
P1 P2
0 Tperiod/2 Tperiod/2 t
45
I. TECHNOLOGY SCALING
Example: Compare 0.5µ technology with tox=100A , 3.3V @ 100 MHz
with 0.1µ technology with tox=50A , 1V @ 5 GHz.
Wp
Keep = K the same in both technologies.
Wn
Solution:
ε ( K + 1)
For 0.1µ CL =0.1u = ox
Wn (0.1μ )
50 A0
ε ( K + 1)
0.5µ CL =0.5u = ox
5Wn (0.5μ )
100 A0
1
PL=0.1u = CL =0.1uVDD 2 L=0.1u f L=0.1u
2
1 ε ox ( K + 1)
= 0
Wn (0.1μ ) (1V) 2 (5000MHz )
2 50 A
1 ε ox ( K + 1)
PL=0.5u = 0
5Wn (0.5μ ) (3.3V) 2 (100 MHz )
2 100 A
PL=0.1u ⎛ 100 A0 ⎞ ⎛ 1 ⎞ ⎛ 0.1μ ⎞ ⎛ 1V ⎞ ⎛ 5000 MHz ⎞
2
=⎜ ⎟⎜ ⎟⎜ ⎟⎜ ⎟ ⎜ ⎟ = 0.368
PL=0.5u ⎝ 50 A0 ⎠ ⎝ 5 ⎠ ⎝ 0.5μ ⎠ ⎝ 3.3V ⎠ ⎝ 100 MHz ⎠
46
CHAPTER III. CMOS LOGIC
in 1
Complementary
pfet tree
in k
out
nfet tree
The nfet tree is formed between the output & ground to produce a logic function at the
output. The pfet tree is formed between VDD & output to replicate the complementary
version of the nfet tree.
In a pfet tree, parallel nfet interconnects are converted into serial pfet interconnects and
serial nfet interconnects are converted into parallel pfet interconnects.
A B
outNAND
A
A B outNAND Comments
0 0 1 pfetA on, pfetB on
0 1 1 pfetA on only
1 0 1 pfetA on only
1 1 0 nfetA on, nfetB on
47
(ii) 2-Input NOR Gate
B
outNOR
A B
A B outNOR Comments
0 0 1 pfetA on, pfetB on
0 1 0 nfetA on only
1 0 0 nfetA on only
1 1 0 nfetA on, nfetB on
Example: out= D + A( B + C )
out
A D
B C
48
Form pfet tree second:
B A
D
out
B A
D
out
A D
B C
Example:
A
0
out
1
B
49
First obtain out = A.EN + B.EN
A EN
out
A B B EN
EN EN
out
nfet-tree pfet-tree
A EN
B EN
out
A B
EN EN
50
C. Fundamentals of Transistor Sizing
Transistor sizing is achieved by determining the maximum charge & discharge paths.
(i) Inverter
wp charge path
in
wn CL
discharge path
P P
Rn= and Rp=
2Wn Wp
A B charge path
Wp Wp
out
A Wn CL
discharge path
B Wn
51
From critical charge path:
P
Rpeq = Rp =
Wp
P P
Rneq = Rn + Rn = 2 =
2Wn Wn
P P
= ⇒ Wp = Wn
Wp Wn
Thus:
A B
Wn Wn
out
A Wn CL
B Wn
P P Wn
=2 ⇒ Wp=
Wp Wn 2
52
(iii) 3-Input NAND Gate
A B C charge path
Wp Wp Wp
out
A Wn CL
discharge path
B Wn
C Wn
P
Rpeq = Rp =
Wp
P
Rneq = Rn + Rn + Rn = 3
2Wn
P 3 P
Therefore, if we want TPH = TPL ⇒ Rpeq = Rneq ⇒ =
Wp 2 Wn
2
Wp= Wn
3
A Wp
charge path
B Wp
out
CL
A B discharge path
Wn Wn
53
TPH = 0.69 C L Rpeq (rise delay)
TPL = 0.69 C L Rneq (fall delay)
2P
Rpeq = Rp + Rp =
Wp
P
Rneq = Rn =
2Wn
2P P
=
Wp 2Wn
Wp = 4Wn
(v) out = D + A ( B + C )
54
TPH = 0.69 C L Rpeq (rise delay)
TPL = 0.69 C L Rneq (fall delay)
3P
Rpeq = Rp + Rp + Rp = 3Rp =
Wp
P P
Rneq = Rn + Rn = 2Rn = 2 =
2 Wn Wn
3P P
If we want TPH = TPL ⇒ Rpeq = Rneq ⇒ =
Wp Wn
Wp=3Wn
Wn2 = Wn
P P P P
Rn3 = Rn + Rn = 2 = ⇒ =
2 Wn Wn 2Wn3 Wn
Wn
Wn3 =
2
Therefore, all the other charge and discharge paths will be equal to the critical charge and
discharge paths, respectively.
55
D. MORE ACCURATE DELAY CALCULATION, ELMORE DELAY
When we were calculating TpH & TpF earlier, we did not take the intrinsic source/drain
capacitances in the delay calculation. Elmore delay calculation considers these intrinsic
capacitances in the following manner:
nfet tree:
out
M1
CL+C1
C2
M2
C3
M3
CN
MN
56
pfet tree:
C0
N1
C1
N2
C2
N3
C(K-1)
NK
out
CL+CK
TpH = 0.69{0.C 0
+ Rp1C1
+ ( Rp1 + Rp 2 ) C 2
+ ( Rp1 + Rp 2 + Rp 3 ) C 3
+.........
+ ( Rp1 + Rp 2 + ....... + Rpk ) (CL + CK )}
57
Example: Determine Elmore delays for 2 NAND with intrinsic S/D capacitance, Ci.
Let Ci be the capacitance per transistor width.
For practical purposes, each transistor has a separate Ci at its source and drain junctions.
A B charge path
out
A CL+(2Wp +Wn )Ci
discharge path
B 2Wn Ci
58
Example: Determine Elmore delays for 4 NAND with intrinsic S/D cap=Ci
A B C D
out
A (4Wp +W n )Ci CL
B 2W n Ci
2W n Ci
C
2W n Ci
D
If we want:
59
Example: Determine Elmore delays for 2 NOR with intrinsic S/D cap=Ci.
A Wp
B 2Wp Ci
Wp
charge path
out
CL+ (Wp +2Wn )Ci
A B
discharge path
Wn Wn
60
Example: Determine Elmore delays for Out = D + A ( B + C )
B Wp A Wp
2Wp Ci
C Wp
3Wp Ci
D Wp
out
CL+Ci(2Wn +Wp )
A Wn D Wn
3Wn Ci
B Wn C Wn
61
E. INTERCONNECT DESIGN STRATEGIES TO REDUCE ELMORE
DELAYS
Consider out = ( A + B + C ) D . There are 2 ways to implement this:
B D
A
out
CL+(2Wp+Wn)Ci
D
4WnCi
A B C
Implementation #1
B D
A
out
A CL+(2Wp+3Wn)Ci
B C
4WnCi
D
Implementation #2
62
CHAPTER IV. PASS-TRANSISTOR LOGIC
When VG = VDD and VD = VDD are applied to the nfet in the above circuit, the initial
voltage across the load capacitor, CL, is zero. Therefore, initially VGS = VDD and VDS =
VDD. This condition induces a maximum current through the nfet (case 1). As this
current charges the output capacitor, CL, and the output voltage rises, the current through
the nfet declines steadily (case 2). When vout = VDD - VT, the current through the nfet
reaches sub-threshold level, which is practically zero (case 3).
Therefore, a logic 1 is applied to the drain (input) of an nfet, one can never reach the full
VDD at the source (output); the maximum attainable value is always VDD - VT.
Final form:
VG = VDD
VDD
VT
vout = VDD - VT
ID= 0 CL
63
Any problem if we apply ground instead of VDD?
ID
VG = VDD
2 1
VGS VGS = VDD
vout = VDD
ID CL
3
VDS
VT VDD /2 VDD
With the above bias across the nfet, output capacitor is discharged completely, and vout
reaches 0 since VGS = VDD at all times. Therefore, if a logic 0 is applied to the source
(input) of an nfet, a logic zero is obtained at the drain (output).
ID CL
3 VSG = VT
VSD
VT V DD /2 V DD
When VG = 0 and vout(0) = VDD are applied to the pfet in the above circuit, the initial VSG
= VDD and VSD = VDD. This condition induces a maximum current through the pfet (case
1). As this current discharges the output capacitor, CL, and the output voltage decreases,
the current through the nfet declines steadily (case 2). When vout = VT, the current
through the pfet reaches sub-threshold level, which is practically zero (case 3).
Therefore, a logic 0 is applied to the drain (input) of a pfet, one can never reach 0 at the
source (output); the minimum attainable value is always VT.
64
Any problem if we apply VDD instead of 0 ?
ID
VSG VG = 0
2 3
V SG = VDD
vout (0) = 0
ID CL
1
VSD
0 V DD /2 VDD
With the above bias across the pfet, output capacitor gets charged completely, and vout
reaches VDD since VSG = VDD at all times. Therefore, if a logic 1 is applied to the source
(input) of a pfet, a logic 1 is obtained at the drain (output).
CONCLUSIONS:
(1)
VDD VDD
VDD
VDD - VT 0
ON ON
CL CL
(2)
0 0
VDD VDD
ON CL ON CL
Because of the level shifting problem in nfet and pfet for VDD and 0, respectively,
transmission gate is developed.
65
B. TRANSMISSION GATES
The transmission gate is the parallel combination of nfet and pfet with an inverter.
out = out =
A A
AB AB
Example: out = A ⊕ B = AB + AB
AB
A
out = A B + A B B out = A B + A B
A
B
B A
AB
66
AB
A
out = A B C
B C
Even though one gets a functionally correct output, one of the biggest drawbacks of serial
transmission gate is that it generates a “slow node” at the output due to high series
resistance. The charge path:
W peq Rp Rp
vout
CL
Rn Rn
Where,
P
Rp =
Wp
P
Rn =
2Wn
Rn R p P
=
Rn + R p W p + 2Wn
R p Rn 1 2
TRISEslow = 2.2CL ( Rpeq + 2 ) = 2.2CL P( + )
R p + Rn Wpeq W p + 2Wn
Rp Rp
vout
CL
Rn Rn
W neq
1 2
TFALLslow = 2.2CL P( + )
2Wneq W p + 2Wn
67
1
Therefore, more serial transmission gate induces more - term, which results in
W p + 2Wn
generating slower output node at vout.
To prevent the slow node formation signals need to be buffered by going to the gate of
the next transistor rather than its drain (or source):
AB
A
C out = A B C
W peq Rp W peq Rp
vab vout
Cinv+Cnox CL
Rn Rn
1 1
TRISEab = 2.2 P (Cinv + Cnox )( + ) < TRISEslow
W peq W p + 2Wn
1 1
TRISEout = 2.2 P (Cinv + Cnox )( + ) < TRISEslow
W peq W p + 2Wn
Similarly,
1 1
TFALLab = 2.2CL P( + ) < TFALLslow
2Wneq W p + 2Wn
1 1
TFALLout = 2.2CL P( + ) < TFALLslow
2Wneq W p + 2Wn
68
Example: Full adder design using transmission gates
A
AB+AB
B
Cin
A
SUM
A
AB+AB
B
Cin
A
Cin
Cout
69