LMC2001_PRE_TEAM_Silicon
The following is actual data and curves taken off the LMC2001 before
the LMC2001 team got complete control of the silicon. This version of
silicon will be referred to as the
Pre_Team_LMC2001_Silicon.
...Don Sauer 10/19/09 dsauersanjose@aol.com
===================Hide_1/F_Noise==========================
Was it possible to completely hide all the 1/f noise of a CMOS input stage?
Up until first silicon, this was for the most part theoretical. Other
choppers amps were not claiming they had killed 1/F noise at that time.
Everyone seem to want to describe it as just shifting 1/f noise and offset
to the clock frequency.
^ dB
Main Amp /|\
___ |\ "High Speed Amp" |____ Low Gain
| |_____|+\ ___ | .\
|___| | | \_____| | | . \
___ | | / |___| |___.__\___\
| |_\___|-/_ 1KHz / Freq
|___| / | |/|\ ^
| | \ Composite /|\ dB
| | \ Amplifier |
| | |\ \ |\ Low speed
| |_|+\ | | \
| | \___| | \
| | / |___\______\
|___|-/ "Precision Amp" 1KHz / Freq
|/
"Chopper"(more like Mod/Demod)
The Pre_Team_LMC2001 was a composite amplifier where a "chopper"
amplifier was being used to dominate gain at frequencies below
1KHz. Any offset and 1/f noise would therefore get attenuated by
how much extra gain the precision amp was adding to the overall
gain. For instant adding 40dB at 10Hz of extra gain should attenuate the
1/f noise of the main amplifier by 40dB at 10Hz. To see this actually
work on first silicon, one only needed to cut out the "chopper" stage.
Noise 1/F Cancelation (Pre_Team_LMC2001)
Voltage ...................................
| . . .
Before . . .
| . . .
| Before . . .
| . . .
100nV|....................................
| Before . .
| . . .
| . Before. .
| . Before .
|After After . Before .
30nV |................After............Before
| . After .
| . . After .
| . . After
| . . .
| . . .
10nV |____________________________________
10Hz 100Hz 1000Hz 10KHz
FREQUENCY
Without the "chopper", the normal 1/f of the CMOS input stage of
the main amplifier comes thru. After the "chopper" signal is
added, it appeared that the noise goes flat all the way down to
DC. Indeed, this is where things got interesting. Things like metal
to metal thermal coupling start to come into the picture when measuring
things at low frequency.
===================Composite_Noise==========================
^ dB
Main Amp /|\
___ |\ "High Speed Amp" |____ Low Gain
| |_____|+\ ___ | .\
|___| | | \_____| | | . \
___ | | / |___| |___.__\___\
| |_\___|-/_ 1KHz / Freq
|___| / | |/|\ ^
| | \ Composite /|\ dB
| | \ Amplifier |
| | |\ \ |\ Low speed
| |_|+\ | | \
| | \___| | \
| | / |___\______\
|___|-/ "Precision Amp" 1KHz / Freq
|/
"Chopper"(more like Mod/Demod)
The Pre_Team_LMC2001 really used more of a modulation method to
modulate DC signal up to a high enough frequency where some high
quality gain could be taken. Then the signal would get
demodulate back to DC and used to cancel out input offset.
In previous choppers, the DC and 1/f would get simply modulated up
to the chopper frequency.
______ ______ ______
VIN1 | |OUT1 | _ |VSUM | | ____
_____|\| \ / |____|\| / |_______|\| \ / |___|OUT2|
_|_ |/| X | |/| _\ | |/| X | |____|
/VIN\ | / \ | |______| | / \ |
\___/ |______| ^ BSUM |______|
_|_ ^ BMUX1 /_\ ^ BMUX2
/// /_\ _|__ 1/F noise /_\
|_____ /_ \ |
_|_ | // \ \ |
VCLK /_ \ | \ \_// |
/| | \ | \____/ |
\ |_|/ | _|_ |
\___/ | /// |
| |________________________|
_|_
///
But in a composite architecture, could the chopper artifacts be
placed at a frequency high enough where the main amplifier would
attempt to swamp out any "Chopper Noise Artifacts" ?
Noise
Voltage Noise Voltage Spectrum (Pre_Team_LMC2001)
60nV |................................................
| . . ^ . .
| . . /|\ . .
| . . | . .
| . . | . .
| . . C . .
50nV |................................H...............
| . . O . .
| . . P . .
|+SD +SD .+SD . P . .
| +SD . E . .
| . R . .
40nV |Ave.....Ave..........+SD........|...............
| . Ave . | . .
| . . . .
|-SD -SD . -SD Ave +SD . .
| . . +SD . .
| . -SD . Ave .+SD +SD .
30nV |........................-SD.......Ave......Ave..
| . . -SD -SD .
| . . . .
| . . . .
| . . . .
| . . . .
20nV |______________________________________________ .
10Hz 100Hz 1KHz 10KHz 100KHz
FREQUENCY
The graph above shows actual data taken off of Pre_Team_LMC2001 silicon
when the spread spectrum circuitry had been disabled. It
was easy to
see the"chopper" artifacts by looking at the input noise voltage.
And there are two different flatband noise levels.
Below 1kHz, the "chopper" amplifier dominates the noise.
Above 1KHz, the main amplifier dominates the noise.
===================Truly_Random_Spread_Spectrum==========================
Noise is practically always defined by the input transistors.
For first silicon, the input CMOS transistors
were not made super large to lower the input noise voltage.
Everything was being tested out for the first time, and more
normal sizes were being used. But later version of this architecture
did appear to attempt to exploit this opportunity.
______ ______ ______
VIN1 | |OUT1 | _ |VSUM | | ____
_____|\| \ / |____|\| / |_______|\| \ / |___|OUT2|
_|_ |/| X | |/| _\ | |/| X | |____|
/VIN\ | / \ | |______| | / \ |
\___/ |______| ^ BSUM |______|
_|_ ^ BMUX1 /_\ ^ BMUX2
/// /_\ _|__ 1/F noise /_\
|_____ /_ \ |
_|_ | // \ \ |
VspreadSpec /_ \ | \ \_// |
/| | \ | \____/ |
\ |_|/ | _|_
\___/ | /// |
| |________________________|
_|_
///
The Pre_Team_LMC2001_Silicon actually used shot noise to spread spectrum
the clock. With the inputs leads biased up incorrectly,
it was possible to see the clock feed-thru at the inputs.
With the spread spectrum cut away, the clock was periodic.
| | | | |
_|\_____|\_____|\_____|\_____|\......
With the spread spectrum in place, the dither was more apparent.
| | | | |
_|\___|\________|\__|\____|\......
The goal was to be able to completely hide all "chopper" artifacts
such that the noise voltage would look flat from DC to 10Mhz.
The Pre_Team_LMC2001_silicon was very encouraging.
________ "The precision Stage"
____________| Clock |________
| |________| |
\|/ \|/
V CMOS|\ _ _ __V_ Gilbert
| \/ \/ \ / \ ___
________| / /\ \__||__/ \/ \_______| |
| | \ \/ / || \ /\ / | |___|
V | /\_/\_/ \____/ _|_
___/\ /\ ___ |/ Blocking ___ Integration
| \/ \/ | Cap _|_ Cap
_|_ _|_ ///
|+IN| |-IN|
|___| INPUTS |___|
Noise Voltage Spectrum (Pre_Team_LMC2001)
Noise ...................................
Voltage | . . .
| . . .
| . . .
| . . .
| . . .
100nV|....................................
| . . .
| . . .
| . . .
| . . .
|Noise Noise Noise . -- Noise .
30nV |........................./..\.......
| . ./ \ .
| . . .
| . . .
| . . .
| . . .
10nV |____________________________________
1000Hz 2000Hz 6000Hz 10KHz
FREQUENCY
When the output of a spectrum analyzer was averaged for a while, it
was possible to see the spread spectrum chopper artifacts
as a mild bump on the input noise curve. Give the composite
architecture, there was great potential to bury all artifacts
of the precision amplifier even deeper inside the main amplifier's
flat band noise.
===================Not_Really_A_Chopper==========================
Modulation Not chopping
___
| |____________________________
|___| | | |
| 1Meg | 1K | Main Amp
|_/\ /\_ |_/\ /\_ |__|\
\/ | \/ | |-\ ___
| | | \__| |
_/\ /\_| _/\ /\_| | / |___|
| \/ | | \/ | __|+/
___ | 1K | | 1Meg | | |/
| |_|_______/|\___|_______/|\__|
|___| | |
V V
To Chopper
The Pre_Team_LMC2001 was using a modulation method in an
attempt to hide input current spikes. To a certain extent, CMOS switch
feed-thru can be minimize through the use of dummy transistors.
But when a CMOS transistor completely turns off, were the charge goes
that formed the channel is not real obvious. There was no real
good way to cancel out all the gate signal feed-thru.
R1 ___ VCC ^
_/\ /\ /\_|IN1|_____________________ /_\
_|_ \/ \/ |___| | | |
/// ^ _| ^ _| |
|_|| MN1 |_|| MN2 |
||__ ||__ __| MN6
CLK1 ||-> CLK2||-> IN3 _| |_
| |________||NM5 ||_
| | ||______|| |
|___________/|\___ ||-> <-|| |
CLK2 | | | |__| |
^ _| ^ _| | IN4 | |
|_||MN3 |_||NM4 |_____/|\______|
||__ CLK1 ||__ _|_
R2 ||-> ||-> / _ \ V1
___ | | \/ \/
_/\ /\ /\__|IN2|_______|____________| I1 /\_/\
_|_ \/ \/ |___| \___/
/// _|_
///
A "chopping like action" was being done without turning off
CMOS transistors. By shifting the resistances of CMOS channels around,
a more of a gain modulation was taking place. The CMOS channels
were designed to alternate between 1Kohm to 1Megohm. The dummy transistors
were biased up the same way in the hope that feed-thru cancellation
would be much better.
(Pre_Team_LMC2001)
Feedthru Noise current @100Kohms
Noise ................................... Noise
Voltage | . . . Current
| . . .
| . . .
| . . .
| . . .
100nV|.................................... 1pA
| . . .
| . . .
| . . .
| . __ . .
|40n 40n 40n FT FT 40n 40n .
30nV |............../......\.............. 0.3pA
| . FT FT. .
| ./ \ .
| . . .
| . . .
| . . .
10nV |____________________________________
1000Hz 2000Hz 6000Hz 10KHz
FREQUENCY
Above is as far as the Pre_Team_LMC2001_silicon got into the challenge
of minimizing input feed-thru current spikes. The potential for improvement
in this area is unknown since this is when the LMC2001 team
started to get control.
===================RRIO_Op_AMP==========================
The Pre_Team_LMC2001 silicon was a full RRIO Op Amp.
It's offset over the full input range is shown below.
INPUT OFFSET (Pre_Team_LMC2001)
+10uV ...............................................
| . . . . . |
| . . . . . |
| . . . . . |
| . . . +SD +SD. |
+5uV |...............................................|
+SD +SD +SD +SD . +SD . . |
| . . . . . |
| . . . . . |
| . . . . . |
0uV |__A____A___A___A_______________________________|
| . . A A A A A . . |
| . . . A . . |
| . . . A. . |
|-SD .-SD -SD . A A A. |
-5uV |................-SD..-SD.-SD...................|
| . . . . . |
| . . . -SD . . |
| . . . . . |
| . . . -SD. . |
-10uV|...............................................|
| . . . . . |
| . . . -SD . |
| . . . . -SD. |
| . . . . . |
-15uV|_______________________________________________|
0 1V 2V 3V 4V 5V
COMMON MODE VOLTAGE
The Pre_Team_LMC2001 silicon was designed to exceed the
input rails by hundreds of millivolts. Apparently junctions
like ESD diodes begin to start introducing input bias
current at common mode voltages below ground.
INPUT OFFSET (Pre_Team_LMC2001) INPUT CURRENT
+10uV ............................................... 1nA
| . . . . . |
|Ibias . . . . . |
| . . . . . |
| . . . . . |
|......Ibias....................................|100pA
| . . . . . |
|Off . Off . Ibias . . |
| . . Off Off . |
| . . . Ibias Off Off |
0uV |...............................................|10pA
| . . . . . Ibias |
| . . . . . |
| . . . . . |
| . . . . . |
|_______________________________________________|
-100mV -66mV -33mV 0V 33mV 66mV 100mV
COMMON MODE VOLTAGE
===================Composite_Amplifier==========================
The composite amplifier architecture makes things very
convenient. The main amplifier can be optimized as a high
speed high quality general purpose RRIO Op Amp. The precision
amplifier was only being asked to remove DC offset and 1/f noise
of the main amplifier. A side benefit was that it greatly
increased things like open loop gain and and input common mode rejection
as well. It looked promising to have the best of everything
without any tradeoffs.
___ |\ Main Amp
| |_____|+\ ___
|___| | | \_____| |
___ | | / |___|
| |_\___|-/_
|___| / | |/|\
| | \
| | \
| | |\ \
| |_|+\ |
| | \___|
| | /
|___|-/ "Chopper"
|/
But there was an unusual feature to this composite architecture.
The "chopper" amplifier could be designed as a pure
single pole Op Amp with it's pole well under 1mHz. If the
main Op Amp was design to have a pole around 1KHz, the
two amplifiers can be made to do a pole zero cancelation
so that the total composite amplifier looks like a perfect
Op Amp with extremely high open loop gain with a high
gain bandwidth.
Gain Band Width Chopper On/Off (Pre_Team_LMC2001)
100dB ................................................
ON . . . .
| . . . .
| ON . . . .
| . . . .
| . . . .
80dB |..........ON....................................
| . ON . . .
| . ON . .
| OFF OFF OFF OFF ON ON .
| . . . .
| . . . ON .
40dB |............................................ON
| . . . .
| . . . .
| . . . .
| . . . .
| . . . .
0dB |_______________________________________________
10Hz 100Hz 1000Hz 10KHz 100KHz
The curve above shows that the Pre_Team_LMC2001 silicon had
around 60dB open loop gain with the "chopper" off. When
the chopper was turned on, first silicon did not have perfect
pole zero cancelation. But things were encouraging.
_________________________________________
| |
| Main Amplifier |\ |
| ____|+\ |
|____________|\ _ _ _|_ | \_______|
VIN | - | \/ \/ \_ /// | / | ___
___________/|\___| /\_/\_/ |_______|-/ |__|OUT|
_|__ | | + |/ INN2| |/ BOPA | |___|
/_ \ | | | |
// \ \ | | |__||________|
\ \_// | | RPZC | || CCOMP |
\____/ | | | |
_|_ | | _/\ /\ /\_|_/\ /\ /\_|
/// | | | \/ \/ RAV\/ \/
| | |_________________
| ___| |
| | CINT __||___|
| | | || |
| | |\ BCHOP RINT | |OUT2
| |__|+\ | |\ |
| | \____/\ /\ /\_|_|-\ |
| | / \/ \/ | \__|
|_____|-/ INN | /
|/ __|+/
_|_ |/
Precision Amplfifier ///
Equivalent circuit
The equivalent circuit above shows how all the poles and zeros are defined.
Since the poles and zero of both amplifiers are set by internal
components, it was possible to set things up so that everything
tracked each other over process and temperature.
POLE ZERO CANCELATION (Pre_Team_LMC2001)
60dB |................................................
| G . . . .
| G . P P . . .
| G . P . .
| G . P . P . .
| G . P . P . .
40dB |...P....PG...P...............P...P...P...P..P...90deg
| G . . .
| . G . . .
| . G . . .
| . G . . .
| . G . . .
80dB |.......................G......... ..............45deg
| . . G .
| . . G . .
| . . G . .
| . . G . .
| . . G . .
60dB |...................................G............ 0deg
| . . . G .
| . . . G .
| . . . .
| . . . .
| . . . .
-20dB|______________________________________________ .
10Hz 100Hz 1KHz 10KHz 100KHz
The pole zero cancelation was less than perfect on first silicon.
This is best seen by the bump in phase of the op amps gain phase curve.
^ dB
/|\
|
|\ Low Speed amp dominates offset, gain, noise
| \/
| \
|___\ high Speed amp dominates Gainbandwidth, slew rate
| .\ /
| . \
|___.__\___\
1KHz / Freq
The Pre_Team_LMC2001's precision amplifier stage was not really designed
to have an actual low frequency pole. In theory, a perfect Op Amp has infinite
open loop gain at DC. But gains higher the 150dB are questionable because
of thermal couple effects whenever two different metals are making electrical
contact. Having a large open loop gain is also creates some IC testing
challenges.
===================High_Gain_High_Bandwidth==========================
Gain Phase (Pre_Team_LMC2001)
60dB |................................................
| G . . . .
| G . . . .
| G . . . .
| G . . . .
| G . . . .
40dB|...P....P..G.P..P..P..P...P..P..................90deg
| . G . P . .
| . G . P .
| . G . ^P .
| . G . /|\P .
| . G . | P .
20dB |.......................G...........|............45deg
| . . G | P .
| . . G | .
| . . G | P .
| . . G \|/ .
| . . G V P .
0dB |...................................G......P..... 0deg
| . . . G P .
| . . . G .
| . . . .
| . . . .
| . . . .
-20dB|______________________________________________ .
10KHz 100KHz 1MHz 10MHz 100MHz
Pre_Team_LMC2001 silicon was designed to be a 100pF Cap stable, low supply current,
10MHz RRIO Op Amp. The target was a CMOS Op Amp with input offsets in microvolts,
have input currents in pico amps, with open loop gain too high to measure,
have zero input offset drift over time and temperature and supply voltage, have flat
input noise voltage with no 1/f or chopper noise, have no noise spikes on the
supplies, used a truly random spread spectrum clock, show low feedthrough noise
current spikes at the input, and be all self contained in a tiny 5 lead package.
Up until the LMC2001 team was formed, meeting all these
goals looked
technically encouraging.