lect2 up120 (100325)
TRANSCRIPT
Lecture 120 – Component Matching (3/25/10) Page 120-1
CMOS Analog Circuit Design © P.E. Allen - 2010
LECTURE 120 – COMPONENT MATCHINGLECTURE ORGANIZATION
Outline• Introduction• Electrical matching• Physical matching• SummaryCMOS Analog Circuit Design, 2nd Edition ReferencePages 56-59 and new material
Lecture 120 – Component Matching (3/25/10) Page 120-2
CMOS Analog Circuit Design © P.E. Allen - 2010
INTRODUCTIONWhat is Accuracy and Matching?
The accuracy of a quantity specifies the difference between the actual value of thequantity and the ideal or true value of the quantity.The mismatch between two quantities is the difference between the actual ratio of thequantities and the desired ratio of the two quantities.
Example:x1 = actual value of one quantityx2 = actual value of a second quantityX1 = desired value of the first quantityX2 = desired value of the second quantityThe accuracy of a quantity can be expressed as,
Accuracy = x - X
X = XX
The mismatch, , can be expressed as,
=
x2x1
-X2X1
X2X1
= X1x2X2x1
- 1
Lecture 120 – Component Matching (3/25/10) Page 120-3
CMOS Analog Circuit Design © P.E. Allen - 2010
Relationship between Accuracy and MatchingLet:
X1 = |x1 - X1| x1 = X1 ± X1
andX2 = |x2 – X2| x2 = X2 ± X2
Therefore, the mismatch can be expressed as,
= X1(X2 ± X2)X2(X1 ± X1) – 1 =
1 ±X2
X2
1 ±X1
X1
– 1 1 ±X2
X2 1 +-
X1X1
– 1
1 ± X2
X2 +-
X1X1
– 1 = ± X2
X2 +-
X1X1
Thus, the mismatch is approximately equal to the difference in the accuracies of x1 and x2assuming the deviations ( X) are small with respect to X.
Lecture 120 – Component Matching (3/25/10) Page 120-4
CMOS Analog Circuit Design © P.E. Allen - 2010
123456789
10
0 X0 1 2 3 4 5 6 7 8 9 10 1211
Num
ber
of S
ampl
es
041005-011314
Characterization of the MismatchMean of the mismatch for N samples-
m = 1N
i=1
N
i
Standard deviation of the mismatch for N samples-
s = = 1
N-1i=1
N( i - m )2
Example:
m = 25340 = 6.325 s = 2.115
Lecture 120 – Component Matching (3/25/10) Page 120-5
CMOS Analog Circuit Design © P.E. Allen - 2010
Motivation for Matching of ComponentsThe accuracy of analog signal processing is determined by the accuracy of gains and timeconstants. These accuracies are dependent upon:
Gain Ratios of components or areasTime constants Products of components or areas
Ratio Accuracy?
Actual Ratio = X1± X1X2± X2 =
X1X2
1±X1
X1
1±X2
X2
X1X2 1±
X1X1 1+-
X2X2
X1X2 1±
X1X1 +-
X2X2
If X1 and X2 match ( X1/X1 X2/X2), then the actual ratio becomes the ideal ratio.
Product Accuracy?
Product accuracy = (X1± X1)(X2± X2) = X1X2 1±X1
X1 1±X2
X2 X1X2 1±X1
X1 ±X2
X2
Unfortunately, the product cannot be accurately maintained in integrated circuits.
Lecture 120 – Component Matching (3/25/10) Page 120-6
CMOS Analog Circuit Design © P.E. Allen - 2010
Switched Capacitor CircuitsSwitched capacitor circuits offer a solution to the product accuracy problem.A switched capacitor replacement of aresistor:
The product of a resistor, R1, and a capacitor,C2, now become,
R1C2 = TcC1 C2 =
1fcC1 C2 =
C2fcC1
The accuracy of the time constant (product) now becomes,C2fcC1 1±
C2C2 +-
C1C1 +-
fcfc
Assuming the clock frequency is accurate and larger than the signal bandwidth, then timeconstants in analog signal processing can be accurately matched by ratios of elements.
060316-06
+
−v1
+
−v2
φ1 φ2
C1+
−v1
+
−v2
R1= C1
Tc
φ1
φ2Tc
Lecture 120 – Component Matching (3/25/10) Page 120-7
CMOS Analog Circuit Design © P.E. Allen - 2010
Types of Mismatches1.) Those controlled or influenced by electrical design
- Transistor operation- Circuit techniques- Correction/calibration techniques
2.) Those controlled or influenced by physical design- Random statistical fluctuations (microscopic fluctuations and irregularities)- Process bias (geometric variations)- Pattern shift (misalignment)- Diffusion interactions- Stress gradients and package shifts- Temperature gradients and thermoelectrics- Electrostatic interactions
Lecture 120 – Component Matching (3/25/10) Page 120-8
CMOS Analog Circuit Design © P.E. Allen - 2010
ELECTRICAL MATCHINGMatching PrincipleAssume that two transistors are matched (large signal model parameters are equal).Then if all terminal voltages of one transistor are equal to the terminal voltages of theother transistor, then the terminal currents will be matched.
Q1 Q2
iC1 iC2iB1 iB2
iE1 iE2
M1 M2
iD1 iD2
041005-02
Note that the terminals may be physically connected together or at the same potential butnot physically connected together.
Lecture 120 – Component Matching (3/25/10) Page 120-9
CMOS Analog Circuit Design © P.E. Allen - 2010
Examples of the Matching Principle
VB
iD1 iD2
M1 M2
M3 M4
+
-Vio
M5M1 M2
M3 M4
VB
iD1 iD2
041005-03
iD1 iD2
M1 M2+
-Vio
M5
Cascode current mirror:The key transistors are M1 and M2. The gates and sources are physically connected
and the drains are equal due to M3 and M4 gate-source drops. As a result, iD1 will bevery close to iD2.
Differential amplifier:When iD1 and iD2 are equal, the fact that the drains of M1 and M2 are equal should
give the smallest value of the input offset voltage, Vio.
Note: Since the drain voltages of M3 and M4 in both circuits are not necessarily equal,the gate-source voltages of M3 and M4 are not exactly equal which cause the drainvoltages of M1 and M2 to not be exactly equal.
Lecture 120 – Component Matching (3/25/10) Page 120-10
CMOS Analog Circuit Design © P.E. Allen - 2010
Gate-Source MatchingNot as precise as the previous principle but useful for biasing applications.A. If the gate-source voltages of two or more FETs areequal and the FETs are matched and operating in thesaturation region, then the currents are related by the W/Lratios of the individual FETs. The gate-source voltagesmay be directly or indirectly connected.
iD1 = K’W1
2L1 (vGS1-VT1)2 (vGS1-VT1)2 =
2K’iD1
(W1/L1)
iD2 = K’W2
2L2 (vGS2-VT2)2 (vGS2-VT2)2 =
2K’iD2
(W2/L2)
If vGS1 = vGS2, then W2
L2iD1 =
W1
L1iD2 or iD1 =
W1/L1
W2/L2iD2
B. If the drain currents of two or more transistors are equal and the trans-istors are matched and operating in the saturation region, then the gate-source voltages are related by the W/L ratios (ignoring bulk effects).
If iD1=iD2, then
vGS1 = VT1+W2/L2
W1/L1(vGS2-VT2) or vGS1 = vGS2 if
W2
L2=
W1
L1
+
-
vGS1
iD1M1
iD2M2
+
-
vGS2
W1L1
W2L2
Fig. 290-02
+
-vGS1
iD1
iD2M2
+
-vGS2
W1L1
W2L2
Fig. 290-0
Lecture 120 – Component Matching (3/25/10) Page 120-11
CMOS Analog Circuit Design © P.E. Allen - 2010
Process Independent Biasing - MOSFETThe sensitivity of the bias points of all transistors depend on both the variation of thetechnological parameters and the accuracy of the biasing circuits.Gate-source voltage decomposition:
The gate-source voltage of the MOSFET can be divided into two parts:1.) The part necessary to form or enhance the channel, VT
2.) The part necessary to cause current toflow, VGS – VT = VON , called theoverdrive.
This overdrive can be expressed,
VON = VDS(sat) = 2ID
K’(W/L)The dependence of the bias point on the technology, VT, can be reduced by making VON= VDS(sat) >> VT.
This implies that small values of W/L are preferable. Unfortunately, this causes thetransconductance to become small if the current remains the same.
Lecture 120 – Component Matching (3/25/10) Page 120-12
CMOS Analog Circuit Design © P.E. Allen - 2010
Doubly Correlated SamplingIllustration of the use of chopper stabilization to remove the undesired signal, vu, from thedesired signal, vin. In this case, the undesired signal is the gate leakage current.
+1
-1t
T fc = 1
vin
vu
voutvB vC
Vin(f)
Vu(f)
VB(f)
ffc0 2fc
f
f
3fcVC(f)
ffc0 2fc 3fc
A1 A2
Clock
VA(f)
ffc0 2fc 3fc
vA
Fig. 7.5-8
Lecture 120 – Component Matching (3/25/10) Page 120-13
CMOS Analog Circuit Design © P.E. Allen - 2010
An Op Amp Using Doubly Correlated Sampling to Remove DC Offsets
051207-01
clkb
clk
clk
clkb
Inn
Inn
Inp
InpVDD
clk clkb
clkclkb
clkb
clkclk
clkb
Inn
Inn
Inp
InpVDD
VDD
M1 M2
M3 M4
Cc
M5
R1
R2
• Chopping with 50% duty cycle• All switches use thick oxide devices to reduce gate leakage• Gain gm1(rds2||rds4)gm5R2Will examine further in low noise op amps.
Lecture 120 – Component Matching (3/25/10) Page 120-14
CMOS Analog Circuit Design © P.E. Allen - 2010
Self-Calibration TechniquesThe objective of self-calibration is to increase the matching between two or morecomponents (generally passive).The requirements for self-calibration:1.) A time interval in which to perform the calibration2.) A means of adjusting the value of one or more of the components.
Fixed Component
AdjustableComponent
Comparisonof values
041007-05
Self-calibration can typically improve the matching by a factor of 2-3 bits (4-8).
Lecture 120 – Component Matching (3/25/10) Page 120-15
CMOS Analog Circuit Design © P.E. Allen - 2010
Example of Capacitor Self-CalibrationConsider the charge amplifier below that should have a gainof unity.
Assume the amplifier has a DC input offset voltage of Vio. The following shows how tocalibrate one (or both) of the capacitors.
+-
VRef
vOUT
C1
C2Vio
VRef -Vio+ -
Vio+-
Autozero Phase
+-
VRef
vOUT
C2
C1
Vio
Vio+-
+
-
Calibration Phase
VRef -Vio
vx
041007-07
In the calibration phase, vx, is:
vx = (VREF-Vio)C2
C1+C2 - (VREF-Vio)
C1C1+C2
= (VREF-Vio)C2-C1C1+C2
The correction circuitry varies C1 or C2 until vx = 0 as observed by vOUT.
+-
vIN vOUT
C1 C2
041007-06
Lecture 120 – Component Matching (3/25/10) Page 120-16
CMOS Analog Circuit Design © P.E. Allen - 2010
Variable ComponentsThe correction circuitry should be controlled by logic circuits so that the correction canbe placed into memory to maintain the calibration of the circuit during application.Implementation for C1 and C2 of the previous example:
C1
S1
C1
S2 S3
C1
2N
SN
C1 1-2K1 2K 2K+1
C1
2K+2
Capacitor C1
C2
S1
C2
S2 S3
C2
2N
SN
C2 1-2K1 2K 2K+1
C2
2K+2
Capacitor C2041007-08
K is selected to achieve the desired tolerance or variationN is selected to achieve the desired resolution (N > K)
Additional circuitry:Every self-calibration system will need additional logic circuits to sense when the valueof vx changes from positive to negative (or vice versa) and to store the switch settings inmemory to maintain the calibration.
Lecture 120 – Component Matching (3/25/10) Page 120-17
CMOS Analog Circuit Design © P.E. Allen - 2010
Basics of Dynamic Element Matching†
Dynamic element matching chooses different, approximately equal-valued elements torepresent a more precise value of a component as a function of time.Goal of dynamic element matching:Convert the error due to element mismatch from a dc offset into an ac signal of equivalentpower which can be removed by the appropriate means (doubly-correlated sampling,highpass filtering of a sigma-delta modulator, etc.)
† L. R. Carley, “A Noise-Shaping Coder Topology for 15+ Bit Converters, IEEE J. of Solid-State Circuits, vol. 24, no. 2, April 1989, pp. 267-273.
R0
S0VRef
i
All resistor are approximately equalvalued to within some tolerance
R1
S1
R2
S2
R3
S3
R4
S4
R5
S5
R6
S6
R7
S7
R8
S8
R9
S9 R0
S0
R1
S1
R2
S2
R3
S3
R4
S4
R5
S5
R6
S6
R7
S7
R8
S8
R9
S9
R0
S0
R1
S1
R2
S2
R3
S3
R4
S4
R5
S5
R6
S6
R7
S7
R8
S8
R9
S9Time t2
Time t1
041010-01
Lecture 120 – Component Matching (3/25/10) Page 120-18
CMOS Analog Circuit Design © P.E. Allen - 2010
How Dynamic Element Matching WorksAssume that we have three approximately equal elements with the following currents:
Element 1 = 0.99mA Element 2 = 1.03mA Element 3 = 0.98mA
Ideal current output level
Error when dynamicelement matching is not used
Error when dynamicelement matching is used
t0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
0
1
2
3
t0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
-1
0
+1
+2
Nor
mal
Err
or (
%)
Elements → 1 1 1 1 1 1 1,2 1,2 1,2 1,2 1,2 1,2 1,2,3
t0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
-1
0
+1
+2
Dyn
amic
Ele
men
tM
atch
ing
Err
or (
%)
Elements → 1 3 2 3 1 2 1,2 2,3 1,3 1,2 1,3 2,3 1,2,3+3
-2
-3 060405-06
Ide
al C
urre
nt (
mA
)
Lecture 120 – Component Matching (3/25/10) Page 120-19
CMOS Analog Circuit Design © P.E. Allen - 2010
Issues of Dynamic Element Matching• The selection of the elements must be truly random for the maximum benefit to occur.• If the number of elements is large this can be an overwhelming task to implement. An
approximation to random selection is the butterfly-type randomizer below:S1
S1
S2
S2
S3
S3
S4
S5
S5
S5
S6
S6
S7
S7
S8
S8
S9
S9
S10
S10
S11
S11
S12
S12
Three-stage, eight-linebutterfly randomizer.Each pair of switchesmarked with the samelabel is controlled toeither exchange the two signal lines or pass them directly to the next stage.
0
1
2
3
4
5
6
7
0
1
2
3
4
5
6
7041010-03
• When using the dynamic element technique, one needs to be careful that the averagingactivity of the dynamic element matching process does not interfere with otheraveraging processes that might be occurring simultaneously (i.e. modulators).
• Other references:1.) B.H. Leung and s. Sutarja, “Multibit - A/D Converter Incorporating A Novel Class of Dynamic Element Matching
Techniques,” IEEE Trans. on Circuits and Systems-II, vol. 39, no. 1, Jan. 1992, pp. 35-51.
2.) R. Baird and T. Fiez, “Linearity Enhancement of Multibit - A/D and D/A Converters Using Data WeightedAveraging,” IEEE Trans. on Circuits and Systems-II, vol. 42, no. 12, Dec. 1995, pp. 753-762.
Lecture 120 – Component Matching (3/25/10) Page 120-20
CMOS Analog Circuit Design © P.E. Allen - 2010
PHYSICAL MATCHINGReview of Physical MatchingWe have examined these topics in previous lectures. To summarize, the sources ofphysical mismatch are:
- Random statistical fluctuations (microscopic fluctuations and irregularities)- Process bias (geometric variations)- Pattern shift (misalignment)- Diffusion interactions- Stress gradients and package shifts- Temperature gradients and thermoelectrics- Electrostatic interactions
Lecture 120 – Component Matching (3/25/10) Page 120-21
CMOS Analog Circuit Design © P.E. Allen - 2010
Rules for Resistor Matching†
1.) Construct matched resistors from the same material. 2.) Make matched resistors the same width. 3.) Make matched resistors sufficiently wide. 4.) Where practical, use identical geometries for resistors (replication principle) 5.) Orient resistors in the same direction. 6.) Place matched resistors in close physical proximity. 7.) Interdigitate arrayed resistors. 8.) Place dummy resistors on either end of a resistor array. 9.) Avoid short resistor segments.10.) Connect matched resistors in order to cancel thermoelectrics.11.) If possible place matched resistors in a low stress area (minimize pieozoresistance).12.) Place matched resistors well away from power devices.13.) Place precisely matched resistors on the axes of symmetry of the die.14.) Consider the influence of tank modulation for HSR resistors (the voltage modulation
of the reverse-biased depletion region changes the sheet resistivity).15.) Sectioned resistors are superior to serpentine resistors.
† Alan Hastings, Art of Analog Layout, 2n d ed, 2006, Pearson Prentice Hall, New Jersey
Lecture 120 – Component Matching (3/25/10) Page 120-22
CMOS Analog Circuit Design © P.E. Allen - 2010
Rules for Resistor Matching – Continued16.) Use poly resistors in preference to diffused resistors.17.) Do not allow the buried layer shadow to intersect matched diffused resistors.18.) Use electrostatic shielding where necessary.19.) Do not route unconnected metal over matched resistors.20.) Avoid excessive power dissipation in matched resistors.
Lecture 120 – Component Matching (3/25/10) Page 120-23
CMOS Analog Circuit Design © P.E. Allen - 2010
Rules for Capacitor Matching†
1.) Use identical geometries for matched capacitors (replication principle). 2.) Use square or octogonal geometries for precisely matched capacitors. 3.) Make matched capacitors as large as possible. 4.) Place matched capacitors adjacent to one another. 5.) Place matched capacitors over field oxide. 6.) Connect the upper electrode of a matched capacitor to the higher-impedance node. 7.) Place dummy capacitors around the outer edge of the array. 8.) Electrostatically shield matched capacitors. 9.) Cross-couple arrayed matched capacitors.10.) Account for the influence of the leads connecting to matched capacitors.11.) Do not run leads over matched capacitors unless they are electrostatically shielded.12.) Use thick-oxide dielectrics in preference to thin-oxide or composite dielectrics.13.) If possible, place matched capacitors in areas of low stress gradients.14.) Place matched capacitors well away from power devices.15.) Place precisely matched capacitors on the axes of symmetry for the die.
† Alan Hastings, Art of Analog Layout, 2n d ed, 2006, Pearson Prentice Hall, New Jersey
Lecture 120 – Component Matching (3/25/10) Page 120-24
CMOS Analog Circuit Design © P.E. Allen - 2010
Mismatched TransistorsAssume two transistors have vDS1 = vDS2, K1’ K2’ and VT1 VT2. Therefore we have
iOiI =
K2’(vGS - VT2)2
K1’(vGS - VT1)2
How do you analyze the mismatch? Use plus and minus worst case approach. Define K’ = K’2-K’1 and K’ = 0.5(K2’+K1’) K1’= K’-0.5 K’ and K2’= K’+0.5 K’
VT = VT2-VT1 and VT = 0.5(VT1+VT2) VT1 =VT -0.5 VT and VT2=VT+0.5 VT
Substituting these terms into the above equation gives,
iOiI =
(K’+0.5 K’)(vGS - VT - 0.5 VT )2
(K’-0.5 K’)(vGS - VT + 0.5 VT)2 =
1 +K’
2K’ 1 -VT
2(vGS-VT)2
1 -K’
2K’ 1 +VT
2(vGS-VT)2
Assuming that the terms added to or subtracted from “1” are smaller than unity givesiOiI ≈ 1 +
K’2K’ 1 +
K’2K’ 1 -
VT
2(vGS-VT)2
1 -VT
2(vGS-VT)2
1 + K’
K’ - 2 VT
(vGS-VT)
If K’/K’ = ±5% and VT/(vGS-VT) = ±10%, then iO/iI 1 ± 0.05 ±(-0.20) = 1±(0.25)
Lecture 120 – Component Matching (3/25/10) Page 120-25
CMOS Analog Circuit Design © P.E. Allen - 2010
Geometric EffectsHow does the size and shape of the transistor effect its matching?Gate Area:
Vth = CVth
W effLeffKp = K’
CKp
W effLeffW/W =
C W/W
W effLeff
where CVth, CKp and C W/W are constants determined by measurement.
Values from a 0.35μm CMOS technology:
Vth,NMOS = 10.6mV·μm
W effLeff Vth,PMOS =
8.25mV·μmW effLeff
and
W
W NMOS = 0.0056·μm
W effLeff
WW PMOS =
0.0011·μmW effLeff
The above results suggest that PMOS devices would be better matched than NMOSdevices in this technology.
Lecture 120 – Component Matching (3/25/10) Page 120-26
CMOS Analog Circuit Design © P.E. Allen - 2010
Pelgrom’s LawSpatial Averaging: Local and randomvariations decrease as the device sizeincreases, since the parameters “averageout” over a greater area.Pelgrom’s Law:
s2( P) = Ap
2
WL + Sp2 Dx
2
where,
P = mismatch in a parameter, PWL = width times the length of the device (effective Pelgrom area)Ap = proportionality constant between the standard deviation of DP and the area of
the deviceDx = distance between the matched devices
Sp = proportionality constant between the standard deviation of P and Dx
As Dx becomes large, the standard deviation tends to infinity which is not realistic.
Threshold mismatch for 0.18 m NMOS
Lecture 120 – Component Matching (3/25/10) Page 120-27
CMOS Analog Circuit Design © P.E. Allen - 2010
Rules for Transistor Matching†
1.) Use identical finger geometries. 2.) Use large active areas. 3.) For voltage matching, keep VGS-VT, small ( i.e. 0.1V). 4.) For current matching, keep VGS-VT, large (i.e. 0.5V). 5.) Orient the transistors in the same direction. 6.) Place the transistors in close proximity to each other. 7.) Keep the layout of the matched transistors as compact as possible. 8.) Where practical use common centroid geometry layouts. 9.) Place dummy segments on the ends of arrayed transistors.10.) Avoid using very short or narrow transistors.11.) Place transistors in areas of low stress gradients.12.) Do not place contacts on top of active gate area.13.) Keep junctions of deep diffusions as far away from the active gate area as possible.14.) Do not route metal across the active gate region.15.) Place precisely matched transistors on the axes of symmetry of the die.16.) Do not allow the buried layer shadow to intersect the active gate area.17.) Connect gate fingers using metal connections.
† Alan Hastings, Art of Analog Layout, 2n d ed, 2006, Pearson Prentice Hall, New Jersey
Lecture 120 – Component Matching (3/25/10) Page 120-28
CMOS Analog Circuit Design © P.E. Allen - 2010
SUMMARY• IC technology offers poor absolute values but good relative values or matching• In analog circuits, gains are determined by ratios (good matching) and time constants
are determined by products (poor matching)• Electrical matching is determined in the electrical design phase
- Matching due to equal terminal voltages- Matching due to process independent biasing- Doubly correlated sampling- Self-calibration techniques- Dynamic element matching
• Physical matching is determined in the physical design phase- Random statistical fluctuations (microscopic fluctuations and irregularities)
- Process bias (geometric variations)- Pattern shift (misalignment)- Diffusion interactions - Stress gradients and package shifts- Temperature gradients and thermoelectrics- Electrostatic interactions