Transcript Simple Structure Balanced Differential Element

Linearized MOSFET Resistors
Dr. Paul Hasler
Review of Gm-C Filters
Gm-C filters: voltage mode/current mode/log-domain
Use amplifier dynamics as filtering elements:
If making 10kHz filter, why make amplifiers run at 10MHz?
Good properties
• Highest bandwidth / power consumed
• Smallest number of elements / area consumed
• Lowest noise levels / power consumed (thermal)
• Utilizes capacitor matching (ie. C4)
• Electronically tunable
Issues for Gm-C Filters
Improvement by
Floating-Gate Techniques
Tuning
Need control schemes (direct or
indirect) – adds significant amount of
control overhead
Mostly compensates: slight
adjustments due to transistor
aging / T changes, etc.
Matching
Huge issue for current-mode techniques
Design to eliminate these issues
Distortion
Techniques to improve linear range, but
at a cost of lower gm/I (lower speed,
higher noise, higher power)
More techniques to improve
linear range
Most Gm-C techniques are fairly recent (80’s-90’s),
and Floating-Gate techniques are even more recent (90’s - ).
Other Filter Techniques
Utilizing higher frequency elements / additional elements,
to improve distortion (as well as 1/f noise, etc.).
Two techniques:
Amplifiers: (Op-amps), that run at much faster frequencies than filter cutoff.
Can use feedback to widen the linear range. Significant power increase.
Oversampling: Using a wider bandwidth than necessary to lower
noise per unit bandwith (and more power) and distortion.
Nonlinear systems can utilize noise shaping (Sigma-Delta Modulators)
Common in sampled data systems.
Two techniques:
Switched Capacitor Blocks
Blocks based upon traditional, discrete RC active fitlers.
How to Build Resistances?
Resistors in a CMOS process
- Sometimes High resistance poly layer in a given process
- Poly, diffusions, or Well, but larger area consumed
Fairly linear, can be large for frequencies under 1MHz.
Not tunable: therefore RC > 20% mismatch, so we have a problem for
precission filters…so either laser trimming,
EEPROM trimming, (could tune cap, but…)
or imprecise filters, like anti-alaiasing filter.
MOSFET as a Resistor
MOSFET as a Resistor
Ohmic Region: how linear will that be, well only over a small region.
We have a gate voltage, so it is tunable, but of course,
we still need a method of tuning.
MOSFET has an ohmic region both in subthreshold
and above threshold operation.
Resistance is not exactly a constant,
except for a fixed source voltage….
resistance changes with source / drain voltage.
Could imagine an nFET and a pFET in parallel,
but still not a precission element.
MOSFET as a Resistor
Two things to improve the situation.
1. Typically built around an amplifier to fix one of the terminals
(mostly op-amps, but could also be a Norton or
transisresistance approach as well)
The amplifier must keep terminals nearly fixed to
eliminate distrotion; therefore, in general the amplifier
must run a lot faster than expected by a simple GmC stage.
2. Can use a combination of MOSFETs to linearize the behavior.
Linearized MOSFET resistors
Simple Structure
+
Vc
Vi
Va
Balanced Differential Element
+
+
Vi
+
Iout
+
Vc
Va
+
Iout
-
-
Vc
Vc
-
Iout
Vi
-
Vc
-
-
Va
Iout
-
Va
+
+
Vc
Vi
-
Linearized MOSFET resistors
In practice, one might use even
lower input impedance elements
+
Vi
Vc
+
+
Iout
-
-
Vc
Vc
-
Iout
+
Vi
-
Iout
Vc
Va
+
+
Iout
-
Va
GND
GND
GND
GND
Basic Resistive Feedback
+
GND
Vi
Vout
Vin
+
Vc
+
Vout
-
-
Vc
Vc
R1
R2
Vi
-
Vout
-
+
Vb
+
Vc
-
-
Vb
Vb
+
Vb
Basic Integrator Structure
C
+
Vi
GND
Vout
Vin
+
Vc
+
Vout
-
-
Vc
Vc
R1
Vi
C
-
Vout
-
C
+
Vc
-
Vb
-
+
Vb
Vb
+
Vb
Ideal Integrator if
+
Vb
-
= Vb
Tow-Thomas SOS (Lowpass)
C1
R3
C2
R1
R
R4
R2
Vin
R
GND
V1
GND
R4 needed for stability
Tuning can be interesting (tuning pots)
All amps must be sufficiently fast
V2
Vout
GND
Tow-Thomas SOS (Lowpass)
C
R
C
R
R4
Vin
R
R
R
GND
Vout
GND
GND
t = RC
Q = R4 / R