6.

775 Midterm Project

65nm CMOS op amp design

Nathan Monroe

[email protected]

4/4/2016

Folded Cascode – Common Source Amplifier

 2Specifications

Specification Requirement This design Notes

1. Settling Time (0.1%) 8ns max up/down 5.3ns (up) 4.8ns down
2. Power Consumption 3mW 2.42mW
3. Output resistance 1Ω 0.28 Ω Closed loop, small sig
4. Thermal Noise 400uV RMS 308uV RMS Output referenced,
10kHz-500MHz
5. Small signal gain 5,000 6,168 No load, DC
6. Output Swing 0.2V – 1.0V 0.068V-1.09V No-load
7. Phase Margin 60o 73.8o 1pF load, gain of 2
8. PSRR+ (vdd) 60dB at DC 100.1dB at DC
40dB at 50KHz 100.0dB at 50KHz
9. CMRR 60dB 80.3dB At Vcm = 0.6V

Component Details (left to right on schematic)

All transistors have L=100nm except M17, which has L=65nm.

RefDes Type Size Notes RefDes Type Size Notes

M19 Nch W=100nm Bias2 gen M14 Pch W=30.4um I source
M18 Nch W=100nm Bias2 gen M7 Pch W=30.4um I source
M6 Pch W=3.04um Bias1 gen M2 Pch W=15.22um Cascode
I7 I source 30uA Reference M21 Pch W=15.22um Cascode
current
M15 Pch W=3.04um I mirror M3 Nch W=15.22um Cascode
I source
M17 Nch W=29.51um Bias3 gen M4 Nch W=15.22um Cascode
I source
M16 Nch W=2.9um Bias M9 Nch W=15.22um Cascode
I source
M11 Nch W=15.22um Diff Pair M10 Nch W=15.22um Cascode
I source
M12 Nch W=15.22um Diff Pair R0 Resistor R=100 ohm Comp
M13 Nch W=30.4um I source C1 Capacitor C=400fF Comp
M8 Pch W=130.6um I source
M5 Nch W=121.6um Gain
 Hand Design
The initial design topology was performed based on consideration of gain, settling time, and phase
margin specifications, and power supply voltage and current constraints. A folded cascode topology was
chosen based on high gain with reasonable device sizes, allowing for high bandwidth.

Gain
(𝑔𝑚 𝑟𝑜 )2
The first stage of this amplifier (folded cascode) provides a gain equal to 4
due to the current gain
of the differential pair and output resistance of cascode stages and active load. The second stage
𝑔 𝑟
provides 𝑚2 𝑜 due to the current gain of the nmos and the parallel output resistance of the gain
(𝑔𝑚 𝑟𝑜 )3
transistor and the active load current source. Therefore the total gain of the amplifier is . The gain
8
target is 5000, I will choose conservatively 6000 for engineering margin. This requires transistors with a
3
gain of √6000 ∗ 8 = 36.3.

Settling Time

Settling time from a step input is based on Gain-bandwidth of the amplifier. Required GBW is
−𝑇𝑠
determined by 𝑘 = 𝑒 𝜏 . This gives a GBW = -ln(0.001)/(8ns) = 875MHz. For adequate phase margin, Ft
of the transistors is desired to be 10X the unity gain bandwidth, so this is set at ~8.7GHz.

Gm requirements

The compensation capacitor is conservatively set to be equal to one half the load capacitance, or 0.5pF.
Gm of the first stage is determined by the corresponding pole, and is equal to

Gm1 = (875MHz) * (2π rad/Hz) * (0.5pF) = 2.74mS.

Gm of the second stage is determined by the corresponding pole, and is equal to

𝐶
Gm2 = 4 ∗ 𝐶 𝑙 ∗ 𝐺𝑚1 = 21.9mS.
𝑐

Based on requirements for Gm (2.74mS and 21.9mS), Ft (8.7GHz) and gain (36.3), the transistors were
chosen to have a length of 100nm and current density of 10uA/um for both PMOS and NMOS. This was
chosen graphically based on small signal parameters derived by OCEAN script, parameters given below.
Vdsat was also taken into consideration to allow for reasonable device headroom within power supply
voltage, in order to meet output swing specification.

Device Gm/um Ft Gain Vgs-Vt for desired Vdsat

current density
NMOS 0.18 mS/um 28GHz 41 382mV 62mV
PMOS 0.12 mS/um 11GHz 31 470mV 150mV
Input stage gain transistor sizing

The width of the input stage M11 and M12 was chosen based on the above gm/um and required Gm.
This results in
2.74𝑚𝑆
𝑊= = 15.22𝑢𝑚
0.18 𝑚𝑆/𝑢𝑚
The wide input transistors also provide for low noise, with noise from the input transistors dominating
overall noise. This is important because noise from the first stage is amplified and directly reflected in
input-referred noise. This will be discussed in more detail later. Based on this width and the chosen
current density, the current in each transistor M11 and M12 is
10𝑢𝐴
15.22𝑢𝑚 ∗ = 152.2𝑢𝐴
𝑢𝑚
First stage Differential Pair tail current source sizing

The current through the tail current source M13 is equal to double the current through each transistor
M11 and M12, or 2*152.2uA = 304.4uA. Based on the chosen current density, this gives the width of
M13 as
304.4𝑢𝐴
𝑊= = 30.4𝑢𝑚
10 𝑢𝐴/𝑢𝑚
First stage cascode / current source transistors sizing

The current through the folded cascode leg of the first stage (M2, M21, M3, M4, M9, M10) is chosen to
be equal to the current in each differential pair transistor, or 152.2uA per leg. This results in a width for
these transistors which is equal to the differential pair transistors, or 15.22um.

First stage PMOS current source transistors sizing

The PMOS current source transistors M14 and M7 supply current which is shared between the
differential pair leg and the cascode leg. Because these legs are equal in current at 152.2uA, the current
supplied by these PMOS transistors is double this current, or 304.4uA. Based on the chosen current
density, this gives the width for M14 and M7 as
304.4𝑢𝐴
𝑊= = 30.4𝑢𝑚
10 𝑢𝐴/𝑢𝑚
Second stage gain transistor sizing

The width of the second stage gain transistor M5 was chosen based on the above gm/um and required
Gm. This results in
21.9𝑚𝑆
𝑊= = 121.6𝑢𝑚
0.18 𝑚𝑆/𝑢𝑚
Based on this width and the chosen current density, the current in the second leg is
10𝑢𝐴
121.6𝑢𝑚 ∗ = 1216𝑢𝐴
𝑢𝑚
Second stage current source transistor sizing

The PMOS current source M8 must supply the second stage with 1216uA (calculated above). Based on
the chosen current density, this gives a width equal to
1216𝑢𝐴
𝑊= = 121.6𝑢𝑚
10 𝑢𝐴/𝑢𝑚
This value was adjusted after simulation to minimize systematic offset.

Current reference and transistor sizing

The current source I7 value was chosen as 30uA for minimal power overhead. The reference transistor
M6 sizing is given by the relative currents and width ratios:
𝑊 𝑊
𝐼𝑟𝑒𝑓 30𝑢𝐴 𝐿 𝑟𝑒𝑓 𝑟𝑒𝑓
= = = 𝐿
𝐼𝑑𝑒𝑠𝑖𝑟𝑒𝑑 304𝑢𝐴 𝑊 𝑑𝑒𝑠𝑖𝑟𝑒𝑑 304
𝐿
Therefore, with L=100nm the width of M6 is equal to 3.04um. The node bias1 supplies the bias to the
PMOS current sources.

Current mirror and input pair current source bias transistor sizing

As described before, current is sunk from the input differential pair by an NMOS transistor M13. M13 is
biased by a current mirror comprising of M15 (PMOS) and M16 (PMOS). These transistors are sized such
that the current through this leg is 30uA for minimal power overhead. This also provides the bias voltage
to M13 based on the W/L ratios as before, at 300uA. Therefore, M15 and M16 are sized to be equal to
M6 at W=3.04uA. The width of M16 is adjusted in the real design, see below. The bias voltage is
approximately 382mV as determined above.

Cascode transistor biasing

Based on the above choice of current density, the desired DC operating point of the first stage output is
382mV. The cascode transistors M3/M4 in the NMOS current sink can be thought of in terms of bias as
source follower transistors. Their bias voltage (bias3) is equal to Vds of the lower transistors M9/M10
plus Vgs required for desired current density. To keep M9/M10 in saturation Vds > 62mV. Therefore,
bias3 > Vds,sat + Vgs = 62mV + 382mV = 444mV. This is the lower bound for bias3. It’s known that the
DC output of the first stage must be equal to 382mV to properly bias the second stage. Because
transistors M3/M4/M9/M10 are carrying equal current densities, and Vd(M4)=382mV, Vd(M9/M10) is
approximately half of this, or 191mV. Therefore, bias3 is approximated to be 382+191 =573mV. The real
value was determined by simulation in order to set the desired operating point of the first stage output,
and was found to be bias3=620mV. Bias3 is generated by a diode-connected transistor M17 inserted
into the current mirror. With the source voltage of M17 equal to 382mV (see above), for M17
Vds=Vgs=620mV-382mV=238mV. As determined by a Vgs versus drain current density from OCEAN plot,
an appropriate choice of size for M17 is L=65nm, W=29.51um. This was also adjusted in simulation. M16
was also adjusted in size based on simulation, due to the addition of M17.
The PMOS cascode transistors M2 and M21 are also biased for the desired current density, with Vgs =
470mV. They also act as source followers, and the bias point is constrained to keep PMOS current source
transistors M14 and M7 in saturation (Vdsat = 150mV). Therefore, the upper bound of the bias point
bias2 is 1200mV – 150mV – 470mV = 580mV. It is known that the drain voltage of M21=382mV for
proper biasing of the second stage. Because PMOS transistors M14/M7/M2/M21 are carrying equal
current densities, it is approximated that their Vds’s are equal. Therefore, the voltage of the internal
node between M7/M21 or M14/M2 is approximately 1200- (1200-382)/2 = 795mV. Therefore, by
subtracting Vgs for the PMOS transistors, the appropriate bias point for transistors M2/M21 is 795-
470mV = 325mV. The final value was determined by simulation to be 300mV. 300mV for bias2 was
generated by a transistor voltage divider from the mid-rail reference of 600mV. Small transistors M18,
M19 of L=W=100nm provide for small current consumption.

Compensation design

Hand design for compensation was performed in a similar methodology to problem set 3. An initial
guess for Cc was Cl/2, or 0.5pF. This conservative estimate provides for approximately 72 degrees of
phase margin at unity gain. The gain of 2 phase margin is even better than this. This value was later
adjusted in simulation to 0.4pF. Rz is constrained based on placement of the nulling resistor for the RHP
zero. It is constrained in the following way, using similar arguments to problem set 3:
1 1
< 𝑅𝑧 <
𝑔𝑚12 3 ∗ 𝑔𝑚5
Therefore, plugging in values

45.6 𝑜ℎ𝑚 < 𝑅𝑧 < 121.6 𝑜ℎ𝑚

Rz was initially chosen to be 100 ohms. This was confirmed in simulation to be an optimal value. These
hand calculations are based on the assumption that the poles are sufficiently split that the lower
frequency pole dominates in phase margin.

1. Settling Time
The op amp was designed with a GBW based on consideration of settling time (see calculations above).
A GBW of 875MHz was chosen. With a DC gain of 5000, this results in a dominant pole of
875MHz/5000=142KHz. Settling time was measured using the following testbench:
Due to minor offset, the final values are 699.9 and 500.1mV, so the 0.1% limits are 699.7-700.1mV and
499.9-500.3mV. Settling time was measured to be 5.3ns up and 4.8ns down to 0.1%. It is expected that
the slew upwards is longer than the slew downwards, as the upwards slew is provided by a current
source while the downwards is provided by the gain stage transistor. This was confirmed in simulation.
Waveforms of the settling time are shown below.

Settling time (both):

Zoomed in settling time (down)

Zoomed in settling time (up)

 2. Power Consumption
Current consumption from the first stage differential pair was estimated to be 2*152.2uA = 304.4uA.
Current consumption from the first stage cascade legs was estimated to be 2*152.2uA = 304.4uA.

Current consumption from the second stage was estimated to be 1.216mA.

Current consumption from two current mirrors at 30uA each is 60uA.

Current consumption from the bias2 voltage divider was estimated to be negligibly small.

Therefore, the estimated current consumption was estimated to be

(304.4uA + 304.4uA + 1216uA + 60uA) = 1884.8uA

Estimated power is then (1.2V * 1.88mA) = 2.25mW.

Power was measured using the same testbench as above, while holding inputs at 0.6V. As seen below,
power was measured to be 2.42mW. Current consumption from the 0.6V supply was measured to be
within specification at 138nA. Real current being slightly above estimates is primarily attributed to the
second stage being biased at a slightly higher current density than originally calculated.
 3. Output Resistance
Open-loop output resistance is the parallel combination of the output resistances of the output stage
transistor and it’s active-load current supply:

𝑅𝑜𝑢𝑡 = 𝑅𝑜𝑢𝑡𝑀5 || 𝑅𝑜𝑢𝑡𝑀8

Based on OCEAN small-signal output, Rout of NMOS at 10uA/um = 220k ohm * um.

Rout of PMOS at 10uA/um = 250k ohm*um. Therefore open loop Rout is estimated to be
220𝑘 ∗ 𝑢𝑚 250𝑘 ∗ 𝑢𝑚
𝑅𝑜𝑢𝑡𝑂𝐿𝑝𝑟𝑒𝑑𝑖𝑐𝑡𝑒𝑑 = || = 920𝑜ℎ𝑚.
121.6𝑢𝑚 130.6𝑢𝑚
Therefore, closed loop Rout in a gain of 1 amplifier is predicted to be
𝑅𝑜𝑢𝑡𝑂𝐿 920
𝑅𝑜𝑢𝑡𝐶𝐿 𝑝𝑟𝑒𝑑𝑖𝑐𝑡𝑒𝑑 = = = 368𝑚 𝑜ℎ𝑚
𝐿𝑜𝑜𝑝 𝑔𝑎𝑖𝑛 2500

Output resistance is measured in two ways, using the following schematic testbench. After biasing the
inverting amplifier at it’s operating point, a small-signal test voltage is applied to the output and a small-
signal test current observed.
The output voltage and current waveforms are as follows:

Output resistance is measured to be

𝑉𝑜𝑢𝑡𝑝𝑝 200𝑢𝑉
𝑅𝑜𝑢𝑡𝑚𝑒𝑎𝑠𝑢𝑟𝑒𝑑 = = = 279𝑚 𝑜ℎ𝑚
𝐼𝑜𝑢𝑡𝑝𝑝 716𝑢𝐴

This is confirmed by HSPICE calculation of small-signal Gds of the two output transistors:
1 1 1 1865 || 1683
𝑅𝑜𝑢𝑡𝑚𝑒𝑎𝑠𝑢𝑟𝑒𝑑 = ( || )= = 176𝑚 𝑜ℎ𝑚
𝐿𝑜𝑜𝑝 𝐺𝑎𝑖𝑛 𝐺𝑑𝑠𝑀5 𝐺𝑑𝑠𝑀8 5000

4. Thermal Noise
For hand calculations of thermal noise, the second stage is neglected- noise is dominated by the first
stage, whose noise is subjected to the gain of the entire amplifier. Input-referred noise spectral density
of the first stage has been derived in the same manner as pset 4 problem 3 and is equal to (derivations
omitted here for brevity):
̅̅̅̅̅̅
𝑉𝑖𝑡 2 ̅̅̅̅̅̅̅̅̅̅̅̅̅2 ∗ ( 𝑔𝑚𝑚9,10 )2 + 𝑉𝑒𝑞
̅̅̅̅̅̅̅̅̅̅̅̅̅̅2 + 𝑉𝑒𝑞
= 2 ∗ (𝑉𝑒𝑞 ̅̅̅̅̅̅̅̅̅̅̅̅̅2 ∗ ( 𝑔𝑚𝑚7,14 )2 )
𝑚11,12 𝑚9,10 𝑚7,14
∆𝑓 𝑔𝑚𝑚11,12 𝑔𝑚𝑚11,12

Substituting in thermal noise,

4𝑘𝑇 ∗ 3 4𝑘𝑇 ∗ 3 4𝑘𝑇 ∗ 3
=2∗( + ∗ 𝑔𝑚𝑚9,10 + ∗ 𝑔𝑚𝑚7,14 )
2 ∗ 𝑔𝑚𝑚11,12 2 ∗ 𝑔𝑚𝑚11,12 2 2 ∗ 𝑔𝑚𝑚11,12 2

6 ∗ 300 ∗ 1.3 ∗ 10−23 6 ∗ 300 ∗ 1.3 ∗ 10−23 6 ∗ 300 ∗ 1.3 ∗ 10−23

=2∗( + + ∗ (30.4 ∗ 10−4 ))
. 00274 . 00274 (. 00274)2
 ̅̅̅̅̅̅
𝑉𝑖𝑡 2 𝑉2
= 5.29 ∗ 10−17
∆𝑓 𝐻𝑧
Multiplying by bandwidth,
̅̅̅̅̅̅
𝑉𝑖𝑡 2 = 5.29 ∗ 10−17 ∗ 499.99 ∗ 106 = 2.65 ∗ 10−8 𝑉 𝑟𝑚𝑠
To calculate output-referred noise, the input referred noise is multiplied by the DC gain of 6168:
̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
2
𝑉𝑜𝑡𝑝𝑟𝑒𝑑𝑖𝑐𝑡𝑒𝑑 = ̅̅̅̅̅̅
𝑉𝑖𝑡 2 ∗ 𝑎𝑣 = 2.65 ∗ 10−8 ∗ 6168 = 132𝑢𝑉 𝑟𝑚𝑠

The choice to refer input noise to the output by multiplying by DC gain is not rigorous, as the amplifier’s
gain rolls off quickly at the first pole (140KHz), and so the gain is much lower over much of the noise
integration bandwidth (up to 500MHz). However, for conservative design and ease of hand calculations
the DC gain was used over the entire frequency spectrum.

Thermal noise was determined empirically using the same testbench setup as in the settling time
measurement, with both inputs held at 600mV. Noise was calculated by OCEAN script, results are given
below for both RMS noise and noise PSD. RMS noise over specified bandwidth was found to be
315.5uV2. This value is slightly higher than was calculated in hand design as it also includes shot noise
and flicker noise, which dominates at low frequencies such as in the specified bandwidth. A small part of
the discrepancy may also be explained by the omission of the second stage in thermal noise calculations.
As expected, the dominating noise sources are the input differential pair and the current sources in the
first stage.
 5. Small Signal Gain

Small signal gain was determined by design, as calculated above. Based on a specified gain of 5000, a
gain of 6000 was chosen for margin. The open-loop gain was measured using the test circuit below.
Gain was measured to be 6168 at DC:

6. Output Swing
Due to the gain of the second stage, the output transistors will saturate before transistors in the first
stage. Therefore, output swing in this topology is largely determined by Vds,sat of the output stage
transistors. This is 62mV and 150mV for the NMOS and PMOS, respectively. Therefore, in a first-order
hand approximation the output swing will be from 62mV to (1200-150)=1050mV. The testbench used is
the same as that used in DC gain measurement, with no load. The noninverting input was held at
600mV, while the inverting input was swept in a DC analysis from 599mV-601mV, while observing the
output voltage (see below, right waveform). By differentiating the output voltage, it is possible to
determine the gain as a function of output voltage (see below, left waveform). As seen below, at 0.2V
and 1.0V the gain of 1000 requirement is easily met, with the real gains being 2919 and 5345,
respectively. The gain of 1000 crossover occurs at 68mV and 1090mV, well within specification and
aligning well with the hand estimation. The output stage was biased at a slightly higher current than was
calculated in hand design (1.21mA for hand calculation vs 1.36mA in simulation). The increased Vdsat
associated with this increased current density likely explains the worse output swing performance than
was predicted.
 7. Phase Margin
Phase margin was predicted to be 72o by design (see above). Phase margin was measured in an open-
loop configuration with 1pF load (see below). A small signal differential input was applied, and output
phase observed. Phase margin was found to be 58o at unity gain and 73.8o at a gain of 2 crossover. The
design was based on a 72o phase margin at unity gain. The true unity gain phase margin is much worse
than this due to the pole associated with the large parasitic capacitance and large output resistance of
the first stage, as a consequence of the cascode topology. The phase margin at gain of 2 crossover is an
improvement from hand calculations because the hand calculations were based on phase margin at
unity gain, and measured phase margin is better at gain of 2 crossover because it occurs at a lower
frequency.
 8. PSRR
As derived in lecture, PSRR stems from two root causes. The first is ripple on the power supply coupling
through the PMOS current sources in the first stage and appearing analogously to a common-mode
input. This impact is identical to that of a common mode input (see below for CMRR calculation)

𝑎+1 = −𝑎𝑣𝑐1 𝑎2 = −0.39

The second cause for PSRR is ripple on the positive rail seen through the voltage divider of the output
stage. Therefore,
𝑟𝑜5 1800
𝑎+2 = = = 0.48
𝑟𝑜5 + 𝑟𝑜8 1800 + 1920
This occurs in the opposite polarity of the first stage’s PSRR impact, so some beneficial cancellation of
PSRR occurs. The total power supply gain is the sum of these two individual gain contributions, a+ = 0.48-
0.39 = 0.09. Therefore, the predicted PSRR at DC is equal to
𝑎𝑣𝑑 6168
𝑃𝑆𝑅𝑅+𝑝𝑟𝑒𝑑𝑖𝑐𝑡𝑒𝑑 = −20 log ( ) = −20 log ( ) = 97𝑑𝐵
𝑎+ 0.09
At 50KHz, capacitive effects are negligible, so a+ is largely unchanged. The dominant pole is at 140KHz,
so the differential gain is largely unchanged as well. Therefore, the predicted PSRR at 50KHz is well
approximated by the PSRR at DC.
PSRR was measured by applying a small signal voltage on the positive supply, VDD (see below) in an AC
analysis. The gain from vdd to output, A+, is defined as the amplitude ratio of vdd and the output node:
|𝑜𝑢𝑡𝑝𝑢𝑡|
𝐴+ =
|𝑣𝑑𝑑|
PSRR is defined as the gain ratio between the differential gain and vdd gain, and the measured values
are:

𝐴𝑣𝑑 0.061
𝑃𝑆𝑅𝑅𝑚𝑒𝑎𝑠𝑢𝑟𝑒𝑑 = −20 log10 ( ) = −20 log10 ( ) = 100.1𝑑𝐵 (𝑎𝑡 𝐷𝐶)
𝐴+ 6168
0.0579
= −20 log10 ( 5800 ) = 100.0𝑑𝐵 (𝑎𝑡 50𝐾𝐻𝑧)
 9. CMRR
For CMRR hand analysis, the input stage was divided into a half-circuit, as seen below (left). It
further simplifies as seen on the right.

Plugging in numbers and iterating KCL on the drain of M12, the gain from common mode input to
output of the first stage is found to be 0.022. Multiplying by the gain of the second stage, the total
common mode gain is

𝑎𝑣𝑐 = 𝑎𝑣𝑐_𝑠𝑡𝑎𝑔𝑒1 ∗ 𝑎𝑣𝑠𝑡𝑎𝑔𝑒2 = 0.022 ∗ 18 = 0.39

Therefore, the CMRR is predicted to be the following:

𝑎𝑣𝑑 0.39
𝐶𝑀𝑅𝑅𝑝𝑟𝑒𝑑𝑖𝑐𝑡𝑒𝑑 = −20log10 ( ) = −20𝑙𝑜𝑔10 ( ) = 83𝑑𝐵
𝑎𝑣𝑐 6168
Common mode rejection ratio was measured using the same test circuit as in PSRR (see above, PSRR
section). A common-mode input voltage was applied in an AC analysis (around Vcm=600mV) and
small-signal output amplitude observed. The common mode gain is defined as the amplitude ratios
of the common mode input and the output:
|𝑜𝑢𝑡𝑝𝑢𝑡|
𝐴𝑣𝑐 =
|𝐶𝑀 𝑖𝑛𝑝𝑢𝑡|

The common mode rejection ratio is defined as the gain ratio of the common and differential modes
and was measured to be
𝐴𝑣𝑑 0.5939
𝐶𝑀𝑅𝑅𝑚𝑒𝑎𝑠𝑢𝑟𝑒𝑑 = −20 log10 ( ) = −20 log10 ( ) = 80.3𝑑𝐵 (𝑎𝑡 𝐷𝐶)
𝐴𝑣𝑐 6168
 Discussion
There were many tradeoffs inherent in this design, as many design variables and specifications are
interconnected. The choice of a folded cascade-common source amplifier allowed for large gain and
output swing while still fitting within the low voltage power supply. The FC-CS topology has a downside
of high power consumption, making the power consumption specification a challenge. In addition, the
high output resistance of the first stage increases noise over many other topologies, forcing wider
transistors and more current in the first stage to compensate and meet specification. The high internal
parasitic capacitance and output resistance associated with the first stage negatively impacts frequency
response and phase margin. Fortunately, despite this it was possible to meet the settling time
specification with good stability.

The choice of compensation capacitor and resistor, chosen for unity-gain stability, resulted in an
excessively conservative phase margin of 74o at a gain of 2 crossover. This is also seen in the step
response, which is overdamped and approaching that of a first order system (90o phase margin). While
settling time specifications were met, settling time could have been improved by altering compensation
to reduce phase margin while still remaining stable.

The design gain of 6000 was somewhat excessive and could be reduced to save area/cost by reducing
device length. This would come at the cost of PSRR and CMRR (both well above specification) while
simultaneously reducing vdsat (improving output swing) and reducing output resistance.
The biasing structure for the cascade transistors could have been optimized. The ½ voltage divider on
the mid-rail supply could have instead been implemented as a ¼ divider from vdd, reducing the need for
the mid-rail supply in the circuit, thus reducing interconnect cost and complexity. In addition, the
decision to derive the bias from the mid-rail supply makes the circuit sensitive to noise on the mid-rail
supply in addition to the power supply. This was not given in the specifications but could be an issue in a
real system. In addition, both the voltage divider scheme for generating bias2 and the diode-connected
transistor for generating bias3 are sensitive to process variation, and a better bias scheme could be used
in future iterations. Despite these possible improvements, a successful design was implemented which
meets the required specifications.