Professional Documents
Culture Documents
Don't Be Fooled by Voltage Reference Long-Term Drift and Hysteresis
Don't Be Fooled by Voltage Reference Long-Term Drift and Hysteresis
The new micropower LT®1461 and LT1790 low dropout reference, this corresponds to a 0.39µV shift after 1000
bandgap voltage references excel not only in temperature hours. This is pretty hard to determine (read impossible)
coefficient and accuracy, but also in long-term drift and if the peak-to-peak output noise is larger than this num-
hysteresis (output voltage shift due to temperature cy- ber. As a practical matter, one of the best laboratory
cling). Long-term drift and hysteresis, which are some- references available has long-term drift of 1.5µV/mo. This
times ignored or wrongly specified by other manufacturers, performance is only available from the best subsurface
can be the accuracy limitations of systems. System cali- zener references such as the LTZ1000, utilizing special-
brations can remove TC and initial accuracy errors, but ized heating techniques.
only frequent calibration can remove the long-term drift
Competitive Reference Measures 500 Times
and hysteresis. Subsurface Zener references, like the
Worse Than Claimed
LT1236, have the best long-term drift and hysteresis, but
they do not offer low output voltage options, low supply Long-term drift data was taken with parts that were
current and low operating supplies like these new band- soldered onto PC boards similar to a “real world” applica-
gap references. tion. These boards were not preconditioned. They were
placed into a constant temperature oven with TA = 30°C
Lies About Long-Term Drift and their outputs were scanned regularly and measured
Some manufacturers are now touting phenomenal long- with an 8.5 digit DVM. Figures 1 and 2 show typical long-
term drift specifications, based on accelerated high tem- term drift of the LT1461S8-2.5 and the SOT-23 LT1790S6-
perature testing. THIS IS A DELIBERATE LIE! Long-term 2.5. Initially, data was taken every hour where the largest
drift cannot be extrapolated from accelerated high tem- changes occur, but after several hundred hours the fre-
perature testing. The only way long-term drift can be quency was lowered to reduce the large number of data
determined is to measure it over the time interval of points. Figure 3 shows long-term drift of a competitive
interest. The erroneous technique produces numbers that reference that specifies long-term drift of 0.2ppm/kHr in
are wildly optimistic and uses the Arrhenius Equation to 200
derive an acceleration factor from elevated temperature 150
readings. The equation is:
DRIFT (ppm)
100
EA 1 1 50
–
AF = e K T 1 T 2 0
–50
where: EA = Activation Energy (Assume 0.7) 0 200 400 600 800 1000 1200 1400 1600
HOURS DN229 F01
K = Boltzmann’s Constant
T2 = Test Condition in °Kelvin Figure 1. LT1461S8-2.5V Long-Term Drift
T1 = Use Condition Temperature in °Kelvin 200
150
To show how absurd this technique is, compare this
DRIFT (ppm)
, LTC and LT are registered trademarks of Linear Technology Corporation. Figure 2. LT1790SOT23-2.5V Long-Term Drift
04/00/229
200
The stabilization time at each temperature was 30 min-
150
utes. The worst-case output voltage changes at 25°C are
DRIFT (ppm)
100 shown in Figures 4 and 5 for the LT1461S8-2.5 and the
50 SOT-23 LT1790S6-2.5. A competitive reference, which
0 makes no mention of hysteresis on its data sheet, was
–50
0 200 400 600 800 1000 1200 1400 1600
also measured and is shown in Figure 6.
HOURS DN229 F03
80
60
PERCENTAGE OF UNITS
its data sheet. Measured data shows this reference to have 85°C TO 25°C – 40°C TO 25°C
50
drift between 60ppm/kHr and 150ppm/kHr or 300 to 750
40
times worse than claimed.
30
several hundred hours of operation. Further changes in Figure 4. LT1461S8-2.5 Industrial Hysteresis
output voltage are typically logarithmic and changes after 50
40
0
When a reference is soldered onto a PC board, the –200 –160 –120 –80 –40 0 40 80 120 160 200
DISTRIBUTION (ppm)
elevated temperature and subsequent cooling cause stress DN229 F05
that influences the output. If the voltage reference is Figure 5. LT1790S6-2.5 Industrial Hysteresis
repeatedly temperature cycled, inelastic stress is applied 50
to the chip and the output voltage does not return to the
PERCENTAGE OF UNITS
40
25°C initial value. The mechanical stress is due to the
30
difference in thermal coefficients of expansion between 85°C TO 25°C – 40°C TO 25°C
20
the silicon chip, plastic package and PC board. This error,
10
known as “thermally induced hysteresis,” is expressed in
0
ppm and cannot be trimmed out because it is variable and –200 –160 –120 –80 –40 0 40 80 120 160 200
DISTRIBUTION (ppm)
has memory of previous temperature excursions. Hyster- DN229 F06
esis is always worse with higher temperature excursions, Figure 6. XX780S8-2.5 Industrial Hysteresis
and differs with die attach and package type.
Conclusion
Hysteresis—Often the “Missing” Spec
Voltage references from Linear Technology are conser-
Most manufacturers ignore hysteresis specifications,
vatively and accurately specified, unlike those from other
but they can be critical in precision designs. To graphi-
manufacturers that intentionally mislead or eliminate key
cally show hysteresis, many references were IR reflow
specifications to cover shortcomings—shortcomings that
soldered onto PC boards and the boards
may cause large errors.
underwent a “heat soak” at 85°C (this ensures that they
all had the same initializing temperature). The tempera- The new LT1461 and LT1790 excel in all specifications
ture was then cycled multiple times between 85°C, 25°C that set system precision. There is nothing left out and
and – 40°C and all 25°C output voltages were recorded. there is nothing hidden.