Accurate temperature sensing with an external P-N junction

How noise affects measurements

Linear Technology devices typically implement a lowpass filter, which filters spikes and noise. However, in some cases filtering results in a significant DC shift.

Figure 9: The graph shows an asymmetrical waveform on the TSENSE pin

The example shown in Figure 9, from an LTC3880, shows an asymmetrical waveform on the TSENSE pin (channel 1) caused by injecting some of the switch node signal into the TSENSE pin. When this is filtered, it results in a DC shift. If temperature is calculated using a ∆V_BE calculation, and the DC shift is the same for both V_BE measurements, the effect will be cancelled out. This means that if the error mechanism is consistent between current measurements, ∆V_BE is robust. If the single V_BE measurement is used, the DC shift from the filtering will be a source of measurement error. (LTC3880 does not support single ∆V_BE measurements)

If the magnitude of noise is very large with respect to ∆V_BE, and the noise is asymmetrical (as in the scope shot) and different between current measurements, ∆V_BE cannot cancel out the noise. In this case a single measurement can produce a more accurate temperature measurement. For example, suppose noise causes an error of 50mV. A typical ∆V_BE is 70mV. The error can be as high as 70%. If a single V_D is used, the error is about 50mV/600mV, or 8%.

Therefore, in systems with symetric noise, the ∆V_BE measurement produces the highest accuracy by eliminating I_S as a source of error. (See ∆V_BE equation). In systems with large non-symetric noise, the V_BE measurement produces the highest accuracy.

Overall, the best accuracy comes from a good layout that ensures near zero noise that is systematic, and uses a ∆V_BE calculation.

Non-symetric noise sources require good layout because the ∆V_BE approach cannot reject them.

Figure 10: Here is a graph showing an example of a coupling problem

An Example Coupling Problem

The example shown in Figure 10 comes from an LTC3880. Signal 1 is the TSENSE signal. When the LTC3880 is applying 32µA, you get the higher signal level, and when it is applying 2µA, you get the lower signal level. The last high and low portions of the waveform are where the two measurements are taken. Signal 2 is the V_OUT of the LTC3880, which is coupling into the 32µA measurement.

Figure 11: A graph showing that the same coupling can occur in the 2 uA measurement

The same coupling can occur in the 2µA measurement as shown in Figure 11. The asymmetry comes from the fact that the coupling affects only one of two measurements, so it is not cancelled by the ∆V_BE calculation. Furthermore, the error will appear random because the output turn-on event and the current forcing mechanism are not synchronized. The only defense against this error is prevention of the coupling by proper layout, or widening the fault limits.

Figure 12: Here we are showing the LTC3883 and 2N3906 PNP current sense

Mitigating error sources

There are two primary methods of preventing errors, both require proper PCB layout. The first involves elimination of shared ground paths. The second involves proper signal trace routing.

Linear Technology data sheets specify how to return current from the collector and base of the temperature measurement transistor to the device. Typically the current returns to a sense ground (SGND), or an amplifier negative (–) input.

The current should return to the device via its own sense trace to ensure there is no shared impedance with high current paths, and to the data sheet specified pin.

Figure 12 shows a LTC3883 and 2N3906 PNP current sense. Q10 in circle 1 is the p-n junction temperature sensor and is filtered by C99. The purpose of C99 is to provide a low AC impedance to prevent any DC offsets from rectification or non-linear waveforms, and to keep coupled noise out of the LTC3883 ADC.

The routing uses two parallel pairs on the same layer so that any coupling from noise sources becomes a common mode signal to the ADC in the LTC3883 and are rejected. The anode trace routes to the sense pin to the LTC3883 Pin 32 shown in  circle 2, and the cathode is routed to SGND: the exposed PAD on the back of the LTC3883. The cathode routing to the exposed PAD ensures no high current from the power ground flows through the sense line.

Figure 13 shows an LTC2991 and two 2N3906 PNP temperature sensors. As in the previous example, capacitor filtering is added near the PNP. However, capacitor filtering was also added at the input of the LTC2991.

Figure 13: In this image we have an LTC2991 and two 2N3906 PNP temperature sensors

The longer trace run offers more opportunity to pick up noise farther from the PNP due to trace inductance. Typically this capacitor is added as an option and installed only if there is a problem. Additionally, notice that the routes avoid switching areas by following the edges of the plane between functional circuits. The routes from the PNP farthest to the right go right between a LTC3883 buck converter below, and a LT1683 isolated boost above.

NOTE: Linear Technology strongly recommends placement of a filter capacitor near the PNP temperature sensor, routing differentially, and avoiding noisy signals. Long routes may pick up more noise, so optionally add a filter capacitor near the device.

Some designs may use a power block with built in temperature diode. Some of these power blocks do not have a pin for the low sense of the diode. These blocks may not have a filter capacitor. In these situations, you can place a filter capacitor on your board as close to the high side diode sense pin of the power block as possible, and try to minimize all noise sources. A low sense line can still be routed from the power ground, but you can’t eliminate the shared current from the switching path, so some noise will be injected. You can mitigate some of the problems that may result by:

1.  Using a slower V_OUT ramp rate when turning on

2.  Adding an offset to the measurement using the proper register (digital power device) to lower the measured temperature

3.  Raising the overtemperature fault limit

4.  Adding a capacitor on the power block