Datasheet

1°C Accurate Remote/Local Temperature
Sensor with SMBus Serial Interface
8 Maxim Integrated
MAX6654
measure ambient temperature; when measuring local
temperature, it senses the temperature of the PCB to
which it is soldered. The leads provide a good thermal
path between the PCB traces and the MAX6654’s die.
Thermal conductivity between the MAX6654’s die and
the ambient air is poor by comparison. Because the
thermal mass of the PCB is far greater than that of the
MAX6654, the device follows temperature changes on
the PCB 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 450µA + 0.4V x 1mA. Package theta J-
A is about 150°C/Ω, so with V
CC
= 5V and no copper
PCB heat sinking, the resulting temperature rise is:
ΔT = 2.7mW x 150°C/W = 0.4°C
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 PCB layout and proper external noise
filtering are required for high-accuracy remote measure-
ments 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°C to +10°C, depending on the frequency and
amplitude (see
Typical Operating Characteristics ).
PCB Layout
1) Place the MAX6654 as close as practical to the
remote diode. In a noisy environment, such as a
computer motherboard, this distance can be 4
inches to 8 inches (typ) or more, as long as the
worst noise sources (such as CRTs, clock genera-
tors, memory buses, and ISA/PCI buses) are avoid-
ed.
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°C 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 PCB contamination must be dealt with careful-
ly since a 20MΩ leakage path from DXP to ground
causes about +1°C 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, PCB-induced thermo-
couples are not a serious problem. A copper-solder
thermocouple exhibits 3µV/°C, and it takes about
200µV of voltage error at DXP-DXN to cause a +1°C
measurement error. So, most parasitic thermocou-
ple 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
aren’t absolutely necessary (as they offer only a
MINIMUM
10MILS
10MILS
10MILS
10MILS
GND
DXN
DXP
GND
Figure 2. Recommended DXP/DXN PCB Traces