Bipolar

ADDENDUM
BIPOLAR DIGITAL DESIGN
Qualitative understanding of the bipolar device

n
Bipolar device models
n
The ECL Gate
n
Advanced bipolar digital design
n
A.1 Introduction
A.2 The Bipolar Transistor
A.2.1 A First Glance at the Device
A.2.2 Static Behavior
A.2.3 Dynamic Behavior
A.2.4 The Actual Bipolar Transistor—Secondary Effects
A.2.5 SPICE Models for the Bipolar Transistor
A.3 The Bipolar ECL Inverter
A.3.1 Issues in Bipolar Digital Design: A Case Study
A.3.2 The Emitter-Coupled Logic (ECL) Gate at a Glance
A.3.3 Robustness and Noise Immunity: The Steady-State Characteristics
A.3.4 ECL Switching Speed: The Transient Behavior
A.3.5 Power Consumption
1
2 Bipolar Digital
A.3.6 Looking Ahead: Scaling the Technology

A.4 Bipolar Gate Design
A.4.1 Logic Design in ECL
A.4.2 Differential ECL
A.4.3 Current Mode Logic
A.4.4 ECL with Active Pull-Downs
A.4.5 Alternative Bipolar Logic Styles
A.5 Exercises and Design Problems

J. Rabaey—Digital ICs-1st Ed. 3
A.1 Introduction
MOS transistors took over the digital integrated circuit market in the 1970s, mainly as a
consequence of their high integration density. Before that time, most digital gates were
implemented in the bipolar technology. The dominance of the bipolar approach to digital
design was exemplified in the wildly and widely successful TTL (Transistor-Transistor
Logic) logic series, which persisted until the late 1980s. Bipolar digital designs occupy
only a small portion of the digital market today. They still are the technology of choice
when very high performance is required, yet even there CMOS is becoming highly com-
petitive. This trend will continue in the future, as the reduced supply voltages of the deep-
submicron technologies make bipolar design exceedingly hard.
Because of this reduced importance, we decided to remove bipolar design from the
2nd Edition of “Digital Integrated Circuits — A Design Perspective”, and to make the
material of the 1st edition freely available at the web-site as a set of addenda. We hope that
this helps to address the concerns of those designers for whom bipolar design is still a
necessity.
In this addendum, we first present a brief overview of the bipolar device and its
models. This is followed by an extensive description and analysis of the Emitter-Coupled
Logic (ECL) gate, the dominant bipolar digital gate at present. After a discussion on how
to build complex logic gates in ECL, the chapter is concluded with an overview of the
BICMOS approach to digital design that combines MOS and bipolar devices into a single
gate.
A.2 The Bipolar Transistor
A.2.1 A First Glance at the Device
Figure A.1a shows a cross section of a typical npn bipolar (junction) transistor structure.
The heart of the transistor is the region between the dashed lines and consists of two np
junctions, connected back to back. In the following analysis, we will consider the ideal-
ized transistor structure of Figure A.1b. The transistor is a three-terminal device, where
the two n-regions, called the emitter and the collector, sandwich the p-type base region. In
contrast to the source and drain regions of the MOSFET, the emitter and collector regions
are not interchangeable, as the emitter is much more heavily doped than the collector.
Depending upon the voltages applied over the device terminals, the emitter-base and
collector-base junctions are in the forward- or reverse-biased condition. Enumeration of
all possible combinations results in Table A.1, which summarizes the operation modes of
the bipolar device. In digital circuits, the transistor is operated by preference in the cut-off
or forward-active mode. Operation in the saturation or reverse regions is, in general,
avoided as the circuit performance in those regions tends to deteriorate.
In a superficial way, the operation of the device can be summarized as follows:
• As both junctions are reverse biased in the cut-off mode, no current flows into or out
of the device. It can be considered off.
4 Bipolar Digital
E B C
p+ n+ p+ n+
p
isolation
p+
n-epitaxy
n+ buried layer
p-substrate
(a) Cross-sectional view
E C
n+ p n
(b) Idealized transistor structure
Figure A.1 npn bipolar transistor.
Table A.1 Modes of operation of the bipolar transistor.
Mode Emitter Junction Collector Junction
Cut-off Reverse Reverse

Forward-active Forward Reverse
Reverse-active Reverse Forward
Saturation Forward Forward
• In the active mode, the transistor acts as a current amplifier. The current flowing
into the base results in a collector current that is β times larger. Furthermore, there
exists an exponential relationship between the emitter-base voltage and the collector
current. This relation is similar to the forward-bias condition of the junction diode.
In the reverse active condition, this current gain is small and virtually nonexistent
(β ≈ 1) in contrast to the forward-active mode, where values of over 100 can be
observed.
• Finally, when the device saturates, a substantial drop in current gain occurs. Typical
for the saturation mode is the low value of the emitter-collector voltage.
Some important differences with the MOS transistor jump immediately into view.
First of all, the exponential relationship between input voltage and output current makes it
possible to drive large currents with small voltage excursions. This has a beneficial impact
on the performance. On the other hand, the control terminal (i.e., the base) of the bipolar
transistor carries an input current when the device is in active or forward mode. This
means that the input resistance of the device is small compared to the MOS transistor. As
will become apparent later, this feature makes the device not as amenable to high-density
digital design.
It should come as no surprise that a complementary device, called the pnp bipolar
junction transistor, can be conceived as well. They have only been used sparingly in digi-
tal design—adding a high-quality pnp to a bipolar process tends to raise the cost substan-
tially—and are therefore treated superficially in the rest of this book. Schematic symbols
for both npn and pnp devices, as well as sign conventions for voltage and currents are pic-
tured in Figure A.2.
C C
VBC – VBC –
IC IC
+ + + +
B B
VCE VCE
IB IB
+ – + –
VBE VBE
– –
IE IE
E E
Figure A.2 Bipolar transistor schematic

(a) npn (b) pnp symbols and sign conventions.
A.2.2 Static Behavior
Forward-Active Region
Figure A.3 shows a cross section of the idealized transistor structure of Figure A.1b as
well as the minority carrier concentrations in the emitter, base, and collector regions. The
concentrations are plotted for the forward-active operation mode. That is, the base-emitter
(be) junction is forward-biased, while the base-collector (bc) junction is in reverse-bias
condition. The subscripts e, b, and c are used to denote the various regions. As we know
from our diode study, the forward bias causes excess minority carriers on the be side,
while the reverse bias at the bc end causes the minority concentration to approach zero.
We assume (without loss of generality) that the short-base diode model is valid for all
junctions.
The law of the junction, described in Chapter 3 of the textbook, is still valid and can
be used to evaluate the minority carrier concentrations at the boundaries of the base region
n b ( 0 ) = n b0 e V BE ⁄ φ T
(A.1)
n b ( W ) = n b0 e VBC ⁄ φT ≈ 0
Obviously, a concentration gradient exists within the base region causing minority
electrons to diffuse from the emitter to the collector end. The width of the base region,
which is substantially below 1 µm in contemporary technologies, is sufficiently smaller
than diffusion length Ln. Consequently, the minority carriers in the base region display a
linear gradient, similar to the case of the short-base diode. The value of this current is
readily computed
6 Bipolar Digital
Minority carrier Depletion

concentration regions
nb(0) B C
E
Collector contact
Emitter contact
pe(x) nb(x) pc0

nb0
pc(x)
pe0
0 W x
WB
Figure A.3 Minority carrier concentration profile in an npn transistor under forward-active conditions.
n b ( 0 ) – n b ( W ) qA E D b
= qA E D b  ----------------------------------
dn b
I diff = qAE D b - = ----------------n b ( 0 )
dx  W  W
(A.2)
qA E D b n b0 V BE ⁄ φT
= ------------------------e
W
where AE is the area of the of the transistor under the emitter, and Db is the diffusion coef-
ficient for minority carriers in the base region.
The following picture now emerges. The forward bias of the be junction reduces the
potential barrier. Electrons, the majority carriers in the emitter, diffuse from the emitter to
the base, where they become minority carriers. As the emitter is more heavily doped than
the base, we assume for the moment that the be junction is a one-sided junction and that
the hole current can be ignored. The concentration gradient in the base causes the injected
(or emitted) electrons to diffuse towards the collector. From Eq. (A.2), it follows that this
diffusion current is an exponential function of the applied base-emitter bias-voltage. Once
arrived at the reverse-biased bc junction, these electrons are swept towards the collector
by the local electrical field (the collector potential is positive with respect to the base). The
current across the junction (or the collector current) is, therefore, a drift current. This
yields the following expression for the collector current
qA E D b n b0 V BE ⁄ φT
IC = ------------------------e = IS e VBE ⁄ φ T (A.3)
W
where IS is the saturation current. Since the thickness of the depletion regions is, generally,
negligible with respect to the base region Wb, it is safe to assume that W ≈ Wb. Be aware
also that the derivation of Eq. (A.3) is based on some extra assumptions:
• The minority carrier concentration in the base region is substantially below the
majority carrier concentration. This is called the low-level injection criterion.
• All voltage drops occur over the depletion regions. This assumes that the neutral
regions are perfect conductors. This simplification will be accounted for later by
adding series resistances.
The main assumption until now, however, was that all electrons make it safely from
the emitter to the collector, or that IC = IE. In reality, this is clearly not the case. A more
accurate picture of the currents flowing in the bipolar npn transistor is pictured in Figure
A.4.
E B C
IE 1 IC
2 3
IB Electrons
Holes
Figure A.4 Components of the base current.
As shown, holes are flowing into the base to supply the following currents:
• Recombination current ¿—Occasionally, minority electrons diffusing through the
base recombine with majority hole carriers. To maintain charge neutrality in the
base region, the vanished holes have to be replaced.
• be-junction hole current ¡—The forward-biased base-emitter junction carries an
electron as well as a hole current. Although the latter is substantially smaller due to
the one-sided nature of the be junction, a small base current is still required to sup-
ply the carriers.
• bc-junction hole current ¬—This current actually flows from the collector to the
base and equals the saturation current of a reverse-biased junction. From the discus-
sion on pn-junction diodes, we know that this component is negligible.
This analysis demonstrates that the base current IB is relatively small; the smaller the
better, actually. It relates to the collector current IC by a constant ratio, called the forward
current gain βF:
I
β F = ----C- (A.4)
IB
8 Bipolar Digital
For typical digital bipolar processes, βF varies between 50 and 100. This means that a
small hole current into the base sustains a large electron current at the collector; hence the
current gain.
The relationship between IE and IB is obtained by enforcing the current conservation
law
IE = IB + IC = IB ( βF + 1 ) (A.5)
Finally, by combining Eq. (A.4) and Eq. (A.5), we can relate IC and IE.
IC βF
- = αF ≤ 1
----- = -------------- (A.6)
IE βF + 1
where αF is called the forward common-base current gain.
Reverse-Active Region
In this operation region, the situation is reversed—the base-collector junction is forward
biased, while the base-emitter junction is in reverse mode. The picture is essentially the
same as in the forward active mode, as shown in Figure A.5. The major difference is that
Minority carrier
concentration
Collector contact
Emitter contact
E B C
pc(x)
nb(W)
nb0 pc0
pe0
pe(x) nb(x)
0 W x
WB
Figure A.5 Minority carrier concentration profile in reverse-active operation mode.
the gradient of the base minority carrier distribution is reversed, so that the diffusion cur-
rent now is directed from collector to emitter. We use a similar approach to derive the cur-
rent expression
qAC D b n b0 VBC ⁄ φT
I E = – ------------------------
-e (A.7)
WB
where AC is the collector area. Notice that we observe the current conventions of Figure
A.2, which explains the negative sign of IE. While this expression looks similar to Eq.
(A.3), some major differences come to light when examining the components of the base
current. While the recombination and base-emitter fractions stay at roughly the same
level, the hole current from the base to the collector is substantial and actually exceeds the
flow of electrons over the junction. The latter is explained by the fact that the base has a
higher doping level than the collector, so that its current in forward bias is dominated by
the flow of holes (Eq. (2.10)). These holes have to be provided by the base current, which
becomes comparable to the emitter current. The reverse current gain βR is, therefore,
small
I
βR = – ----E ≈ 1 (A.8)
IB
and so is the reverse common-base current gain
I βR
α R = ----E- = --------------
-«1 (A.9)
IC βR + 1
Saturation Region
In this mode of operation, both junctions are forward-biased (VBE and VBC > 4 φT). We
only consider the case where the emitter junction has a stronger bias (or VBE > VBC). The
opposite case is called the reverse-saturation condition and occurs rarely in digital cir-
cuits.
Under this condition, excess minority carriers are present at both the emitter and col-
lector boundaries of the base region, although a somewhat higher concentration is present
at the emitter end. The resulting minority carrier concentration is plotted in Figure A.6.
Minority carrier concentration
B
nb(0)
Collector contact
E nb(x)
Emitter contact
C
QA nb(W)
pc(x)
pe(x) QS
pc0
pe0 nb0
0 W x
WB
Figure A.6 Minority carrier concentration in saturation region (VBE > VBC > 4φT).
Due to the short base width, a linear carrier gradient is still appropriate, and a diffusion
current flows from emitter to collector, albeit of a smaller value than in the forward-active
region. An important increase in the base charge is furthermore apparent. This gives rise
to an increase in the recombination component of the base current. The combination of
both factors indicates that the current gain is reduced in the saturation region, and is sub-
stantially smaller than βF. A simple application of Kirchoff’s voltage law also shows that
saturation condition corresponds to a small value of VCE.
V CE = V BE – V BC (A.10)
with VBE > VBC > 4φT. When the device is in deep saturation, VCE(sat) generally ranges
between 0.1 to 0.2V.
10 Bipolar Digital
Cut-off Mode
Finally, in cut-off mode, both diodes are reverse-biased. The corresponding concentration
profile is shown in Figure A.7. No excess base charge is present. The currents into the ter-
minals are limited to the saturation currents of the reverse-biased diodes and are extremely
small. The transistor is considered to be in the off-state.
Minority carrier concentration
Collector contact
E B C
Emitter contact
nb0 pc0
pe(x) pe0
nb(x) pc(x)
0 W x
WB
Figure A.7 Minority carrier profile in cut-off mode.
A Global View
The different operation modes of the bipolar transistors can be unified into a single set of
equations, called the Ebers-Moll model. The model combines both forward and reverse
currents, and in its most general form is expressed by the following set of equations:
I I
IC = I F – -----R- I E = -----F- – I R IB = IE – IC (A.11)
αR αF
where
V BE ⁄ φ T V BC ⁄ φ T
IF = IS ( e – 1) IR = IS ( e – 1) (A.12)
and where IS, αF, and αR are defined by Eqs. (A.3), (A.6), and (A.9) respectively. This
form is often used for the computer representation of transistor large-signal behavior.
We now have all the information needed to plot the I-V characteristics of the bipolar
transistor. The behavior of an MOS transistor was completely described by the variable set
(ID, VGS, and VDS), two of which could be chosen independently. In the bipolar case, the
presence of the base current introduces one extra independent parameter. Therefore, a dual
set of plots is needed to completely characterize the device. This is illustrated in the fol-
lowing pair of graphs. Figure A.8 plots IC as a function of VCE with IB as a parameter for
both the reverse and the forward operation modes. Observe the reduced current gain in
reverse operation. Figure A.9 shows IC as a function of VBE, plotted on a logarithmic scale.
The reduced slope of the currents at higher current levels is due to a number of secondary
effects, which will be introduced later in the chapter.
0 IB = 25 µA 15
Forward operation
IB = 50 µA
Active
IB = 75 µA IB = 100 µA
IB = 100µA 10
IC(mA)
IB = 75 µA
IC (mA)
–0.25
IB = 50 µA
5
Reverse operation IB = 25 µA
Saturation
–0.5 0
–3.0 –1.0 0.0 2.0
VCE (V) VCE (V)
Figure A.8 Collector current as a function of VCE with IB as a parameter.
100
IC
ln(I) (A)
IB
10–5 IB r B
10–10
0.2 0.8 1.2 Figure A.9 Collector and base current as a function
VBE (V) of VBE (for a constant VCE of 3 V).
Problem A.1 Current Gain of Bipolar Transistor.

Determine βF and βR of the device plotted in Figure A.8.
Problem A.2 pnp Transistor Characteristics.

Plot, approximately, the I-V characteristics of a pnp transistor.
Model for Manual Analysis

The Ebers-Moll model, while being accurate, tends to be too complex to be useful in the
first-order analysis of a bipolar circuit. A simplified model of the bipolar transistor,
intended for manual analysis, is therefore at hand. A summary of such a model, catego-
rized per operation region, is given below (similar models can be established for the
reverse operation modes).:
• Cut-off—VBE < VBE(on), VBC < VBC(on)
All currents are zero: IC = IB = 0
12 Bipolar Digital
• Forward-active—VBE ≥ VBE(on), VBC < VBC(on)

A first-order model for large signal computations consists of a base-emitter diode
and a current source (Figure A.10a). An even simpler version of this model, which is often
useful, replaces the input diode by a voltage source with a fixed voltage VBE(on), which
ranges between 0.6 and 0.75 V. This represents the fact that the base-emitter voltage var-
ies only a little in the forward-active region due to the steep exponential characteristic
(Figure A.10b).
• Forward-saturation—VBE ≥ VBE(sat), VBC ≥ VBC(on), VBE > VBC
During saturation, the collector-emitter voltage is assumed to be fixed at a voltage
VCE(sat), which produces the model of Figure A.10c.
Example A.1 Static npn Model for Manual Analysis

The npn transistor of Example A.3 is to a first degree modeled by the following parameters:
βF = 100, IS =1.E−17 A, VBE(on) = 0.7 V, VBE(sat) = 0.8 V, VCE(sat) = 0.1 V.
IB IB
B C B C
+
VBE βFIB VBE(on) + βFIB
–
–
IB = IS(eVBE/φT – 1)
E E
(a) Forward-active (b) Forward-active (simplified)
IB
B C
VBE(sat) + + VCE(sat)
– –
IC < βFIB
E
(c) Forward-saturation
Figure A.10 First-order large signal models for npn bipolar transistor
A.2.3 Dynamic Behavior
As we have done for diodes and MOS devices, we model the dynamic behavior of the
bipolar transistor by a set of capacitances. In the bipolar device, these capacitive effects
originate from three sources: the base-emitter and base-collector depletion regions, the
collector-substrate diode, and the excess minority carrier charge in the base. We discuss
each of them individually.
The Base-Emitter and Base-Collector Depletion Capacitances

Depending upon the operation region, the be and bc junctions are either forward- or
reverse-biased. This has an important impact on charge storage in the depletion region
and, hence, on the equivalent capacitance. The nature of the junction capacitance and its
model should be familiar by now, as it has been discussed extensively in previous sec-
tions. To combat the nonlinearity of the junction capacitance, the linearized large-scale
model of Eqs. (2.17)–(2.18) (repeated below for clarity) is often used. In determining the
value of the Keq multiplier factor, one should be careful in delineating exactly the range of
the voltage swing over the diode, since this affects the accuracy.
C eq = K eq C j0 (A.13)
where
– φ 0m
- [ ( φ – V high ) 1 – m – ( φ 0 – V low ) 1 – m ]
K eq = --------------------------------------------------- (A.14)
( V high – V low ) ( 1 – m ) 0
Both junction capacitances are thus adequately modeled by the following parameters:
• Their zero-bias value (Cbe0 and Cbc0, respectively). Although the value of the collec-
tor capacitance is typically a trifle larger than its counterpart at the emitter side, both
values are comparable. For instance, for an npn device in an advanced bipolar pro-
cess with emitter size of 0.6 µm × 2.4 µm, Cbe0 = 6.7 fF, and Cbc0 = 7.5 fF
[Yamaguchi88].
• The junction grading coefficients mbe and mbc.
• The built-in potentials.
The Collector-Substrate Capacitance

As can be observed from the cross-sectional view of the bipolar transistor (Figure A.1),
the n-type collector is isolated from the substrate by an inverse-biased diode. The model is
complicated by the presence of the n+ buried layer and the p+ isolation regions. This
results in a parasitic capacitance, which is, once again, a nonlinear junction capacitance.
Since the area of the collector-substrate junction is considerably larger than the emitter
area, the zero-bias value of this parasitic capacitance can be substantial and might domi-
nate the performance of the device. The value of Ccs0 typically ranges between 1 and 2.5
times the base-emitter capacitance. Advanced processes reduce its impact by providing
14 Bipolar Digital
trench isolation between devices (i.e., a vertical plug of SiO2 inserted between transistors),
or even by using isolating substrates.
The Base Charge—An Introduction

When deriving the diode current model, we were able to relate the diode current to the
excess minority charge in the neutral regions. This proved to be beneficial, especially
when studying the dynamic characteristics of the device. A similar approach works for the
bipolar transistor as well. The important parameter here is the excess minority carrier
charge in the base region, which is a strong function of the operation mode.
1. Forward-active
Assuming that the minority carriers in the base region display a linear gradient, as
shown in Figure A.3, the total excess minority charge in the base is readily computed.
n b ( 0 ) + n b ( W ) – 2n b0 nb ( 0 )
Q F = qA E W B  ---------------------------------------------------
- ≈ qA E W B ------------
- (A.15)
 2  2
this yields the following relationship between the collector current of Eq. (A.3) and the
excess base charge,
Q
I C = ------F- (A.16)
τF
with τF the mean forward-transit time, which can be interpreted as the mean time for the
minority carriers to diffuse from the emitter to the collector.
W B2
τ F = ---------
- (A.17)
2D B
Observe the similarity between the forward transit time and the mean transit time of the
short-base diode, Eq. (2.20). Keeping this transit time as short as possible is clearly a good
idea, as it results in a larger current for a smaller base charge. This can be accomplished be
reducing the base width to a maximal extent. This trend is apparent in contemporary bipo-
lar processes, where WB varies between 50 and 100 nm. Values of τF range between 5 and
30 psec.
In a similar fashion, we can relate the base current IB to the base charge
Q
I B = -------F- (A.18)
τ BF
with τBF the minority carrier lifetime in the base region during forward-active operation. It
accounts for the three factors shown in Figure A.4. Taking into account that the base and
collector currents are related by the forward current gain, a relationship between the base
transit and lifetimes is easily derived.
τ BF
β F = -------
- (A.19)
τF
2. Reverse-active
A similar analysis yields the following charge-driven model for the reverse currents.
Q Q Q τ BR
IE = – ------R- IC = – ------R- – -------R- and β R = -------
- (A.20)
τR τ R τ BR τR
3. Forward-saturation
This excess base charge in the saturation region can be divided into a forward and a
reverse component, as shown by the dotted lines in Figure A.6. This results in the follow-
ing expression for IC,
I C = ------F- – ------R- ------ = ------F- – Q R  ----- + --------

Q Q 1 Q 1 1
(A.21)
τ F τR α R τF  τ R τ BR
It is conceptually more interesting to consider a different subdivision as illustrated

by the shadings of grey in Figure A.6.
The first component, called the critical base charge QA, has a triangular shape. It
represents the charge needed to bring the transistor to the edge-of-saturation (eos), which
is the boundary condition between forward-active and saturation. At the eos, the equations
governing the forward-active mode are still valid.
Q
I C ( eos ) = ------A- = β F I B ( eos ) (A.22)
τF
where IB(eos) is the corresponding base current.
The second component, QS, is called the overdrive charge and is the result of both
junctions being in forward bias. Being rectangular, it does not contribute to the gradient of
the excess minority carriers, nor to the collector current. It arises from the pushing of more
current into the base than is required to reach the saturation condition. The current in
excess of the IB(eos) is called the excess base current IBS and is related to the overdrive
charge by a time constant τS, called the saturation time constant,
Q
I BS = -----S- (A.23)
τS
which is a weighted average of τF and τR [Hodges88].
αF ( τF + αRτR ) αR
τ S = ----------------------------------- ≈ --------------
-τ (A.24)
1 – αF α R 1 – αR R
(assuming that τF << τR and αF ≈ 1).
With the transistor in saturation, an expression for the (static) base current is now
readily derived
Q Q
I B = I B ( eos ) + I BS = -------A- + ------S (A.25)
τ BF τ S
16 Bipolar Digital
It is, especially, the overdrive charge that makes the saturation region an operation mode
to avoid in digital design. Recall from the discussion of diode dynamics that it takes time
to remove or provide this minority excess charge. Bringing a transistor in and out of satu-
ration is a slow operation that is inconsistent with the high performance requirements we
impose on digital circuits.
Changing the Base Charge—The Diffusion Capacitance

From the above, it becomes apparent that a considerable amount of excess minority carrier
charge is stored in the base region of an active or saturated bipolar transistor. Changing the
operation mode of the device requires the addition or the removal of this charge.
As with the diode, this minority excess charge can be modeled by capacitances in
parallel with the be junction (representing the forward charge QF) and the bc junction (rep-
resenting the reverse charge QR). The value of these capacitors is a strong function of the
transistor operation mode and the current levels. A large-signal approach, in the sense of
Eqs. (2.23)–(2.24), is therefore appropriate. Observe that the base charge is an exponential
function of the voltage over the junctions.
As an example, during forward-active operation QR is ignorable. Only a single
capacitor between base and emitter, representing QF = ICτF, has to be considered. This
capacitor is often called the diffusion capacitance Cd, as it represents the carriers diffusing
through the base region. Similar to the approach used for the diode, the small-signal value
of Cd is readily computed.
I
C d = τ F ----C- (A.26)
φT
The large-signal value (called CD) is comparable to Eq. (2.24). During saturation,
the base charge can be represented as the sum of the forward and reverse charges. For
manual analysis, it is often more convenient to divide the base charge into the overdrive
and the edge-of-saturation (eos) charge in correspondence with the model discussed in the
previous section. Bringing a transistor out of saturation simply requires the removal of the
overdrive charge.
Bipolar Transistor Capacitive Model

The effects of all the above contributions are combined into a single capacitive model for
the bipolar transistor, as shown in Figure A.11. Its components can be identified based on
the preceding discussion. Having an intuitive feeling for what components are important
for each operation condition is what makes a good bipolar designer.
Bipolar Transistor Switching Time—A Case Study

Consider the simple circuit of Figure A.12. We analyze the case where the transistor is ini-
tially on, or V0 > VBE(on), and in forward-active operation mode. At time t = 0, the input
voltage drops to 0, and the device is turned off.
Similar to the diode case, the turn-off process proceeds in two phases. In a first step,
the base charge is removed, turning the transistor off. During this phase, the base-emitter
QR Cbc
Ccs
QF S
Cbe
E Figure A.11 Capacitive model of an npn bipolar transistor.
VCC
RB
V0
Vin
+
–
t=0 Figure A.12 Example of npn-transistor-based
circuit.
voltage remains relatively constant. In the second phase, the base-emitter voltage drops to
zero. This requires the (dis)charging of the junction capacitances. We analyze each of
these phases individually. Again, be aware that this break-up is a simplification.
1. Removal of the base charge
Before time t = 0, the excess minority-carrier charge in the base equals
QF(t = 0) = QF0 = IC(t = 0) × τF
where IC(t = 0) is expressed by the following equation:

IC(t = 0) = βF × IB(t = 0) = βF(V0 – VBE(on))/RB
We may assume that during the base-charge removal the base-emitter voltage remains rel-
atively constant and is approximately equal to VBE(on). This means the space charge of the
be and bc junctions is constant also, and not a factor at this point. The length of this inter-
val is thus uniquely defined by the time it takes to remove the base charge. Under the
assumption of a fixed base-emitter voltage, the base current that accomplishes this is con-
stant and equals
IB = (0 − VBE(on))/RB
This current actually serves two purposes: (1) to sustain the recombination in the base and
(2) to remove the base charge. This dual purpose is expressed by the following equation,
Q dQ F
I B = -------F- + (A.27)
τ BF d t
18 Bipolar Digital
Solving the differential equation, taking the initial condition for QF into account, yields
QF(t),
– t ⁄ τBF
Q F ( t ) = τBF [ I B – ( I B ( t = 0 ) – I B )e ] (A.28)
and the time to remove the excess charge (QF = 0).
IB ( t = 0 ) – IB
t base = τ BF ln -------------------------------- (A.29)
–IB
The recombination factor in Eq. (A.27) is often small compared to IB and actually drops
off very rapidly. It can be ignored for a first-order analysis, which means that all base cur-
rent is assumed to be used for base-charge removal. This yields a simplified expression for
tbase.
∆Q Q F0 IB ( t = 0 ) C D ∆V
- = – τ BF ---------------------
t base = ----------F = – -------- - = -------------- (A.30)
IB IB IB IB
Observe especially the last component of Eq. (A.30). CD stands for the large-signal equiv-
alent capacitance (defined over the range ∆V). This equation states that, if the recombina-
tion can be ignored, the base charge removal can be modeled as the discharging of a
capacitor.
2. Changing the space charge of the be and bc junctions
Once the transistor is off, the rest of the transient is devoted to discharging the junc-
tion capacitances to 0 volt. During this time span, the circuit is modeled as a first-order
RC-circuit. The time to reach 90% of the final value is a simple function of the time con-
stant of the network
t space = 2.2R B ( C be + C bc ) (A.31)
Example A.2 Dynamic Behavior of an npn Transistor

Assume that the bipolar transistor in Figure A.12 is characterized by the following parame-
ters: βF = 100, VBE(on) = 0.75 V, τF = 10 psec, Cbe0 = 20 fF, Cbc0 = 22 fF, Ccs0 = 47 fF, m =
0.33, φ0 = 0.75 V. The following circuit parameters are further assumed: VCC = 2 V, V0 = 1 V,
and RB = 5 kΩ.
First we compute the initial values of the currents. Since the VCE of the transistor is
equal to 2 V, the transistor is, obviously, in the forward-active mode. The base current is
expressed as
IB (t = 0) = (1 – 0.75) / (5 × 103) = 50 µA
After the switching of the input, the base current changes direction, or
IB = (–0.75) / (5 × 103) = −150 µA
The time to remove the base charge can now be computed with the aid of Eq. (A.29).
tbase = (100 × 10 psec) ln (200 µA / 150 µA) = 288 psec
The simplified expression of Eq. (A.30), on the other hand, yields 333 psec. This is suffi-
ciently close to be useful as a first-order approximation.
As mentioned, the circuit behaves as an RC-equivalent network during the space-charge
removal. To approximate the time constant of the network, it is necessary first to linearize the
junction capacitances over the voltage range of interest. Observe that the collector voltage
remains fixed during the transient, so that Cbc can be considered as a grounded capacitor. The
voltage over the base-emitter junctions decreases from 0.75 V to 0 V, while the voltage over
the bc junction increases from 1.25 V to 2 V. Injecting those values into Eq. (2.18) yields the
following values of Keq: Keq(be) = 1.5 and Keq(bc) = 0.68. The value of tdepl can now be com-
puted
tdepl = 2.2 × 5 kΩ × (1.5 × 20 fF + 0.68 × 22 fF) = 495 psec
which is considerably larger than the base-charge removal period.

These results are validated by the SPICE simulations shown in Figure A.13. During the
removal of the base charge, a fast decrease in the collector current can be observed. At the end
of the base-charge removal phase, the collector current drops to 0, the transistor goes into the
cut-off mode, and the base-emitter voltage decays to 0 V in an exponential manner. The sim-
ulated values of tbase and tdepl equal 302 psec and 429 psec (the transition between both
regions is somewhat fuzzy and is determined here by the point where the extrapolated linear
collector current crosses the zero-axis). The discrepancy between the predicted and simulated
values of tdepl is due to the linearization, which is more accurate in the reverse- than the for-
ward-biased regions of the diodes.
0.8 5
4
0.6
IC (mA)
VBE (V)
3
0.4 tbase
2
0.2
tdepl 1
0.0 0 0.2 0.4 0.6 0.8 1 00 0.2 0.4 0.6 0.8 1

t (nsec) t (nsec)
Figure A.13 Simulated transient response of the circuit of Figure A.12.
Problem A.3 Turning on the Bipolar Transistor

Derive the time it takes to turn on the npn transistor in the example of Figure A.12, assuming
the same device and circuit parameters and assuming a step from 0 V to 1 V at the input.
A.2.4 The Actual Bipolar Transistor—Secondary Effects
In the previous sections, we have sketched a model of an ideal npn device. As always, the
actual transistor is somewhat more complex, and some second-order parameters have to
be considered.
20 Bipolar Digital
The Early Voltage

In the analysis of the forward-active operation mode, we assumed the base-collector junc-
tion to be reverse-biased and to have no effect on the value of the collector current (Eq.
(A.3)). This is a useful approximation for first-order analysis, but is not strictly true in
practice. The reason for this is easily understood. Until now, we ignored the width of the
depletion regions with respect to the width of the base region WB. In other words, we
assumed that the width of the neutral base region W is equal to the base width WB. For thin
base regions, this is not really the case. Furthermore, an increase in the collector-base volt-
age increases the reverse bias and, consequently, the width of the depletion region. Since
this decreases the effective base width, it reduces the forward transit time τF and raises the
collector current IC.
The collector current therefore increases as a function of the collector voltage, as is
shown in Figure A.14, which plots the collector voltage as a function of VCE with VBE as a
parameter. It is interesting to observe that all characteristics extrapolate to a single point
on the VCE-axis, called the early voltage VA. For integrated circuit transistors, VA varies
from 15 V to 100 V. The inverse of the early voltage is analogous to the channel-length
modulation parameter λ in the MOS transistor.
IC
Forward-
active VBE3
Saturation
VBE2
VBE1
Figure A.14 Bipolar transistor
characteristics showing the early
VCE voltage effect.
VA
The collector-current Eq. (A.3) can be adapted to compensate for the early effect.
Due to additional computational complexity this model is rarely used in manual calcula-
tions and is reserved for computer models.
I C = I S  1 + --------
V CE V BE ⁄ φ T
- e (A.32)
 VA 
Transistor Breakdown Voltages

The mechanism of avalanche breakdown in a pn-junction was described in section 2.3.4
Similar effects occur at the be and bc junctions of the bipolar transistor. This limits the
voltages that can be applied to the device. The breakdown voltages depend upon the tran-
sistor configuration and are commonly expressed as BVCBO (common base configuration
breakdown voltage), VCE0 (common emitter configuration) and BVEBO (base-emitter break-
down). The values of the breakdown voltages vary from 6–8 V (BVEBO) to several tens of
volts for the others. Since breakdown creates excessive power dissipation and might lead
to device destruction when sustained, it is avoided in digital bipolar design.
Parasitic Resistances
Parasitic resistances are produced by the finite resistance of the neutral regions of the tran-
sistor, as shown in Figure A.15. While rE is normally very small (1–5 Ω), rB and rC can be
substantial and have a significant impact on the device performance due to the high dop-
ing level of the emitter region.
E B C
p+ rE n+ p p+ n+
rC1
isolation
n-epitaxy rB p+
rC3
n+ buried layer
rC2
p-substrate
Figure A.15 npn transistor structure, showing parasitic resistances.
The collector resistance consists of three components, labeled rC1, rC2, and rC3. Of
these, the latter is in general the dominant component. In advanced processes, rC3 is
reduced by adding a low-resistance n+ region below the collector (called a deep collector),
shown in dotted lines in Figure A.15. The value of rC varies between 20 Ω (with deep col-
lector) to almost 1 kΩ (without). The value of rB ranges from 50 Ω to 500 Ω and unfortu-
nately varies with the collector current due to an effect called current crowding. This
effect causes the transistor action to occur at the edges of the emitter area, instead of the
central portion, which results in a varying distance between the active base region and the
base contact.
The presence of these parasitic resistances is particularly noticed under high collec-
tor-current conditions. The voltage drop over the resistances reduces the voltage differ-
ences at the internal device terminals and, as a result, the transistor currents. Simulation
accuracy is strongly contingent upon a careful modeling of the resistances. The designer
can, if necessary, reduce the parasitic effects by modifying the transistor structure. Avail-
able options include increasing the emitter area, and providing multiple base and emitter
terminals.
βF Variations
The ideal model states that in the forward-active region, the current gain βF is a constant.
This parameter, in fact, does vary with the operating conditions of the device as shown in
Figure A.16, which plots the values of ln(IC,IB) as a function of VBE. It can be observed
that for intermediate values, a constant value of βF can indeed be observed. At low current
values, a degradation of the current gain occurs. This is attributed to an increase in the
base current, caused by the recombination of carriers in the be depletion region. This
effect is present at all current levels, but only has an impact under very low current condi-
tions. At the other end of the spectrum, the collector current drops below the ideal current
as a result of high-level injection effects—as the injected minority carrier density in the
22 Bipolar Digital
base approaches the majority carrier density, the hole current from base to emitter
becomes substantial leading to a decline in the collector current.
ln (I)
IKF
IC
High-level injection
βF
Recombination IB
VBE (linear) Figure A.16 Deterioration of the current

gain at low and high current levels.
In addition to high-level injection, the βF at high currents is also affected by the on-
set of the Kirk effect or base stretching. This occurs when the level of injected electrons
into the collector becomes comparable to the donor-atom doping density of the collector
region. This causes a collapse of the space-charge region, and the base stretches out into
the collector regions, resulting once again in a current-gain degradation.
The effects of the low- and high-level injection can modeled by modulating the
emission coefficient n in the base and collector currents Eq. (A.33). For instance, the high-
level injection effect is modeled by modifying n from the standard value of 1 to 2. The
point of onset of the high-level injection effects is called the knee current (IKF).
I C = I S e V BE ⁄ nφ T (A.33)
The impact of some of the high-current and parasitic resistance effects is observed
in the I-V plots of Figure A.9. Notice especially how the voltage drop over the base resis-
tance causes both the collector and base currents to degrade for high values of VBE.
A.2.5 SPICE Models for the Bipolar Transistor
Earlier versions of SPICE supported two separate models for the bipolar transistor: the
Ebers-Moll (E-M) and the charge-based Gummel-Poon (G-P) model. Both were merged in
later versions into the modified Gummel-Poon model which incorporates various exten-
sions to model high-bias conditions [Getreu76]. Under default conditions for certain
parameters, this model automatically reduces to the simple E-M model.
The complete model for an npn transistor, as implemented in SPICE, is shown in
Figure A.17 (The model of a pnp device is obtained by reversing the polarities of VBE,
VBC, VCE, IC and IB). Under normal operating conditions, the values of the current sources
are given by the modified Ebers-Moll equations
V BC I S V BC ⁄ ( nR φ T )
I C = I S ( e V BE ⁄ ( nF φ T ) – e V BC ⁄ ( n R φT ) )  1 – --------
- – ------ ( e – 1) (A.34)
VA  βR
and
rC
Ccs
+ VBC –
S
QBC +
QBE VCE
IC
B –
rB IB
+ VBE –
rE
E Figure A.17 SPICE model for an npn transistor.
I I
I B = -----S- ( e VBE ⁄ ( nF φ T ) – 1 ) + -----S- ( e V BC ⁄ ( nR φT ) – 1 ) (A.35)
βF βR
where IS is the saturation current, VA the early voltage, and nF and nR the forward and
reverse current-emission coefficients. As can be observed, the model superimposes the
forward- and reverse-operation conditions. The base-width modulation effect is incorpo-
rated in the collector current through the early voltage VA.
This model does not take into account the second-order effects occurring at low cur-
rent levels (depletion-layer recombination) or high injection levels. These are properly
accounted for in the Gummel-Poon model, which relates all terminal voltages and currents
to the base charge. A complete description of the model is beyond the scope of this text.
The intrinsic capacitances and resistances are modeled by the expressions presented
in Sections 2.2.4 and A.2.3. QBE and QBC represent the combined effects of the depletion
and base charge for the be and the bc junctions, respectively.
Table A.2 lists the main SPICE model parameters. The parameters covering the par-
asitic resistive and capacitive effects have been transferred to a separate table for the sake
of clarity (Table A.3).
In contrast to the MOS transistor case, not many parameters can be defined for the
individual transistor. The only parameter that can be defined at the instantiation time of a
device is the AREA factor, which determines how many of the bipolar junction transistors
of type MODNAME are put in parallel to create that particular device. The default value
is 1. This parameter is not used very often. The common practice is to provide a different
model for every possible device dimension, which may be used. This is acceptable, because
a typical bipolar digital design only uses a small range of device dimensions, contrary to the
MOS practice, where transistor dimensions can vary widely, even in a single circuit.
Example A.3 SPICE Bipolar Junction Transistor Model

The model for a bipolar npn transistor with an emitter area of 2 µm × 3.75 µm, implemented
in a process technology with a minimum line-width of 1 µm, is included. We use this as the
generic bipolar transistor in the rest of the text, unless specified otherwise.
* SPICE Model for 2 x 3.75 npn transistor
24 Bipolar Digital
Table A.2 Main SPICE bipolar junction transistor model parameters.
Parameter Name Symbol SPICE Name Units Default Value
Transport Saturation Current IS IS A 1.0E−16

Maximum Forward Current Gain βF BF − 100
Forward Current-Emission Coefficient nF NF − 1
Forward Early Voltage VAF VAF V ∞
Maximum Reverse Current Gain βR BR − 1
Reverse Current-Emission Coefficient nR NR − 1
Reverse Early Voltage VAR VAR V ∞
Corner for Forward Beta High-Current IKF IKF A ∞
Roll-offa
be Junction Leakage Saturation Currenta ISE A 1.0E−13
be Junction Leakage Emission Coeff. NE − 1.5
(low-current condition)a
Corner for Reverse Beta High Current IKR IKR A ∞
Roll-offa
bc junction leakage saturation currenta ISC A 1.0E−13
bc junction leakage emission coeff. NC − 2
(low-current condition)a
Ideal Forward Transit Time τF TF sec 0
Ideal Reverse Transit Time τR TR sec 0
a. Gummel-Poon Model Parameter
Table A.3 SPICE Parameters for parasitics (resistances, capacitances).
Parameter Name Symbol SPICE Name Units Default Value
Emitter Resistance rE RE Ω 0
Collector Resistance rC RC Ω 0
Zero-Bias Base Resistance rB RB Ω 0
Minimum Base Resistance RBM Ω RB
Current where RB falls halfway to RBM IRB A ∞
Zero-Bias be-Junction Capacitance Cbe0 CJE F 0
be-Junction Grading Coeff. mbe MJE − 0.33
be-Junction Built-in Voltage φbe VJE V 0.75
Zero-Bias bc-Junction Capacitance Cbc0 CJC F 0
bc-Junction Grading Coeff. mbc MJC − 0.33
bc-Junction Built-in Voltage φbc VJC V 0.75
Zero-Bias Collector-Substrate Cap. Ccs0 CJS F 0
cs-Junction Grading Coeff. mcs MJS − 0
cs-Junction Built-in Voltage φcs VJS V 0.75
.MODEL BF=100 BR=1 IS=1.E-17 VAF=50

+ TF=10E-12 TR=5E-9 IKF=2E-2 IKR=0.5
+ RE=20 RC=75 RB=120
+ CJE=20E-15 VJE=0.8 MJE=0.5 CJC=22E-15 VJC=0.7 MJC=0.33
+ CJS=47E-15 VJS=0.7 MJS=0.33
An example of an actual transistor instantiation is given below. Transistor Q1 is an NPN
device with its collector, base and emitter terminals connected to nodes 2, 1, 0, respectively.
Transistor Q2 consists of four transistors of type NPN, connected in parallel, and has its sub-
strate connected to node 5.
Q1 2 1 0 NPN
Q2 2 1 0 5 NPN 4
A.3 The Bipolar ECL Inverter
While we learned in the previous section that the CMOS inverter has an almost ideal dc
characteristic and low power consumption, it also became clear that the speed of the
inverter is restricted by the fact that the maximum current is proportional to VDD2 (or VDD
under velocity-saturation conditions). This puts a firm upper bound on the achievable
clock speeds. The bipolar device, on the other hand, is known to have an exponential rela-
tionship between collector current and base-emitter voltage, which means that a small
change in voltage can provide a large current. It is therefore reasonable to assume that
bipolar gates should be capable of achieving superior propagation delays.
Although the TTL gate (transistor-transistor-logic) has long been the flagship of the
bipolar digital logic gates, its importance has been reduced dramatically in the last decade
as CMOS has become more and more competitive. Most modern bipolar digital designs
are implemented in a circuit style called emitter-coupled logic (or ECL). We will, there-
fore, concentrate on this logic family. Following the scheme of the CMOS discussion, we
will first describe the ECL gate in a qualitative fashion, followed by a quantitative analysis
of its static and dynamic behavior, as well as the power dissipation. A discussion of the
effects of technology scaling concludes the section.
A.3.1 Issues in Bipolar Digital Design: A Case Study
Before engaging in a discussion of the ECL gate, some issues in the design of bipolar
gates should be highlighted. The particular nature of the bipolar junction transistor and the
lack of a good complementary device in most standard bipolar manufacturing processes
translate into gate topologies that are dramatically different from the CMOS structures.
The most important features can be summarized as follows:
• The high transconductance of the bipolar devices translates to a large variation in
collector current for a small change in the input voltage.
• The input resistance of the transistor is finite. This means that a fan-out gate presents
a current-sink for the driving gate. This is in contrast to CMOS designs, where fan-
out presents only a capacitive load.
26 Bipolar Digital
• The presence of an excessive amount of base-charge makes the transistor very slow
when operated in the saturation mode. Saturating the device should, therefore, be
avoided as far as possible.
To illustrate these observations, let us first examine a simple bipolar gate, called the
RTL (resistor-transistor logic) gate. A schematic of such a structure is shown in Figure
A.18. It can be readily seen from the circuit that with the input voltage Vin less than the
turn-on voltage VBE(on) of the transistor, the transistor is in cut-off mode, the collector cur-
rent IC essentially equals zero, and the output voltage Vout is equal to the supply voltage
VCC. When the input voltage is increased above VBE(on), the transistor turns on and enters
the forward-active mode. Further increasing the input voltage causes the output voltage to
drop due to the increasing collector current IC. This drop is swift, as the bipolar transistor
is characterized by a large gain in the forward-active mode. With sufficient input voltage,
when the output voltage has fallen sufficiently, the transistor will enter the saturation
region. The output voltage Vout remains fixed at VCE(sat) ≈ 0.1 V, the saturation collector-
emitter voltage. To enter this mode of operation, we require (as a first approximation) that
Vin ≥ VBE(sat) (≈ 0.8 V). Observe that the above analysis uses the simple model for manual
analysis of the bipolar transistor, presented in Figure 2.42.
Vcc
RC
RB Vout
Vin Q1
Figure A.18 Resistor-transistor logic (RTL) inverter.
From the above, it becomes obvious that the gate of Figure A.18 acts as an inverter.
A rough sketch of the voltage-transfer characteristic is shown in Figure A.19, where linear
interpolations are used to join the two major breakpoints of the graph.
Example A.4 VTC of an RTL Inverter

The VTC of an RTL inverter is derived for the following component and parameter values:
VCC = 5 V, RC = 1 kΩ, and RB = 10 kΩ. The bipolar transistor model presented in Chapter 2 is
used for Q1. Initially, it is assumed that the fan-out of the inverter is set to 0.
• For Vin < VBE(on), Q1 is in cut-off, IC = 0, and Vout = VOH = VCC.
• In the low output mode, Q1 is in saturation (or Vout = VCE(sat) ≈ 0.1 V), if the condition VBE
≥ VBE(sat) is fulfilled. The boundary condition on Vin for this to be valid can be derived by
analyzing the circuit with Q1 at the edge of saturation. This translates to the following con-
dition on Vin,
Vout
VCC
VTC of nonsaturating gate
VCE(sat)
VBE(on) Vin(eos) Vin
Cutoff Saturation
Figure A.19 Approximated voltage transfer
Forward-active characteristic of inverter.
IC V CC – VCE ( sat )
IB = -----
- = ----------------------------------
βF RC βF
(A.36)
= V BE ( eos ) + I B R B = V BE ( eos ) + ------B  ----------------------------------
R V CC – VCE ( sat )
Vin ( eos )
RC  βF 
Assuming VCE(sat) = 0.1 V, VBE(eos) = 0.8V, and βF = 100, Eq. (A.36) evaluates to Vin(eos) =
1.29 V.
The simulated VTC of the RTL inverter is plotted in Figure A.20 (consider only the
case of zero fan-out at present). The observed breakpoints equal approximately 0.7 V
(~VBE(on)) and 1.4 V, which are close to the predicted values. The major discrepancy is the
value of V0L, which equals 0.5 V in contrast to the predicted value of 0.1 V. This difference is
caused by the nonzero value of the series collector and emitter resistances (RE = 20 Ω and RC
= 75 Ω). In the saturation mode, the collector current IC equals approximately VCC/RC = 5 mA.
This causes an extra voltage drop of 5 × 10−3(75 + 20) = 0.48 V over the collector-emitter ter-
minals of Q1.
It can be observed that the RTL inverter displays some major dc problems:
1. The VTC is asymmetrical. The NML is less than or equal to 0.2 V. Noise and distur-
bances on the GND line or input signal could easily result in faulty operation.
2. To obtain a reasonable low value of VOL, it is necessary to drive the transistor into
saturation. From the earlier remarks, this can be labeled as unfortunate as the exces-
sive base charge has a significant impact on switching performance. One option is to
choose the resistor values so that saturation never occurs for the voltage range of
interest, which is equivalent to requiring that Vin(eos) > VCC in Eq. (A.36). This
approach causes a deterioration of VOL and reduces the gain in the transient region.
The corresponding VTC is plotted in shaded lines in Figure A.19.
3. The output impedance of the gate is equal to RC when the output is high. Reducing
RC is not a reasonable option, as this raises the value of the collector current IC and
28 Bipolar Digital
5.0
FO = 0
FO = 1
FO = 2
4.0
FO = 5
3.0
Vout (V)
2.0
1.0
0.0
0.01.0 2.0 3.0 4.0 5.0
Vin (V)
Figure A.20 Simulated voltage-transfer characteristic of RTL inverter for different fan-out
configurations.
increases power consumption and switching noise.1 Furthermore, in contrast to the

CMOS inverter, the gate has a finite input impedance. When Q1 is on, current flows
into its base through RB. The VOH of the inverter (and the NMH) is, hence, sensitive
to loading as is illustrated in Figure A.21. This effectively restricts the fan-out of the
inverter to approximately three gates.
Vcc Vcc
Vout = VCC - RCIB
RC RC
IB
RB
Q1 Q2
RB
Figure A.21 Effect of fan-out on VOH.
1
Switching noise is the noise present on the supply rails resulting from the switching of large currents in
the presence of parasitic resistances and inductors.
Example A.5 Effect of Fan-out on the VTC of the RTL Inverter

Suppose that the output of the RTL gate of the previous example is connected to N identical
gates and that its output is high. In this mode, each of the connecting gates will sink an
amount of current Iin into its base as illustrated in Figure A.21.
V out – V BE ( sat )
I in = ---------------------------------
- (A.37)
RB
This results in the following value of Vout as a function of the fan-out N:
V out – VBE ( sat )

Vout = V CC – N ⋅ R C ⋅ Iin = V CC – N ⋅ R C ⋅ ---------------------------------
-
RB
(A.38)
VCC + N ( RC ⁄ R B )VBE ( sat )
= ------------------------------------------------------------
-
1 + N ( RC ⁄ R B )
For large values of N, Vout (= VOH) will eventually approach VBE(sat) ≈ 0.8 V, which means that
the NMH is reduced to zero (or is even negative)! For smaller values of N, the situation is not
as bad, however. For instance, N = 5 results in an output voltage of VOH = 3.6 V, as is con-
firmed by the simulations of Figure A.20.
The main problem of the RTL gate is, however, evident in its transient behavior.
Consider, for instance, the tpLH. The propagation delay can be decomposed into two ele-
ments:
1. Initially, Q1 is in deep saturation. Turning off the device requires the removal of the
base charge (through RB). This is a slow operation.
2. As in the CMOS case, a major part of the switching time consists of charging up the
load capacitance CL from VOL to VOH. In the RTL case, this has to happen through
the load resistance RC, and the delay is determined by the time-constant RC CL. Mak-
ing RC small increases IC(sat) [= (VCC – VCE(sat))/RC]. This raises both the power con-
sumption as well as the first component of the propagation delay. Indeed, the charge
stored in the base region is proportional to the collector current (IC = QF/τF, as dis-
cussed in Chapter 2).
Similar arguments can be made regarding tpHL. The situation is not as bad in this
case, as CL is discharged by Q1 instead of through a resistor. This is considerably faster.
The dominant part of the propagation delay is the buildup of the base charge to bring Q1
into saturation. Observe that avoiding saturation is not an option, since it results in a sub-
stantial degradation of the dc behavior.
Finally, note that the gate consumes static power when the output is low, because a
direct path between VCC and GND exists through the resistor and transistor.
Example A.6 Transient Response of the RTL Inverter

The transient response of the RTL inverter of the previous example (for a fan-out of 1) is sim-
ulated using SPICE and is shown in Figure A.22 (a manual analysis of the transient behavior
of a bipolar circuit is complex and is, therefore, delayed until later in the chapter). The
response is clearly asymmetrical: tpLH = 420 psec, and tpHL = 165 psec. The resulting propaga-
tion delay is equal to 292.5 psec, which is inferior to the performance of the CMOS inverter!
30 Bipolar Digital
5.0
4.0 Vin
3.0
Vout (V)
2.0
1.0
Figure A.22 Transient

0.0 0 0.5 1 1.5 2 response of an RTL inverter (for
t (nsec) a fan-out of 1).
An important part of this delay is due to the base-charge buildup and removal. A fast bipolar
gate should avoid having its transistors going into saturation, since this is where the major
base-charge buildup happens.
In summary, a more effective bipolar gate should address the dc and ac issues raised
above. In the next section, we discuss how this is accomplished in the ECL gate.
A.3.2 The Emitter-Coupled Logic (ECL) Gate at a Glance
The ECL gate differs fundamentally from both CMOS and RTL inverters. It has been con-
ceived with utmost performance in mind. One of the means to accomplish this is to keep
the logic swing low. The ECL gate, hence, operates typically with a swing of only 0.5 V
(compare this to the 3 to 5 V CMOS structures). Ensuring a reasonable noise margin under
those conditions is nontrivial and requires careful circuit design.
The resulting gate is rather complex and is a composition of three components: the
differential pair, the output driver, and the bias network. Each of these affects one partic-
ular aspect of the gate functionality and performance.
The Differential Pair (or Current Switch)

The function of the input stage of the ECL gate is to provide maximal noise margins, low
noise sensitivity, and fast switching in the presence of a small logic swing. A schematic of
this portion of an ECL gate is plotted in Figure A.23. This structure is well known in the
analog world, where it is called an emitter-coupled pair. One of the two inputs is con-
nected to a reference voltage Vref that is generated by the bias network (discussed later).
Assuming that both transistors Q1 and Q2 are identical, the circuit is biased in a completely
symmetrical fashion if Vin is set equal to Vref as typically would be the case in an analog
design: Vbe1 = Vbe2, IC1 = IC2 = IEE/2 and Vout1 = Vout2 = VCC − RCIEE/2. This bias condition
Vcc Vcc
RC RC
Vout1 Vout2
Vin Q1 Q2 Vref
Vx
IEE
VEE Figure A.23 The differential pair.
turns out to be rather useless for digital purposes, where a key requirement is that two eas-
ily distinguishable operation modes are produced.
Let us now increase (decrease) Vin by a small amount with respect to Vref. This
changes the current balance in the differential pair and increases (decreases) the fraction
of IEE that is routed through Q1. A critical observation is that this current fraction is an
exponential function of the voltage difference between Vin and Vref. This is expressed in
Eq. (A.39), which is based on the familiar collector-current equation of the bipolar transis-
tor (2.63) (with φT equal to kT/q, the thermal voltage).
V in – V x
------------------- V in – Vref
I C1 φ ----------------------
e T –1 φT
------
- = --------------------------
Vref – V x
≈ e (A.39)
I C2 --------------------
-
φ
e T –1
At room temperature, a 60 mV increase in Vin causes a tenfold increase of IC1/IC2,
while a 120 mV increase is sufficient to set IC1 to 100 IC2. With IC2 only 1% of IC1, it can
be stated that, for all practical purposes, Q2 is essentially off, and that Q1 is carrying all of
IEE. As a result, approximate values for Vout1 and Vout2 can be derived.
V out1 = V CC – I EE R C (logic low)

(A.40)
V out2 = V CC (logic high)
A similar argument can be used to analyze the case where Vin drops below Vref.
Here, all the current is diverted to Q2, resulting in a high Vout1 and a low Vout2. A number
of conclusions can be drawn from this simple analysis:
1. The ECL gate provides differential (or complementary) outputs. Both the output sig-
nal (Vout1) and its inverted value (Vout2) are simultaneously available. This is a dis-
tinct advantage, as it eliminates the need for an extra inverter to produce the
complementary signal. This also prevents some of the time-differential problems
introduced by additional inverters. For example, in logic design it often happens that
both a signal and its complement are needed simultaneously. When the complemen-
tary signal is generated using an inverter, the inverted signal is delayed with respect
32 Bipolar Digital
to the original (Figure A.24a). This causes timing problems, especially in very high-
speed designs. The differential output capability of ECL avoids this problem (Figure
A.24b).
Vout2
Vout1 Vout2 Vin

Vin
Vout1
Vout1 Vout2
Vout1 Vout2
(a) Single-ended (b) Differential

Figure A.24 Advantage of differential (b) over single-ended (a) gate.
2. Another advantage of the differential operation is an increased noise immunity, as

common mode noise signals—i.e., signal disturbances common to both Vin and
Vref—are rejected to a large degree.
3. The transition region from one logic state to the other is centered around Vref and is,
to the first order, independent of the transistor parameters. To determine the bound-
aries of the transition region, an often-used simplification is to place them at the
points where the current ratio drops to 1%. This is only a qualitative approxima-
tion, which does not strictly follow the definitions of VIH and VIL (gain = −1). A
more detailed analysis is needed to produce precise values and will follow later.
Using the simplified definition, the transition width evaluates to approximately
240 mV.
V IL = V ref – 120mV
(A.41)
V IH = V ref + 120mV
In the case of the static CMOS and the RTL gates, the logic swing approximates the
voltage difference between the supply rails. This is not so for the ECL gate, where
the difference between VOH and VOL is determined by the bias current IEE and the
load resistance RC, as was illustrated in (A.40). The reduced voltage swing results in
a faster operation, as the time to (dis)charge the load capacitor is reduced.
4. The current drawn from the supply is always constant and equal to IEE. Therefore,
the differential pair consumes a constant amount of static power equal to IEE(VCC −
VEE). This static power consumption precludes the use of ECL for the design of very
complex VLSI circuits (> 100,000 gates), as the removal of the excessive heat
becomes either impossible or excessively expensive. Therefore, the use of ECL-
style circuits is mostly confined to the design of very high-speed building blocks
such as CPUs for mainframe computers.
On the other hand, the fact that the current drawn from the supply is constant and
independent of the switching operation reduces the amount of switching noise on
the supply lines. This is further aided by the low logic swing, which reduces the cur-
rent spikes caused by charging and discharging the load capacitance. As a result,
ECL circuits can operate reliably, even with small noise margins.
Furthermore, it can be observed that the output voltages (Vout1 and Vout2) are defined
with respect to VCC and are, to a first degree, not influenced by the value of VEE.
Noise on the latter supply therefore has virtually no effect on the signal levels. On
the other hand, VCC has to be as clean as possible.
5. With a proper choice of the bias current, the input voltage levels, and the resistor
values, it can be assured that neither Q1 nor Q2 ever goes into saturation. Actually,
in normal operation, both transistors are always forward-active. This speeds up the
device operation, as the time to alter the base charge is drastically reduced.
The Output Driver

The differential pair can be used in a stand-alone configuration. While it is very fast, this
structure suffers from one of the problems that plagued the RTL inverter. Its VTC is very
sensitive to output loading, since in the high state all current has to be provided by RC. To
alleviate the problem, an output driver is inserted between the outputs of the differential
pair and the input of the next gate. The main task of this driver is to reduce the output
impedance of the gate, so that ample current drive is available. At the same time, it acts as
a level shifter. It aligns the output logic levels symmetrically around Vref, making them
compatible with the input logic levels of a fan-out gate. The emitter-follower circuit mod-
ule performs all these functions and is, therefore, the ideal candidate for the output driver.
The revised circuit is shown in Figure A.25. It causes a level shift on the output sig-
nal equal to VBE(on). It also reduces the output impedance for the high logic level from RC
to approximately RC /(βF+1), thus providing ample current-drive capability (with βF the
forward-current gain of the bipolar device). To complete the circuit, a resistor RB has been
added at the input terminal. RB serves as a current path to VEE from the base of Q1 or Q2.
This helps to speed up the removal of the base charge QF during the switching of the dif-
ferential pair. For a very high speed design, RB also acts as termination for the transmis-
sion line formed by the interconnect. We will discuss this in depth in Chapter 8.
Design Consideration
The differential pair draws a constant current from the supply. This is obviously not true for the
emitter-followers, whose current levels are substantially different between the high and the low
output states. Large current peaks occur during the switching events, while (dis)charging the
load capacitances. This induces large fluctuations in the supply voltages and can cause mal-
34 Bipolar Digital
Vcc Vcc Vcc Vcc
RC RC
VC1 VC2
Q3 Q4
Vin Q1 Q2 Vref
Vx
Vout1 Vout2
RB
IEE
VEE
Figure A.25 ECL gate consisting of
differential pair and emitter-followers.
VEE
functioning of the circuit, given the small noise margins. To avoid this cross-coupling, a com-
mon practice is to provide a separate VCC supply for the output drivers.
The Bias Network

The purpose of the bias is to generate the reference voltage Vref. The prime requirement for
this reference is that it is to be centered as much as possible in the middle of the logic
swing, independent of the operating temperature. As was mentioned in Chapter 2, the volt-
age drop over a forward-biased base-emitter junction is a strong function of temperature.
Typically, VBE changes linearly with the temperature, or ∆VBE = −k∆T, with k approxi-
mately equal to 1.5 mV/°C. This drift causes the voltage-transfer characteristic of the ECL
gate and its noise margins to vary as a function of temperature as well. The ambient tem-
perature in ECL devices can be quite high, due to the large power dissipation of the gate.
Therefore, the bias network has to be designed so that Vref tracks those changes and the
noise margins are kept symmetrical over a wide range of temperatures. An example of
such a network is shown in Figure A.26. The value of the reference voltage is easily
derived.
V ref = V B5 – V BE ( on )
R1 (A.42)
- ( V – 2V D – V EE )
V B5 = V CC – -----------------
R 1 + R 2 CC
Transistor Q5 is connected in the emitter-follower configuration, reducing the output
impedance of the reference network. Changes in temperature affect the VBE(on) of Q5 and
Q2. These changes are tracked to a first order by the change in the voltage drops across
diodes D1 and D2. Actual implementations of this circuit have shown that it ensures the
centering of Vref between VOH and VOL for a temperature range from −30°C to +85°C.
In advanced ECL configurations, the bias network is also used to generate other ref-
erence voltages, such as the bias voltage for the current source. Fortunately, a single bias
VCC
R1
Q5
Vref D1
D2
R3 R2
Figure A.26 Voltage reference circuit (as used in the ECL 10K
VEE series from Motorola)
network can serve multiple gates, reducing the overall power consumption and the area
overhead.
After this qualitative analysis of the structure and properties of the basic ECL gate,
we will now derive the quantitative dc and ac characteristics of the gate. Before proceed-
ing, we would like to mention that there exist many different ECL families. The most
famous are the ECL 10K and 100K series and the MECL I, II, and III circuits [Hodges88].
The major differences between these are the values of the resistors, the type of current
source used and the construction of the bias network. Other variants of the gate are used in
modern, integrated ECL circuits used in instrumentation circuits, CPUs, and even micro-
processors (see, for example, [Jouppi93]). Instead of studying all those versions separately,
we discuss only one single gate structure in detail. Similar approaches can then be
employed to analyze the characteristics of related structures or even to design custom ECL
gates, whose current levels are optimized for the required performance or power
consumption.
A.3.3 Robustness and Noise Immunity: The Steady-State Characteristics
This section will analyze, in detail, how a typical ECL gate structure, as shown in Figure
A.27, can operate with a low voltage swing, and yet maintain sufficient noise immunity.
The gate topology is similar to the one presented before, with the exception of the current
source. The latter is implemented as a resistance RS in series with a transistor Q6, whose
base is dc-biased at VCS. To a first order, this produces the following bias current,
( V CS – V BE ( on ) – V EE )
I EE = ---------------------------------------------------- (A.43)
RS
For a manual analysis, we may assume that IEE is constant over the range of input
voltages. We will derive the VTC assuming a fan-out of 1, or, in other words, with the out-
put stage loaded with the resistance RB.
Suppose first of all that all transistors are operating in the forward-active region.
Under those conditions, the following circuit equations are valid:
36 Bipolar Digital
Vcc Vcc Vcc Vcc
RC RC
Q3 Q4
Vout1 Vin Q1 Q2 Vref Vout2

Vx
RB
RB RB
VEE Q6 VCS
VEE VEE
RS
VEE
Figure A.27 ECL gate with transistor/resistor current source.
V in – Vref
I C1 ----------------------
φ
------- = e T
I C2
βF
-≈I
I C1 + I C2 = I EE --------------
β F + 1 EE
(A.44)
V C1 = V CC – R C ( I C1 + I B3 )
V out1 = V C1 – V BE ( on )
V o1 – V EE
I B3 = --------------------------
-
( βF + 1 )R B
These equations can be solved to yield the following equations for IC1 and Vout1 (similar
equations can be derived for IC2 and Vout2).
IEE e x
I C1 = -------------
- (A.45)
1 + ex
with x = (Vin − Vref)/φT, and
( β F + 1 )R B V EE R C
V out1 = ---------------------------------------
- V CC – V BE ( on ) – R C I C1 + --------------------------
-
( β F + 1 )R B + R C ( β F + 1 )R B (A.46)
≈ V CC – V BE ( on ) – R C I C1 (for R B ( β F + 1 ) » R C )
The important parameters of the voltage-transfer characteristic can now be derived.
Nominal Voltage Levels

For Vin >> Vref, IC1 = IEE, and VOL ≈ VCC − VBE(on) − RCIEE. For Vin << Vref, IC1 ≈ 0, and VOH
≈ VCC − VBE(on) − Vb. Vb is the voltage drop over RC, caused by the base current of Q3 (=
RCIB3). This factor can normally be ignored for manual analysis. The logic swing of the
gate hence equals
V swing = R C I EE (A.47)
Noise Margins
Deriving exact expressions for VIH and VIL is somewhat more involved, due to the heavily
nonlinear current relations. As mentioned earlier, one common approach to circumvent
this analysis is to use revised definitions of VIH and VIL. Specifically, VIL is defined as the
point where transistor Q1 carries 1% of IEE, while at VIH Q1 carries 99% of the total cur-
rent. The corresponding values of Vin can be derived from Eq. (A.45):
I C1 e x - = α = 0.01
------- = -------------
I EE 1 + ex
(A.48)
α
V IL, IH = V ref ± φ T ln  ------------
 1 – α
At room temperature (φT = 26 mV), we find that VIL = Vref − 120 mV and VIH = Vref +
120 mV. This results in a very narrow transition region of 240 mV.
The results ensuing from this approach tend to differ considerably from those
obtained using the traditional unit gain definition. Consequently, it is worthwhile to derive
more accurate expressions. In correspondence to earlier derivations, we use the small-sig-
nal approach. A number of simplifications have to be considered to make the analysis
tractable.
• Combining the small-signal models of the differential pair and the emitter-follower
yields rather intractable equations. The model can be simplified in an important way
by observing that the input impedance of the emitter-follower is approximately
equal to its load impedance, multiplied by (βF + 1). This technique, illustrated in
Figure A.28, is called the resistance reflection rule2 and will be quite useful for the
computation of the propagation delay of the ECL gate later in the chapter. Further-
more, the gain of the emitter-follower is approximately equal to 1 and will have vir-
tually no effect on the overall gain of the gate. Using this approach, we can
eliminate the emitter-follower from the small-signal model and replace it by a resis-
tor R′B = RB(βF + 1) in parallel with RC.
• We assume that, at VIH and VIL, the differential amplifier is sufficiently balanced so
that the small-signal gain of the amplifier approximates the gain of the amplifier
under symmetrical biasing conditions. Although the actual situation deviates sub-
stantially from this assumption (at VIH and VIL, an important imbalance exists
2
For an in-depth derivation of this rule, please refer to textbooks on analog circuit design, such as Micro-
electronic Circuits by Sedra and Smith [Sedra87, p. 457].
38 Bipolar Digital
Vcc
Q4
RB (βF + 1)
VEE
RB
Figure A.28 Approximation of the loading effect of
VEE the emitter-follower using the reflection rule.
between IC1 and IC2), it produces a reasonable approximation useful for manual
analysis.
Under those assumptions the gain of the amplifier equals g = –gmRL/2, with RL the
load resistance of the amplifier (= RC || RB(βF + 1)) and gm the transconductance of Q1
(= IC1/VT).3 The values of VIH and VIL can now be found by setting g to –1 and plugging in
the appropriate value of IC1 (Eq. (A.45)).
I EE ( R C R B ( β F + 1 ) ) V swing 
V IH,IL = V ref ± V T ln  --------------------------------------------------- – 1 ≈ V ref ± φ T ln  -------------- –1 (A.49)
 2φ T   2φ T 
As expected, VIH and VIL differ only a couple of φT (= 26 mV at room temperature)

from Vref (for Vswing varying between 250 mV and 1V). Their values depend on the value
of the logic swing IEERC in a logarithmic fashion, or, in other words, they are rather insen-
sitive to voltage variations. It is worth observing that a more precise analysis, which does
not employ the second assumption, yields only a slightly different equation:
V IH,IL = V ref ± φ T acosh  -------------- – 1

V swing
(A.50)
2φ T
Switching Threshold
Finally, VM can be determined by setting Vin equal to Vout in Eq. (A.46). This leads, once
again, to a hopelessly complex expression that is difficult to solve analytically. Some
interesting information can be obtained by realizing that the preferred value of VM equals
Vref. Under this circumstance, IEE is split equally over both branches of the current switch,
and a dc bias condition for the gate can be derived:
R C I EE
V ref = V M ≈ V CC – V BE ( on ) – -------------- (A.51)
2
3
For more information on the small-signal model of the differential pair, please refer to standard analog
design handbooks, such as [Sedra87, pp. 494–496].
Other dc bias conditions can be derived that help to define the values of the resistors
and the current source. For example, RC and IEE have to be chosen so that transistors Q1
and Q2 are biased in the forward-active region over the complete range of input voltages.
The results of the above analysis are summarized in Figure A.29, where an asymp-
totic picture of the ECL voltage transfer characteristic is drawn. Notice that for large val-
ues of the input signal Vin transistor Q1 eventually saturates, causing output Vout1 to track
Vin in a linear fashion (with Vout1 = Vin − VBE(sat) + VCE(sat)). This operation mode is of no
concern as it is out of the normal input voltage range.
Vout +/– n φT
Vout2
VCC – VBE(on)
Q1 saturates
Vout1
VCC – VBE(on) – IEERC
Vref Vin
Figure A.29 Approximate VTC of ECL gate.
Example A.7 VTC of ECL Gate

An ECL gate was designed with the following parameters: VCC = 0 V, VEE = −5 V, Vref = −
0.95 V, IEE = 0.5 mA, RC = 1 kΩ, and RB = 50 kΩ. The βF of the bipolar devices equals 100,
while IS is set to 10−17A. The bipolar transistors are fabricated using the bipolar process
described in Chapter 2. All transistors are minimum-size devices with Aemitter = 2.0 µm × 3.75
µm. The VTC of the device is computed using both manual analysis and SPICE.
First, the exact value of the VBE(on) of the emitter-follower transistors has to be derived.
This voltage varies with the collector current. A reasonable approximation is obtained by set-
ting IC to (5 V – 0.7 V) / 50 kΩ ≈ 0.1 mA, which equals the collector current of the emitter-
follower in the high output state. This results in the following value of VBE (for φT = 26 mV):
I
V BE ( on ) = φ T ln  ----C- = 0.778V .
 IS 
The important parameters of the voltage-transfer characteristic can now be computed.

40 Bipolar Digital
V OH = V CC – V BE ( on ) = – 0.78V·
V OL = V CC – V BE ( on ) – IEE R C = – 1.28V·
VIH = – 0.95 + 0.026ln  ----------------------

0.5 – 1 = – 0.89 V
 2 × 0.026  (A.52)
VIL = – 0.95 – 0.026ln  ----------------------

0.5 – 1 = – 1.01V
 2 × 0.026 
VM = V ref = – 0.95 V
The VTC has also been computed using SPICE and is plotted in Figure A.30. The
important parameters have been extracted and are compared with the manual results in Table
A.4. An excellent correspondence is observed[.
–0.70
–0.80 Vout1 Vout2
–0.90
Vout (V)
–1.00
–1.10
–1.20
–1.30
–1.5 –1.3 –1.1 –0.9 –0.7 –0.5
Vin (V)
Figure A.30 Simulated voltage-transfer characteristic of ECL gate.
Table A.4 DC-parameters of ECL inverter (manual versus computed).
Manual SPICE
VOH –0.78 –0.77

VOL –1.28 –1.26
VIH –0.89 –0.87
VIL –1.01 –1.03
VM –0.95 –0.96
NML 0.27 0.23
NMH 0.11 0.1
Design Considerations
Compared to the CMOS inverter, the small logic swing (0.49 V) and noise margins are appar-
ent. The smaller swing has the advantage of increased switching speed. The smaller noise mar-
gins can be tolerated because the logic levels are robust and not particularly dependent on
process parameters. For instance, the VBE of a bipolar process is far more predictable than the
threshold of an MOS transistor. The reduced voltage swing also helps to keep the noise levels
down. In fast ECL circuits, however, with many simultaneous switching actions, it is essential
to keep the noise within bounds.
At this point, it is worth observing that all voltage levels (such as VOH and VOL) in an
ECL gate are defined with respect to VCC and are only affected in a minor way by VEE. It makes
sense to assign VCC the best available fixed potential, which is generally the ground potential of
the power supply, which is, in general, the “most clean” supply rail. Therefore, ECL circuits
will normally set VCC to 0 V. This requires that VEE be at a negative potential, which explains
the choice of the supply voltages in the example (VCC = 0 V, VEE = −5 V). The standard value
used for the 10k and 100k series was −5.2 V. Current ECL circuits operate at lower supply
voltages ranging from −3 V to −4.5 V.
Finally, it is worth mentioning that the dc characteristics of the above gate are virtually
unaffected by fan-out. This is due to the very low output impedance of the emitter-follower.
A.3.4 ECL Switching Speed: The Transient Behavior
Studying the propagation delay of the ECL gate is a considerably more involved task than
the analysis of the CMOS inverter. The structure is inherently more complex, as it con-
tains internal nodes, each of which could dominate the performance. This is illustrated in
Figure A.31, which shows the equivalent circuit to be considered when computing the
propagation delay between the input and the inverting output Vout1 for a single fan-out.
Getting an accurate prediction of the response is only possible through SPICE simulation.
Unfortunately, this does not give an insight into the mechanisms governing the transitions.
Such an insight can only be obtained by studying a tractable circuit model, obtained
through significant simplifications:
• All capacitances are linearized over the appropriate voltage range and lumped
together into a limited number of capacitances.
• Internal nodes are eliminated (as much as possible) using the reflection rule (intro-
duced in the dc analysis).
• The transient response is decomposed into several steps, based on the dominant
effects governing the response at each step. Although in reality these steps do par-
tially overlap, this approach yields reasonable accuracy, while simplifying the anal-
ysis substantially. In particular, we isolate the switching of the differential pair from
the (dis)charging of the load capacitors. This important assumption dramatically
42 Bipolar Digital
Vcc Vcc Vcc
RC Cbc RC
VC1
Cbc Q3 Cbc
Ccs Vout1
Cbe
Vin Q1 Q1 FAN-OUT
Vx
Cd Cbe Cd Cbe
IEE RB IEE
VEE VEE VEE

Figure A.31 Circuit model of ECL gate with single fan-out, used in propagation delay analysis.
simplifies the analysis, while enabling the identification of the dominant perfor-
mance parameters.
Switching the Differential Pair

The analysis of the large-signal transient response of the current switch is intricate and
yields complex expressions.4 Complicating the derivation is the impact of the base and
emitter resistances, the feedback effect offered by the coupled emitters, and the coupling
between the transistor terminals. To avoid this multiple-page numerical art-work, we use a
simple heuristic model. In Chapter 2, it was derived that the transient response of a bipolar
transistor is dominated by the base and depletion-region charge. The same is true for the
differential pair. To illustrate this statement, let us study the simplified network of Figure
A.32. Some of the most important transient parameters are annotated on the diagram.
Notice that in this model, the collectors are kept at VCC, which effectively eliminates the
impact of the discharging of the load capacitances.
VCC
IC1 IC2
Cbc1
rB
Q1 Q2 Vref
Vin
CD1 Cbe1 Vx
IEE
VEE Figure A.32 Switching the differential pair.
4
For a detailed analysis, please refer to [Embabi93, pp.215–224]
The transient response of the current switch is simulated for varying values of the
current source IEE. The resulting collector currents for both branches are plotted in Figure
A.33. The switching time is a strong function of IEE for large values of the current source,
which suggests that the base charge is the dominant factor under those circumstances. On
the other hand, the delay is independent of IEE for smaller current values. This means that
the depletion charge of the be and bc junctions has become the most important factor.
Varying the value of the junction capacitances in the simulation confirms this assumption.
Observe the large spikes in the collector currents during the switching of the input signal.
This is due to the capacitive coupling between base and emitter of Q1. A simple model can
now be constructed for both mechanisms.
IC1
Collector current (normalized)
1 IC2
10 mA 5 mA
1 mA
0.5 mA
Figure A.33 Simulated collector
–1 currents of differential pair (final value
0 0.1 0.2 normalized to 1), with IEE as a
Time (nsec) parameter.
(Dis)charging the Depletion Capacitances (space charge). Based on our common

practice, we first replace all nonlinear junctions capacitors by linearized equivalents.
This is accomplished by multiplying the zero-bias capacitance with the Keq factor (Eq.
(2.18)), which is a function of the applied voltage range. The voltage at the base of Q1
swings between VOL to VOH, while the emitter voltage moves between Vref − VBE(on) to
VOH – VBE(on). The collector voltage stays constant at VCC. Knowledge of these voltage
ranges allows for the computation of the Keq values:
K eq, be = K eq ( V 1, V 2 ) = K eq ( V OL – V ref + V BE ( on ), V BE ( on ) )
(A.53)
K eq, bc = K eq ( V 1, V 2 ) = K eq ( V OL – V CC, V OH – V CC )
The lumping of the junction capacitances into a single component requires some
extra caution. The Cbc1 undergoes a full Vswing during the transient period. The total volt-
age swing over Cbe1, on the other hand, is limited to Vswing/2 (assuming that Vref is placed
in the middle of the voltage swing), as the emitter tracks the base once the transistor is on.
Its impact is thus half of the other capacitor (an inverse Miller effect). Taking this into
account, the following expression for the equivalent input capacitance can be derived over
the Vswing voltage range.
C in, j = K eq ,bc C bc1 + K eq ,be C be1 ⁄ 2 (A.54)

44 Bipolar Digital
The circuit diagram of Figure A.32 suggests that the (dis)charging of the depletion
capacitances is dominated by the rBCin,j time-constant. Achieving 90% of the final value
requires approximately 2.2 time-constants, if the voltage waveform at the base is exponen-
tial (which is approximately the case). Under these conditions, the time to change the
space charge is adequately modeled by the following expression:
t space = 2.2r B C in ,j = 2.2r B ( K eq ,bc C bc1 + K eq ,be C be1 ⁄ 2 ) (A.55)
(Dis)charging the Diffusion Capacitance (base charge). The diffusion capacitance CD1
represents the excess-charge storage in the transistor base (introduced in Chapter 2). CD1
is placed at the base of the transistor and stores an amount of charge equal to ∆QF over the
voltage range of interest. From Chapter 2, recall the following expression,
∆Q
C D = ----------F- (A.56)
∆V
During a single transition, the current through Q1 evolves from virtually nonexisting
to IEE, or ∆QF = QF = τFIEE. The voltage swing at the base during that period is, obviously,
Vswing. Hence,
∆Q τ F I EE
C D1 = ----------F- = -------------- (A.57)
∆V V swing
It is worth observing that the diffusion capacitance is proportional to the value of the
current source. This proportionality is clearly manifested in the simulation results of Fig-
ure A.33. Remembering that the voltage swing can be expressed as a function of IEE as
well (Eq. (A.47)) yields another interesting relation. Eq. (A.58) states that the diffusion
capacitance of Q1 is only a function of the forward transit time τF and the collector resis-
tance RC (a design parameter!).
τ F I EE τF
C D1 = -------------- = ------ (A.58)
R C I EE RC
Deriving an exact expression of the time to change the base charge is difficult, espe-
cially due to the impact of the emitter-coupling (and other factors, such as emitter resis-
tance). The time is proportional, though, to the time-constant of the input circuit, formed
by rB and CD1. This suggests the following model for tbase:
t base = αr B C D1 (A.59)
where α is an empirical factor, and depends upon circuit and device parameters. From
simulations, we derive that α approximately equals 2 and 5 for the 50% and 90% points,
respectively.
Propagation Delay of the Differential Pair. The overall switching time of the differen-
tial pair can now be expressed as the sum of the space- and base-charge components
(assuming that both mechanisms happen consecutively). Remember from the simulations
that the response is typically dominated by a single mechanism.
t dp = t space + t base = r B ( 2.2C in ,j + αC D1 ) (A.60)

Example A.8 Switching the Differential Pair

Consider the ECL gate of Example A.7. The bipolar transistor model defined in Chapter 2
provides the extra parameter values needed for the transient analysis:
rB = 120 Ω, rE = 20 Ω, and rC = 75 Ω; Cbe = 20 fF, Cbc = 22 fF, and Ccs = 47 fF.
mbe = 0.5, mbc = mcs = 0.33, τF = 10 psec.
With IEE = 0.5 mA (or Rc = 1 kΩ) and τF = 10 psec, CD1 equals 10 fF. Cbe and Cbc are defined
to be 20 and 22 fF, while their Keq-factors evaluate to 3.35 and 0.75, respectively (with φ0 =
0.7 V).5 This yields the following capacitance values:
CD1 = 10 psec/ 1kΩ = 10 fF; Cin,j = 3.35 × 20 / 2 + 0.75 × 22 = 50 fF
The time it takes for the collector currents to reach 90% of their value is now approxi-
mated using Eq. (A.60) (and assuming that α = 5).
tdp = 120 Ω × (2.2 × 50 fF + 5 × 10 fF) = 20 psec
(Dis)Charging the Load Capacitances

To derive the second component of the transient response, we assume that the current
switch has already reverted state. This means that the current-switch transistors are either
off, or carry the complete current IEE. Before addressing the analysis of the propagation
delay, it is good practice to derive the capacitance values first, as we did for the MOS
inverter.
Deriving the Load Capacitances. As illustrated in Figure A.31, the capacitance model
of the ECL inverter is complex and consists of many contributions. To simplify the analy-
sis, we gather all the individual contributions into two lumped capacitors: the collector
capacitance CC and the load capacitance CL.
1. The collector capacitance CC = Ccs1 + Cbc1 + Cbc3 + CD3.
The first three factors in CC represent the depletion capacitances of the collector-
substrate junction of Q1 and the base-collector junctions of Q1 and Q3. Since the
emitter-follower transistor Q3 is assumed to be always on and in forward-active
mode, the voltage over the base-emitter junction is constant, and its depletion charge
remains unchanged. This explains the absence of Cbe3 in the list. The equivalent
junction capacitances are easily computed once the voltage excursions are known.
The base of Q1 is assumed to stay constant, while its collector voltage VC swings
between VCC to VCC – Vswing. The collector of Q3 stays at VCC. No Miller effect
occurs, and the capacitances can simply be added, weighted with the appropriate Keq
factors.
CD3 models the change in base charge of Q3 during a transition:
∆QF = τ F (VOH − VOL)/RB = τFVswing/RB,
5
In the computation of Keq for Cbe, it is assumed that the maximum forward bias is equal to φ0.
46 Bipolar Digital
where VOH/RB and VOL/RB approximate the collector currents of Q3 in the high and
low output state, respectively. The equivalent diffusion capacitance is derived by
dividing ∆QF by the voltage swing of interest:
∆Q τ F V swing τ
C D3 = ----------F = --------------------- = -----F- (A.61)
∆V R B V swing RB
Observe the similarity to Eq. (A.58).
2. The load capacitance CL = CW + Cin
The load of the gate equals the sum of the wiring capacitance CW and the input
capacitance of the fan-out gates. The latter is a complex combination of space- and
base-charge capacitance, and varies with the direction of the transition. To sim-
plify the analysis, we just assume a fixed, average value in this study.
Example A.9 Load Capacitances of an ECL Gate

Consider, again, the ECL gate of Example A.7. The values of the load capacitances are
readily computed.
1. CC—Due to the high value of RB, Cd3 evaluates to 0.2 fF and can be ignored. The Keq
factors for Cbc1, Cbc3, and Ccs1 equal 0.84, 0.91, and 0.51, respectively (the low value
for CCS is due to the strong reverse bias on that junction, with the substrate at −5 V).
This yields the following value for the collector capacitor.
CC = 0.51 × 47 + 0.84 × 22 + 0.91 × 22 + 0.2 fF = 62.7 fF
2. CL—We assume a value of 60 fF here, which approximates the average input capaci-
tance of a similar ECL gate.
Discharging the Load Capacitances. The time it takes to discharge the collector and
output nodes is determined by two competing mechanisms: the discharging of the load
capacitance CL through the load resistor RB, and the discharging of the collector capaci-
tance by the current source IEE. Both processes are illustrated in the (simplified) transistor
network of Figure A.34. Depending upon the values of the resistances, capacitances, and
current levels, either one of them can be dominant.
Vcc Vcc
Rc
VC1
Q3
IEE Vout1
CC
VEE RB
VEE
CL Figure A.34 Discharging the collector
VEE and load capacitances.
When RBCL << RCCC, the discharging of the collector node through IEE dominates
the performance, and the discharge time is proportional to the time-constant RCCC.
Observe that value of IEE sets the dc voltage levels, but does not influence the discharge
time. If this argument does not seem obvious, a useful exercise is to construct the Theve-
nin equivalent circuit of the network, consisting of IEE, VCC, and RC.
A more accurate derivation takes the base current of Q3 into account. Even for very
small values of RB CL, Vout1 can never fall faster than Vc1, since this causes the emitter-fol-
lower to turn on. The nodes, therefore, discharge in unison. The collector and output nodes
are tightly coupled as the output node follows the collector node with a (fixed) voltage dif-
ference of VBE(on). This operation is adequately modeled by the equivalent circuit of Figure
Vcc
Rc VBE(on)
Vout1
CC(βF+1)
RC/(βF+1)
+ – IEQ Vout1
ib
RB
IEE
CL
CC βFib RB CL
VEE VEE
(a) Simplified circuit (b) Norton equivalent circuit of (a)

Figure A.35 Equivalent circuits for the computation of tpHL, for RBCL << RCCC, with IEQ = IEE(βF + 1) −
VEE /RB − (VCC − VBE(on))(βF + 1)/RC.
A.35a (where the base-emitter diode is modeled as a voltage source), or by the even sim-
pler Norton equivalent circuit model of Figure A.35b. The same circuit could have been
derived in a single step by using the reflection rule to eliminate the internal node. Divide
all impedances at the base of Q3 by (βF + 1) and move them to the emitter node. Notice
that this means that Cc is multiplied with (βF + 1). The discharge time is approximated by
the time-constant of the resulting circuit:
t discharge1 = 0.69  R B || --------------

RC 
- [ C + CC ( βF + 1 ) ] (A.62)
 β F + 1 L
For large values of RCCC, this expression simplifies to 0.69RCCC, as was projected.
When RBCL >> RCCC, the collector voltage drops faster than the output node, which
means that the emitter-follower transistor Q3 shuts off (until the output voltage gets
clamped to its final value by the follower, as shown in Figure A.36). In this case, the prop-
agation delay is dominated by the time it takes to discharge CL through RB. The effect of
the clamping makes it necessary to use the average current approach to determine the
delay, as the RC-technique used in Eq. (A.62) would predict a considerably different
result.
48 Bipolar Digital
1 V OH – V EE V OH – V swing ⁄ 2 – V EE V swing

- + ----------------------------------------------------- = ------  V OH – V EE – --------------
I av = ---  ------------------------
1
2 RB RB  RB  4 
C L ( V swing ⁄ 2 ) 0.5C L R B V swing V swing  (A.63)
t discharge2 = --------------------------------- = ---------------------------------------------- ≈ 0.5C L R B  ------------------------
-
I av V swing  V CC – V EE
V OH – V EE – --------------
4
Vout1
Clamping by emitter-follower
Figure A.36 Effect of voltage clamping on

Discharge to VEE by RB discharge waveform. When only the RC-
network would be considered, the output
would discharge to VEE. This is prevented by
t
the emitter-follower.
The actual discharge time is approximated by taking the worst case of both
scenarios:
t discharge ≈ max ( t discharge1, t discharge2 ) (A.64)
with tdischarge2 the time it takes to discharge the collector node and tdischarge1 the time to dis-
charge the load capacitance through RB.
Charging the Load Capacitances. A similar approach can be taken to estimate the time
it takes to charge the load capacitances CC (through RC) and CL (through Q3). The process
is illustrated in the circuit model of Figure A.37. To obtain accurate expressions for the
Vcc Vcc
Rc
Q3
Vout1
CC
RB
CL
VEE Figure A.37 Equivalent circuit for the computation of tpLH.
delay, we resort again to the reflection rule and the use of equivalent circuits, as shown in
Figure A.38a and b. The propagation delay of the resulting circuit is easily derived:
RC
t ch arg e = 0.69  --------------
- || R B [ C C ( βF + 1 ) + C L ] (A.65)
 βF + 1 
which is similar to the expression obtained for tdischarge2 above.

Vcc
VBE(on) Vout1
Rc
Vout1
CC(βF+1)
RC/(βF+1)
+ –
ib
RB
CL
IEQ
CC βFib RB CL
VEE
(a) Simplified circuit (b) Norton equivalent circuit of (a)

Figure A.38 Equivalent circuits for the computation of tpLH, with IEQ = −VEE/RB − (VCC − VBE(on)) ×
(βF + 1)/RC.
Example A.10 ECL Transient Response

The total propagation delay of the ECL gate of Example A.7 can now be computed (recall
that RC = 1kΩ, and RB = 50 kW).
Consider first the tpHL. The time to switch the differential pair was computed in Exam-
ple A.8 and equals 20 psec. The second factor of the delay is the time to discharge the node
capacitance. Since RBCL >> RCCC, the output node time-constant dominates, and the dis-
charge time is given by Eq. (A.63).
tdischarge = (0.5 × 60 fF × 50 kΩ × 0.5 V) / (−0.7 + 5 − 0.125 V) = 180 psec.
Adding the two components leads to the following approximation:

tpHL = 20 psec + 180 psec = 200 psec.
The low-to-high transition consists of turning off the current switch and charging the
load capacitances. The charge time is approximated by Eq. (A.65), with the collector node
presenting the largest delay.
tcharge = 0.69 RCCC = 0.69 × 1 kΩ × 62.7 fF = 43.3 psec.
Combining this with the delay of the differential pair yields

tpLH = 20 + 43.4 = 63.3 psec
and
tp = (200 + 63.3)/2 = 132 psec
The simulated transient response of the circuit is shown in Figure A.39. The extracted values
of tpHL and tpLH respectively equal 163 psec and 92 psec, yielding tp = 127.5 psec, which is
consistent with the estimated results.
A Global View of the Transient Response

The results of this extensive analysis are summarized in Table A.5. Observe again that
these expressions are first-order models and are by no means intended to be accurate.
50 Bipolar Digital
–0.70
Vin Vout1
–0.90
V (Volt)
–1.10
–1.30 Figure A.39 Simulated

0 0.2 0.4 0.6 0.8 1.0 transient response of ECL
t (nsec) inverter.
SPICE simulations are indispensable if precise numbers are required. The most important
function of the presented model is to provide a clear insight into the parameters dominat-
ing the performance of the ECL structure.
The analysis shows that the ECL gate exhibits asymmetrical high-to-low and low-
to-high transitions. The difference between charge and discharge times is most obvious for
large load values. In that case, the rise time is determined by the CLRC/βF time constant,
Table A.5 Propagation delay of ECL gate: summary.
Switching the
differential pair Capacitive (dis)charge
 V swing 
tpHL RBCL >> RCCc r B ( 2.2C in ,j + αC D1 ) 0.5 C L R B  --------------------------
-
 V CC – V EE
 RC 
RBCL << RCCc - || R B ( C C ( β F + 1 ) + C L )
0.69  ---------------
 βF + 1 
 RC 
tpLH r B ( 2.2C in ,j + αC D1 ) - || R B ( C C ( β F + 1 ) + C L )
0.69  ---------------
 βF + 1 
while the fall time at the output is set by CLRB. For CL smaller or comparable to CC, tpHL
and tpLH are approximately equal, and the delays are dominated by the CCRC time constant.
The former case occurs most often in discrete ECL gates or in gates driving large capaci-
tive loads.The latter situation occurs in gates internal to modules such as adders or CPUs.
In these circumstances, the propagation delay is mainly dominated by the intrinsic capaci-
tances, modeled by CC.
In the presented example, the propagation delay is dominated by the capacitive (dis)charge time.
This causes the propagation delay to increase linearly with the capacitive load, as in the CMOS
case. It is important to notice, however, that the ECL gate is less sensitive to capacitive loading
than the CMOS counterpart. During a low-to-high transition, CL is charged through the emitter-
follower, which means that a substantial charging current is available (or the effective capacitive
load is divided by the current gain of the emitter-follower). The discharge time for large load
capacitances is dominated by the RBCL time-constant, which can be made very small by reducing
the value of RB.
Compared to the propagation delay of the CMOS inverter (225 psec), a speedup of 1.8 is
obtained for a single fan-out. Considerable improvements over this result can be attained. The
gate design presented in the example is far from optimal. Its delay can be lowered by increas-
ing the current levels in both the differential pair and the emitter-follower. This effectively
trades off an improvement in performance against an increased power consumption. For a
given logic swing, increased current levels mean reduced values of resistors RC and RB. From
Table A.5, it is apparent that this translates to a reduction in capacitive (dis)charge times.
At this point, you probably wonder how low a delay can be obtained and if an optimum
current level can be defined. In effect, increasing the collector current also increases the base
charge in the bipolar transistors, and, consequently, the turn-on and turn-off times of the
devices. This is clearly demonstrated in Eqs. (A.58) and (A.61), which state that the diffusion
capacitances are inversely proportional to resistor values. At some current level, the diffusion
capacitance becomes dominant, and the delay levels off and even starts to grow. Other second-
order effects, such as the Kirk effect and the collector resistance come into play at higher cur-
rent levels and cause a further degradation. To illustrate this behavior, we have plotted the sim-
ulated tpLH of our ECL gate as a function of IEE (we have kept RB constant for this analysis). A
200
150
100
tpLH (psec)
50
0
0 5 10 15 20 Figure A.40 tpLH of ECL
inverter as a function of the IEE
IEE (mA) bias current.
sharp decrease in the delay can initially be observed due to a reduction in the dominant time-
52 Bipolar Digital
constant RCCC. For higher current levels, the time to turn-off Q1 starts to dominate, and the
delay rises again (in accordance with rBCD1).
In summary, careful optimization of the gate structure can substantially reduce the delay.
Combining this with a state-of-the-art bipolar technology (the parasitic capacitances can be as
low as 10 fF in a 0.8 µm bipolar process) leads to ECL propagation delays ranging from 40 to
70 psec. This is between 3 to 4 times faster than its CMOS equivalent.
A.3.5 Power Consumption
The ECL gate clearly consumes static power. Sources of this static consumption are the
emitter-coupled pair, the bias network, and the output stage. The dissipation of the latter
depends upon the value of the pull-down resistor and the termination mechanism used.
The power consumption of the bias network, on the other hand, can be distributed over
multiple gates. These observations are summarized in the following equation,
 -------------------------- – V EE
V OH + V OL
 I bias 2
P stat = ( V CC – V EE )  I EE + ---------- + 2 -----------------------------------------  (A.66)
 N RB 
 
where N equals the number of gates serviced by a single bias network. The factor 2 in the
power-consumption expression for the emitter-follower is due to the complementary out-
puts.
Furthermore, dynamic power is consumed during switching. During each switching
event, both internal and external capacitances are (dis)charged, requiring a charge transfer
equal to Vswing(CC + CL). In contrast to the CMOS inverter, the voltage swing is consider-
ably smaller than the supply voltage. This results in a slight modification in the expression
for the dynamic power consumption (compare this to the expression in Eq. (3.42), derived
for the CMOS inverter).
P dyn = C T ( V CC – V EE )V swing f (A.67)
with CT the sum of all capacitances switched.
The static power factor generally dominates the overall dissipation. In fact, it can be
shown that for the dynamic power to become the dominating factor, it is necessary to
switch the gate faster than the propagation delay would allow.
Example A.11 Power Dissipation of the ECL Inverter

The power dissipation of the inverter of Example A.7 is analyzed. We ignore the power con-
sumed in the bias network.The static power consumption is computed with the aid of Eq.
(A.66).
Pstat = 5 (0.5 mA + 0.162 mA) = 3.3 mW
Even when switching the gate at the maximum allowable speed (f = 1/tp), the dynamic con-
sumption is still smaller,
Pdyn = (60 fF + 72.8 fF) × 5V × 0.5V / 127.5 psec = 2.6 mW
Combining the two factors yields a considerable consumption of approximately 6 mW.

The power-delay product of the gate equals 760 fJ, which is almost identical to the number
obtained for the CMOS inverter. The reduced switching delay is offset by the increased
power consumption. Notice, however, that PDPs as low as 120 fJ have been reported in state-
of-the-art 0.6 µm bipolar processes. The high power dissipation obviously constrains the
number of gates that can be integrated on a single die.
Design Consideration
Finally, one more observation is worth mentioning. During the qualitative analysis of the gate,
we have already noted that the differential pair draws a constant current from the supply, hence
introducing virtually no switching noise on the supply lines. This is also true for the bias net-
work, but is not the case for the emitter-follower. Depending upon the values of the terminating
resistor and the load capacitance, the switching causes large stepwise current variations in the
collectors of the emitter-follower transistors. As will be discussed in later chapters, parasitic
inductances can translate those current fluctuations into supply noise (ringing, voltage spikes),
which might cause the circuit to fail. Therefore, ECL circuits often employ two VCC’s: a
“clean” VCC, which connects to the differential pairs and the bias networks, and a “dirty” VCC,
which feeds the emitter-followers. This separation avoids the feeding back of switching tran-
sients into differential pairs that could prove disastrous given the low noise margins.
A.3.6 Looking Ahead: Scaling the Technology
When discussing the influence of technology scaling on the performance of a CMOS pro-
cess, we introduced the ideal scaling model. In this model, both dimensions and voltages are
scaled in a similar way, keeping the electrical fields in the devices approximately constant.
Bipolar scaling is considerably more complex than MOS scaling, as more device
parameters are involved in the design. Also, power-supply voltage and logic swing cannot
be reduced much, since they already approach their lower limit at room temperature. The
on-voltage of the base-emitter junction is a built-in parameter, which is virtually unaf-
fected by scaling. The full-scaling model is therefore not appropriate for the ECL inverter,
and the fixed-voltage scaling model must be used. Consequently, both electrical field
strengths and current densities increase when scaling.
A typical bipolar scaling model is presented in Table A.6. Voltages and currents are
kept to a constant level. This means that the current density increases with a squared fac-
tor. Capacitances, consisting of junction and diffusion capacitances, are scaled linearly,
which requires a scaling of the base width as well as the doping levels in base and collec-
tor regions. Thus, bipolar scaling means not only shrinking the lateral dimensions, but also
the vertical profile of the transistor.
This model is illustrated in Figure A.41, where the projected gate delay of an ECL
circuit is plotted as a function of the current level and feature size. The plot also provides
54 Bipolar Digital
Table A.6 Scaling model for bipolar inverter (from [Tang89]).
Parameter Scaling Factor

AE 1/S2
wB 1/S0.8
Vsupply, Vswing 1
J S2
I 1
Cd,Cj 1/S
tp 1/S
P 1
insight in the number of gates that can be integrated on a single die (for a maximum con-
sumption of 2 Watts).
Figure A.41 Projected gate delay of scaled

ECL circuit in function of current level and
feature size (from [Tang89]).
A.4 Bipolar Gate Design
In the subsequent sections, we first discuss how the generic ECL gate, introduced in Chap-
ter 3, can be extended to implement complex logic functions. A number of modifications
to the generic gate are introduced to increase its performance or to improve its scalability.
The section concludes with a short overview of some alternative bipolar logic styles.
A.4.1 Logic Design in ECL
To recapitulate, the basic structure of the ECL gate is repeated in Figure A.42. It consists
Vcc Vcc Vcc Vcc
RC RC
VC1 VC2
Q3 Q4
Vin Q1 Q2 Vref
Vout1 Vx Vout2
RB
IEE
VEE
VEE Figure A.42 Basic structure of the ECL gate.
of a bias network generating a reference voltage Vref, a differential pair Q1-Q2, and a pair
of emitter-follower output drivers. From a logic point of view, one of the useful properties
of the basic ECL gate is that it uses a complementary logic style, which means that the
complement of each logic signal is available. This is similar to the CMOS CPL design
technique discussed earlier. This approach avoids the need for extra inverters, as is the
case in inverting logic families such as complementary CMOS. It has been observed that
for a given logic function, static CMOS takes approximately twice as many gates as ECL.
Example A.12 Advantages of Complementary Logic

A 4-bit ALU in CMOS requires 98 gates using NAND logic, while an ECL implementation
can be conceived with only 57 gates. Even more important, the average number of cascaded
logic stages in a typical signal path equals 6.1 for CMOS and 2.9 for ECL. This reduction of
the critical path by a factor of 2.1 has a significant impact on performance [Masaki92].
One of the simplest gates to implement in ECL is the OR/NOR function, which can
be realized by adding extra input transistors in parallel, as shown in Figure A.43. Turning
one of the two input transistors on is sufficient to divert the bias current to the left branch
of the current switch and to pull VC1 down. This is equivalent to realizing a NOR gate. The
complementary node VC2 goes high at the same time, implementing the OR function. The
static and dynamic characteristics of the standard ECL gate are only marginally affected
by this modification.
Realizing an (N)AND operation is somewhat more complex. One approach to
implement complex functions is to use the wired-OR configuration, as shown in Figure
A.44. Simply wiring the gate outputs together (also called dotting) is equivalent to an
“OR-ing.” It is sufficient for one of the connected outputs to be high for the combined
result to be high. The resulting gate implements (A + B) + (C + D) = (A + B) ⋅ (C + D).
The advantage of this approach is that complex gates can be constructed on the fly by sim-
ply connecting or wiring the outputs of basic logic OR structures. Observe that the com-
bined gate is not complementary anymore, since only (A + B) ⋅ (C + D) is available, not its
56 Bipolar Digital
Vcc Vcc
RC RC
VC1 VC2
NOR OR
A B Vref
Vx
VEE
RB IEE
Figure A.43 Two-input OR/NOR gate in ECL.
VEE The emitter-followers have been omitted for
VEE simplicity .
Vcc Vcc
Vcc Vcc Vcc Vcc
RC RC
A B Vref Vref C D
RB (A+B)(C+D)
IEE IEE
VEE
VEE VEE
Figure A.44 Wired-OR configuration, realizing (A + B) · (C + D).
complement. Figure A.44 also shows that in case only one output of an ECL gate is used,
it is perfectly reasonable to omit the collector resistance at the other side.
The ECL circuits discussed so far are of the so called single-ended nature. The cen-
tral component of the gate is a current switch, with one side connected to the input(s) and
the other side to a reference voltage centered in the middle of the logic swing. To ensure
sufficient noise margins, it is essential that the low and high input levels differ a number of
φT from the reference voltage. Remember from Chapter 3 that the transition region of the
ECL inverter approximately equals 120 mV to 240 mV. For that reason, the logic swing of
a single-ended family is normally chosen to range between 500 and 800 mV. The extra
safety margin is necessary to accommodate variations in Vref, supply voltage, process
parameters, or temperature.
A.4.2 Differential ECL
Since the propagation delay of a digital gate is directly proportional to the logic swing, it is
attractive to reduce the swing even more. This is made possible by a simple modification of
the generic ECL structure. Instead of connecting the second input of the current switch to a
reference voltage, it can be driven by the inverted value of the first input, as shown in Fig-
ure A.45. Such a gate has a reduced transition region. While one input goes up, the second
Vcc Vcc Vcc Vcc
RC RC
VC1 VC2
Q3 Q4
Vin Q1 Q2 Vin
Vout Vx Vout
RB RB
IEE
VEE VEE
VEE Figure A.45 Differential ECL buffer.
one goes down. This effectively doubles the voltage swing observed by the differential
pair.
This observation is confirmed by a first-order derivation of the VIH and VIL of the
inverter/buffer. The ratio of the currents between the left and right branches of the current
switch can be expressed by Eq. (A.68).
V in – V in
I C1 I S e V in ⁄ φ T ---------------------
------- = ------------------- = e φT (A.68)
I C2 I S e V in ⁄ φ T
The current ratio is an exponential function of the difference between the signal and
its inverse, which is twice as large as the difference between the signal and a fixed refer-
ence voltage, assuming that the inverse signal changes at the same rate. Using the alterna-
tive definition of VIL and VIH (IC = 1% or 99% of IEE, respectively), the width of the
transient region can be computed as follows
V in – V in
---------------------
I C1
------- e φ T - = α = 0.99
= --------------------------
I EE V in – V in
---------------------
1 + e φT (A.69)
α  = 120 mV (at room temperature)

V IH – V IL = φ T ln  ------------
 1 – α
As expected, a reduction by a factor of 2 with respect to the single-ended structure is

obtained. This, in turn, allows for a reduction in voltage swing. Values as small as 200 mV
are not uncommon!
The differential approach offers several important advantages:

58 Bipolar Digital
• The sensitivity to supply noise is reduced. Local drops in the supply voltages affect both
Vin and Vin in a similar way. This has no impact on the operation of the gate, since only
the difference between input signals is important. This is not true in the single-ended
case, where the input signal is compared to a fixed reference voltage that varies as a
function of temperature, process parameters, and noise levels.
• In addition, the switching noise introduced on the supply lines is substantially reduced in
the differential case as a result of the reduced voltage swing and the balanced load. For
the single-ended case, it is assumed that only one side of the gate is connected to an
emitter-follower, which causes large variations in the supply current during switching.
The supply current in the differential case is approximately constant. This is confirmed
in Figure A.46, which compares the supply current (or power) of the single-ended and
differential logic structures for both a positive and negative signal transition (from
[Greub91]).
Figure A.46 Instantaneous power consumption (or supply current) for differential and single-ended
buffers (from [Greub91]). Parameters: VCC − VEE = 5 V, IEE = 400 µA, Iemitter-follower = 800 µA,
Vswing(differential) = Vswing(single-ended)/2 = 250 mV.
• Because differential logic circuits tend to require a lower number of cascaded gates for a
given function, they are generally faster than their single-ended counterparts.
The differential approach also has some potential deficiencies. The implementation
of a fully differential style requires complementary logic networks similar to the DCVSL
CMOS logic discussed in Chapter 4. This translates to a larger number of transistors, as
demonstrated in Figure A.47, which shows the schematic diagrams of a three-input OR
and AND gate. The OR/NOR structure requires six transistors in contrast to the four
needed in the single-ended case.
Differential logic structures are based on the same fundamental concept as the sim-
ple ECL inverter, namely, current steering. Based on the value of the inputs, the current
Vcc Vcc Vcc Vcc
ABC ABC A+B+C A+B+C
C C C C
B B B B
A A A A
IEE IEE
VEE VEE
(a) Three-input AND/NAND (b) Three-input OR/NOR

Figure A.47 AND and OR gates in differential ECL (emitter-followers are omitted).
provided by the current source is guided along either the left or the right branch of the dif-
ferential pair, causing one output to go low and the other to remain high. This is easily ver-
ified for the gates presented in Figure A.47. Be aware that a current path has to be
available for all possible operation conditions!
While the idea seems simple enough, there is a catch, as illustrated in Figure A.48a.
For performance reasons, Q3 should not go into saturation (see Section 3.4.1). Conse-
quently when Q3 is on (or Vin3 is high), it holds that VC3 ≥ VB3.To turn Q2 on, VB2 must be
at VC3 + VBE(on) ≥ VB3 + VBE(on) or, Vin2 ≥ Vin3 + VBE(on). This means that all inputs connect-
ing to transistors at level 2 (Figure A.48a) need a dc offset of at least VBE(on) with respect to
inputs to level 3 devices.6 Similar offsets are needed when connecting to devices placed
higher in the input network. Achieving these offsets is accomplished by adjusting the out-
Vcc Vcc
Level-shifter
RC
Q4
Vout1 Level 1 Vcc Vcc
Q1
Vin1
Q5 Level 1
Level 2
Vout2 Q2 Level 2
Vin2 Vout2
Q6 Vout1
Q3 RB RB
Vout3 Level 3
Vin3
VEE VEE
IEE
RB
VEE VEE
(a) Multilevel emitter-follower output stage (b) Level 1-to-2 converter

Figure A.48 Adjusting voltage levels in differential ECL using level shifters.
60 Bipolar Digital
put stage of the (preceding) gate, as demonstrated in Figure A.48a. Connecting the base
and collector of transistors Q4 and Q5 turns those devices into diodes with an on-voltage of
VBE(on). By tapping the outputs at the various diode positions, suitable input signals can be
generated for all required levels.
This level-shifting creates extra complexity, since multiple wires may need to be
routed for the same signal, depending upon the fan-out of the gate. Its also puts a restric-
tion on the number of transistors that can be stacked. The propagation delay is a function
of the output level, because signals lower on the stack have a higher delay. Adding too
many layers results in an intolerable performance degradation. Furthermore, the number
of layers is restricted by the available voltage range between the supply rails. For instance,
the level 3 signals in Figure A.48a range from VCC − 3VBE(on) (high) to VCC − 3VBE(on) −
Vswing (low). A typical differential ECL library (e.g. [Tektronix93]) uses at most three
layers.
Another way to address noncompatible signal levels, while avoiding the complex,
multilevel output emitter-follower, is to insert level-shifting circuits whenever needed. A
level 1-to-level 2 converter is shown in Figure A.48b. In this way, all standard logic gates
can be designed with a sole level 1 output, and level-shifting buffers are introduced only
when connecting to multilevel gates.
One additional property of the differential logic style is worth mentioning. The sin-
gle-ended ECL gate tends to exhibit radically different values for its output rise and fall
times, because these are set by different circuit elements. This results in different values of
tpLH and tpHL, because the switching point of the gate is determined by comparing the input
signal with the fixed reference voltage. In the differential circuit, the switching is deter-
mined by the crossing of the input signal and its inverse. Although these signals may have
very different slopes, this does not affect the delay of the gate, which is thus independent
of the direction of the transition. Figure A.49 compares the delay of a single-ended versus
a differential buffer as a function of the capacitance. While the delay of the single-ended
version is a strong function of the transition direction, only one delay exists for the differ-
ential case. Notice also the smaller delay of the differential buffer caused by the reduced
swing. Finally, observe that the value of the load capacitance has only a minor impact on
the delay in contrast to CMOS.
Example A.13 Differential ECL Buffer

Figure A.50 shows the schematics and layout of a differential ECL buffer ([Tektronix93]). Its
unloaded delay is 180 psec and increases by 330 psec for every pF of load capacitance for a
total current of 0.6 mA. The input capacitance of the gate equals 50 fF. For a 5 V supply volt-
age, this translates into a PDP of 600 fJ (unloaded).
Example A.14 Differential ECL Gate

The circuit schematics of a four-input multiplexer is shown in Figure A.51. Verify its func-
tionality by checking how the current flows for each of the combinations of the signals A and
B. The logic swing at the output equals 400 µA × 625 Ω = 250 mV. To function correctly, the
6
Because ECL signals are normally referenced with respect to VCC, it is common to refer to level 1
signals as those that are one VBE below VCC , level 2 as 2 VBE’s below VCC , and so on.
input signals i0–3 must be at level 1 (or VCC – VBE(on)), while the B and A signals should be at
levels 2 and 3, respectively. The static power consumption of the differential switch is easily
computed and equals 400 µA × 5 V = 2 mW.
Vcc Vcc
625 Ω 625 Ω
Out
Out
i3 i3 i2 i2 i1 i1 i0 i0
B B B B
A A
400 µA
VEE = –5 V
Figure A.51 Differential ECL four-input multiplexer (emitter-followers have been omitted); VCC =
0 V.
Figure A.49 Performance comparison of a differential and a single-ended buffer (from [Tektronix93]). Process
parameters: fT = 12 GHz, AE,min = 0.6 µm × 2.4 µm, worst-case modeling. Both cells use 100 µA tree and 100 µA
emitter-follower currents.
62 Bipolar Digital
Vcc
R2
R1 R2
4 kΩ 4 kΩ R1
Q3 Q6
Q4 VCC
Inh Inl R4
Q1 Q2 Outl Outh
Q1
Q3
Q2
Vbias Q7
Q5 Q6 Q4
Vbias Q7
R3 R4 R5
8 kΩ 8 kΩ 8 kΩ Q5
VEE
VEE VEE VEE
R5
R3
Figure A.50 Schematics and layout of differential ECL buffer.

The load resistors are implemented in polysilicon. As they are
located in the routing channel, they do not add area overhead.
Courtesy of Tektronix.
Problem A.4 Differential ECL Gate

Derive the circuit topology for a two-input XOR gate in both differential and single-ended
design styles.
A.4.3 Current Mode Logic
The emitter-follower output structure of the ECL gate makes it possible to drive large
capacitive loads with small delay. Additionally, the output structure supports a large fan-
out, which would otherwise be impossible due to the finite input impedance of the ECL
gate, formed by the base of the nonsaturating differential pair input transistor in parallel
with the pull-down resistor. On the negative side, the output stage adds a considerable area
overhead, as is apparent in Figure A.50. It also consumes a substantial amount of power,
which prevents the realization of complex integrated circuits. For instance, a 300 MHz 32-
bit microprocessor implemented in the single-ended ECL logic style [Jouppi93] has been
reported. The power dissipation of the 486k transistor device equals 115 W! To disperse
the excess heat generated at these power levels, special (and expensive) cooling
approaches are an absolute necessity.
The capacitive loads on the internal nodes of a complex logic block such as an adder
tend to be small, and the fan-out is generally restricted to 1 or 2. For these small loads, it is
reasonable to eliminate the emitter-follower. When a bipolar gate only consists of a differ-
ential pair, the logic style is called current mode logic (CML). A differential version of a
CML gate is obtained by eliminating transistors Q3 and Q4, as well as the pull-down resis-
tors in Figure A.45. Eliminating the follower results in a drastic reduction in power con-
sumption, while keeping the propagation delay reasonable for small loads. Figure A.52
plots the propagation delay of a differential ECL and a CML buffer as a function of the
logic swing for a fan-out of 1 (with and without the extra load of an interconnect wire of
0.5 mm long). CML is actually faster for small logic swings and small loads. Its delay
increases rapidly with the logic swing, as the transistors start to saturate during the switch-
ing. On the other hand, the static power dissipation of the CML gate is only 2 mW versus
the 10 mW of the ECL gate. CML is, therefore, only used within cells where the fan-out is
small, the interconnect length is short, and power dissipation and cell area is an issue.
Figure A.52 Propagation delays of ECL and CML buffers versus logic swing for different loads (unloaded and
loaded with a 500 µm interconnect wire). Design parameters: fan-out = 1, IEE = 400 µA, Ifollower = 800 µA (from
[Greub91]).
Example A.15 CML Gate Characteristics

A differential CML buffer (obtained by removing the emitter-followers in Figure A.45) has
been designed using our generic bipolar technology. The value of the current source is set to
0.4 mA and RC = 625 Ω. The supply voltages VCC and VEE are set to 0 V and -1.7 V, respec-
tively. This translates into a voltage swing of 0.25V with VOH = 0 V and VOL = –0.25 V.
Observe that the differential nature of the gate eliminates the need for output level-shift-
ing, necessary in single-ended gates.
64 Bipolar Digital
The VTC of the gate is plotted in Figure A.53a. The simulated values of the logic levels
are consistent with the manual analysis. The values of VIH and VIL (using the unity gain defini-
tion) evaluate to −0.085 V and −0.165 V, which results in both a low and a high noise margin
of 0.085 V. Compared to the ECL gate of Example 3.13, all dc parameters (width of transition
region, logic swing, and noise margins) have been halved. It should be mentioned that the dc
analysis was performed with perfectly complementary inputs (Vin2 = −0.25 V − Vin1).
Vout2
Vin1 Vout1
–0.05
0.0
Vout (V)
Vout (V)
–0.1
–0.15
Vout2
–0.2
Vout1
–0.25 –0.3
–0.25 –0.15 –0.05 0 1 2 3 4 5
Vin1 (V) t (nsec)
(a) VTC (b) Transient response

Figure A.53 VTC and transient response of CML gate (Vin2 = −0.25 − Vin1).
For our standard bipolar technology, the load capacitance (for a single fan-out) can be
computed by the same techniques used in Example 3.15 for the ECL gate. Assuming a fan-out
capacitance of 60 fF, we can approximate the capacitance at the output node as follows:
CL = Cfan-out + CC = Cfan-out + (Ccs1 + 2 Cbc1) =
60 fF + 0.67 × 47 fF + 2 × 1.01 × 22 fF = 136 fF
Notice that Cbc is accounted for twice to incorporate the Miller effect. The weighting
functions are the Keq factors. Ignoring the turn-on (off) time of the transistors, the propagation
delay can now be approximated:
tp = 0.69 RC CL = 0.69 × 625 × 136 = 59 psec
The simulated transient response of the circuit is shown in Figure A.53b. The values of
tpLH and tpHL evaluate to 69 and 63 psec respectively, or tp = 66 psec. 7 This number rapidly
degrades for larger fan-out values. For instance, for a fan-out of 3, the delay climbs to 102
psec!
The static power consumption of the gate is independent of the logic state and equals
0.4 mA × 1.7 V = 0.68 mW. When switching the gate at the fastest possible speed, the
dynamic power consumption evaluates to
Pdyn = CL (VCC – VEE) Vswing / tp = 0.98 mW
7
These numbers are obtained using the traditional definitions of tpHL and tpLH . As mentioned earlier,
these definitions do not make much sense for a differential logic family, where the propagation delay should be
determined by measuring the time between the cross-over points of the differential input and output signals.
Observe that this expression takes into account that at least one collector gets charged for
every transition. The differential nature of the structure effectively doubles the dynamic con-
sumption.
Combining the obtained numbers results in a PDP of 102 fJ (for a fan-out of 1), which
compares very favorably to the 750 fJ obtained for both the CMOS and ECL inverters dis-
cussed in Chapter 3.
A.4.4 ECL with Active Pull-Downs
The realization of a fast, low-power ECL-type gate is hampered by the following con-
straints or restrictions.
• In order to reduce tpHL, the value of RB has to be kept small; tpHL normally dominates
the transient performance.
• Small values of RB result in high values of the static power consumption.
Scaling the technology to smaller lithographic dimensions has only a minor impact on this
picture.
• The supply voltage of the traditional, single-ended ECL gate cannot be dropped
much below 3 VBE (≈ 2.5 V). VOH equals VCC – VBE; the voltage drop between base
and emitter of the input transistor adds another VBE, and an additional VBE is needed
to ensure the proper operation of the current source of the differential pair. Since
VBE is a built-in voltage that is only marginally affected by technology parameters,
this voltage is not reduced in scaled technologies. The power consumption/gate can
hence only be reduced by reducing the current levels. This is in contrast to CMOS,
where the supply voltage is expected to keep dropping as the technology feature
sizes are reduced.
• RB is not a function of technology parameters and is mainly set by the allowable
power-consumption level. Fitting more gates on a die requires a reduction of the
consumption/gate, which means a higher value for RB. This adversely affects the
performance (or keeps it constant at best), as tpHL ~ RB CL. CL might decrease pro-
portional to the scaling factor, although this depends upon what portion of the load
is composed of wiring capacitance.
Both of the above arguments, nonscalability of the supply voltage and the pull-down
resistance, represent major obstacles to a continued performance improvement of the ECL
logic family unless some substantial changes are made to the structure. Short of completely
eliminating the emitter-follower (as is the case in CML), one approach is to replace the
pull-down resistor by an active network [Chuang92]. Such an active pull-down network
provides ample current when switching, but operates at reduced current levels in standby.
An example of such a circuit is shown in Figure A.54. This ac-coupled active pull-
down circuit utilizes a capacitor to strongly turn on the pull-down npn-transistor during a
negative-going output transition, and to turn it off during a positive one. The steady-state
current of the device is set by the dc-bias network. This approach reduces the stand-by cur-
rent while substantially improving the performance. The capacitor completely blocks dc
66 Bipolar Digital
signals. Extra biasing devices are needed, however, to establish the steady-state current in
the output stage. This approach has been used to realize the ECL buffers with a propaga-
tion delay of 23 psec, which is among the fastest achieved in room-temperature silicon
[Toh89].
Vcc
Vcc Vcc Vcc
RC RC
Vin Vout
Vref
IEE CC
VEE VEE Figure A.54 Ac-coupled,

VEE active pull-down ECL buffer.
Another approach is to present a variable load to the emitter-follower. The circuit in

Figure A.55 uses an NMOS device to achieve that goal. This approach can be seen as an
introduction to the BiCMOS technology, which merges bipolar and MOS transistors in the
same process and is treated in Section 5.3. The NMOS device is turned on more strongly
during a high-to-low transition, providing more pull-down current and hence increased
performance. In the high-output state, the gate-source voltage of the transistor and the
static power consumption are reduced. It has been reported that for the same performance
(of 200 psec), the structure presented consumes four times less power than the ECL circuit
with the resistive pull-down [Chen92].
2.5 V
2.5 V 2.5 V
RC RC
Vref
2.5V Vin Vout
1.3 V
Figure A.55 2.5 V

Current source ECL gate with NFET
NMOS pull-down pull-down.
Observe how the circuit of Figure A.55 is designed to operate at the minimum possi-
ble supply voltage (approximately three times VBE). The signal swing is approximately
equal to VBE and is centered around 1.3 V (1.5 VBE below the supply voltage VCC). The
diode in the output stage reduces the dc standby current. The current source of the input
stage is implemented by a bipolar current mirror, which reduces the required voltage drop
over the source. The reference voltage of 1.3 V is chosen to avoid the saturation of the cur-
rent-source transistor.
Problem A.5 Active Emitter-Follower Load

Instead of using a resistive load for the emitter-follower circuit, a current source can also be
employed, as shown in Figure A.56. The same circuit was shown earlier in the differential
buffer of Figure A.50. Discuss the impact of this option on performance and power
consumption.
Vcc
Vcc Vcc
RC RC
Vin Vref Vout
Vs
Rs Rs
VEE VEE Figure A.56 Current-source
pull-down.
Various other approaches have been proposed to improve the performance of the
ECL structure or to reduce its power consumption [Chuang92]. While these techniques
have resulted in spectacular performances at reasonable power levels, whether these
approaches will help ECL to compete with deep submicron CMOS in times to come is an
open question [Masaki92].
A.4.5 Alternative Bipolar Logic Styles
A large variety of bipolar digital gates have been proposed over time, covering a wide
span of speed and power requirements. The most popular among them has been the tran-
sistor-transistor-logic (TTL) gate. TTL dominated the discrete logic component market,
which supports individually packaged NAND and NOR gates, multiplexers or bus-driv-
ers, for more than two decades. This picture has changed in the 1980s due to two major
factors:
1. Discrete logic gates in static CMOS became competitive in speed at a lower power
cost.
2. The advent of programmable logic components such as PLDs and FPGAs, discussed
in Chapter 11, made it possible to program complex random logic functions (equiv-
alent to hundreds of TTL gates) on a single component. This results in a large reduc-
tion in board real-estate cost, while adding flexibility.
68 Bipolar Digital
The last factor in particular has influenced the demise of TTL. Due to its historical impact
and for the sake of completeness, a brief examination of the basic TTL structure is worth
the effort however. It consists of three units, as shown in Figure A.57:
Vcc Vcc
Vcc
Vout
Vin1 Vin2
Figure A.57 Generic TTL NAND

VEE VEE structure. The arrows denote the
direction of the signal transitions for
Input stage Phase splitter Output stage one of the input signals going low.
• The input stage, which consists of a multi-emitter bipolar transistor and performs an
AND-ing of the inputs: both inputs of the transistor have to be high for its collector
to be high.
• The phase splitter, which generates two signals with opposite phases. These signals
are used to drive the output stage.
• The totem-pole (push-pull) output structure. Only one of the bipolar transistors of
the structure is ON in steady-state mode, since the controlling signals have opposite
phases. In contrast to the emitter-follower output stage, this structure has the advan-
tage that no static power is consumed, while ample current drive is still available.
The disadvantage is the need for a phase splitter. Saturation of the output transistors
also degrades the performance. Elaborate techniques (using, for instance, Schottky
Barrier diodes—see Appendix D) have been devised to avoid saturation.
Because the TTL gate is an intricate composition, its actual operation is complex,
but not particularly fast. A detailed analysis of its behavior would lead us astray and would
contribute little to the understanding of contemporary digital design approaches. The
interested reader should refer to the numerous available textbooks and reference works
that treat TTL design in extensive detail (e.g., [Hodges88]).
One legacy of the TTL era has endured. The TTL logic levels have become a de
facto standard due to their widespread usage. Input and output signals of integrated cir-
cuits must often still comply with this standard. An overview of these requirements is
given in Table A.7.
Table A.7 TTL signal requirements, valid for popular TTL families such as low-power Schottky (LS) and fast
(F) TTL (from [Buchanan90]).
VIL(max), V VIH(min), V VOL(max), V VOH(min), V
0.8 2.0 0.5 2.5

Besides ECL and TTL, a number of other bipolar logic families have emerged and
vanished over the years. Three examples are integrated injection logic (I2L or MTL),
Schottky transistor logic, and integrated Schottky logic ([Hodges88]). A detailed discus-
sion is not warranted since none are actively used in current integrated circuit designs.
Vcc Vcc
RC
Q2
Vin Q1
Vout
RB
RE CE
VEE
Speed-up capacitor
Figure A.58 Nonthreshold logic
VEE (NTL) gate structure.
One gate structure, called nonthreshold logic (NTL), is gradually gaining some
acceptance in high-performance designs, due to its very low power-delay product
([Ichino87]), and is therefore worth analyzing. The basic structure of an NTL gate is
shown in Figure A.58. It consists of an input (logic) stage and an emitter-follower output
structure. The input structure resembles the RTL gate, briefly discussed in Chapter 3,
where it was concluded that its performance suffers from the saturation of the pull-down
device. This is avoided in the NTL gate by adding the emitter-degenerating resistor RE,
which ensures that Q1 stays in the forward-active mode for the voltage input range of
interest. The emitter-follower is added to provide additional fan-out drive capability, elim-
inating another deficiency of the RTL gate. The dc parameters can be derived by ignoring
the base current of the emitter-follower.
V OH = V CC – V BE ( on )
R
V OL = V CC – V BE ( on ) – ------C ( V OH – V BE ( on ) )
RE (A.70)
R E – R C 2R C – R E
- V +  ---------------------
=  ------------------ - V
 R E  CC  R E  BE ( on )
The gain of the gate in the transient region is approximately proportional to (−RC/RE), as
can be derived using the small-signal approach. A large value of the ratio enhances the
voltage gain and the noise margin, but increases the gate delay. RC = 2RE seems to be a
good compromise. Under those conditions, VOL evaluates to 3VBE(on) – VCC, and the total
signal swing equals 2VCC – 4VBE(on). As VOL has to be larger than 0, this translates to the
constraint that VCC has to be kept smaller than 3 VBE(on). Ensuring that Q1 does not saturate
over the input range of interest puts a lower bound on VCC of approximately 2 VBE(on).
A remarkable feature of this gate is that the voltage-transfer characteristic displays
only a single strong nonlinearity, corresponding to the turning on of Q1. For the rest of the
operation range Q1 stays in the forward-active mode, because saturation is avoided during
70 Bipolar Digital
normal operation. Hence the name nonthreshold logic. The lack of a well-defined low out-
put level makes this approach particularly sensitive to noise. The attentive reader will have
observed already that the gate retains the regenerative property as long as RC > RE.
Example A.16 VTC of an NTL Gate

The simulated VTC of an NTL gate is shown in Figure A.59 for the following parameters:
VCC = 1.9 V, RC = 2 RE = 0.5 kΩ. Assume that Vbe(on) = 0.75 V. The manually derived values
for VOL and VOH (0.35 V and 1.15 V, respectively) closely correspond to the simulated values
marked on the simulation. This corresponds to a signal swing of 0.8 V. The value of VIL
equals 0.8 V, while VIH cannot be defined.
1.5
VOH
1.0
Vout (V)
VOL
0.5
Q1 saturates
0.0
0.0 0.5 1.0 1.5 2.0 Figure A.59 Voltage-transfer
Vin (V) characteristic of an NTL gate.
While the emitter degeneration acts as a negative feedback and helps create the non-
threshold characteristic, its effects are not desirable from a transient point of view. Raising
the voltage at the base of Q1 also raises its emitter voltage, reducing the current available
for discharging the collector capacitance. The effects of this negative feedback can be
reduced by adding an extra decoupling capacitor CE at the emitter node, as shown in Fig-
ure A.58. The larger the capacitance, the larger the decoupling effect and the speed-up.
This capacitor can be implemented with a reverse-biased diode.
Example A.17 Transient Response of the NTL Gate

The simulated transient response of the NTL gate is shown in Figure A.60 (for a fan-out of 1
and a pull-down resistance of 2 kΩ). The response is shown without and with a decoupling
capacitance of 0.3 pF. It can be observed that the decoupling capacitance only affects the
high-to-low transition. The simulated value of the propagation delay with decoupler equals 41
psec.
The static power consumption of the gate can be derived. For a low input, only the
emitter-follower consumes static power, while both the logic and the output structures are
consuming power for a high input. Adding all factors results in a static consumption of 2.28
mW. The dynamic consumption, when clocked at the maximum possible rate, equals 2.3 mW.
Reduction of the consumption is possible by reducing the voltage swing, the supply voltage,
Without decoupler
Vout (V) 1.0
0.8
0.6
With decoupler Figure A.60 Transient response of an

0.4 NTL gate (without and with an emitter-
0 0.1 0.2 0.3 0.4 0.5 decoupling capacitance of 0.3pF).
t (nsec)
and the circuit performance. For this particular version of the gate, the PDP evaluates to
376 pJ.
The NOR operation is the logic function most easily implemented in NTL, and is
constructed by a parallel connection of the input transistors. More complex functions can
be realized by transistor stacking. This requires extra level shifters, while saturation
should be carefully avoided.
REFERENCES
[Antognetti88] P. Antognetti and G. Masobrio (eds.), Semiconductor Device Modeling with SPICE,
McGraw-Hill, 1988.
[Banzhaf92] W. Bhanzhaf, Computer Aided Analysis Using PSPICE, 2nd ed., Prentice Hall, 1992.
[Chen90] J. Chen, CMOS Devices and Technology for VLSI, Prentice Hall, 1990.
[Getreu76] I. Getreu, “Modeling the Bipolar Transistor,” Tektronix Inc., 1976.
[Gray69] P. Gray and C. Searle, Electronic Principles, John Wiley and Sons, 1969.
[Gray93] P. Gray and R. Meyer, Analysis and Design of Analog Integrated Circuits, 3rd ed., John
Wiley and Sons, 1993.
[Haznedar91] H. Haznedar, Digital Microelectronics, Benjamin/Cummings, 1991.
[Hodges88] D. Hodges and H. Jackson, Analysis and Design of Digital Integrated Circuits, 2nd ed.,
McGraw-Hill, 1988.
[Howe95] R. Howe and S. Sodini, Microelectronics: An Integrated Approach, forthcoming, Prentice
Hall, 1995.
[Hu92] C. Hu, “IC Reliability Simulation,” IEEE Journal of Solid State Circuits, vol. 27, no. 3, pp.
241–246, March 1992.
72 Bipolar Digital
[Hu93] C. Hu, “Future CMOS Scaling and Reliability,” IEEE Proceedings, vol. 81, no. 5, May
1993.
[Jensen91] G. Jensen et al., “Monte Carlo Simulation of Semiconductor Devices,” Computer Phys-
ics Communications, 67, pp. 1–61, August 1991.
[Ko89] P. Ko, “Approaches to Scaling,” in VLSI Electronics: Microstructure Science, vol. 18, chap-
ter 1, pp. 1–37, Academic Press, 1989.
[Muller86] R. Muller and T. Kamins, Device Electronics for Integrated Circuits, 2nd ed., John
Wiley and Sons, 1986.
[Nagel75] L. Nagel, “SPICE2: a Computer Program to Simulate Semiconductor Circuits,” Memo
ERL-M520, Dept. Elect. and Computer Science, University of California at Berkeley, 1975.
[Sedra87] A. Sedra and K. Smith, Microelectronic Circuits, 2nd ed., Holt, Rinehart and Winston,
1987.
[Sheu87] B. Sheu, D. Scharfetter, P. Ko, and M. Jeng, “BSIM: Berkeley Short-Channel IGFET
Model for MOS Transistors,” IEEE Journal of Solid-State Circuits, vol. SC-22, no. 4, pp.
558–565, August 1987.
[Sze81] S. Sze, Physics of Semiconductor Devices, 2nd ed., John Wiley and Sons, 1981.
[Thorpe92] T. Thorpe, Computerized Circuit Analysis with SPICE, John Wiley and Sons, 1992.
[Toh88] K. Toh, P. Koh, and R. Meyer, “An Engineering Model for Short-Channel MOS Devices,”
IEEE Journal of Solid-Sate Circuits, vol. 23. no. 4, pp 950–957, August 1988.
[Tsividis87] Y. Tsividis, Operation and Modeling of the MOS Transistor, McGraw-Hill, 1987.
[Yamaguchi88] T. Yamaguchi et al., “Process and Device Performance of a High-Speed Double
Poly-Si Bipolar Technology Using Borsenic-Poly Process with Coupling-Base Implant,” IEEE
Trans. Electron. Devices, vol. 35, no 8, pp. 1247–1255, August 1988.
[Weste93] N. Weste and K. Eshragian, Principles of CMOS VLSI Design: A Systems Perspective,
Addison-Wesley, 1993.
A.5 Exercises and Design Problems
For all problems, use the device parameters provided in Chapter 2 (Sections 2.2.5, 2.3.5, and A.2.5)
and the inside back book cover, unless otherwise mentioned. Also assume T = 300 K by default.
1. [E,SPICE,2.23]
a. Consider the circuit of Figure A.61. Using the simple model, with VDon = 0.7 V, solve for
ID .
b. Find ID and VD using the ideal diode equation. Use Is = 10–14 A and T = 300 K.
c. Solve for VD1, VD2, and ID using SPICE.
d. Repeat parts b and c using IS = 10–16 A, T = 300K, and IS = 10–14A, T = 350 K.
2. [M, None, 2.2.3] For the circuit in Figure A.62, Vs = 3.3 V. Assume AD = 12 µm2, φ0 = 0.65
V, and m = 0.5. NA = 2.5 E16 and ND = 5 E15.
a. Find ID and VD.
b. Is the diode forward- or reverse-biased?
c. Find the depletion region width, Wj, of the diode.
d. Use the parallel-plate model to find the junction capacitance, Cj.
e. Set Vs = 1.5 V. Again using the parallel-plate model, explain qualitatively why Cj
increases.
R1 = 2 kW
+ ID R2 = 2 kW
5V +
– VD
–
Figure A.61 Resistor diode circuit.
Rs = 2 k¾
+ –
Vs ID VD
– +
Figure A.62 Series diode circuit
3. [C, None, 2.2.3] For the circuit in Figure A.63, sketch iR(t), vd(t), and vo(t) showing quantita-
tive values for the asymptotes and time constants. Assume VD(on) = 0.7 V and the minority car-
rier transit time τT = 50 nsec. Further assume a short-base model with a high hole-injection
efficiency. Neglect the effects of junction capacitance.
C = 2 pF vi 50 nsec
1.7
vo t (nsec)
+ vd – –3.3
vi iR R = 2 k¾
Figure A.63 Series diode circuit
4. [E, SPICE, 2.3.2] Figure A.64 shows NMOS and PMOS devices with drains, source, and gate
ports annotated. Determine the mode of operation (saturation, triode, or cutoff) and drain cur-
rent ID for each of the biasing configurations given below. Verify with SPICE. Use the fol-
lowing transistor data: NMOS: k'n = 60 µA/V2, VT0 = 0.7 V, λ = 0.1 V–1, PMOS: k'p = 20
µA/V2, VT0 = –0.8 V, λ = 0.1 V–1. Assume (W/L) = 1.
a. NMOS: VGS = 3.3 V, VDS = 3.3 V. PMOS: VGS = –0.5 V, VDS = –1.5 V.
b. NMOS: VGS = 3.3 V, VDS = 2.2 V. PMOS: VGS = –3.3 V, VDS = –2.6 V.
c. NMOS: VGS = 0.6 V, VDS = 0.1 V. PMOS: VGS = –3.3 V, VDS = –0.5 V.
D S
G G
ID ID
S D Figure A.64 NMOS and PMOS devices.
5. [E, SPICE, 2.3] Using SPICE plot the I-V characteristics for the following devices.
a. NMOS W = 2.4 µm, L = 0.6 µm
b. NMOS W = 12.8 µm, L = 3.6 µm
c. PMOS W = 2.4 µm, L = 0.6 µm
74 Bipolar Digital
d. PMOS W = 12.8 µm, L = 3.6 µm

6. [E, SPICE, 2.3] Indicate on the plots from problem 5.
a. the regions of operation.
b. the effects of channel length modulation.
c. Which of the devices are in velocity saturation? Explain how this can be observed on the I-
V plots.
7. [M, None, 2.3] Repeat problem 5 using hand analysis using the LEVEL 1 model. Explain the
differences with the results of problem 5 and suggest approaches to reduce the discrepancies.
8. [E, None, 2.3.2] You are given an NMOS device with the following process parameters: NA =
1016 cm–3 (substrate), ND = 1020 cm–3 (gate), NSS = 1011 cm–3 (surface states), and xj = 0.5 µm.
Assume there has been no threshold channel implant.
a. Due to process variations in tox you observe values of VT0 between –0.1 V and 0 V. What
is the range of oxide thicknesses?
b. You desire to make a low-VT process with VT0,min = 0.3 V. What channel implant dose is
required?
9. [M, None, 2.3.2] Given the data in Table A.8 for an NMOS transistor with k′ =20 µA/V2, cal-
culate VT0, γ, λ, 2|φf|, and W / L:
Table A.8 Measured NMOS transistor data
VGS VDS VBS ID (µA)
1 3 5 0 1210
2 5 5 0 4410
3 5 10 0 5292
4 5 5 –2 3265
5 5 5 –5 2381
10. [M, None, 2.3] The data points in Table A.9 were measured for an MOS transistor:
a. What type of device is this: PMOS or NMOS, VT > 0 or < 0?
b. Is this device velocity saturated? Why or why not.
c. Derive the values of k = k′ (W / L), VT0, and λ . Determine the operation region for each of
the rows in the table.
d. Suppose that tox is reduced. How do k, Vto and λ change?
Table A.9 Measured MOS transistor data.
VGS(V) VDS(V) VSB(V) ID (µA)
1 –2 3 0 0
2 0 1 0 47.86
3 0 5 0 62.11
4 1 3.5 0 192.52
5 1 5 0 208.76
Table A.9 Measured MOS transistor data.
VGS(V) VDS(V) VSB(V) ID (µA)
6 5 5 0 1182
11. [M, None, 2.3.2] Consider the circuit configuration of Figure A.65.
a. Write down the equations (and only those) which you need to determine the voltage at
node X. Do NOT plug in any values yet. Assume that λp = 0.
b. Draw the (approximative) load lines for both MOS transistor and resistor. Mark some of
the significant points.
c. Determine the required width of the transistor (for Leff = 1.2 µm) such that X equals 1.5 V.
d. We have, so far, assumed that M1 is a long-channel device. Redraw the load lines assum-
ing that M1 is velocity-saturated. Will the voltage at X rise or fall?
3V
I R1 = 500 k¾
X
M1
Figure A.65 MOS circuit.
12. [M, None, 2.3.2] The circuit of Figure A.66 is known as a source-follower configuration. It
achieves a DC level shift between the input and output. The value of this shift is determined
by the current I0. Assume LD = 0.15 µm, γ = 0.543, 2|φf| = 0.6 V, VT0 = 0.74 V, k′ = 19.6
µA/V2, and λ = 0.
a. Derive an expression giving Vi as a function of Vo and VT(Vo). If we neglect body effect,
what is the nominal value of the level shift performed by this circuit.
b. The NMOS transistor experiences a shift in VT due to the body effect. Find VT as a func-
tion of Vo for Vo ranging from 0 to 3 V with 0.5 V intervals. Plot VT vs. Vo.
c. Plot Vo vs. Vi as Vo varies from 0 to 3 V with 0.5 V intervals. Plot two curves: one neglect-
ing the body effect and one accounting for it. How does the body effect influence the oper-
ation of the level converter? At Vo (body effect) = 3 V, find Vo (ideal) and, thus, determine
the maximum error introduced by body effect.
VDD = 5 V
Vi 5.4/1.2
Vo
I0 = 58.8 mA
Figure A.66 Source-follower level converter.
13. [M, SPICE, 2.3.2] Problem 13 uses the MOS circuit of Figure A.67.
76 Bipolar Digital
a. Plot Vout vs. Vin with Vin varying from 0 to 5 volts (use steps of 1 V). VDD = 5 V.
b. Repeat a using SPICE.
c. Repeat a and b using a MOS transistor with (W / L) = 4.8 / 1.2. Is the discrepancy between
manual and computer analysis larger or smaller. Explain why.
VDD
8 k¾
Vout
Vin M1
20/5
Figure A.67 MOS circuit.
14. [M, None, 2.3.3] Compute the gate and diffusion capacitances for transistor M1 of Figure
A.67. Assume that drain and source areas are rectangular, and are 20 µm wide and 5 µm long.
Use the parameters of Example 3.5 to determine the capacitance values. Assume mj = 0.5 and
mjsw = 0.33. Also compute the total charge stored at node In, for the following conditions:
a. Vin = 5 V, Vout = 5 V, 2.5 V, and 0 V.
b. Vin = 0 V, Vout = 5 V, 2.5 V, and 0 V.
15. [C, SPICE, 2.5] Though impossible to quantify exactly by hand, it is a good idea to understand
process variations and be able to at least get a rough estimate for the limits of their effects.
a. For the circuit of Figure A.67, calculate nominal, minimum, and maximum currents in the
NMOS device with Vin = 0 V, 2.5 V and 5 V. Assume 3σ variations in VT0 of 25 mV, in k′
of 15%, and in lithographic etching of 0.15 µm.
b. Analyze the impact of these current variations on the output voltage. Assume that the load
resistor also can vary by 10%. Verify the result with SPICE.
16. [M, None, 2.3.4] Short-channel effects:
a. Use the fact that current can be expressed as the product of the carrier charge per unit
length and the velocity of carriers (IDS = Qv) to derive IDS as a function of W, Cox, VGS –
VT , and carrier velocity v.
b. For a long-channel device, the carrier velocity is the mobility times the applied electric
field. The electrical field, which has dimensions of V/m, is simply (VGS – VT ) / 2L. Derive
IDS for a long-channel device.
c. From the equation derived in a, find IDS for a short-channel device in terms of the maxi-
mum carrier velocity, vmax.
d. Based on the results of b and c describe the most important differences between short-
channel and long-channel devices.
17. [C, None, 2.3.4] Another equation, which models the velocity-saturated drain current of an
MOS transistor is given by
1  µ 0 C ox
-  -------------- W
Idsat = ------------------------------------------------------- ----- ( V – V T ) 2
1 + ( VGS – V t ) ⁄ ( E sat L )  2  L GS
Using this equation it is possible to see that velocity saturation can be modeled by a MOS
device with a source-degeneration resistor (see Figure A.68).
a. Find the value of RS such that IDSAT(VGS, VDS) for the composite transistor in the figure
matches the above velocity-saturated drain current equation. Hint: the voltage drop across
RS is typically small.
b. Given Esat = 1.5 V/µm and k′ = µ0Cox = 20 µA/V2, what value of RS is required to model
velocity saturation. How does this value depend on W and L?
VD
VG W/L
V′S
RS
Figure A.68 Source-degeneration model of
VS velocity saturation.
18. [C, SPICE, 2.2.3] The circuit in Figure A.69 shows an NMOS device with its parasitic diode.
The NMOS transistor has kn = µn Cox W / L = 6000 µA / V2, VT0 = 0.7 V, and γ = 0, and is off
(Vg = 0 V) until time t1 when the gate voltage steps to 5 V. The parasitic diode has AD = 100
µm2, φ0 = 0.65 V, and m = 0.33. IBIAS = 500 µA.
a. Solve for the initial conditions on ID, VD, and IM1 (at time t1–).
b. Find the minority carrier charge in the diode, QD.
c. Find the time for the NMOS device to remove this excess minority carrier charge from the
diode. During this interval, assume that VD is constant, and model the NMOS device as a
constant current source. Verify your results with SPICE.
d. Find the final values of ID, VD, and IM1 (after the NMOS device has been on for a long time).
e. Find the diode junction capacitance, Cj, linearized over the values of VD found in parts a
and d.
f. Find the time to reach 90% of the final value of VD. During the diode space-charge
removal, model the NMOS device as a current source of value equal to the average current
at the interval endpoints. Compare your results with those of SPICE.
VDD
Vg
5V
IBIAS
0V
t0 t1
–
Vg id vd
iM1 +
Figure A.69 NMOS with parasitic diode.
19. [E, SPICE, 2.3.2] Figure A.70 shows npn and pnp devices biased with bias resistors RB and
RC. Determine the mode of operation (forward-active, saturation, or cut-off) and collector
current IC for the resistor values given below. Verify with SPICE. Use the following transistor
data. npn: βF = 100, VBE(on) = 0.7 V, VBE(sat) = 0.8 V, VCE(sat) = 0.2 V. pnp: βF = 30, VBE(on) = −
0.7 V, VBE(sat) = −0.8 V, VCE(sat) = −0.2 V.
a. RB = 5 kΩ, RC = 1 kΩ.
78 Bipolar Digital
b. RB = 50 kΩ, RC = 0.5 kΩ.

c. RB = 10 kΩ, RC = 250 Ω.
5V 5V
RC IC
IB RB
IB RB
IC
RC
Figure A.70 npn and pnp devices with their
biasing schemes.
20. [M, None, 2.5] Assuming 3σ process variations of 5% in RC, 10% in RB, and 20% in βF,
repeat Problem 19. For each case, find the region of operation for both devices, and the mini-
mum and maximum values for IB, IC, and VCE.
21. [M, None, 2.4.3] In this problem, we consider the delays involved in turning on the bipolar
transistor. Sketch vBE(t), QF(t), and iC(t) for the circuit of Figure A.71. Annotate the critical
delay components on your graphs. Use the transistor data of Example A.2.
VCC = 2 V
1
Vin 0
R = 5k¾
Figure A.71 Circuit for analyzing turn-on
transient of bipolar transistor.
22. [C, None, 2.4.3] Consider adding a capacitor in parallel with the base resistor in the circuit of
Problem 21 (Figure A.72). The capacitor is referred to as a speed-up capacitor, and it allows
the output current to achieve its steady-state value almost instantaneously. How? What value
of C is required to achieve this effect. Repeat problem 21 using this value for C.
VCC = 2 V
C
1
Vin 0
R = 5 k¾
Figure A.72 Bipolar transistor circuit with
speed-up capacitor.
23. [M, None, 2.4.4] The deviation in the actual collector current of a bipolar transistor, ICA, from
the ideal value, I CI = I S e +VBE ⁄ φT , due to parasitic and high-level injection effects can be
modeled as a base resistance, RB.
a. Derive an expression for RB as a function of ICI and ICA.
b.

Bipolar

Uploaded by

Copyright:

Available Formats

Bipolar

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Bipolar

Uploaded by

Copyright:

Available Formats

ADDENDUM

BIPOLAR DIGITAL DESIGN

Qualitative understanding of the bipolar device

A.3.6 Looking Ahead: Scaling the Technology

A.5 Exercises and Design Problems

A.2 The Bipolar Transistor

A.2.1 A First Glance at the Device

(a) Cross-sectional view

(b) Idealized transistor structure

Figure A.1 npn bipolar transistor.

Table A.1 Modes of operation of the bipolar transistor.

Mode Emitter Junction Collector Junction

Cut-off Reverse Reverse

Figure A.2 Bipolar transistor schematic

A.2.2 Static Behavior

Minority carrier Depletion

pe(x) nb(x) pc0

Figure A.4 Components of the base current.

Figure A.5 Minority carrier concentration profile in reverse-active operation mode.

Minority carrier concentration

Figure A.8 Collector current as a function of VCE with IB as a parameter.

Problem A.1 Current Gain of Bipolar Transistor.

Problem A.2 pnp Transistor Characteristics.

Model for Manual Analysis

• Forward-active—VBE ≥ VBE(on), VBC < VBC(on)

Example A.1 Static npn Model for Manual Analysis

(a) Forward-active (b) Forward-active (simplified)

A.2.3 Dynamic Behavior

The Base-Emitter and Base-Collector Depletion Capacitances

The Collector-Substrate Capacitance

The Base Charge—An Introduction

I C = ------F- – ------R- ------ = ------F- – Q R  ----- + --------

It is conceptually more interesting to consider a different subdivision as illustrated

Changing the Base Charge—The Diffusion Capacitance

Bipolar Transistor Capacitive Model

Bipolar Transistor Switching Time—A Case Study

E Figure A.11 Capacitive model of an npn bipolar transistor.

where IC(t = 0) is expressed by the following equation:

and the time to remove the excess charge (QF = 0).

t space = 2.2R B ( C be + C bc ) (A.31)

Example A.2 Dynamic Behavior of an npn Transistor

IB = (–0.75) / (5 × 103) = −150 µA

which is considerably larger than the base-charge removal period.

0.0 0 0.2 0.4 0.6 0.8 1 00 0.2 0.4 0.6 0.8 1

Figure A.13 Simulated transient response of the circuit of Figure A.12.

Problem A.3 Turning on the Bipolar Transistor

A.2.4 The Actual Bipolar Transistor—Secondary Effects

The Early Voltage

Transistor Breakdown Voltages

Figure A.15 npn transistor structure, showing parasitic resistances.

VBE (linear) Figure A.16 Deterioration of the current

A.2.5 SPICE Models for the Bipolar Transistor

E Figure A.17 SPICE model for an npn transistor.

Example A.3 SPICE Bipolar Junction Transistor Model

Table A.2 Main SPICE bipolar junction transistor model parameters.

Parameter Name Symbol SPICE Name Units Default Value

Transport Saturation Current IS IS A 1.0E−16

a. Gummel-Poon Model Parameter

Table A.3 SPICE Parameters for parasitics (resistances, capacitances).

Parameter Name Symbol SPICE Name Units Default Value