2癈 Accurate Remote/Local Temperature
Sensor with SMBus Serial Interface
8  _______________________________________________________________________________________
measure ambient temperature; when measuring local
temperature, it senses the temperature of the PC board
to which it is soldered. The leads provide a good ther-
mal path between the PC board traces and the
MAX6690s die. Thermal conductivity between the
MAX6690s die and the ambient air is poor by compari-
son. Because the thermal mass of the PC board is far
greater than that of the MAX6690, the device follows
temperature changes on the PC board with little or no
perceivable delay.
When measuring temperature with discrete remote sen-
sors, the use of smaller packages, such as SOT23s,
yields the best thermal response times. Take care to
account for thermal gradients between the heat source
and the sensor, and ensure that stray air currents
across the sensor package do not interfere with mea-
surement accuracy. When measuring the temperature
of a CPU or other IC with an on-chip sense junction,
thermal mass has virtually no effect; the measured tem-
perature of the junction tracks the actual temperature
within a conversion cycle.
Self-heating does not significantly affect measurement
accuracy. Remote-sensor self-heating due to the diode
current source is negligible. For the local diode, the
worst-case error occurs when autoconverting at the
fastest rate and simultaneously sinking maximum cur-
rent at the ALERT output. For example, at an 8Hz rate
and with ALERT sinking 1mA, the typical power dissi-
pation is V
CC
x 450礎 + 0.4V x 1mA. Package theta J-
A is about 150癈/? so with V
CC
= 5V and no copper
PC board heat sinking, the resulting temperature rise is:
擳 = 2.7mW x 150癈/W = 0.4癈
Even with these contrived circumstances, it is difficult
to introduce significant self-heating errors.
ADC Noise Filtering
The ADC is an integrating type with inherently good
noise rejection, especially of low-frequency signals such
as 60Hz/120Hz power-supply hum. Micropower opera-
tion places constraints on high-frequency noise rejection;
therefore, careful PC board layout and proper external
noise filtering are required for high-accuracy remote
measurements in electrically noisy environments.
High-frequency EMI is best filtered at DXP and DXN with
an external 2200pF capacitor. This value can be
increased to about 3300pF (max), including cable
capacitance. Capacitance >3300pF introduces errors
due to the rise time of the switched current source.
Nearly all noise sources tested cause the ADC measure-
ments to be higher than the actual temperature, typically
by +1癈 to +10癈, depending on the frequency and
amplitude (see Typical Operating Characteristics).
PC Board Layout
1)  Place the MAX6690 as close as practical to the
remote diode. In a noisy environment, such as a
computer motherboard, this distance can be 4in to
8in (typ) or more, as long as the worst noise
sources (such as CRTs, clock generators, memory
buses, and ISA/PCI buses) are avoided.
2)  Do not route the DXP-DXN lines next to the deflec-
tion coils of a CRT. Also, do not route the traces
across a fast memory bus, which can easily intro-
duce +30癈 error, even with good filtering.
Otherwise, most noise sources are fairly benign.
3)  Route the DXP and DXN traces in parallel and in
close proximity to each other, away from any high-
voltage traces, such as +12VDC. Leakage currents
from PC board contamination must be dealt with
carefully since a 20M?leakage path from DXP to
ground causes about +1癈 error.
4)  Connect guard traces to GND on either side of the
DXP-DXN traces (Figure 2). With guard traces in
place, routing near high-voltage traces is no longer
an issue.
5)  Route through as few vias and crossunders as pos-
sible to minimize copper/solder thermocouple
effects.
6)  When introducing a thermocouple, make sure that
both the DXP and the DXN paths have matching
thermocouples. In general, PC-board-induced ther-
mocouples are not a serious problem. A copper-
solder thermocouple exhibits 3礦/癈, and it takes
about 200礦 of voltage error at DXP-DXN to cause
a +1癈 measurement error. So, most parasitic ther-
mocouple errors are swamped out.
7)  Use wide traces. Narrow traces are more inductive
and tend to pick up radiated noise. The 10mil
widths and spacings recommended in Figure 2
arent absolutely necessary (as they offer only a
MINIMUM
10mils
10mils
10mils
10mils
GND
DXN
DXP
GND
Figure 2. Recommended DXP/DXN PC Traces