
LTC2641/LTC2642
17
26412fc
For more information www.linear.com/LTC2641
Op Amp Specifications and Bipolar DAC Accuracy
The op amp contributions to unipolar DAC error discussed
above apply equally to bipolar operation. The bipolar ap-
plication circuit gains up the DAC span, and all errors, by
a factor of 2. Since the LSB size also doubles, the errors
in LSBs are identical in unipolar and bipolar modes.
One added error in bipolar mode comes from IB (IN–),
which flows through RFB to generate an offset. The full
bias current offset error becomes:
VOFFSET = (IB (IN–) RFB – IB (IN+) ROUT 2) [Volts]
So:
VI IN
kI IN
k
V
OFFSET
BB
REF
=
()
+
() – () .
–
28
12 4
33
[]
LSB
Settling Time with Op Amp Buffer
When using an external op amp, the output settling time
will still include the single pole settling on the LTC2641/
LTC2642 VOUT node, with time constant ROUT (COUT +
CL) (see Unbuffered VOUT Settling Time). CL will include
the buffer input capacitance and PC board interconnect
capacitance.
Theexternalbufferamplifieraddsanotherpoletotheoutput
response, with a time constant equal to (fbandwidth/2
π).
For example, assume that CL is maintained at the same
value as above, so that the VOUT node time constant is
83ns = 1μs/12. The output amplifier pole will also have a
time constant of 83ns if the closed-loop bandwidth equals
(1/2
π 83ns) = 1.9MHz. The effective time constant of
two cascaded single-pole sections is approximately the
root square sum of the individual time constants, or √2
83ns = 117ns, and 1/2 LSB settling time will be ~12
117ns = 1.4μs. This represents an ideal case, with no slew
limiting and ideal op amp phase margin. In practice, it
will take a considerably faster amplifier, as well as careful
attention to maintaining good phase margin, to approach
the unbuffered settling time of 1μs.
The output settling time for bipolar applications (Figure 3)
will be somewhat increased due to the feedback resistor
network RFB and RINV (each 28k nominal). The parasitic
capacitance,CP,ontheopamp(–)inputnodewillintroduce
a feedback loop pole with a time constant of (CP 28k/2).
A small feedback capacitor, C1, should be included, to
introduce a zero that will partially cancel this pole. C1
should nominally be <CP, typically in the range of 5pF
to 10pF. This will restore the phase margin and improve
coarse settling time, but a pole-zero doublet will unavoid-
ably leave a slower settling tail, with a time constant of
roughly (CP + C1) 28k/2, which will limit 16-bit settling
time to be greater than 2s.
Reference and GND Input
The LTC2641/LTC2642 operates with external voltage
references from 2V to VDD, and linearity, offset and
gain errors are virtually unchanged vs VREF. Full 16-bit
performance can be maintained if appropriate guidelines
are followed when selecting and applying the reference.
The LTC2641/LTC2642’s very low gain error tempco of
0.1ppm/°C, typical, corresponds to less than 0.5LSB
variation over the –40°C to 85°C temperature range. In
practice, this means that the overall gain error tempco
will be determined almost entirely by the external refer-
ence tempco.
TheDACvoltage-switchingmode“inverted”resistorladder
architectureusedintheLTC2641/LTC2642exhibitsarefer-
ence input resistance (RREF) that is code dependent (see
the Typical Performance curves IREF vs Input Code).
In unipolar mode, the minimum RREF is 14.8k (at code
871Chex, 34,588 decimal) and the the maximum RREF is
300k at code 0000hex (zero scale). The maximum change
in IREF for a 2.5V reference is 160A. Since the maximum
occurs near midscale, the INL error is about one half of the
change on VREF, so maintaining an INL error of <0.1LSB
requiresareferenceloadregulationof(1.53ppm2/160A)
=19[ppm/mA].Thisimpliesareferenceoutputimpedance
of 48mΩ, including series wiring resistance.
To prevent output glitches from occuring when resistor
ladder branches switch from GND to VREF, the reference
input must maintain low impedance at higher frequencies.
A 0.1μF ceramic capacitor with short leads between REF
and GND provides high frequency bypassing. A surface
mount ceramic chip capacitor is preferred because it has
the lowest inductance. An additional 1μF between REF
and GND provides low frequency bypassing. The circuit
will benefit from even higher bypass capacitance, as long
applicaTions inForMaTion