Overview of VLSI Design Flow
Chip manufacturing process:
d
e
s
i
g
n
s
t
y
l
e
{
f
u
l
l
c
o
s
t
u
m
e
:
d
e
s
i
g
n
e
v
e
r
y
t
h
i
n
g
s
t
a
n
d
a
r
d
c
e
l
l
b
a
s
e
d
e
s
i
g
n
(
A
S
I
C
)
:
u
s
e
p
r
e
−
d
e
s
i
g
n
g
a
t
e
design\,\, style \begin{cases} full\,\,costume &\text{: } design\,\, everything \\ standard\,\, cell\,\, base \,\,design\,(ASIC) &\text{: } use\,\, pre-design\,\, gate \end{cases}
designstyle{fullcostumestandardcellbasedesign(ASIC): designeverything: usepre−designgate
full costume: minimized area, power, maximized speed, expensive and long-period
ASIC: just cheap and short-period
Cost of an Integrated Circuit
cost per IC = variable cost per IC + (fixed cost / volume)
variable cost per IC=(cost of die +cost of test +cost of packaging)/final test yield
f
i
n
a
l
t
e
s
t
Y
i
e
l
d
Y
=
N
u
m
b
e
r
o
f
g
o
o
d
c
h
i
p
s
p
e
r
w
a
f
e
r
T
o
t
a
l
n
u
m
b
e
r
o
f
c
h
i
p
p
e
r
w
a
f
e
r
∗
100
%
D
i
e
c
o
s
t
=
W
a
f
e
r
c
o
s
t
D
i
e
s
p
e
r
w
a
f
e
r
×
D
i
e
y
i
e
l
d
D
i
e
s
p
e
r
w
a
f
e
r
=
π
×
(
w
a
f
e
r
d
i
a
m
e
t
e
r
2
)
2
d
i
e
a
r
e
a
−
π
×
w
a
f
e
r
d
i
a
m
e
t
e
r
2
×
d
i
e
a
r
e
a
D
i
e
y
i
e
l
d
=
(
1
+
d
e
f
e
c
t
s
p
e
r
u
n
i
t
a
r
e
a
×
d
i
e
a
r
e
a
α
)
−
α
α
≈
3
final\;test\;Yield\,\, Y=\frac {Number\,\, of\,\, good \,\,chips\,\, per\,\, wafer}{Total\,\, number\,\, of\,\,chip\,\,per\,\,wafer} *100\%\\ Die \,\,cost=\frac {Wafer \,\,cost}{Dies\,\, per \,\,wafer \times Die \,\,yield}\\ Dies\,\,per\,\,wafer=\frac{\pi \times (\frac {wafer\,\,diameter} 2)^2}{die\,\, area} - \frac {\pi \times wafer\,\,diameter}{\sqrt{2 \times die \,\,area}}\\ \\ Die \,\,yield=(1+\frac {defects\,\,per\,\,unit\,\,area\times die\,\,area}{\alpha})^{-\alpha}\\ \alpha\approx3
finaltestYieldY=TotalnumberofchipperwaferNumberofgoodchipsperwafer∗100%Diecost=Diesperwafer×DieyieldWafercostDiesperwafer=dieareaπ×(2waferdiameter)2−2×dieareaπ×waferdiameterDieyield=(1+αdefectsperunitarea×diearea)−αα≈3
Example:
we have a 4-inch wafer fab and an 8-inch wafer fab.
Calculate the cost per IC of each design at each fabrication plant (4 calculations, 2 per 4-inch and 2 per 8-inch). Identify the minimum cost fabrication facility for each design.
The microprocessor die size is 3.2
c
m
2
cm^2
cm2 and the projected volume is 50,000,000 units over the lifetime of the product. The cost of packaging the microprocessor is $25.00 per part, cost of testing per part is $2.78, and the non-recurring engineering design cost is $200,000,000.00.
The ASIC design is
0.25
c
m
2
0.25 cm^2
0.25cm2 and the projected volume is only 1,000,000 units. The cost of packaging the ASIC is $0.67, cost of testing is $0.69, and the design cost is $2,000,000.00.
For both designs, the functional test yield rate is 95% and α=3. Information on the fabrication options is listed in the table below.
Ans:
Microprocessor with 4-inch wafer:
cost per IC = variable cost per IC + (fixed cost / volume)=variable cost per IC + (
2
∗
1
0
8
/
5
∗
1
0
7
2*10^8/5*10^7
2∗108/5∗107)
variable cost per IC= (cost of die +cost of test +cost of packaging)/final test yield
cost of test = 25
cost of packaging= 2.78
cost of die
=
W
a
f
e
r
c
o
s
t
D
i
e
s
p
e
r
w
a
f
e
r
×
D
i
e
y
i
e
l
d
=
150
D
i
e
s
p
e
r
w
a
f
e
r
×
D
i
e
y
i
e
l
d
=\frac {Wafer \,\,cost}{Dies\,\, per \,\,wafer \times Die \,\,yield}=\frac {150}{Dies\,\, per \,\,wafer \times Die \,\,yield}
=Diesperwafer×DieyieldWafercost=Diesperwafer×Dieyield150
final test yield = 95%
D
i
e
s
p
e
r
w
a
f
e
r
=
π
×
(
10.16
2
)
2
3.2
−
π
×
10.16
2
×
3.2
Dies\,\,per\,\,wafer=\frac{\pi \times (\frac {10.16} 2) ^2}{3.2} - \frac {\pi \times 10.16}{\sqrt{2 \times 3.2}}
Diesperwafer=3.2π×(210.16)2−2×3.2π×10.16
Die yield=
(
1
+
d
e
f
e
c
t
s
p
e
r
u
n
i
t
a
r
e
a
×
d
i
e
a
r
e
a
α
)
−
α
=
(
1
+
0.5
×
3.2
3
)
−
3
(1+\frac {defects\,\,per\,\,unit\,\,area\times die\,\,area}{\alpha})^{-\alpha}=(1+\frac {0.5\times3.2}{3})^{-3}
(1+αdefectsperunitarea×diearea)−α=(1+30.5×3.2)−3
Die yield=0.28
Dies per wafer=12
final test yield = 95%
cost of die=44.64
cost of packaging=2.78
cost of test = 25
variable cost per IC = 76.23
cost per IC = 76.23+
2
∗
1
0
8
/
5
∗
1
0
7
≈
2*10^8/5*10^7\approx
2∗108/5∗107≈ $80.23
Microprocessor with 8-inch wafer:
Die yield=
(
1
+
0.56
×
3.2
3
)
−
3
(1+\frac {0.56\times3.2}{3})^{-3}
(1+30.56×3.2)−3= 0.245
Dies per wafer=
π
×
(
20.32
2
)
2
3.2
−
π
×
20.32
2
×
3.2
\frac{\pi \times (\frac {20.32} 2) ^2}{3.2} - \frac {\pi \times 20.32}{\sqrt{2 \times 3.2}}
3.2π×(220.32)2−2×3.2π×20.32= 76
final test yield = 95%
cost of die=
800
76
×
0.25
\frac {800}{76 \times 0.25}
76×0.25800=42.9
cost of packaging=2.78
cost of test = 25
variable cost per IC = (42.9+2.78+25) / 0.95 = 73.45
cost per IC = 74.45+
2
∗
1
0
8
/
5
∗
1
0
7
≈
2*10^8/5*10^7\approx
2∗108/5∗107≈ $78.45
ASIC with 4-inch wafer:
cost per IC = variable cost per IC + (fixed cost / volume)=variable cost per IC + (
2
∗
1
0
6
/
1
0
6
2*10^6/10^6
2∗106/106)
variable cost per IC= (cost of die +cost of test +cost of packaging)/final test yield
cost of test = 0.69
cost of packaging= 0.67
cost of die
=
W
a
f
e
r
c
o
s
t
D
i
e
s
p
e
r
w
a
f
e
r
×
D
i
e
y
i
e
l
d
=
150
D
i
e
s
p
e
r
w
a
f
e
r
×
D
i
e
y
i
e
l
d
=\frac {Wafer \,\,cost}{Dies\,\, per \,\,wafer \times Die \,\,yield}=\frac {150}{Dies\,\, per \,\,wafer \times Die \,\,yield}
=Diesperwafer×DieyieldWafercost=Diesperwafer×Dieyield150
final test yield = 95%
D
i
e
s
p
e
r
w
a
f
e
r
=
π
×
(
10.16
2
)
2
0.25
−
π
×
10.16
2
×
0.25
Dies\,\,per\,\,wafer=\frac{\pi \times (\frac {10.16} 2) ^2}{0.25} - \frac {\pi \times 10.16}{\sqrt{2 \times 0.25}}
Diesperwafer=0.25π×(210.16)2−2×0.25π×10.16
Die yield=
(
1
+
d
e
f
e
c
t
s
p
e
r
u
n
i
t
a
r
e
a
×
d
i
e
a
r
e
a
α
)
−
α
=
(
1
+
0.5
×
0.25
3
)
−
3
(1+\frac {defects\,\,per\,\,unit\,\,area\times die\,\,area}{\alpha})^{-\alpha}=(1+\frac {0.5\times0.25}{3})^{-3}
(1+αdefectsperunitarea×diearea)−α=(1+30.5×0.25)−3
Die yield=0.88
Dies per wafer= 279
final test yield = 95%
cost of die= 0.61
cost of packaging=0.69
cost of test = 0.67
variable cost per IC = 2.07
cost per IC = 2.07+
2
∗
1
0
6
/
1
0
6
≈
2*10^6/10^6\approx
2∗106/106≈ $4.07
ASIC with 8-inch wafer:
Die yield=
(
1
+
0.56
×
0.25
3
)
−
3
(1+\frac {0.56\times0.25}{3})^{-3}
(1+30.56×0.25)−3= 0.87
Dies per wafer=
π
×
(
20.32
2
)
2
0.25
−
π
×
20.32
2
×
0.25
\frac{\pi \times (\frac {20.32} 2) ^2}{0.25} - \frac {\pi \times 20.32}{\sqrt{2 \times 0.25}}
0.25π×(220.32)2−2×0.25π×20.32= 1206
final test yield = 95%
cost of die=
800
1206
×
0.87
\frac {800}{1206 \times 0.87}
1206×0.87800= 0.76
cost of packaging= 0.69
cost of test = 0.67
variable cost per IC = (0.76+0.69+0.67) / 0.95 = 2.23
cost per IC = 2.23+
2
∗
1
0
8
/
1
0
6
≈
2*10^8/10^6\approx
2∗108/106≈ $4.23
A new 12-inch fabrication facility for production of the microprocessor part. The 12-inch fabrication facility will cost an additional $1,000,000,000.00 to construct for this project. Develop a constraint on the defect density of this new facility that will ensure cost savings over the previous minimum cost solution. Volume cost for 12-inch wafers is estimated to be
$1500/wafer.
Microprocessor with 12-inch wafer(new):
find defect density
x
x
x
Die yield=
(
1
+
x
×
3.2
3
)
−
3
(1+\frac {x\times3.2}{3})^{-3}
(1+3x×3.2)−3
Dies per wafer=
π
×
(
30.48
2
)
2
3.2
−
π
×
30.48
2
×
3.2
\frac{\pi \times (\frac {30.48} 2) ^2}{3.2} - \frac {\pi \times 30.48}{\sqrt{2 \times 3.2}}
3.2π×(230.48)2−2×3.2π×30.48= 190
final test yield = 95%
cost of die=
1500
190
×
D
i
e
y
i
e
l
d
\frac {1500}{190 \times Die \;yield}
190×Dieyield1500
cost of packaging=2.78
cost of test = 25
variable cost per IC = (cost of die+2.78+25) / 0.95
cost per IC = variable cost per IC+
(
2
∗
1
0
8
+
1
0
9
)
/
5
∗
1
0
7
<
(2*10^8+10^9)/5*10^7<
(2∗108+109)/5∗107< $78.45
d e f e c t d e n s i t y x ≤ 0.42 defect\;density \;x \le 0.42 defectdensityx≤0.42
Fan-out: the number of load gates
N
N
N that are connected to the output of the
driving gate
Fan-in: the number of inputs to the gate
propagation delay
t
p
t_p
tp:how quickly it responds to a change at its input(s)
measured between the 50% transition points of the input and output waveforms
t
p
=
t
p
H
L
+
t
p
L
H
2
t_p=\frac {t_{pHL}+t_{pLH}} 2
tp=2tpHL+tpLH
Period of ring oscillator
T
T
T
odd number(
N
N
N) of inverter
T
=
2
×
t
p
×
N
T=2\times t_p \times N
T=2×tp×N
noise margin
V
O
H
V_{OH}
VOH: Maximum output voltage when the output level is logic “1”
V
O
L
V_{OL}
VOL: Maximum output voltage when the output level is logic “0”
V
I
L
V_{IL}
VIL: Maximum input voltage which can be interpreted as logic “0”
V
I
H
V_{IH}
VIH: Maximum input voltage which can be interpreted as logic “1”
Propagation delay of First-order RC network
The time to reach the 50% point:
t
=
l
n
(
2
)
τ
=
0.69
τ
t=ln(2)\tau=0.69\tau
t=ln(2)τ=0.69τ
Time to the 90% point:
t
=
l
n
(
9
)
τ
=
2.2
τ
t=ln(9)\tau=2.2\tau
t=ln(9)τ=2.2τ
τ
=
R
C
,
t
i
m
e
c
o
n
s
t
a
n
t
\tau=RC, time\;constant
τ=RC,timeconstant
logic change from 0 to 1, this process needs power:
E
0
→
1
=
C
L
⋅
V
d
d
2
E_{0\rightarrow1}=C_L\cdot V_{dd}^2
E0→1=CL⋅Vdd2
logic change from 1 to 0 do not need energy
power dissipation of capacitor:
E
c
a
p
=
1
2
C
L
⋅
V
d
d
2
E_{cap}=\frac 1 2C_L\cdot V_{dd}^2
Ecap=21CL⋅Vdd2
CMOS Fabrication Process Technology
Photolithography: The technique to accomplish selective masking so that a desired processing step can be selectively applied to the remainingregions.
Photolithography invovle:
- Oxidation layering
- Photoresist coating
- Stepper exposure
- Photoresist development and bake
- Acid Etching
- Spin, rinse, and dry
- Various process steps
- Photoresist removal (or ashing)
Some important techniques used:Diffusion and Ion Implantation, Deposition, Etching, Planarization
Simplified process:
MOS Transistor Theory
static behaviour
Built-in potential: under zero bias, there exists a voltage ϕ 0 \phi_0 ϕ0 across the junction. ϕ 0 = ϕ T l n [ N A N D n i 2 ] \phi_0=\phi_Tln[\frac {N_AN_D}{n_i^2}] ϕ0=ϕTln[ni2NAND] n i n_i ni: intrinsic carrier concentration in a pure sample of the semiconductor and equals approximately 1.5 ∗ 1 0 10 c m − 3 1.5*10^{10} cm^{-3} 1.5∗1010cm−3 at 300 K for silicon.
Ideal diode equation:
I
D
=
I
S
(
e
V
D
/
ϕ
T
−
1
)
I_D=I_S(e^{V_D/ \phi_T}-1)
ID=IS(eVD/ϕT−1)
ϕ
T
\phi_T
ϕT: termal voltage, equal to 26 mV at room temperature.
I
S
I_S
IS: saturation current, constant value.
Junction Capacitance: the space-charge region contains few mobile carriers, it acts as an insulator with a dielectric constant
ε
s
i
\varepsilon_{si}
εsi of the semiconductor material. The n- and p-regions act as the capacitor plates.
C
j
=
C
j
0
(
1
−
V
D
/
ϕ
0
)
m
C_j=\frac {C_{j0}} {(1-V_D/\phi_0)^m}
Cj=(1−VD/ϕ0)mCj0
m
=
0.5
m=0.5
m=0.5 :abrupt junction
m
=
0.33
m=0.33
m=0.33 :linear junction
C
j
0
C_{j0}
Cj0: the capacitance under zero-bias conditions,
C
j
0
=
A
D
(
ε
s
i
q
2
N
A
N
D
N
A
+
N
D
)
ϕ
0
−
1
C_{j0}=A_D\sqrt{(\frac{\varepsilon_{si}q} 2\frac{N_AN_D}{N_A+N_D})\phi_0^{-1}}
Cj0=AD(2εsiqNA+NDNAND)ϕ0−1
ϕ
0
\phi_0
ϕ0: Built-in potential (defined previously)
MOS Transistor under Static Conditions
Threshold Voltage:
V
T
=
V
T
0
+
γ
∗
(
∣
−
2
ϕ
F
+
V
S
B
∣
−
∣
−
2
ϕ
F
∣
)
V_T=V_{T0}+\gamma*(\sqrt{|-2\phi_F+V_{SB}|}-\sqrt{|-2\phi_F|})
VT=VT0+γ∗(∣−2ϕF+VSB∣−∣−2ϕF∣)
V
T
0
V_{T0}
VT0: Threshold voltage for substrate bias voltage
V
S
B
=
0
V_{SB}=0
VSB=0
ϕ
F
\phi_F
ϕF: Fermi Potential,
ϕ
F
=
−
ϕ
T
l
n
(
N
A
n
i
)
\phi_F=-\phi_Tln(\frac {N_A}{n_i} )
ϕF=−ϕTln(niNA), sometimes 0.55V
N
A
N_A
NA: substrate doping
γ
\gamma
γ: body-effect coefficient,
γ
=
2
q
N
A
ε
S
i
/
C
o
x
\gamma=\sqrt{2qN_A\varepsilon_{Si}}/C_{ox}
γ=2qNAεSi/Cox
the voltage-current relation of the transistor in linear region/resistive region (
V
D
S
<
V
G
S
−
V
T
V_{DS}<V_{GS}-V_T
VDS<VGS−VT)
I
D
=
k
n
′
W
L
[
(
V
G
S
−
V
T
)
V
D
S
−
V
D
S
2
2
]
=
k
n
[
(
V
G
S
−
V
T
)
V
D
S
−
V
D
S
2
2
]
I_D =k'_n\frac W L[(V_{GS}-V_T)V_{DS}-\frac {V_{DS}^2} 2]\\ =k_n[(V_{GS}-V_T)V_{DS}-\frac {V_{DS}^2} 2]
ID=kn′LW[(VGS−VT)VDS−2VDS2]=kn[(VGS−VT)VDS−2VDS2]
the voltage-current relation of the transistor in saturation region(
V
D
S
>
V
G
S
−
V
T
V_{DS}>V_{GS}-V_T
VDS>VGS−VT)
I
D
=
k
n
′
2
W
L
(
V
G
S
−
V
T
)
2
I_D=\frac {k'_n} 2 \frac W L(V_{GS}-V_T)^2
ID=2kn′LW(VGS−VT)2
k
n
′
k'_n
kn′: process transconductance parameter,
k
n
′
=
μ
n
C
o
x
k'_n=\mu_nC_{ox}
kn′=μnCox
μ
n
\mu_n
μn: mobility parameter, expressed in
m
2
/
V
⋅
s
m^2/V\cdot s
m2/V⋅s
C
o
x
C_{ox}
Cox: capacitance per unit area presented by the gate oxide.
C
o
x
=
ε
o
x
t
o
x
C_{ox}=\frac {\varepsilon_{ox}} {t_{ox}}
Cox=toxεox
ε
o
x
\varepsilon_{ox}
εox: oxide permittivity,
3.5
×
1
0
−
11
3.5\times 10^{-11}
3.5×10−11
t
o
x
t_{ox}
tox: thickness of the oxide
k
n
k_n
kn: gain factor
W
W
W: channel width
L
L
L: channel length
V
G
S
V_{GS}
VGS: voltage apply between gate and source
V
D
S
V_{DS}
VDS: voltage apply between drain and source
V
T
V_T
VT:threshold voltage (define previously)
channel length modulation
increasing
V
D
S
V_{DS}
VDS causes the depletion region at the drain junction to grow,reducing the length of the effective channel, the current increases when the length factor
L
L
L decrease
I
D
=
I
D
′
(
1
+
λ
V
D
S
)
I_D=I_D'(1+\lambda V_{DS})
ID=ID′(1+λVDS)
I
D
′
I_D'
ID′: the current expressions derived earlier
λ
\lambda
λ: empirical parameter
short channel effect
- velocity saturation
- subthreshold current
- hot electron
- DIBL: brain induced barrier lowering
velocity saturation:
when the electrical field along the channel reaches a critical value
ξ
c
\xi_c
ξc, the velocity of the carriers tends to saturate due to scattering effects.
Velocity-saturation effects are more pronounced in NMOS short-channel transistors.
drain current in the resistive/linear region with velocity saturation effect
I
D
S
A
T
=
υ
s
a
t
C
o
x
W
(
(
V
G
S
−
V
T
)
−
V
D
S
A
T
)
=
κ
(
V
D
S
A
T
)
μ
n
C
o
x
W
L
[
(
V
G
S
−
V
T
)
V
D
S
A
T
−
V
D
S
A
T
2
2
]
V
D
S
A
T
=
κ
(
V
G
S
−
V
T
)
(
V
G
S
−
V
T
)
=
L
ξ
c
I_{DSAT}=\upsilon_{sat}C_{ox}W((V_{GS}-V_T)-V_{DSAT})\\ =\kappa(V_{DSAT})\mu_nC_{ox}\frac WL[(V_{GS}-V_T)V_{DSAT}-\frac {V_{DSAT}^2}2]\\ V_{DSAT}=\kappa(V_{GS}-V_T)(V_{GS}-V_T)=L\xi_c
IDSAT=υsatCoxW((VGS−VT)−VDSAT)=κ(VDSAT)μnCoxLW[(VGS−VT)VDSAT−2VDSAT2]VDSAT=κ(VGS−VT)(VGS−VT)=Lξc
υ
s
a
t
\upsilon_{sat}
υsat: saturation velocity,
=
μ
n
ξ
c
=\mu_n\xi_c
=μnξc, approximately equals
1
0
5
10^5
105 m/s.
κ
(
V
D
S
A
T
)
\kappa(V_{DSAT})
κ(VDSAT):
=
1
1
+
(
V
D
S
A
T
/
ξ
c
L
)
=\frac 1 {1+(V_{DSAT}/\xi_cL)}
=1+(VDSAT/ξcL)1, measure of the degree of velocity saturation
subthreshold conduction
the current does not drop abruptly to 0 at
V
G
S
=
V
T
V_GS=V_T
VGS=VT. MOS transistor is already partially conductiong for voltages below the threshold voltage. This effect is called subthreshold or weak-inversion conduction.
current in subthreshold exponential region:
I
D
=
I
S
e
V
G
S
n
k
T
/
q
(
1
−
e
V
D
S
k
T
/
q
)
I_D=I_Se^{\frac {V_{GS}}{nkT/q}(1-e^{\frac {V_{DS}}{kT/q}})}
ID=ISenkT/qVGS(1−ekT/qVDS)
I
S
,
n
I_S, n
IS,n: empirical parameters
slope factor
S
S
S: inverse rate of decline of the current with respect to
V
G
S
V_GS
VGS below
V
T
V_T
VT, measures by how much
V
G
S
V_GS
VGS has to be reduced for the drain current to drop by a factor of 10:
S
=
n
(
k
T
q
)
l
n
10
S=n(\frac {kT} q)ln10
S=n(qkT)ln10
unit:
m
V
/
d
e
c
a
d
e
mV/decade
mV/decade
hot electron/ hot carrier
Caused by high electric fields. Electrons and holes gaining high kinetic energies in the electric field (hot carriers) may be injected into the gate oxide, and cause permanent changes in the oxide-interface charge distribution, degrading the current-voltage characteristics of the MOSFET.
DIBL: drain-induced barrier lowering
If the drain voltage is increased, the potential barrier in the channel decreases, leading to drain-induced barrier lowering. The reduction of the potential barrier eventually allows electron flow between the source and the drain, even if the gate-to-source voltage is lower than the threshold voltage. The channel current that flows under these conditions (
V
G
S
<
V
T
0
V_{GS}<V_{T0}
VGS<VT0) is called the sub-threshold current.
dynamic behaviour
channel capacitance
total:
C
=
C
o
x
⋅
W
⋅
L
D
C=C_{ox}\cdot W\cdot L_D
C=Cox⋅W⋅LD
C
o
x
C_{ox}
Cox:gate oxide capacitance per unit area
W
W
W: channel width
L
D
L_D
LD: lateral diffusion
operation region | C g s C_{gs} Cgs- between gate and source | C g d C_{gd} Cgd- between gate and drain | C g b C_{gb} Cgb- between gate and body |
---|---|---|---|
Cutoff | 0 | 0 | C o x W ( L − 2 L D ) C_{ox}W(L-2L_D) CoxW(L−2LD) |
Resistive/linear | 1 2 C o x W ( L − 2 L D ) \frac1 2 C_{ox}W(L-2L_D) 21CoxW(L−2LD) | 1 2 C o x W ( L − 2 L D ) \frac1 2 C_{ox}W(L-2L_D) 21CoxW(L−2LD) | 0 |
Saturation | 2 3 C o x W ( L − 2 L D ) \frac 2 3 C_{ox}W(L-2L_D) 32CoxW(L−2LD) | 0 | 0 |
L
−
2
L
D
L-2L_D
L−2LD: effective channel length
p-mos pull up; n-mos pull down
Source-drain Resistance
R
S
,
D
=
R
□
L
S
,
D
W
+
R
C
R_{S,D}=R_{\square}\frac {L_{S,D}} W+R_C
RS,D=R□WLS,D+RC
L
S
,
D
L_{S,D}
LS,D: length of the source or drain region
W
W
W: width of the transistor
R
□
R_{\square}
R□: sheet resistance
R
C
R_C
RC: contact resistance
Circuit Characteristic and Performance Estimation
latch up
must be prevented by reducing substrate noise, it will ruin all devices
Why latch up will cause problem: the creation of a low-impedance possitive-feedback path between power supply rails as a result of triggering a parasitic device. The noise in this circuit will be amplifid and produce excessisve current.
wire capacitancce
capacitance of wire interconnect:
wire parallel plate capacitance:
C
=
ε
o
x
t
o
x
(
W
∗
L
)
C=\frac {\varepsilon_{ox}} {t_{ox}}(W*L)
C=toxεox(W∗L)
L: wire length
W: wire width
C
t
o
t
a
l
=
C
p
a
r
a
l
l
e
l
+
C
f
r
i
n
g
e
=
w
ε
o
x
t
o
x
+
2
π
ε
o
x
l
g
(
t
o
x
/
H
)
C_{total}=C_{parallel}+C_{fringe}=\frac {w\varepsilon_{ox}} {t_{ox}}+\frac {2\pi\varepsilon_{ox}}{lg(t_{ox}/H)}
Ctotal=Cparallel+Cfringe=toxwεox+lg(tox/H)2πεox
H: interconnect thickness
w
w
w:
w
=
W
−
H
/
2
w=W-H/2
w=W−H/2
ε
o
x
t
o
x
\frac {\varepsilon_{ox}} {t_{ox}}
toxεox. area capacitance(parallel-plate) are expressed in
a
F
/
μ
m
2
aF/\mu m^2
aF/μm2, while the fringe capacitance(given in the shaded rows) are in
a
F
/
μ
m
aF/\mu m
aF/μm
wire resistance
R
=
ρ
L
H
W
=
R
□
L
W
R=\frac {\rho L}{HW}=R_\square \frac LW
R=HWρL=R□WL
resistance (per unit length) at high frequencies(with skin effect, consider skin depth)
r
(
f
)
=
π
f
μ
ρ
2
(
H
+
W
)
r(f)=\frac {\sqrt{\pi f \mu \rho}}{2(H+W)}
r(f)=2(H+W)πfμρ
μ
\mu
μ: permeability of the surrounding dielectric,
4
π
×
1
0
−
7
H
/
m
4\pi\times 10^{-7} H/m
4π×10−7H/m
lumped model
how to calculate RC delay of a tree-structured network
τ = C 1 R 1 + C 2 ( R 1 + R 2 ) + C 3 ( R 1 + R 3 ) + C 4 ( R 1 + R 3 + R 4 ) + C i ( R 1 + R 3 + R i ) \tau=C_1R_1+C_2(R_1+R_2)+C_3(R_1+R_3)+C_4(R_1+R_3+R_4)+C_i(R_1+R_3+R_i) τ=C1R1+C2(R1+R2)+C3(R1+R3)+C4(R1+R3+R4)+Ci(R1+R3+Ri)
τ S D = C 1 R 1 + C 2 ( R 1 + R 2 ) + C 3 ( R 1 + R 2 + R 3 ) + C 4 ( R 1 + R 2 + R 3 ) + C 5 R 1 + C 6 R 1 + C 7 ( R 1 + R 2 ) + C 8 ( R 1 + R 2 + R 3 ) \tau_{SD}=C_1R_1+C_2(R_1+R_2)+C_3(R_1+R_2+R_3)+C_4(R_1+R_2+R_3)+C_5R_1+C_6R_1+C_7(R_1+R_2)+C_8(R_1+R_2+R_3) τSD=C1R1+C2(R1+R2)+C3(R1+R2+R3)+C4(R1+R2+R3)+C5R1+C6R1+C7(R1+R2)+C8(R1+R2+R3)
Time-Constant of Resistive-Capacitive Wire
τ
D
N
=
R
C
N
+
1
2
N
=
r
c
L
2
N
+
1
2
N
\tau_{DN}=RC\frac{N+1}{2N}=rcL^2\frac{N+1}{2N}
τDN=RC2NN+1=rcL22NN+1
wire delay:
r
c
2
l
2
\frac {rc}2l^2
2rcl2
r
r
r: resistance per unit length
c
c
c: capacitance per unit length
9 stages or 11 stage RC
simplified model:
π
\pi
π model or T model
in a chip:
top layer is the most thick layer
global signal travel same distance
If gate delay>wire delay, we can ignore wire delay
*diffusion equation of voltage node at node i of distributed RC line * ∂ 2 V ∂ x 2 = r c ∂ V ∂ t \frac {\partial^2V}{\partial x^2}=rc\frac{\partial V}{\partial t} ∂x2∂2V=rc∂t∂V
impedence of transmission line is not related to the length
impedence:
Z
=
Z
L
−
Z
0
Z
L
+
Z
0
Z=\frac{Z_L-Z_0}{Z_L+Z_0}
Z=ZL+Z0ZL−Z0
Z
L
Z_L
ZL: load
Z
0
Z_0
Z0: characteristic impedence
telegraph equation:
V
r
=
(
Z
L
−
Z
0
Z
L
+
Z
0
)
V
i
n
V_r=(\frac{Z_L-Z_0}{Z_L+Z_0})V_in
Vr=(ZL+Z0ZL−Z0)Vin
when impedence matching:
V
r
=
0
V_r=0
Vr=0
lattice diagram
Reflection Coefficient:
T
Z
L
=
Z
L
−
Z
0
Z
L
+
Z
0
=
ρ
L
T
Z
S
=
Z
S
−
Z
0
Z
S
+
Z
0
=
ρ
S
T_{Z_L}=\frac{Z_L-Z_0}{Z_L+Z_0}=\rho_L\\ T_{Z_S}=\frac{Z_S-Z_0}{Z_S+Z_0}=\rho_S
TZL=ZL+Z0ZL−Z0=ρLTZS=ZS+Z0ZS−Z0=ρS
incoming signal:
V
i
n
=
Z
0
Z
S
+
Z
0
V
s
=
V
i
n
i
t
i
a
l
V_{in}=\frac{Z_0}{Z_S+Z_0}V_s=V_{initial}
Vin=ZS+Z0Z0Vs=Vinitial
(
R
L
=
Z
L
;
R
S
=
Z
L
)
(R_L=Z_L ; R_S=Z_L)
(RL=ZL;RS=ZL)
a
=
V
i
n
i
t
i
a
l
b
=
a
ρ
L
c
=
b
ρ
s
d
=
c
ρ
L
e
=
d
ρ
s
f
=
e
ρ
L
a=V_{initial}\\ b=a\rho_L\\ c=b\rho_s\\ d=c\rho_L\\ e=d\rho_s\\ f=e\rho_L
a=Vinitialb=aρLc=bρsd=cρLe=dρsf=eρL
example:
R
S
=
5
Z
0
,
R
L
=
∞
,
V
S
=
5
V
R_S=5Z_0,\;R_L=\infty,V_S=5V
RS=5Z0,RL=∞,VS=5V
a
=
V
i
n
i
t
i
a
l
=
Z
0
Z
S
+
Z
0
V
s
=
1
6
∗
5
V
=
0.8333
V
ρ
L
=
Z
L
−
Z
0
Z
L
+
Z
0
=
1
b
=
a
ρ
L
=
5
6
∗
1
=
0.8333
V
ρ
S
=
Z
S
−
Z
0
Z
S
+
Z
0
=
2
3
c
=
b
ρ
s
=
5
6
∗
2
3
=
0.5556
d
=
c
ρ
L
=
5
9
∗
1
=
0.5556
e
=
d
ρ
s
=
5
9
∗
2
3
=
0.3704
f
=
e
ρ
L
=
0.3704
g
=
f
ρ
s
=
0.2469
h
=
g
ρ
L
=
0.2469
a=V_{initial}=\frac{Z_0}{Z_S+Z_0}V_s=\frac 16*5V=0.8333V\\ \rho_L=\frac{Z_L-Z_0}{Z_L+Z_0}=1\\ b=a\rho_L=\frac 56*1=0.8333V\\ \rho_S=\frac{Z_S-Z_0}{Z_S+Z_0}=\frac23\\ c=b\rho_s=\frac56*\frac23=0.5556\\ d=c\rho_L=\frac59*1=0.5556\\ e=d\rho_s=\frac59*\frac23=0.3704\\ f=e\rho_L=0.3704 g=f\rho_s=0.2469\\ h=g\rho_L=0.2469
a=Vinitial=ZS+Z0Z0Vs=61∗5V=0.8333VρL=ZL+Z0ZL−Z0=1b=aρL=65∗1=0.8333VρS=ZS+Z0ZS−Z0=32c=bρs=65∗32=0.5556d=cρL=95∗1=0.5556e=dρs=95∗32=0.3704f=eρL=0.3704g=fρs=0.2469h=gρL=0.2469
A
=
a
=
0.8333
B
=
a
+
b
+
c
=
2.2216
C
=
a
+
b
+
c
+
d
+
e
=
3.1476
A=a=0.8333\\ B=a+b+c=2.2216\\ C=a+b+c+d+e=3.1476
A=a=0.8333B=a+b+c=2.2216C=a+b+c+d+e=3.1476
A
′
=
a
+
b
=
1.6666
B
′
=
a
+
b
+
c
+
d
=
2.7772
C
′
=
a
+
b
+
c
+
d
+
e
+
f
=
3.5180
A'=a+b=1.6666\\ B'=a+b+c+d=2.7772\\ C'=a+b+c+d+e+f=3.5180
A′=a+b=1.6666B′=a+b+c+d=2.7772C′=a+b+c+d+e+f=3.5180
example:
R
S
=
75
Ω
,
Z
0
=
50
Ω
,
R
L
=
∞
,
V
S
=
2
V
R_S=75\Omega,\;Z_0=50\Omega,\;\;R_L=\infty,V_S=2V
RS=75Ω,Z0=50Ω,RL=∞,VS=2V
a
=
V
i
n
i
t
i
a
l
=
Z
0
Z
S
+
Z
0
V
s
=
2
5
∗
2
V
=
0.8
V
ρ
L
=
Z
L
−
Z
0
Z
L
+
Z
0
=
1
b
=
a
ρ
L
=
4
5
∗
1
=
0.8
V
ρ
S
=
Z
S
−
Z
0
Z
S
+
Z
0
=
1
5
c
=
b
ρ
s
=
4
5
∗
1
5
=
0.16
d
=
c
ρ
L
=
0.16
∗
1
=
0.16
e
=
d
ρ
s
=
0.16
∗
1
5
=
0.032
f
=
e
ρ
L
=
0.032
a=V_{initial}=\frac{Z_0}{Z_S+Z_0}V_s=\frac 25*2V=0.8V\\ \rho_L=\frac{Z_L-Z_0}{Z_L+Z_0}=1\\ b=a\rho_L=\frac 45*1=0.8V\\ \rho_S=\frac{Z_S-Z_0}{Z_S+Z_0}=\frac15\\ c=b\rho_s=\frac45*\frac15=0.16\\ d=c\rho_L=0.16*1=0.16\\ e=d\rho_s=0.16*\frac15=0.032\\ f=e\rho_L=0.032
a=Vinitial=ZS+Z0Z0Vs=52∗2V=0.8VρL=ZL+Z0ZL−Z0=1b=aρL=54∗1=0.8VρS=ZS+Z0ZS−Z0=51c=bρs=54∗51=0.16d=cρL=0.16∗1=0.16e=dρs=0.16∗51=0.032f=eρL=0.032
A
=
a
=
0.8
B
=
a
+
b
+
c
=
1.96
C
=
a
+
b
+
c
+
d
+
e
=
2.152
A=a=0.8\\ B=a+b+c=1.96\\ C=a+b+c+d+e=2.152
A=a=0.8B=a+b+c=1.96C=a+b+c+d+e=2.152
A
′
=
a
+
b
=
1.6
B
′
=
a
+
b
+
c
+
d
=
1.92
C
′
=
a
+
b
+
c
+
d
+
e
+
f
=
1.984
A'=a+b=1.6\\ B'=a+b+c+d=1.92\\ C'=a+b+c+d+e+f=1.984
A′=a+b=1.6B′=a+b+c+d=1.92C′=a+b+c+d+e+f=1.984
Circuit Simulation and Combinational Circuit Design
CMOS saturation diagram of NOT gate
V
I
L
:
V_{IL}:
VIL: PMOS in linear NMOS in saturation
V
I
H
:
V_{IH}:
VIH: PMOS in saturation NMOS in linear
V
O
H
:
V_{OH}:
VOH: PMOS in linear NMOS off
V
O
L
:
V_{OL}:
VOL: PMOS off NMOS linear
at middle point ©,
V
i
n
=
V
o
u
t
=
V
T
L
(
l
o
g
i
c
t
h
r
e
s
h
o
l
d
)
=
V
D
D
2
V_{in}=V_{out}=V_{TL}(logic\;threshold)=\frac {V_{DD}}2
Vin=Vout=VTL(logicthreshold)=2VDD
k
R
=
k
n
k
p
k_R=\frac {k_n}{k_p}
kR=kpkn
for fast raise time and middle logic threshold point: w p L n = 2.5 ∗ w n L n \frac {w_p}{L_n}=2.5*\frac{w_n}{L_n} Lnwp=2.5∗Lnwn
multi-stage fanout
capacitance between driver and receiver:
C
L
=
g
s
p
+
c
g
d
p
+
c
g
d
n
+
c
g
s
n
+
c
w
C_L=_{gsp}+c_{gdp}+c_{gdn}+c_{gsn}+c_{w}
CL=gsp+cgdp+cgdn+cgsn+cw
C
g
p
=
2.5
C
g
n
C_{gp}=2.5C_{gn}
Cgp=2.5Cgn
total delay:
τ
p
=
0.69
c
L
(
R
e
q
N
+
R
e
q
P
2
)
\tau_p=0.69c_L(\frac{R_{eqN}+R_{eqP}}2)
τp=0.69cL(2ReqN+ReqP)
width:
w
p
=
β
∗
w
n
w_p=\beta*w_n
wp=β∗wn
t
p
=
0.345
[
(
1
+
β
)
(
c
g
d
N
+
c
g
s
N
)
+
c
w
]
R
e
q
N
(
1
+
β
)
t_p=0.345[(1+\beta)(c_{gdN}+c_{gsN})+c_w]R_{eqN}(1+\beta)
tp=0.345[(1+β)(cgdN+cgsN)+cw]ReqN(1+β)
τ
=
τ
i
n
t
+
S
⋅
C
L
\tau=\tau_{int}+S\cdot C_L
τ=τint+S⋅CL
t
p
=
0.69
R
e
q
c
i
n
t
(
1
+
C
e
x
t
/
C
i
n
t
)
t_p=0.69R_{eq} c_{int} (1+C_{ext}/C_{int} )
tp=0.69Reqcint(1+Cext/Cint)
C
e
x
t
C_{ext}
Cext: external load capacitance
energy store in cap
1
2
C
L
V
D
D
2
\frac 12C_LV_{DD}^2
21CLVDD2
if we need big driver, we should gradually increase size of TR, optimize total delay gate 1 gate 2 gate 3 gate 4 … gate n
τ
=
N
⋅
R
1
S
C
1
\tau=N\cdot R_1SC_1
τ=N⋅R1SC1
R
1
R_1
R1: the resistor of the first transistor
C
1
C_1
C1: capacitor of the first transistor
maximum number of fan-out(also the minimum delay): 4
sizing factor
F
F
F and effective fanout
f
f
f
f
=
C
L
/
C
g
,
1
N
=
F
N
f=\sqrt[N]{C_L/C_{g,1}}=\sqrt[N]F
f=NCL/Cg,1=NF
N
N
N: number of fan-out
C
g
,
1
C_{g,1}
Cg,1:input capacitance of the first inverter (minimally-sized device)
minimum delay
t
p
=
N
t
p
0
(
1
+
F
N
/
γ
)
t_p=Nt_{p0}(1+\sqrt[N]F/\gamma)
tp=Ntp0(1+NF/γ)
t
p
0
t_{p0}
tp0: intrinsic delay of inverter (independent of sizing)
γ
\gamma
γ:
c
i
n
t
c
g
\frac {c_{int}}{c_g}
cgcint, proportionality factor, which is only a function of technology and is close to 1
fanout for best power consumption
C t o t = C g 1 [ ( 1 + γ ) ( 1 + f ) + F ] energy dissipation: E = V d d 2 C g 1 ( ( 1 + γ ) ( 1 + f ) + F ) C_{tot}=C_{g1}[(1+\gamma)(1+f)+F]\\ \text{energy dissipation:}\\ E=V_{dd}^2C_{g1}((1+\gamma)(1+f)+F) Ctot=Cg1[(1+γ)(1+f)+F]energy dissipation:E=Vdd2Cg1((1+γ)(1+f)+F)
CMOS circuit power consumption
- dynamic power/switching power
- short circuit power
- leakage power/static power
- subthreshold current
- BTBT current (diode leakage current)
- Gate tunneling
energy consumed per switching period:
E
d
p
=
t
s
c
V
D
D
I
p
e
a
k
E_{dp}=t_{sc}V_{DD}I_{peak}
Edp=tscVDDIpeak
t
s
c
t_sc
tsc: time both devices are conducting,
V
D
D
−
2
V
T
V
D
D
t
r
i
s
i
n
g
\frac {V_{DD}-2V_T}{V_{DD}}t_{rising}
VDDVDD−2VTtrising
average power consumption:
P
d
p
=
C
s
c
V
D
D
f
P_{dp}=C_{sc}V_{DD}f
Pdp=CscVDDf
power delay product PDP
C
L
V
D
D
2
/
2
C_LV_{DD}^2/2
CLVDD2/2
energy delay product EDP
P
D
P
∗
t
p
=
α
∗
C
L
2
V
D
D
3
2
(
V
D
D
−
V
T
)
PDP*t_p=\frac {\alpha*C_L^2V_{DD}^3}{2(V_{DD}-V_T)}
PDP∗tp=2(VDD−VT)α∗CL2VDD3
α
\alpha
α: technology parameter
techniques to reduce power
- MTVT: multi-threshold voltage technique
- clock gating
- pipelining
- parallel processing
- coding technique
- thermometer code
MTVT: threshold voltage higher, harder to turn on transistor, react slower, but less leakage (high threshold can reduce leakage).
technology scaling
Parameter | Relation | Full Scaling | General scaling | Fixed-Voltage scaling |
---|---|---|---|---|
Area/Device | W L WL WL | 1 / S 2 1/S^2 1/S2 | 1 / S 2 1/S^2 1/S2 | 1 / S 2 1/S^2 1/S2 |
Intrinsic Delay | R o n C g a t e R_{on}C_{gate} RonCgate | 1 / S 1/S 1/S | 1 / S 1/S 1/S | 1 / S 1/S 1/S |
Intrinsic Energy | C g a t e V 2 C_{gate}V^2 CgateV2 | 1 / S 3 1/S^3 1/S3 | 1 / S U 2 1/SU^2 1/SU2 | 1 / S 1/S 1/S |
Intrinsic Power | Energy/Delay | 1 / S 2 1/S^2 1/S2 | 1 / U 2 1/U^2 1/U2 | 1 |
Power Density | P/Area | 1 | S 2 / U 2 S^2/U^2 S2/U2 | S 2 S^2 S2 |
(未完)