IEEE Solid-States Circuits Magazine - Summer 2021 - 20

generation of the reference current at
its source, that puts severe constraints
on the system reference block as well
as the temperature and process variation
of the currents themselves.
My fix for the manufacturing problem
is to use another transistor in
place of the resistor in the ZGA, as
shown in Figure 4. Here, the role of
the resistor is played by transistor
" Dirty " System
Reference Current
" Clean " Bias Current
M2
Magic dc
Voltage
+
-
" Bias Rail "
M1
M2. This is perfect because the source
resistance of M2
transconductance of M1
will always track the
regardless
of process variations, current levels,
and temperature because the two
transistors are identical and they carry
the same current.
This configuration will, of course,
completely kill the large-signal peaking
behavior discussed earlier, but the
small-signal, zero-gain effect will
be dialed in precisely. If this can be
made to work, we can convert a contaminated
system reference current
(the equivalent of lead to an alchemist)
into spurious-free, clean internal
bias currents (gold!).
The fly in this ointment
is the
" magic dc voltage " identified in Figure
4. This voltage must match the
steady-state Vgs of M2
to set the right
ZGA
Local Current Sources
FIGURE 4: The amazing noise-eating bias concept. (All transistors are the same size.)
voltage on the bias rail. No problem:
we'll just set up a replica transistor
operating at the same current as M2
,
and, voilà, its Vgs will be the magic
voltage source!
Not so fast there, cowboy. The
VDD
Incoming System
Reference ac " Noise "
M3
C1
M4
R1
100 k
ZGA
(a)
10
-10
-20
-30
-40
-50
-60
-70
10 kHz 100 kHz 1 MHz 10 MHz 100 MHz 1 GHz 10 GHz
Reference Current Noise Frequency
(b)
FIGURE 5: The (a) implementation and (b) simulated performance of the amazing noiseeating
bias circuit: 60 dB of reference noise rejection is achieved for 100-MHz interference.
(All transistors are the same size.)
20 SUMMER 2021
IEEE SOLID-STATE CIRCUITS MAGAZINE
10 p
M1
C2
Bias Rail
10 p
Local Current Source
ac 1
dc 25 µA
M2
Measure ac
current here.
magic voltage source cannot be modulated
by the small-signal noise on
the incoming system reference current.
Remember that M1
and M2
must
see the same small voltage perturbations
on their gates so that you have
the two gain paths of +1 and -1 to
cancel each other out. If the magic
voltage source is also modulated, you
have a top path gain of +2. (You might
consider making the bottom path -2,
but this is left as an exercise for the
reader ... which is code for " it proved
too tricky for the author. " ) So how do
we make a " clean " voltage source of
just the right value?
My best attempt at this is shown
in Figure 5(a). Veterans will recognize
this as a Wilson current mirror.
The " magic dc voltage " is made
by a low-pass-filtered version of
Vgs3. (The resistor is not critical
and could be made with a long, thin
transistor if that's advantageous in
the process at hand.) The spurious
rejection performance is shown in
Figure 5(b).
The 10-pF capacitor to ground on
the bias rail in Figure 5(a) serves two
purposes. First, it absorbs transients
Noise in Local
Current Source Output (dB)
IEEE Solid-States Circuits Magazine - Summer 2021

Table of Contents for the Digital Edition of IEEE Solid-States Circuits Magazine - Summer 2021
