THE BICMOS APPROACH - .xyzlibvolume6.xyz/medicalelectronics/btech/semester6/vlsidesign/basic... · THE BICMOS APPROACH ... Design in the BiCMOS technology. ... sumption makes very

1

A D D E N D U M

B

T H E B I C M O S A P P R O A C H

B.1 Introduction

B.2 The BiCMOS Gate at a Glance

B.3 The Static Behavior and Robustness Issues

B.4 Performance of the BiCMOS Inverter

B.5 Power Consumption

B.6 Technology Scaling

B.7 Designing BiCMOS Digital Gates

B.8 Summary

B.9 To Probe Further

B.10 Exercises and Design Problems

n

The BiCMOS gate

n

Design in the BiCMOS technology

2 The BiCMOS Approach

B.1 Introduction

Complementary MOS offers an inverter with near-perfect characteristics such as high,symmetrical noise margins, high input and low output impedance, high gain in the transi-tion region, high packing density, and low power dissipation. Speed is the only restrictingfactor, especially when large capacitors must be driven. In contrast, the ECL gate has ahigh current drive per unit area, high switching speed, and low I/O noise. For similar fan-outs and a comparable technology, the propagation delay is about two to five timessmaller than for the CMOS gate. However, this is achieved at a price. The high power con-sumption makes very large scale integration difficult. A 100k-gate ECL circuit, forinstance, consumes 60 W (for a signal swing of 0.4 V and a power supply of 4 V). Thetypical ECL gate also has inferior dc characteristics compared to the CMOS gate—lowerinput impedance and smaller noise margins.

In recent years, improved technology has made it possible to combine complimen-tary MOS transistors and bipolar devices in a single process at a reasonable cost. A cross-section of a typical BiCMOS process is shown in Figure 1. A single n-epitaxial layer isused to implement both the PMOS transistors and bipolar npn transistors. Its resistivity ischosen so that it can support both devices. An n+-buried layer is deposited below the epi-taxial layer to reduce the collector resistance of the bipolar device, which simultaneouslyincreases the immunity to latchup. The p-buried layer improves the packing density,because the collector-collector spacing of the bipolar devices can be reduced. It comes atthe expense of an increased collector-substrate capacitance.

This technology opens a wealth of new opportunities, because it is now possible tocombine the high-density integration of MOS logic with the current-driving capabilities ofbipolar transistors. A BiCMOS inverter, which achieves just that, is discussed in the fol-lowing section. We first discuss the gate in general and then provide a more detailed dis-cussion of the steady-state and transient characteristics, and the power consumption. Thesection concludes with a discussion of the usage of BiCMOS and the future outlook. Mostof the techniques used in this section are similar to those used for CMOS and ECL gates,so we will keep the analysis short and leave the detailed derivations as an exercise.

Figure B.1 Cross-section of BiCMOS process (from [Haken89]).

J. Rabaey—Digital ICs-1st Ed. 3

B.2 The BiCMOS Gate at a Glance

As was the case for the ECL and CMOS gates, there are numerous versions of the BiC-MOS inverter, each of them with slightly different characteristics. Discussing one is suffi-cient to illustrate the basic concept and properties of the gate. A template BiCMOS gate isshown in Figure 0.1a. When the input is high, the NMOS transistor M1 is on, causing Q1to conduct, while M2 and Q2 are off. The result is a low output voltage ( Figure 0.1b). Alow Vin, on the other hand, causes M2 and Q2 to turn on, while M1 and Q1 are in the off-state, resulting in a high output level ( Figure 0.1c). In steady-state operation, Q1 and Q2are never on simultaneously, keeping the power consumption low. An attentive readermay notice the similarity between this structure and the TTL gate, described in the adden-dum on bipolar design. Both use a bipolar push-pull output stage. In the BiCMOS struc-ture, the input stage and the phase-splitter are implemented in MOS, which results in abetter performance and higher input impedance.

The impedances Z1 and Z2 are necessary to remove the base charge of the bipolartransistors when they are being turned off. For instance, during a high-to-low transition onthe input, M1 turns off first. To turn off Q1, its base charge has to be removed. This hap-pens through Z1.. Adding these resistors not only reduces the transition times, but also hasa positive effect on the power consumption. There exists a short period during the transi-tion when both Q1 and Q2 are on simultaneously, thus creating a temporary current pathbetween VDD and GND. The resulting current spike can be large and has a detrimentaleffect on both the power consumption and the supply noise. Therefore, turning off thedevices as fast as possible is of utmost importance.

The following properties of the voltage-transfer characteristic can be derived byinspection. First of all, the logic swing of the circuit is smaller than the supply voltage.

Vin

VDD

Vout

Q2

Q1

Z2

Z1

M1

M2

Figure B.2 The BiCMOS gate.

Vout

Q1

Z1

Z1

M1

VDD

Vout

Q2

Z2

M2

(a) A generic BiCMOS gate

(b) Equivalent circuit for high-input signal

(c) Equivalent circuit for low-input signal


Consider the high level. With Vin at 0 C, the PMOS transistor M2 is on, setting the base ofQ2 to VDD. Q2 acts as an emitter-follower, so that Vout rises to VDD – VBE(on) maximally.The same is also true for VOL. For Vin high, M1 is on. Q1 is on as long as Vout > VBE(on).Once Vout reaches VBE(on), Q1 turns off. VOL thus equals VBE(on).

1 This reduces the total volt-age swing to VDD – 2VBE(on), which causes not only reduced noise margins, but alsoincreases the power dissipation. Consider for instance the circuit of Figure 0.2, where theBiCMOS gate is shown with a single fan-out for Vin = 0. The output voltage of VDD –VBE(on) fails to turn the PMOS transistor of the subsequent gate completely off, sinceVBE(on) is approximately equal to the PMOS threshold. This leads to a steady-state leakagecurrent and power consumption. Various schemes have been proposed to get around thisproblem, resulting in gates with logic swings equal to the supply voltage at the expense ofincreased complexity. Some of these schemes will be discussed later. Aside from this dif-ference, the VTC of the BiCMOS inverter is remarkably similar to that of CMOS.

The propagation delay of the BiCMOS inverter consists of two components: (1)turning the bipolar transistors on (off), and (2) (dis)charging the load capacitor. From ourdiscussion of the RTL gate (Chapter 3), we learned how important it was to keep the bipo-lar transistors out of the saturation region. Building and removing the base charge of a sat-urated transistor requires a considerable amount of time and results in a slow gate. One ofthe attractive features of the BiCMOS inverter is that the structure prevents both Q1 andQ2 from going into saturation. They are either in forward-active mode or off. For the highoutput level, Q2 remains in the forward-active mode when VOH is reached. The PMOStransistor M2 acts a resistor, ensuring that the collector voltage of M2 is always higher thanits base voltage ( Figure 0.1c). Similarly, at the low-output end, M1 acts as a resistorbetween the base and the collector of Q1, preventing the device from ever saturating ( Fig-ure 0.1b). The base charge is, therefore, kept to a minimum, and the devices are turned onand off quickly.

Consequently, it is reasonable to assume that for typical capacitive loads, the delayis dominated by the capacitor (dis)charge times. To analyze the transient behavior of the

1 Given enough time, the output voltage will eventually reach the ground rail. Once Q1 is turned off, aresistive path to ground still exists through M1-Z1. Due to the high resistance of this path, this takes a substantialamount of time. It is therefore reasonable to assume that VOL = VBE(on).

VDD

Z2

Z1

M1

M2

VDD

VDD – VBE(on)

Q2

M2

VBE(on)

Ileakage

Figure B.3 Increased power consumption due to reduced voltage swing.


inverter, assume that the load capacitance CL is the dominating capacitance. Consider firstthe low-to-high transition. In this case, the equivalent circuit of Figure 1a is valid. Q1 isswitched off fast, as its base charge is removed through Z1. The load capacitor CL ischarged by the current multiplier M2-Q2. The source current of M2 is fed into the base ofQ2 and multiplied with the βF of Q2 (assuming that Q2 operates in the forward-activeregion). This produces a large charging current of (βF + 1) (VDD − VBE(on) − Vout) / Ron(with Ron the equivalent on-resistance of the PMOS transistor). During the high-to-lowtransition, the equivalent circuit of Figure 1b is valid. Q2 is turned off through Z2. Onceagain, the combination M1-Q1 acts as a βF current multiplier. Assuming that the resistanceof M2 in the forward-active mode equals Ron, the discharge current equals (βF + 1) (Vout −Vbe(on)) / Ron (assuming that Ron << Z1). The current multiplication factor makes the BiC-MOS gate more effective than the CMOS inverter for large capacitive loads.

In summary, the BiCMOS inverter exhibits most of the properties of the staticCMOS inverter. In addition, it displays excellent capacitance-driving capabilities as aresult of the push-pull bipolar output stage. The price is a slightly more complex gate anda more complex and expensive fabrication process.

B.3 The Static Behavior and Robustness Issues

The use of resistive elements makes the BiCMOS gate of Figure 0.1 unattractive for realdesigns. A number of slightly modified, more popular circuits are shown in Figure 1. Inthe first circuit (a), the impedances Z1 and Z2 are replaced by active impedances (or tran-sistors) that are only turned on when needed. It still has the unfortunate property that adiode voltage drop is lost at the high end of the output range and to a lesser degree also atthe low end. Circuit (b) has similar properties. The main difference between the twotopologies resides in the transient behavior. Circuit (c) remedies the voltage-dropproblem.

VDD

Vout

Q2

Q1

Z1

M2

VDD

Vout

Q2

Q1

Z2

M1

VDD

CLCL

(b) tpHL(a) tpLH

Figure B.4 Transient behavior of BiCMOS inverter.


Deriving the other parameters of the VTC of the BiCMOS inverter manually is trulycomplex due to the large number of devices and their interplay. We restrict ourselves toSPICE simulations.

Example B.1 VTC of a BiCMOS Inverter

The voltage-transfer characteristic of the inverter of Figure 1b is simulated using SPICE. Acontrived BiCMOS process is employed that merges the MOS devices and bipolar transistorsdescribed by the models of Chapter 2. The NMOS and bipolar transistors are minimum size,while the PMOS transistors are made twice as wide as the NMOS devices. The supply voltageVDD is set at 5 V.

The resulting VTC is shown in Figure 0.3. The complex shape of the curve is causedby the complex interactions among the large number of active devices present in the circuit.To clarify the behavior, we have also plotted the dc transfer characteristics for the base volt-ages of transistors Q1 and Q2. In the transient region between 2 V and 3.5 V, none of the bipo-lar transistors are really on. Also, the PMOS device M1 only turns on after M3 turns off andwhen Vbase2 is sufficiently below Vout. This causes Q1 to turn on and creates an additional dropin the output voltage around Vin ≈ 3.5 V. Notice, furthermore, that VOH is higher thanexpected. This results from the fact that Q2 still carries some emitter current when the base-emitter voltage is smaller than VBE(on). The following dc parameters can be extracted:

VOH = 4.64 V; VOL = 0.05 V

VIL = 1.89 V; VIH = 3.6 V

VM = 2.34 V

NML = 1.84 V; NMH = 1.04 V

Figure B.5 Alternative topologies for BICMOS inverters.

Vin

Vout

Q2

Q1

M1

M2

(a)

M3

M4

VDD

Vin

Vout

Q2

Q1

M1

M2

(b)

M3

M4

VDD

Vin

Vout

Q2

Q1

M1

M2

(c)

M3

VDD

R1


Although the noise margins are not as good as for the CMOS inverter, they are stillwithin the acceptable range. Actually, the projected value of VIH is open for discussion. Wecould also pick the first break-point in the VTC (Vin ≈ 2.5 V), which yields even better noisemargins.

An example of a BiCMOS inverter that does not suffer from a reduced voltageswing is shown in Figure 1c. The resistor R1 (in combination with M2) provides a resistivepath between VDD and Vout and slowly pulls the output to VDD once Q2 is turned off, asdemonstrated in Figure 2. Full-rail BiCMOS circuits are the subject of active research.

B.4 Performance of the BiCMOS Inverter

The BiCMOS inverter exhibits a substantial speed advantage over CMOS gates whendriving large capacitive loads. This results from the current-multiplying effect of the bipo-lar output transistors. As in the ECL case, deriving accurate expressions for the propaga-tion delay is nontrivial. The gate consists of a large number of active devices (up to six)and contains a number of internal nodes, each of which could have a dominant effect on

Figure B.6 Simulated voltage-transfer characteristic for BiCMOS inverter of Figure 1b.0.0 1.0 2.0 3.0 4.0 5.0

Vin (V)

0.0

1.0

2.0

3.0

4.0

5.0V o

ut (V

)Vout

Vbase2

Vbase1

Vout

Q2

M2

VDD

R1

Q1 on

Q1 off

t

Vout

Figure B.7 Low-to-high transition in full-swing BiCMOS circuit.

VDD – VBE(on)

(b) Transient response(a) Equivalent circuit


the transient response. Although detailed studies have been presented in the literature(e.g., [Rosseel88]), we restrict ourselves to a simplified analysis. This establishes a first-order model for the delay. SPICE simulations can then be used to establish a more quanti-tative result.

Consider first the low-to-high transition in the circuit of Figure 0.1a. Assume thatthe input signal is switching very fast and that its rise/fall times can be ignored. After turn-ing off M1, the impedance Z1 allows the base charge of Q1 to drain to ground. Since thetransistor was operating in forward-active mode, the stored charge is small, and Q1 turnsoff fast. To a first order, we can therefore assume that this has no impact on the propaga-tion delay and that Q1 is turned off instantaneously. Under those conditions, the transientbehavior can be modeled by the equivalent circuit of Figure 1a.

The propagation delay consists of two components. First, the capacitor Cint has to becharged to VBE(on) through M2 to turn on Q2. Once this point is reached, Q2 acts as an emit-ter-follower, and CL gets charged. Approximative expressions can be derived for bothtime intervals:

(B.1)

with Icharge1 the average charging current.

(B.2)

As Z2 is normally a large resistor, the latter component of the charging current can beignored. The PMOS device operates in saturation mode in this time interval, providingample current; therefore, tturn-on is small.

To compute the second component of the propagation delay, where Q2 acts as anemitter-follower, we can use the reflection rule (similar to the analysis of the ECL gate)to merge the internal and external circuit nodes into a single node. CL now appears in par-

Figure B.8 Equivalent circuits for transient analysis.

VDD

Vout

Q2

M2

CL

Z2 Cint

Vout

Q1

M1

VDD

CL

(b) tpHL

Z1 Cint

(a) tpLH

tturn-onCintVBE on( )

Icharge1---------------------------=

Icharge1IM2 Vint = 0( ) IM2 Vint = VBE on( )( ) VBE on( ) Z2⁄–+

2-----------------------------------------------------------------------------------------------------------------------=


allel with Cint, but its value is divided by (βF + 1). This is equivalent to stating that thebase current of Q2 is multiplied by that factor. The corresponding delay is now readilycomputed:

(B.3)

Icharge2 equals the average charging current during that interval. This consists primarily ofthe current through M2 (ignoring the current loss through Z2). The value of Vswing is deter-mined by gate topology, but normally equals VDD – 2 VBE(on), as was apparent from the dcanalysis. The value of Icharge2 is comparable to the average PMOS charging current, asobserved in a CMOS inverter with similar-size devices.

The overall value of the low-to-high propagation delay is obtained by combiningEq. (0.2) and Eq. (0.3).

(B.4)

This delay consists of two components:

1. A fixed component that is proportional to Cint and is normally small. Cint is a lumpedcapacitance, composed of contributions of the PMOS device (diffusion capacitance)and the bipolar transistor (be- and bc-junction capacitance and base-charge capaci-tance).

2. The second component is proportional to the load capacitance CL. The loading effectis substantially reduced by the (βF + 1) current multiplier introduced by the bipolartransistor.

It is interesting to compare this result with the delay of a CMOS inverter, assumingsimilar-size MOS transistors. The following linear approximation of the delay of theCMOS inverter is valid.

(B.5)

In comparing Eqs. (0.4) and (0.5), we realize that the values of the coefficients areapproximately equal (a ≈ c and b ≈ d), as determined by the current through the PMOSand the voltage swing, which are of the same order in both designs. Cint is substantiallylarger in the BiCMOS case due to the contributions of the bipolar device. These observa-tions allow us to draw an approximate plot of tpLH versus the load capacitance CL for boththe CMOS and BiCMOS gates ( Figure 0.4).

For very low values of CL, the CMOS gate is faster than its BiCMOS counterpartdue to the smaller value of Cint. For larger values of CL, the bipolar output transistors eas-ily provide the extra drive current, and the BiCMOS gate becomes superior. Although the

Icharge

CintCL

βF 1+---------------+

Vswing

2--------------

Icharge2----------------------------------------------------=

tpLH tturn-on tcharge+CintVBE on( )

Icharge1---------------------------

CintCL

βF 1+---------------+

Vswing

2--------------

Icharge2----------------------------------------------------+= =

a= Cint×b C× L

βF 1+---------------+

tpLH CMOS( ) c Cint× d CL×+=


cross-over point Cx is technology-dependent, it typically ranges from CL ≈ 50 to 250 fF.As a result, BiCMOS inverters are normally used as buffers to drive large capacitances.They are not very effective for the implementation of the internal gates of a logic structure(such as an ALU), where the associated load capacitances are small. One must alsoremember that the complexity of the BiCMOS gate incurs an important area overhead.Consider carefully when and where to use BiCMOS structures.

A similar analysis holds for the high-to-low transition. It is assumed that Q2 turnsoff instantaneously, as its base charge is quickly removed through Z2. The resulting equiv-alent circuit is shown in Figure 1b. Once again, the delay consists of two contributions:

1. Turning on Q1. This requires the charging of the internal capacitance Cint throughthe NMOS device.

2. Discharging CL through the combined network of NMOS and bipolar transistor.Ignoring the current loss through Z1, all the drain current of M1 sinks into the base ofQ1. Assuming forward-active operation, this results in a collector current βF timeslarger. The total discharge current equals (βF + 1) INMOS.

Hence, the following approximative expression is valid

(B.6)

Eq. (0.6) closely resembles the one derived for the tpLH. It is worth mentioning thatCint is not constant and changes between turn-on and discharge modes. An average valueover the complete operation range produces acceptable first-order results.

Example B.2 Propagation Delay of a BiCMOS Inverter

The propagation delay of the BiCMOS buffer of Example 0.1 is simulated using SPICE for aload of 1 pF. The result is plotted in Figure 2 and compared with the performance of a CMOSinverter (for a similar load). The propagation delay of 0.86 nsec for the BiCMOS gate com-pares favorably to the 6.0 nsec of the CMOS inverter.

Notice the reduced voltage swing of the BiCMOS gate. The loss at both the high andlow levels is, however, substantially smaller than the 0.7 V (VBE(on)) suggested by the first-

CMOS

BiCMOS

tpLH

CLCx

~ 1/IPMOS

~ 1/(βFIPMOS)

Figure B.9 Propagation delay of BiCMOS and CMOS gates as a function of CL.

tpHL tturn-on tdischarge+CintVBE on( )

Icharge3---------------------------

CintCL

βF 1+---------------+

Vswing

2--------------

Icharge4----------------------------------------------------+= =


order model and is approximately equal to 0.4 V. For very low capacitive loads, the CMOSgate is approximately 5.5 times faster than its BiCMOS counterpart. This is illustrated in Fig-ure 0.5, where the propagation delays of the CMOS and BiCMOS gates are plotted as a func-tion of CL. The cross-over point, where BiCMOS becomes faster than CMOS, is situatedaround 100 fF. Notice that for CL values below 1 pF the propagation delay of the BiCMOSgate is virtually independent of the load capacitance. For those load values, the capacitance ofthe internal nodes (Cint) dominates the performance, and the factor attributable to CL is negli-gible. The measured slope of the CMOS curve is approximately 64 times steeper, which issomewhat lower than the expected value of βF + 1 (or 101). The discrepancy is due to a num-ber of inefficiencies in the BiCMOS gates, such as the VBE losses.

The analysis derived above is correct as long as the current flowing through thebipolar transistors is limited. Large currents might adversely affect the speed of the gatedue to the second-order effects listed below.

0 10 20 30t (nsec)

0.0

2.0

4.0

6.0

V (V

olt)

Vout (BiCMOS)

Vout (CMOS)

Vin

Figure B.10 Transient response of BiCMOS and CMOS inverters for a load of 1 pF.

0.0 2.0 4.0 6.0 8.0 10.0CL (pF)

0.0

2.0

4.0

6.0

t p (n

sec)

BiCMOS

CMOS

Figure B.11 Simulated propagation delays of CMOS and BiCMOS gates as a function of CL.


• Collector Resistance Rc—The equivalent circuits of Figure 1 ignore the presenceof the collector resistance rc between the extrinsic collector contact and the intrinsiccollector-base junction. The voltage drop over rc causes the transistor to saturateeven though the extrinsic VCE is larger than 0.7 V, as is guaranteed by the BICMOSbuffer design. For instance, a collector resistance of 100 ohms conducting a tran-sient current of 1mA causes a voltage drop of 0.1 V. When driving large capacitiveloads, currents in excess of 5 mA are regularly observed. The transistor accordinglysaturates, causing a deterioration of the propagation delay; tp is then composed ofthe time to get the transistor into saturation, followed by a discharging of the loadcapacitance with a time constant rCCL (Eq. (0.7)). This problem can be avoided byincreasing the size of the transistor, decreasing rC.

(B.7)

• High-Level Injection—This effect occurs when the density of electrons transportedacross the collector-base space is comparable to the doping of the collector. Theresulting base push-out effectively increases the width of the base and degrades theswitching performance of the transistor. A typical parameter to quantify the onset ofhigh-level injection is the knee current Ik , which is (in practice) the value of the col-lector current at which the forward current gain βF is reduced to 50% of its value.Similar to the degradation caused by the collector resistance, high-level injectioneffects can be avoided by increasing the emitter area of the transistor.

B.5 Power Consumption

The BiCMOS gate performs in the same manner as the CMOS inverter in terms of powerconsumption. Both gates display almost no static power consumption, while the dynamicdissipation is dominated by the (dis)charging of the capacitors. When driving small loads,the latter factor is slightly larger for the BiCMOS gate, due to the increased complexity ofthe gate. On the other hand, when driving very large capacitors, BiCMOS becomes favor-able. To achieve comparable performance, CMOS drivers consist of a cascade of gradu-ally increasing inverters (discussed in Chapter 8). The power dissipated in charging theinternal capacitances becomes an important fraction of the overall consumption. This cas-cading is, generally, not needed in BiCMOS.

The short-circuit currents during switching might be smaller or larger for BiCMOS,depending upon the level of circuit optimization. The superior current-driving capabilitiesof the bipolar transistors produce steeper signal slopes and, consequently, a faster transi-tion through the transition region. This potential advantage is, however, easily annihilatedby intrinsic RC delays in the gate. A small differential delay might cause the bipolar tran-sistors to be on simultaneously for a longer time, causing a large direct current to flow(remember the high transconductance of the bipolar transistors). All in all, only precisesimulations that include parasitic capacitances and resistances can tell exactly which gateis more power efficient.

tp HL LH,( ) tturn-on tsat αrCCL+ +=


B.6 Technology Scaling

Because the BiCMOS technology is a merger of CMOS and bipolar technologies, the scal-ing behavior of the BiCMOS gate is determined by the combined properties of both tech-nologies. Scaling down the dimensions generally results in improved performance.

Unfortunately, the BiCMOS gate inherits one of the most important deficiencies ofthe bipolar technology: built-in voltages such as VBE(on) do NOT scale. The performanceof the BiCMOS gate suffers in an important way when the supply voltage is reduced. Con-sider the equivalent circuit of Figure 1b. The current through M1 during the discharging ofCL is proportional to (VGS – VT) = (Vin – VBE(on) – VT). Vin itself suffers from a VBE(on) lossfor most BiCMOS gate topologies. Therefore, INMOS ~ (VDD − 2VBE(on) − VT) ≈(VDD – 2.2 V). This leaves ample current drive for VDD = 5 V. For lower supply voltages, asubstantial degradation in performance can be observed that eventually causes the CMOSgate to be faster even for large capacitive loads. This is illustrated in Figure 1, where thepropagation delays of CMOS and BiCMOS gates are plotted as a function of VDD (for twodifferent technologies; see [Raje91]). We can see that using BiCMOS does not makemuch sense for supply voltages below 3 V.

This deficiency ultimately hampers the future usefulness of BiCMOS, because voltagescaling is an absolute necessity for submicron devices. The conception of a low-voltageBiCMOS structure is currently a hot research topic [Embabi93].

B.7 Designing BiCMOS Digital Gates

The analysis of a number of industrial BiCMOS designs rapidly reveals that the BiCMOSgate is almost uniquely used for buffering or driving purposes. When driving large fan-

Figure B.12 Propagation delays of CMOS and BiCMOS gates as a function of VDD for 2.0 and 0.5 µm technologies(from [Raje91]). The BiCMOS gates analyzed are the structures of Figure 1a (called BiCMOS) and 1b (calledMBiCMOS).

8

6

4

2

02 3 4 5

BiCMOSMBiCMOSCMOS

Mea

sure

d D

elay

(ns)

Supply (V)

3 pF

1 pF

(a) 2µ technology

2

1.5

1

0.5

02 3 4 5

BiCMOSMBiCMOSCMOS

Sim

ulat

ed D

elay

(ns)

Supply (V)

3 pF

1 pF

(b) 0.5µ technology. Input cap = 33 fF, all gates.

Load Load


out, high-capacitive busses, and off-chip signals, the bipolar output stage helps to providelarge currents, using only a small area and consuming less power compared to the CMOSbuffer. Therefore, the BiCMOS design approach has its major impact in circuits such asmemories and gate arrays where large capacitive loads are common. The topic of drivinglarge capacitances is discussed in detail in Chapter 8, while memories and gate arrays aretreated in Chapters 10 and 11.

The limited usage of the bipolar device seems to be a waste of a valuable resource.Once the step is made to the more expensive BiCMOS technology, it is worth exploitingits capabilities to a maximal degree. This requires some rethinking of traditional designapproaches, which might explain the reluctance of the designer to freely mix MOS andbipolar transistors in a design. The proliferation of BiCMOS circuit into the design offunctions such as ALUs is also hampered by the reduced packing density. MOS devices ofthe same type can be placed in the same well, which means that the distances between thedevices are short. On the other hand, bipolar transistors must be placed in separate n-wells,which significantly reduces the device density. This constraint can be somewhat relaxedby merging npn transistors and PMOS devices in the same well.

Extending the basic BiCMOS inverter gate to more complex logic operations isstraightforward. The logic function only affects the CMOS part of the gate, while thebipolar output circuitry remains unchanged. An example of a two-input NAND gate isshown in Figure 0.6. Both pull-up and pull-down networks are implemented using the tra-ditional CMOS approach. Extension to other gates is trivial.

The most important issue is to determine when it is beneficial to employ such a gatein a combinational circuit. As established above, the BiCMOS approach is advantageousfor a large load capacitance (or, equivalently, fan-out). An additional advantage is that thebipolar output stage isolates the internal, logic nodes of the gate from the load. This allowsfor an optimization of the MOS logic transistors based on the gate topology only (forinstance, using a progressive sizing), regardless of the load, because the latter is taken careof by the bipolar output buffer. This addresses a major disadvantage of the complexCMOS gate, as pointed out in Chapter 4. For a BiCMOS structure to be the gate of choicerequires either a large fan-out, or a complex gate. For instance, it has been shown[Rosseel88] that the BiCMOS two-input NAND gates becomes superior over its CMOS

Vout

Q2

Q1

VDD

A

B

A B

Figure B.13 Two-input BiCMOS NAND gate.


equivalent for a fan-out of four gates. This crosspoint comes even earlier for more com-plex gates.

Some caution is advisable. Due to the lower packing density of the bipolar transistorand the more complex nature of the gate, an area penalty is involved. Only when speed isan issue (for instance on the critical path of a design) and the load capacitance is substan-tial should a BiCMOS gate be considered.

More innovative structures are possible as well. One possible application of a mixedbipolar-MOS design has been shown in the addendum on advanced bipolar design, wherea MOS transistor was used as an active pull-down in an ECL gate. Another possible optionis to use bipolar amplifiers to speed up the performance of traditional CMOS circuits. Forinstance, a faster circuit can be conceived by reducing the signal swing in the critical pathand propagating that reduced swing to the next stage through a bipolar amplifier[Rosseel89]. The larger transconductance of the bipolar device makes it more sensitive tosmaller signal swings. Getting a similar gain requires substantially larger MOS devices.

Example B.3 BiCMOS Inverter Layout

Figure 1 shows the layout of a BiCMOS inverter/driver using the circuit topology of Figure1a (from [Embabi93]). Observe how the PMOS transistor M2 and the bipolar device Q2 aremerged in the same n-well. This saves the isolation area that would be needed if two separatewells were used. Still, the bipolar output stage represents a substantial area overhead. Frominspection of the layout, the device sizes can be derived: M1 = (32/2), M2 = (56/2), M3 = M4 =(6/2) (all sizes in λ). The emitter areas for Q1 and Q2 equal 3 × 10 × 2 λ2. Notice how M1 andM2 are implemented as two parallel transistors, while the emitters of the bipolar devices areimplemented as three separate strips. This helps to adapt the aspect ration of the cell.

It is left as an exercise for the reader to simulate the steady-state and transient perfor-mance of the cell.

B.8 Summary

The following concepts were introduced in this chapter:

• The BiCMOS gate combines the density of CMOS with the current-drive capabili-ties of bipolar. This results in a fast gate structure that outperforms CMOS, espe-cially when driving large capacitances or when the capacitive load is unpredictable.Examples of such structures are memories and gate arrays, where BiCMOS hasresulted in important speed-ups. One can argue that pure CMOS designs can achievesimilar driving capabilities in approximately the same area. Accomplishing thisrequires a careful optimization (using a chain of inverters with gradually increasingtransistor sizes,) and typically consumes more power.

• The performance gain is obtained by using a push-pull bipolar output stage, whichdelivers a βF current gain with respect to similar CMOS structure. However, thisrequires a more expensive technology and a more complex gate structure.

• One of the main challenges facing BiCMOS design is to maintain that performancegain at lower voltage levels.


B.9 To Probe Further

A number of specialized textbooks have been published on BiCMOS design, a number ofwhich are listed below. Excellent overviews of the state-of-the-art techniques can be foundin [Elmasry94], [Embabi93], and [Long90]. Once again, the IEEE Journal of Solid-StateCircuits and the ISSCC conference proceedings are the common source to consult regard-ing the latest developments in this technology.

REFERENCES

[Abe89] M. Abe et al., “Ultrahigh-Speed HEMT LSI Circuits”, in Submicron Integrated Circuits,ed. R. Watts, Wiley, pp. 176-203, 1989.

[Alvarez89] A. Alvarez, BiCMOS Technology and Applications, Kluwer Academic Publishers, Bos-ton, 1989.

[Asbec84] P. Asbec et al., “Application of Heterojunction Bipolar Transitsors to High-Speed, SmallScale Digital Integrated Circuits,” IEEE GaAs IC Symposium, pp. 133–136, 1984.

[Bakoglu90] H. Bakoglu, Circuits, Interconnections and Packaging for VLSI, Addison-Wesley,1990.

Figure B.14 Layout of BiCMOS inverter of Figure 1a (from[Embabi93]).

VDD

GND

n-well

M4

M1

M2

M4

Input

Output

Q1

Q2


[Buchanan90] J. Buchanan, CMOS/TTL Digital Systems Design, McGraw-Hill, 1990.[Chen92] C. Chen, “2.5 V Bipolar/CMOS Circuits for 0.25 µm BICMOS Technology,” IEEE Jour-

nal of Solid-State Circuits, vol. 27, no. 4, April 1992.[Chuang92] C.T. Chuang, “Advanced Bipolar Circuits,” IEEE Circuits and Systems Magazine, pp.

32–36, November 1992.[Curtice80] W. Curtice, “A MESFET Model for Use in the Design of GaAs Integrated Circuits,”

IEEE Trans. Microwave Theory and Tech., vol. MTT-28, pp. 448–456, 1980.[Dingle78] R. Dingle et al., “Electron Mobilities in Modulation-Doped Semiconductor Hetero-junc-

tion Superlattices,” Appl. Phys. Letters, vol. 33, no. 7, pp. 665–667, October 1987.[Elmasry94] M. Elmasry, ed., BiCMOS Integrated Circuit Design, IEEE Press, 1994. [Embabi93] S. Embabi, A. Bellaouar, and M. Elmasry, Digital BiCMOS Integrated Circuit Design,

Kluwer Academic Publishers, Boston, 1993.[Ghoshal93a] U. Ghoshal, L. Huynh, T., Van Duzer, and S. Kam, “Low-Voltage, Nonhysteretic

Operation of CMOS Transistors at 4K,” IEEE Electron Device Letters, 1993.[Ghoshal93b] U. Ghoshal, D. Hebert, and T. Van Duzer, “Josephson-CMOS Memories,” 1993

ISSCC Conference, vol. 36, pp. 54–55, 1993.[Greub91] H. Greub et al., “High Performance Standard Cell Library and Modeling Technique for

Differential Advanced Bipolar Current Tree Logic,” IEEE Journal of Solid-State Circuits, vol.26, no. 5, pp 749–762, May 1991.

[Haken89] R. Haken et al., “BiCMOS Process Technology,” in [Alvarez89], pp. 63–124, 1989.[Hasuo89] S. Hasuo and T. Imamura, “Digital Logic Circuits,” Proc. of the IEEE, vol. 77, no. 8, pp.

1177–1193.[Hodges88] D. Hodges and H. Jackson, Analysis and Design of Digital Integrated Circuits,

McGraw-Hill, 1988.[Ichino87] H. Ichino, “A 50-psec 7K-gate Masterslice Using Mixed Cells Consisting of an NTL

Gate and LCML Macrocell,” IEEE Journal of Solid-State Circuits, SC-22, pp. 202–207, 1987.[Josephson62] B. Josephson, “Possible New Effects in Superconductive Tunneling,” Phys. Letters,

vol. 1, pp. 251, 1962.[Jouppi93] N. Jouppi et al., “A 300 MHz 115 W 32b Bipolar ECL Microprocessor,” Dig. Technical

Papers ISSCC Conf., pp 84–85, 1993.[Kanopoulos89] N. Kanopoulos, Gallium Arsenide Integrated Circuits: A Systems Perspective,

Prentice Hall, 1989.[Kotani90] S. Kotani et al., “A 1 GOPS 8b Josephson Signal Processor,” Dig. Tech. Papers ISSCC

Conf., pp. 148–149, February 1990.[Likharev91] K. Likharev and V. Semenov, “RSFQ Logic/Memory Family: A New Josephson-Junc-

tion Technology for Sub-Terahertz-Clock-Frequency Digital Systems,” IEEE Trans. AppliedSupercond., vol. 1, pp. 3–28, March 1991.

[Long90] S. Long and S. Butner, Gallium Arsenide Digital Integrated Circuit Design, McGraw-Hill, 1990.

[Masaki92] A. Masaki, “Deep-Submicron CMOS Warms Up to High-Speed Logic,” IEEE Circuitsand Devices Magazine, pp. 18–24, November 1992.

[Mehra94] R. Mehra, “Digital Filter Design with High Performance Superconducting Technology,”Masters thesis, University of California, Berkeley, 1994.

[Milutinovic90] V. Milutinovic, ed., Microprocessor Design for GaAs Technology, Prentice Hall,1990.


[Raje91] P. Raje et al., “MBiCMOS: A Device and Curcuit Technique Scalable to the Sub-micron,Sub-2V Regime,” Digest of Technical Papers ISSCC Conf., vol. 34, pp. 150–151, February1991.

[Rocchi90] M. Rocchi, High Speed Digital IC Technologies, Artech House, 1990.[Rosseel88] G. Rosseel et al., “Delay Analysis for BiCMOS Drivers,” BCTM, pp. 220–222, 1988.[Rosseel89] G. Rosseel et al., “A single-ended BiCMOS Sense Circuit for Digital Circuits,” Pro-

ceedings ISSCC Conference, pp. 114–115, February 1989.[Singh86] H. Singh et al., “A Comparative Study of GaAs Logic Families Using Universal Shift

Resistors and Self-Aligned Gate Technology,” IEEE GaAs IC Symposium, pp. 11–15, 1986.[Sze69] S. Sze, Physics of Semiconductor Devices, Wiley Interscience, 1969.[Tektronix93] GST-1 Standard Cell IC User Documentation, Tektronix, Portland, 1993.[Toh89] K. Toh et al., “A 23 psec/2.1 mW ECL Gate with an ac-coupled Active Pull-down Emitter-

follower Stage,” IEEE Journal of Solid State Circuits, SC-24, no. 5, pp 1301–1305, 1989.[Van Duzer89] T. Van Duzer, “Superconductor Digital IC’s,” in VLSI Handbook, Ed. J. De Gia-

como, McGraw-Hill, pp. 16.1–16.21, 1989.

B.10 Exercises and Design Problems

For all problems, use the device parameters provided in Chapter 2 (as well as the book cover), unlessotherwise mentioned.

1. [M&D, SPICE,] Determine the transistor sizes for the BiCMOS driver of Figure 1a so that atpHL = tpLH = 5 nsec is achieved for load capacitance CL = 40 pF, while minimizing area. VDD =5 V. Refine your design with SPICE. Explain any deviations.

2. [M, None] Consider the circuit configuration of Figure 0.7.a. Determine the values of VOL and VOH.b. Explain why Q1 normally operates in forward-active mode during the low-to-high

transition.c. In reality, the bipolar transistor can saturate due to the parasitic collector rc. Explain why. d. For a supply voltage of 3.3 V, determine the maximum value of rc so that Q1 never satu-

rates during the low-to-high transition.

3. [E, None] Consider again Figure 0.7.a. Derive an expression for tpLH as a function of VDD. Assume that λ = 0.

Figure B.15 Bipolar-MOS circuit.

VDD

In

CL = 1 pF

Out

1.81.2

1.81.2

Q1


b. Compare the tpLH of this gate to an equivalent CMOS inverter, where the PMOS-bipolarcombination is replaced by a single PMOS device with (W/L) = (1.8/1.2). Determinewhich one is faster (for VDD >> VT) and derive an approximate expression for the speedratio. Perform only a qualitative analysis.

4. [C, SPICE] A BiCMOS gate is given in Figure 0.8.a. What is the function of the gate?b. Hand calculate VOH, VOL, VIL, and NML. VIL is defined as the point where M3 and Q2 turn

on. Draw the VTC of the circuit. Assume a sharp transition at VIL on the VTC curve.c. Use SPICE to plot the VTC. Determine VOH, VOL, VIH, VIL, VM, NMH, and NML from the

plot.d. Compare the results of parts (b) and (c). Do hand-calculated VOH and VOL agree with

SPICE? If not, give two reasons for the difference.

5. [E, SPICE] For the circuit of Figure 0.8, plot tp as a function of output loading for values ofCL between 0 and 10 pF. (Use SPICE to find tpHL and tpLH for several data points.) Computethe slope of the curve. If SPICE experiences convergence errors, try: .option method=gearmaxord=3.

M1 (4.8/1.2)

M2 (2.4/1.2)

M3 (2.4/1.2)

M4 (2.4/1.2)

Q1

Q2

Vout

Vin

VDD = 3 V

Figure B.16 BiCMOS gate.

THE BICMOS APPROACH - .xyzlibvolume6.xyz/medicalelectronics/btech/semester6/vlsidesign/basic... · THE BICMOS APPROACH ... Design in the BiCMOS technology. ... sumption makes very

Documents