3.1.2 Calculating Total System Accuracy
The voltage seen by the LM95245 also includes the IFRS volt-
age drop of the series resistance. The non-ideality factor, η,
is the only other parameter not accounted for and depends
on the diode that is used for measurement. Since ΔVBE is
proportional to both η and T, the variations in η cannot be
distinguished from variations in temperature. Since the non-
ideality factor is not controlled by the temperature sensor, it
will directly add to the inaccuracy of the sensor. For the for
Intel processor on 65nm process, Intel specifies a +4.06%/
−0.897% variation in η from part to part when the processor
diode is measured by a circuit that assumes diode equation,
Equation 4, as true. As an example, assume a temperature
sensor has an accuracy specification of ±1.0°C at a temper-
ature of 80°C (353 Kelvin) and the processor diode has a non-
ideality variation of +4.06%/−0.89%. The resulting system
accuracy of the processor temperature being sensed will be:
TACC = + 1.0°C + (+4.06% of 353 K) = +15.3 °C
and
TACC = - 1.0°C + (−0.89% of 353 K) = −4.1 °C
TrueTherm technology uses the transistor equation, Equation
4, resulting in a non-ideality spread that truly reflects the pro-
cess variation which is very small. The transistor equation
non-ideality spread is ±0.39% for the 65nm thermal diode.
The resulting accuracy when using TruTherm technology im-
proves to:
TACC = ±0.75°C + (±0.39% of 353 K) = ± 2.16 °C
Intel does not specify the diode model ideality and series re-
sistance of the thermal diodes on 45nm so a similar compar-
ison cannot be calculated, but lab experiments have shown
similar improvement. For the 45nm processor the ideality
spread as specified by Intel is -0.399% to +0.699%. The re-
sulting spread in accuracy when using TruTherm technology
with the thermal diode on Intel processors with 45nm process
is:
TACC = -0.75°C + (-0.39% of 353 K) = -2.16 °C
to
TACC = +0.75°C + (+0.799% of 353 K) = +4.32 °C
The next error term to be discussed is that due to the series
resistance of the thermal diode and printed circuit board
traces. The thermal diode series resistance is specified on
most processor data sheets. For Intel processors in 45 nm
process, this is specified at 4.5Ω typical with a minimum of
3Ω and a maximum of 7Ω. The LM95245 accommodates the
typical series resistance of Intel Processor on 45 nm process.
The error that is not accounted for is the spread of the
processor's series resistance. The equation used to calculate
the temperature error due to series resistance (TER) for the
LM95245 is simply:
(6)
Solving for RPCB equal to -1.5Ω to 2.5Ω results in the addi-
tional error due to the spread in this series resistance of
-0.93°C to +1.55°C. The spread in error cannot be canceled
out, as it would require measuring each individual thermal
diode device. This is quite difficult and impractical in a large
volume production environment.
Equation 6 can also be used to calculate the additional error
caused by series resistance on the printed circuit board. Since
the variation of the PCB series resistance is minimal, the bulk
of the error term is always positive and can simply be can-
celled out by subtracting it from the output readings of the
LM95245.
Processor Family Transistor Equation ηT,
non-ideality
Series
R,Ω
min typ max
Intel Processor on
45 nm process
0.997 1.001 1.008 4.5
Intel Processor on
65 nm process
0.997 1.001 1.005 4.52
Note: NA= Not Available at publication of this document.
3.2 PCB LAYOUT FOR MINIMIZING NOISE
30015117
FIGURE 9. Ideal Diode Trace Layout
In a noisy environment, such as a processor mother board,
layout considerations are very critical. Noise induced on
traces running between the remote temperature diode sensor
and the LM95245 can cause temperature conversion errors.
Keep in mind that the signal level the LM95245 is trying to
measure is in microvolts. The following guidelines should be
followed:
1. VDD should be bypassed with a 0.1 µF capacitor in
parallel with 100 pF. The 100 pF capacitor should be
placed as close as possible to the power supply pin. A
bulk capacitance of approximately 10 µF needs to be in
the near vicinity of the LM95245.
2. A 100 pF diode bypass capacitor is recommended to filter
high frequency noise but may not be necessary. The
LM95245 can handle capacitance up to 3.3 nF (see
Typical Performance Curve "Remote Temperature
Reading Sensitivity to Thermal Diode Filter
Capacitance"). Place the filter capacitors close to the
LM95245 pins and make sure the traces to this capacitor
are matched.
3. Ideally, the LM95245 should be placed within 10 cm of
the Processor diode pins with the traces being as
straight, short and identical as possible. Trace resistance
of 1Ω can cause as much as 0.62°C of error. This error
can be compensated by using simple software offset
compensation.
4. Diode traces should be surrounded by a GND guard ring
to either side, above and below if possible. This GND
guard should not be between the D+ and D− lines. In the
event that noise does couple to the diode lines it would
be ideal if it is coupled common mode. That is equally to
the D+ and D− lines.
5. Avoid routing diode traces in close proximity to power
supply switching or filtering inductors.
6. Avoid running diode traces close to or parallel to high
speed digital and bus lines. Diode traces should be kept
at least 2 cm apart from the high speed digital traces.
www.national.com 22
LM95245