CMOS Low Noise Amplifier Design Optimization Technique by oox83341


									                    CMOS Low Noise Amplifier Design
                       Optimization Technique
        Trung-Kien Nguyen, Nam-Jin Oh, Hyung-Chul Choi, Kuk-Ju Ihm, and Sang-Gug Lee
                                     Information and Communications University
                                119 Munjiro, Yuseong-gu, Daejeon, 305-714 South Korea

Abstract                                                       terminal of the cascode transistor to that of common
                                                               source transistor. The description of the methodology for
  In this paper, a set up noise parameter expression and
                                                               LNA design optimization and the proposed LNA are
the third order intermodulation product expression (IM3)
                                                               discussed in detail in section 2 and 3, respectively. The
for a power-constrained simultaneous noise and input
                                                               proposed LNA for 5 GHz WLAN applications is
matching low noise amplifier design optimization
                                                               fabricated based on 0.18 m CMOS technology.
technique are introduced. Based on these expressions,
                                                               Measured results show 20 dB power gain, 1.5 dB NF and
the methodology to design LNA to archive the power-
                                                               –5 dBm IIP3. The proposed LNA dissipates the DC
constrained simultaneous noise and input matching as
                                                               current of 3 mA at supply voltage of 2.5 V.
well as satisfy the linearization condition is explained. In
additional, the power gain is enhanced by using a very           2. Methodology for Low Noise Amplifier Design
simple positive feedback. The proposed LNA for 5GHz
                                                               A. Noise Optimization Analysis
WLAN applications is fabricated based on 0.18 m
CMOS technology. Measured results show 20 dB power               Figure 1-a shows the schematic of a cascode LNA
gain, 1.5 dB NF and –5 dBm IIP3. The proposed LNA              topology that is adopted to explain the PCSNIM LNA
dissipates DC current of 3 mA at supply voltage of 2.5         design technique. The LNA shown in Fig. 1-a differs by
V.                                                             one additional capacitor Cex in comparison with the
                                                               typical cascode LNA.
1. Introduction
   With the recent proliferation of wireless transceiver                                                                        id
applications, there is an extensive effort to develop low                                                                   M2
cost, highly integrated RF circuits. CMOS has become a                                     Matching Circuit
competitive      technology     for    radio    transceiver
implementa-tion due to the technology scaling, higher                                                                       M1
level of integrability, lower cost, etc. [1]. In typical
                                                                                Zs                        Cex
receiver architectures, a low noise amplifier (LNA) is the                                 Rs
one of the most critical blocks that determines the                                  vs                                         Ls
sensitivity of wireless receiver systems [2-[4]. Normally,                                              (a)
LNA design involves the tradeoff between noise figure
(NF), gain, linearity and power consumption.                               Lg             Matching Circuit                                   id
Consequently, the goal of LNA design is to achieve                                                                          +
simultaneous noise and input matching at any given
                                                               Zs                                                           vgs
amount of power dissipation as well as satisfy the                    Rs
                                                                                          Cex                 gg      Cgs            gmvgs
                                                                                                   2                                               2
linearization conditions. The LNA design optimization                                             ing                                             ind
technique proposed in [4] can be applied for power-              2
                                                                                 Zs Zin
constrained simultaneous noise and input matching.                                                                                     Ls
However, as discussed in [4], the fully potential of this                                               (b)
technique is not provided clearly. This paper attempts to      Fig. 1. Simple cascode LNA to adopt the PCSNIM technique
analyze and provide clear and perspe-ctive understanding       (a) and its small-signal equivalent circuit (b)
one of the LNA design optimization techniques, namely            This LNA topology was first introduced in [3] as a
power-constrained simultaneous noise and input                 solution to reduce the noise figure of the LNA at low
matching technique. The analyses are based on the noise        power dissipation, however, the potential and the
parameter expressions and the expression for the third         theoretical analysis as a power-constrained (i.e., low
order intermodulation product (IM3). By using those            power) simultaneous noise and input matchable LNA
expressions, the design principle, advantages and              topology has not been recognized. Fig. 1-b shows the
practical limitation for the mentioned LNA technique are       simplified small-signal equivalent circuit of the cascode
explained. In additional, in this paper, the power gain of     amplifier shown in Fig. 1-a for the noise analysis. In Fig.
the LNA is improved by using simple positive feedback          1-b, the effects of common-gate transistor M2 on the
technique. The simple positive feedback is implemented         noise and frequency response are neglected [2]. The
by one additional capacitor connected from drain               noise parameter expressions for a circuit with series
feedback, shown in Fig. 1-b, can be obtained by applying                                                                                                   g m Ls
                                                                                                                                                                      Re[Z s ]                              (9)
the Kickoff’s law [1]. The results are simple enough to                                                                                                     Ct
provide useful insights as shown below [5]                                                                                                                    1
                                                                                                                                                     sLs               Im Z s                              (10)

                                                                                                            eff                 As mentioned above, for the advanced CMOS
                                         1 s 2Ct Lg                       Ls 1 | c |
                                                                                                            5                 technology parameters, (8) is approximately equal to
                           gd 0                                                                                               (10). Therefore, (10) can be dropped, which means that
            1                                                                                      2
F 1         2                                                 2                           eff                           (1)   for the given value of Ls, the imaginary value of the
           gm Rs                             sCt Rs                1 |c|
                                                                                          5                                   optimum noise impedance becomes approximately equal
                                                                                                                              to that of the input impedance with opposite sign. Now
                                         1 c
                                                             gm sCt
                                                                                  Rs2 sL2
                                                                                                                              then, the design parameters that can satisfy (7)-(9) are
                                 5                                                                                            VGS, W (or Cgs), Ls, and Cex. Since there are three
                                                             2                                                                equations and four unknowns, (7)-(9) can be solved for
                                 Fmin         1                                   ( 1 | c |2 )                          (2)
                                                              5       T
                                                                                                                              an arbitrary value of Zs, by fixing the value of one of the
                                                                                                                              design parameters that can be the power dissipation or
                                                              j                       c                                       VGS. In other word, this LNA design optimization
                            5 (1 c )                               Cgs                        5
     Zopt                                                                                                         sLs   (3)   technique allows to design simultaneous noise and input
                                                                                                        2                     matching at any given amount of power dissipation.
                   Cgs                               2
                             5 (1 c )                             C gs                        5                               B Linearity Analysis

                                                                          1                                                     In RF circuit design, the linearity is another important
                                                         Rn                                                             (4)   aspect to consider. Since LNA is the first block in the
                                                                                                                              typical receiver system, the linearity of the LNA is
where Ct = Cgs1+Cex
                                                                                                                              commonly estimated by the third order intermodulation
  As can be seen from (2) and (4), Fmin and Rn are not
                                                                                                                              product. Two signals of adjacent channels Asin 1 and
affected by the addition of Cex. In other word, by using
Cex, the minimum noise figure and the noise resistance                                                                        Asin 2 will generate products IM3 such as Asin(2 1- 2)
expressions for power-constrained simultaneous noise                                                                          and Asin(2 2- 1) at the output of nonlinear circuit. IM3
and input matching technique are the same as those in                                                                         usually calculated in the literature as the ratio of intermo-
[3]. From Fig. 1-(b), the input impedance of the LNA is                                                                       dulation of the third order and the response magnitude of
given by                                                                                                                      the fundamental frequency which is given by
                                                              1           g m Ls                                                                                    3 2 A3 2       1        2
                                Z in         sLs                                                                        (5)                            IM 3           A                                    (11)
                                                             sCt           Ct                                                                                       4      A1
  In (5), the source degeneration generates real part at                                                                      where A1, A3 are the first order and third order coefficient
the input impedance. This is important because there is                                                                       of Volterra series.
no real part in the input impedance without degeneration
                                                                                                                                           Lg         Matching Circuit                               gm2
while there is in the optimum noise impedance.                                                                                                                                                             id
Therefore, Ls helps to reduce the discrepancy between                                                                                                  Zin
the real parts of the optimum noise impedance and the                                                                           '                                                  vgs
                                                                                                                               Zs                               Cex         Cgs                      Yo1
LNA input impedance. Furthermore, from (5), the                                                                                       Rs                                                   gm1vgs
imaginary part of Zin is changed by sLs, and this is                                                                             vs                                                -
followed by nearly the same change in Zopt in (3),
especially with advanced technology considering the
value of c is higher than 0.4 (e.g., c 0.5 with 0.25 m
                                                                                                                              Fig. 2 Circuit model for nonlinear analysis
technology), and becomes lower than 1 [6].
  Now, for the circuit shown in Fig. 1-(a), the conditions                                                                       For linearity analysis purpose, the equivalent small
that allow the simultaneous noise and input matching are                                                                      signal circuit of LNA in Fig. 1 is depicted in Fig. 2.
                                                                                                                              Now, M2 can be considered and modeled by the series
                                                              2                                                               trans-conductance gm2, assuming rds2 >> Rout. In this case,
                                         5 (1                c )
                                                                                                        Re[Z s ]        (7)   the effects of the Cgs2 and Cgd1 have been neglected. The
                                                     Ct                                                                       output admittance seen at the drain of M1, Yo1, is added in
          C gs                       2
                                                                          c                                                   the model with the purpose to identify the output
                 5 (1            c )                 C gs                         5
                                                                                                                              contribution. Using the Kickoff’s law in the model of
                           Ct                                                                                                 Fig. 6 the input signal can be written as
                   j                         c                                                                                                  vin s vgs a1 s id a2 s               (12)
                           Cgs                           5
                                                                                                  sLs       Im Z s      (8)   Where              a1 s        sCt Z in        sLs       1                   (13)
   C gs                      2
                                                                  c                                                                                                          Yo1
            5 (1           c )               C gs                         5                                                                           a2 s          sLs 1                                  (14)
When the effective mobility reduction is taken into                                                   The qualitative description of the proposed design
account, the current between the source and drain                                                     process would be as follows. First, choose the DC bias,
terminals of the transistor M1 is given as                                                            VGS, for example the bias point that provides minimum
                                       Wvsat Cox Vgs Vt
                                                                                                      Fmin. Second, choose the transistor size, W, based on the
                             I ds                                                              (15)   power constraint, PD. Third, choose the additional
                                           Vgs Vt        1   2 Lvsat                                  capacitance, Cex, as well as the degeneration inductance,
where     1      o       2 Lvsat           and Vgs VGS vgs                                            Ls, to satisfy (7), (9), and s2CtLs = -1 conditions (as
  Here, VGS is the DC bias voltage of the transistor, vgs is                                          mentioned, to improve the linearity of circuit the
the small signal between gate and source, and vsat is the                                             condition s2CtLs = -1 need to be satisfied). With the given
carrier velocity saturation. Using (12), the Volterra series                                          Ls the condition Im[Zin*] = Im [Zopt] is automatically
expression of id is derived as                                                                        satisfied. At this point, the simultaneous noise and input
                                2                3                                                    matching is achieved. As the last step, if there exists any
        id A1 s vin A2 s1 ,s2 vin A3 s1 ,s2 ,s3 vin    (16)
                                                                                                      mismatch between Zin and Zs’, as shown in Fig. 1 (b), an
Here the coefficients of order higher than three are                                                  impedance matching circuit can be added.
ignored. Usually, the adjacent channel frequencies 1                                                     This design optimization technique suggest that, by
and 2 providing the intermodulation products are very                                                 using an additional capacitor, Cex, the LNA can be
close to the fundamental frequency therefore s s1 s2                                                  designed to archive power-constrained simultaneous
can be assumed. The |IM3| at (2 1 - 2) is                                                             noise and input matching as well as satisfy the
                                A2 A1 s                                 2
                                                                                                      linearization condition. The limitations of the PCSNIM
              IM 3                    3
                                        a1 s 3 g3                    2 g2 B                    (17)   technique are high Rn and low effective cut-off
                                2 g m1
                                                                                                      frequency. High Rn can be a serious limitation for the
     B   2 sLs       s a2              s   A1      s     2sLs 2s a2 2s A1 2s (18)                     practical high yield LNA design.
                                                         g m1
                                A1 s                                                           (19)   3. Gain Enhancement Technique and Proposed LNA
                                                a1 s      g m1a2 s
                                                                                                         One of the simple ways to improve the power gain of
              4K            2
                         L2vsat                                     4K          2
g2                   0
                                                  , g3                   0
                                                                                               (20)   LNA is using positive feedback. In this paper, the
         Vgs Vt          1          2 Lvsat                  Vgs Vt          1   2 Lvsat              positive feedback is realized by Cf shown in Fig. 3-a.
                                                                                                      This phenomenon can be understood by another point of
                          s s1 s2
                                                                                                      view as the form of oscillator. In Fig. 3-a, Cgs1, Cf, and
where g2 and g3 are the second and third degree                                                       M2 constitutes an oscillator topology with inductive
coefficients of the transistor nonlinear Taylor expansion.                                            termination at the output [1]. The effect of the positive
The B coefficient is the second-order interaction of the                                              feedback will increase maximum available gain of the
products 2 , 1- 2, and 2- 1. A1(s) is the transcon-                                                   cascode amplifier at high frequencies. Note that no
ductance of the circuit. Substituting (19) into (17), it                                              additional active device is used therefore no more DC
shows the dependence of |IM3| with inverse of the term                                                power is dissipated and no noise is contributed. The limit
                                                                Yo1                                   to amount of feedback is governed by stability
                                    sCt Rin       sLs g m1 1                                   (21)   consideration. To ensure the stability condition, Gtol must
                                                                                                      always positive. This technique is first introduced in [7];
As can be seen in (17), the linearity can be improved by                                              however, the reported results are simulation-based only.
using different ways. Revising (17), the |IM3| can be                                                 This paper tries to realize this idea in term of measured
lowered with the reduction of a1(s), g3, or with the                                                  results. The simplified proposed LNA is shown in Fig. 3-
increase (21). As shown in (13), with inductive                                                       b. The proposed LNA is implemented by combining the
degeneration the s2CtLs term will cancel the “1” term,                                                PCSNIM design technique described in previous section
and as a result a1(s) is reduced. This indicates that the                                             and the gain enhancement technique shown in Fig. 3-a.
selected topology is more adequate to keep the |IM3|                                                  In the Fig. 3-b, the simple Lo-Co network represents the
small in comparison with resistive and capacitive                                                     output-matching network and Lo is implemented by off
degeneration topology, where such cancellation does not                                               chip inductor.
exist. The joint effect of g3 and g2 coefficients in |IM3| is                                                             VDD                                      VDD
inversely dependent on the bias (Vgs-Vt), indicating that
the linearity can be improved by increasing gate source                                                               Lo         Co   RFout                        Lo    Co   RFout
voltage. However, increasing the gate source voltage will
increase the power dissipation. With large Yo1 and gm1                                                     Vbias                                     Vbias
                                                                                                                                M1                                      M2
values and small gm2 value (21) is increased such that the                                                                            Cf                                      Cf
linearity will be increased. For the same reason, any
                                                                                                                   Cgs1                               Lg
increase in Ct, preserving the matching condition in the                                                                                                                M1
input circuit, also improves the linearity.                                                                          i1                       RFin      Cex
C. Design Consideration                                                                                                                                                  Ls
  In this section, the overall consideration for LNA                                                                      (a)                                (b)
design to obtain power-constrained simultaneous noise                                                 Fig. 3 Gain enhancement technique and the proposed LNA
and input matching as well as linearization is described.
4. Measurement Results                                                  In this paper, a very simple and insightful set of noise
                                                                     parameter expressions and the third order intermodu-
  To demonstrate the potential of power-constrained
                                                                     lation product for the power-constrained simultaneous
simultaneous noise and input matching optimization
technique and the gain enhancement technique, the                    noise and input matching LNA design optimization
current dissipation of the proposed LNA is fixed at 3                technique is newly introduced. Based on those expres-
mA. Three LNA versions are fabricated based on 0.18                  sions, the design principle, advantage, and the limitation
  m CMOS technology, the first circuit is simple cascode             for the power-constrained simultaneous noise and input
inductive degeneration topology, the second one simple               matched technique are explained. To demonstrate the
cascode with Cex and the third one is the proposed LNA               potential of this design technique, the proposed LNA is
shown in Fig. 3. Note that, all the circuits are designed at         designed and optimized for 5 GHz WLAN applications.
the same power dissipation. A Comparison of measured                 The measured results show good agreement with
NF results are shown in Fig. 4. As can be shown in Fig.              theoretical analysis.
4, by using the power-constrained simultaneous noise                                            40
and input matching technique, the obtained NF is lower
than that for the case of simple cascode inductive                                              10          Fundamental
degeneration. The main reason of the improvement in NF

                                                                           Output Power [dBm]
can be understood as the discrepancy between real parts
of input and noise matching conditions
                     3.5                                                                        -50

                      3                                                                                              IM3
 Noise Figure [dB]

                                     With Cex
                                     Without Cex                                      -110
                      2                                                                               -40      -30         -20          -10        0           10
                                                                                                                         Input Power [dBm]
                                                                     Fig. 6 IIP3 of the proposed folded cascode LNA

                           4   4.5         5         5.5   6   6.5
                                            Freq [GHz]                                                                     (a)    (b)
Fig. 4 Measured NF of LNAs

 Power Gain [dB]

                                                                     Fig. 7 Microphotograph of the three LNA: (a) simple cascode,
                                                                     (b) simple cascode includes Cex, and (c) proposed LNA
                                        With Cf
                                        Without Cf                                                                      References
                                                                     [1]                        B. Razavi, “CMOS technology characterization for analog and RF
                                                                                                design,” IEEE Journal of Solid- State Circuits, Vol. 34, pp. 268-276,
                       0                                                                        March 1999.
                                                                     [2]                        S. P. Voinigescu et al., “A Scalable High-Frequency Noise Model for
                           4   4.5          5        5.5   6   6.5                              Bipolar Transistors with Application Optimal Transistor Sizing for
                                            Freq [GHz]                                          Low-Noise Amplifier Design,” IEEE J. Solid- state Circuits, Vol. 32,
                                                                                                pp 1430-1439, Sep 1997.
Fig. 5 Measured power gain of LNAs                                   [3]                        D. K. Shaeffer et al., “A 1.5V, 1.5 GHz CMOS Low Noise Amplifier”,
                                                                                                IEEE Journal of Solid-Stage Circuits, Vol. 32, pp 745-758, May 1997.
  Fig. 5 shows the measured results comparison of two                [4]                        G. Girlando et al., “Noise Figure and Impedance matching in RF
LNAs simple cascode and proposed LNA shown in Fig.                                              Cascode Amplifiers,” IEEE Transaction on Circuits and Systems-II,
3-b. As can be seen from Fig. 5, the power gain is                                              Vol. 46, pp. 1388-1396, Nov. 1999.
                                                                     [5]                        Trung-Kien Nguyen, et al., “CMOS Low Noise Amplifier Design
improved by 3 dB compare to that of simple cascode                                              Optimization Techniques,” Accepted to be published on IEEE
topology. Fig. 6 shows the measured result of input third                                       Transactions on Microwave Theory and Technique, May 2004.
order intermodulation product of the proposed LNA. The               [6]                        G. Knoblinger et al., “A New Model for Thermal Channel Noise of
                                                                                                Deep-Submiron MOSFET and its Applications in RF-IC Design,”
proposed LNA has power gain of 20 dB, NF of 1.5 dB at                                           IEEE Journal of Solid- State Circuits, Vol. 36, pp. 831-837, May
5.25 GHz and IIP3 of –5 dBm. The microphotograph of                                             2001.
the three circuits is shown in Fig. 7.                               [7]                        K. L. Chan, et al., “1.5 V 1.8 GHz Bandpass Amplifier,” IEE
                                                                                                Symposium of Circuits Devices and Systems pp. 331-333, Dec. 2000.
5. Conclusion

To top