You are on page 1of 5

A C ircu it for All Seasons

Behzad Razavi

The Bridged T-Coil

T
The bridged T-coil, often simply called
the T-coil, is a circuit topology that ex-
tends the bandwidth by a greater fac-
tor than does inductive peaking. Many
high-speed amplifiers, line drivers, L
–M

L
L+M L+M
1
2
mLK
1
2
mLK

and input/output (I/O) interfaces in –M 1 – m2


L
4m K
today’s wireline systems incorporate
on-chip T-coils to deal with parasitic
Cg Cg mCk
capacitances. In this article, we intro-
duce and analyze the basic structure
and study its applications.

Figure 1:  A bridged T-coil circuit described by Ginzton et al. in 1948 [1].
Brief History
The T-coil circuit can be traced back
to the 1948 classic paper on distrib- GaAs realizations appeared in the late consists of two mutually coupled
uted amplifiers by Ginzton et al. [1]. 1980s and early 1990s [4], [5]. With inductors and a bridge capacitor
The authors call the structure the the RF circuits revolution in the 1990s (Figure 2). The coupling polarity mat-
“bridged-tee connection” and pre- and the tremendous work on inte- ters and the two inductances are com-
sent it along with its equivalent cir- grated inductors, the T-coil was bound monly chosen to be equal. With cer-
cuits, as shown in Figure 1. to find its way to CMOS chips as well. tain loads attached to this circuit, the
The use of T-coils for bandwidth Of course, the finite Q and parasitic impedance seen at node 1 or 2 and
enhancement was pioneered by Tek- capacitances of on-chip structures the transfer function from either of
tronix engineers in the late 1960s [2]. would introduce new issues. More- these nodes to node 3 present inter-
The need for fast “vertical” amplifiers over, a well-defined coupling factor esting properties.
for the front end of oscilloscopes had would need to be created between two As an example, consider the sim-
led to many new wide-band circuit spiral inductors. In 2003, two papers ple common-source stage shown in
techniques, and Tektronix design- described the design of integrated Figure 3(a) with a load capacitance C L .
ers saw the significant advantage of T-coils and their use in broadband At high frequencies, the small-signal
T-coils. The instrumentation manu- drivers [6] and electrostatic discharge drain current of M 1 is shunted by C L,
facturer guarded the design details (ESD) protection circuits [7]. causing | Vout | to fall. We can place an
of T-coil circuits as a trade secret for inductor in series with R D [Figure 3(b)]
many years [2]. It was only in 1990 Basic Idea so that the series impedance of R D and
that Dennis Feucht, a former Tek- The bridged T-coil is a special case L D increases with frequency, thereby
tronix engineer, provided the T-coil of two-port bridged-T networks. It forcing a greater current through C L
design equations in his book [3]. and lessening the gain roll-off. Alterna-
The early T-coil implementations tively, we can insert a T-coil circuit in the
were based on discrete, off-chip induc- signal path as illustrated in Figure 3(c).
CB
tors or transformers, suffering from We are interested in the transfer func-
board parasitics, bond wire induct- L1 L2 tion Vout /Vin and its behavior as a
ances, and unwanted couplings to and 1 2 function of component values.
from other signals. A few integrated The transfer function can be
3 derived using the extra element
Digital Object Identifier 10.1109/MSSC.2015.2474258 theorem [8] or the D-Y transformation
Date of publication: 2 December 2015 Figure 2:  A basic bridged T-coil structure. [9] and is as follows:

IEEE SOLID-STATE CIRCUITS MAGAZINE fa l l 2 0 15 9


which agree with those in [3]. It is
VDD interesting to note that g increases
RD with k, i.e., a tighter coupling trans-
lates to a more damped response.
VDD
L2
LD Bandwidth Advantage
VDD Vout
CB As mentioned above, the bridged
L1 CL
RD RD T-coil improves the speed to a
Vout Vout greater extent that does inductive
M1 CL CL peaking. We formulate this advan-
Vin Vin M1 Vin M1
tage by considering the 3-dB band-
widths in the two cases. From (10),
(a) (b) (c)
the T-coil bandwidth is expressed as

Figure 3:  A common-source stage with (a) a simple resistive load, (b) inductive peaking,
and (c) T-coil peaking. ~ 2BW, T - coil = [1 - 2g 2
(16)
+ (1 - 2g 2) 2 + 1 ] ~ 2n
Vout where k is the coupling factor and = [1 - 2g 2 + (1 - 2g 2) 2 + 1 ]
(s) =-g m R D 2
Vin equal to M/ L 1 L 2 = M/L, , and # .
  (1 - k) LC L (17)
# a 2 s 2 +a 1 s +1 ,
b 4 s 4 +b 3 s 3 +b 2 s 2 +b 1 s +1
 k C = (1 + k) L - 2C . (9)
(1) 1+k
L B
R 2D We replace k from (14) and L from
where (13), obtaining
The resulting second-order transfer
a 2 = (L 1 + L 2 + 2M) C B (2) function assumes the form [8]
~ 2BW,T - coil = 4g [1 - 2g 2
a 1 = (L 2 + M) /R D (3) 1 .
Vout ~ 2n + (1-2g 2) 2 +1 ]
b 4 = C B C L (L 1 L 2 - M 2) (4) (s) = - g m R D 2 , RD CL
Vin s + 2g~ n s + ~ n2
b 3 = C B C L R D (L 1 + L 2 + 2M) (5) (10) (18)
b 2 = C B (L 1 + L 2 + 2M) + C L L 2 (6) where
b 1 = R D C L . (7) For example, if g = 2 /2, then
~ 2n = 2 (11) ~ 2BW,T - coil =2 2 /(R D C L).2.83/(R D C L).
(1 - k) LC L
Here, M denotes the mutual induc- R D C L - (1 + k) L/R D Remarkably, the T-coil multiplies
tance between L 1 and L 2 with the g= . (12) the original bandwidth by a factor
2 (1 - k) LC L
polarity shown in Figure 3(c). This of 2.83. By comparison, the induct-
transfer function does not offer much For design purposes, we select a ively peaked stage of Figure 3(b)
intuition but a special case thereof value for the damping factor, g, and exhibits a bandwidth of approxi-
is more mathematically manageable wish to determine the other circuit mately 1.8/ (R D C L) for g = 2 /2. (A
and practically attractive. We assume parameters. Solving the above equa- more accurate comparison should
L 1 = L 2 = L and choose the values tions, [8] finds that take the time-domain overshoot
such that the zeros in (1) are canceled into account as well.)
R 2D C L
4 c
by two of the poles. As shown in [8], L1 = L2 = 1 + 1 2 m (13)
4g
this can be accomplished if two con- 4g 2 - 1 ESD Protection
ditions hold, namely, k= (14) In addition to broadening the band-
4g 2 + 1
width, T-coils can also create a con-
C B = 1 1 - k , (8) C B = C L 2 , (15)
CL 4 1+k 16g stant, resistive input impedance in
the presence of a heavy load capaci-
tance, a situation commonly encoun-
tered in ESD protection circuits. For
CB example, in the input network shown
L1 L2 in Figure 4(a), where R T is a termin-
Vout ation resistor, the ESD device cap-
CESD RT CESD acitance, C ESD, degrades the input
Zin RT
Zin matching, thus causing reflections.
(a) (b) On the other hand, if a bridged T-coil
is inserted as shown in Figure 4(b),
Figure 4:  (a) An input network with an ESD device and (b) an input network using a T-coil Z in can be made equal to R T at all
for broadband matching. frequencies [7]. We can intuitively see

10 fa l l 2 0 15 IEEE SOLID-STATE CIRCUITS MAGAZINE


this property at the two extremes:
at very low frequencies, L 1 and L 2
short R T to the input, and at high fre- RT
quencies, C B does the same. It can be
proved that Z in = R T at all frequen-
cies if L 1 = L 2 and the pole-zero can- R2
cellations leading to (10) also hold.
Output Driver L2
In other words, the conditions stipu- CB
lated by (13)–(15) apply here as well.
An intuitive argument can explain Iout CESD L1
Rout
why the T-coil network cannot have
zeros in this case. If the circuit does
contain a zero, then Z in must still be
equal to the termination resistance at
the zero frequency, s z . Now suppose Zout
we drive the circuit of Figure 3(c)
with an input of the form exp (s z t), Figure 5:  An output driver using a T-coil.
obtaining Vout = 0. Thus, C L can be
removed. In other words, at s = s z,
the drain load reduces to R D in ser-
ies with the parallel combination of
C B and L 1 + L 2 + 2M. This combin-
ation cannot have a zero impedance
at s ! 0 and hence Z in ! R D .
Output drivers using ESD pro-
tection can benefit from T-coils in a
similar manner. Shown in Figure 5,
such an arrangement assumes an
infinite output impedance for the
driver stage and presents a resist-
ance equal to R T to the outside
world. If the output impedance,
E F
R out, is not sufficiently high, a small
resistance, R 2, can be placed in ser-
ies with L 2 to compensate for its
E F
effect [10], [7]. This resistance is
(a) (b)
given by R T / (R out /R T - 1) .

T-Coil Implementation Figure 6:  (a) The implementation of T-coil and (b) a distributed model for circuit simulations.
In the special case where L 1 = L 2, the
inductors lend themselves to a simple
implementation in the form of a sym-
metric spiral [Figure 6(a)] [7]. Here, VDD
the line spacing is chosen to yield RD
the desired mutual coupling, and
the outer dimension and the num- L2
ber of turns to provide the required Vout Series CB
inductance. To include the parasitic CB Peaking
L1 CL L1 L2
resistances and capacitances of the Vout
spiral in simulations, a distributed LS
CESD1 CESD2
model can be constructed as shown in RT
Series
Figure 6(b). Note that the interwinding Vin M1 C1 LS Peaking
capacitance is also taken into account.
As a first-order approximation, this (a) (b)
capacitance appears between E and
F and can be subtracted from the Figure 7:  The use of series peaking and T-coils in (a) a gain stage with a high output ca-
bridge capacitance, C B . pacitance and (b) an input network with high ESD capacitance.

IEEE SOLID-STATE CIRCUITS MAGAZINE fa l l 2 0 15 11


Combination with Other Techniques
The bridged T-coil network can be
VDD combined with other high-speed
RD RD topologies so as to achieve greater
bandwidths. For example, since the
input impedance of the second-order
L2 L2
T-coil circuit is constant, one can
readily add series peaking in the input
CB CB signal path. Illustrated in Figure 7,
L1 L1
this combination proves useful in two
cases: 1) if a stage incorporates a large
transistor [Figure 7(a)], suffering from
CL CL a high output capacitance, C 1, or 2) if
an input network must accommodate
CN a large ESD capacitance [Figure 7(b)],
in which case both capacitors can
represent ESD devices. Series peak-
ing can also be applied to output net-
Figure 8:  The addition of a negative capacitance generator to T-coil network. works such as that in Figure 5.
Differential circuits can combine
T-coils with other differential tech-
niques. For example, as shown in
400 Figure 8, a negative capacitance gen-
350 erator using a cross-coupled pair can
be added in parallel with the load
300
CK capacitance [6], thereby improving
250
Ron (Ω)

the overall speed. To avoid signifi-


M1 200 cant overshoot in the time response,
Vin Vout 150 we choose C N . C B /4.
C1
M2 100
Questions for the Reader
CK 50
1) Use a power dissipation argument
0.0 0.2 0.4 0.6 0.8 1.0 1.2
Vin (V) to determine the transfer function
(a) (b) of the circuit shown in Figure 4(b).
2) In Figure 7(a), how should L S be
chosen if the damping factor of
Figure 9: The on-resistance of complementary switches as a function of input voltage. the series peaking network must
remain around 2 /2?

Answers to Last Issue’s Questions


1) In Figure 9, we write R on = R 0 +
CK CK
VDD
R 1 cos 2~ in t +R 2 cos 4~ in t +g and
assume R 1 . R 0 . Suppose we de-
VDD
P fine the small-signal bandwidth
M3 M8 M14 M10
of the sampler as ~ 3dB = (R 0 C 1) -1 .
X Determine the ratio of ~ in to this
CK Cb bandwidth if the third-order dis-
tortion given by R 1 C 1 ~ in /2 must
M9 remain lower than –60 dB. This
M12
example demonstrates the sever-
Vout
Vin M11 ity of the variable on-resistance.
For the distortion to remain
C1 below –60 dB, we must have
R 1 C 1 ~ in /2 1 10 -3 . Replacing
R 1 C 1 . R 0 C 1 with 1/~ 3dB, we
have ~ in /~ 3dB 1 1/500. This ex-
Figure 10: A bootstrap circuit. ample shows that the sampler’s

12 fa l l 2 0 15 IEEE SOLID-STATE CIRCUITS MAGAZINE


small-signal bandwidth must potential difference to remain [6] S. Galal and B. Razavi, “10-Gb/s limiting
amplifier and laser/modulator driver in 0.18
be far greater than the input less than VDD, VX must not μm CMOS technology,” IEEE J. Solid-State Cir-
frequency. exceed 2VDD - VTH . cuits, vol. 38, pp. 2138–2146, Dec. 2003.
[7] S. Galal and B. Razavi, “Broadband ESD
2) To which node(s) should the protection circuits in CMOS technology,”
n -wells of M 3 and M 8 in Figure 10 References IEEE J. Solid-State Circuits, vol. 38, pp.
[1] E. L. Ginzton, W. R. Hewlett, J. H. Jasberg, 2334–2340, Dec. 2003.
be connected? and J. D. Noe, “Distributed amplification,” [8] J. Paramesh and D. J. Allstot, “Analysis
They should be connected to Proc. IRE, vol. 36, pp. 956–969, Aug. 1948. of the bridged T-coil circuit using the
[2] J. Williams, ed. Analog Circuit Design: Art, extra-element theorem,” IEEE Trans. Cir-
node P to ensure the source and Science, and Personalities, Oxford, U.K.: cuits Syst.-II, vol. 53, pp. 1408–1412, Dec.
drain junctions of these transis- Butterworth-Heinemann, 1991. 2006.
[3] D. Feucht, Handbook of Analog Circuit De- [9] S. C. D. Roy, “Comments on the ‘Analysis
tors are not forward biased. sign. New York: Academic Press, 1990. of the bridged T-coil circuit using the ex-
3) How high can VX in Figure 10 go [4] C. Hutchinson and W. Kennan, “A low tra-element theorem’,” IEEE Trans. Circuits
noise amplifier with gain control,” in Proc. Syst.-II, vol. 54, pp. 673–674, Aug. 2007.
to avoid stressing M 14 ? IEEE GaAs IC Symp., 1987, pp. 119–122. [10] T. T. True, “Bridged-T termination net-
When M 14 is off, its source [5] L. Selmi, D. Estreich, and B. Ricco, “Small- work,” U.S. patent 3,155,927, Nov. 3, 1964.
signal MMIC amplifiers with bridged T-coil
voltage reaches approximately matching networks,” IEEE J. Solid-State Cir-
VDD - VTH . For the source-drain cuits, vol. 27, pp. 1093–1096, July 1992. 

editor’s note (Continued from p. 4) circuit intuitions (Continued from p. 8)

■■ Willy Sansen, in “Minimum Power in Analog Ampli- This equation, along with equations for fp1 and fz,
fying Blocks: Presenting a Design Procedure,” can now be used to form the equation for the overall
answers questions he received from his 2015 ISSCC voltage transfer function of the two-stage amplifier.
plenary talk. It is worth noting that as we increase C 12, fp1 and
■■ Behzad Razavi continues his column series “A Cir- fp2 (as found by their respective equations) will move
cuit for All Seasons” by providing an article that dis- farther apart, a phenomenon referred to as pole split-
cusses the bridged T-coil. This article fits well into ting [1], [2].
this issue’s feature of wireline communications due In summary, Miller’s approximation uses the dc gain
to the use of the T-coil for extending the bandwidth of the amplifier to provide a relatively accurate estima-
of a circuit. tion of its dominant pole (i.e., the circuit bandwidth).
■■ Ali Sheikholeslami provides another piece in his well-re- This approximation, however, becomes inaccurate
ceived series, “Circuit Intuitions.” In this issue, he contin- when determining the second pole of the amplifier;
ues discussing Miller’s theorem, its uses and shortcom- other intuitive methods exist for this purpose.
ings when analyzing circuits. As usual (and the e-mail we For further discussions and intuition into Miller’s
receive would support this), the article provides useful theorem, we refer the readers to [3].
insight into circuit analysis and design.
■■ Finally, Marcel Pelgrom discusses “The Next Hype” References
[1] A. S. Sedra and K. C. Smith, Microelectronic Circuits, 7th ed. Lon-
in his column, which is always an entertaining arti- don, U.K.: Oxford Univ. Press, 2014.
cle that provokes thought. It’s one of my favorite [2] B. Razavi, Fundamentals of Microelectronics. New York: Wiley,
2008.
reads in each magazine issue. I hope you agree! [3] B. Mazhari, “On the estimation of frequency response in ampli-
We hope you enjoy reading IEEE Solid-State Circuits Mag- fiers using Miller’s theorem,” IEEE Trans. Education, vol. 48, no.
3, pp. 559–561, Aug. 2005.
azine. Please send comments to me at r­jacobbaker@
gmail.com. 

IEEE SOLID-STATE CIRCUITS MAGAZINE fa l l 2 0 15 13

You might also like