Power MOSFET temperature measurements

Three temperature-sensitive electrical parameters are compared as thermometers for power MOSFET devices. The parameters are the forward drain-body diode voltage, the source-gate voltage, and the on-resistance. The results are also compared with temperatures measured with an infrared microradiometer. The procedure, apparatus, and circuits required to use each of the parameters as a thermometer are described. Some general considerations for measuring the temperature of power semiconductor devices are also discussed. Each parameter is found to be satisfactory for measuring the temperature of power MOSFETs. The sourcegate voltage measures a temperature nearest to the peak device temperature, and the drain-body diode voltage shows the least variation in calbiration from device to device.


INTRODUCTION
is a device specification that permits the user to establish the temperature for any power level. In order for a device manufacturer to specify and for a user to verify the device thermal resistance, an accurate, reliable method for measuring the device temperature is required. Ideally, the method would be nondestructive. Traditionally, temperature-sensitive electrical parameters (TSEPs) are used to measure the temperature of a semiconductor device.
The purpose of this paper is to provide a compari son of three techniques using three different TSEPs for measuring the temperature of power MOS FETs and to compare each with the temperature determined using an infrared microradiometer (IRM). The three TSEPs used in this study as device thermometers are the source-gate voltage measured at a low level drain current, the forward voltage of the drain-body diode, and the device on-resistance. In addition, a brief discussion of some general considerations for power semiconduc tor device temperature measurement is given.
Power Metal Oxide Semiconductor Field Effect Tran sistors (power MOSFETs) are relatively recent innovations. These devices are rapidly finding uses in many new as well as established applica tions because of their fast switching speeds and their relative freedom from thermal instabilities as compared to bipolar transistors. Because power MOSFETs are relatively new devices and because they are still in a state of evolution, measure ment methods for characterizing many of their important electrical and thermal properties are in a rudimentary phase.
One of the most important parameters of any semi conductor device is its operating temperature. The operating temperature is important in deter mining the reliability and operating life of the device, and it also has a strong influence on many of the electrical parameters of the device. The thermal resistance of a device, R$ (°C/W), is given by: AT where AT R (°C) is the rise in device temperatare with respect to the temperature of a speci fied reference point and P (W) is the power dissi pated by the device. Ideally, thermal resistance

TEMPERATURE-SENSITIVE ELECTRICAL PARAMETERS
A temperature-sensitive electrical parameter is usually used as a thermometer to nondestructively measure the temperature of a semiconductor device. Many device electrical parameters are temperature sensitive, but the TSEP must satisfy several cri teria if it is to be considered as a practical device thermometer. First, its variation with temperature must be large enough to be readily measured and to provide sufficient temperature resolution for the application at hand.  Each of the TSEPs examined in this study will be discussed with reference to this figure.

Drain-Body Diode Forward Voltage, V D B
The drain-body diode is formed at the junction of the opposite conductivity type drain and body regions (p-type body and n-type drain for the nchannel device of fig. 1). The forward voltage of the diode can be directly sensed at the sourcedrain terminals because the source metallization overlaps and makes electrical contact to the body region. The source-metal contact to the body provides a low resistance path for "stray" minori ty carriers that may enter the body region (such as due to leakage or avalanche current generated at the drain-body junction) and thus inhibits their injection into the source. Injection of minority carriers into the source can turn on the parasitic bipolar transistor comprised of the source-body-drain causing the gate to "lose con trol" of the MOSFET. The metal contacting both body and source thus enhances the safe operating limits of the MOSFET.
A substantial body of knowledge exists concerning the temperature sensitivity of a forward-biased diode (1). For constant current: Over the temperature range of interest, the drain resistivity temperature variation is dominated by the temperature dependence of the bulk mobility which is dominated by lattice scattering. Com puted values of the temperature coefficient of resistance (1/p)•(dp/dTj vary from about +0.007/°C for 1 ft*cm n-type material to about +0.008/°C for 100 ft*cm n-type material at room temperature (3). A less than 1-percent change in drain region re sistance (and thus RDS^0 11 )) occurs for a 1 °C change in temperature. For a "typical" device with R DS (on) = 1 ft, the variation of R D s(°n) with temperature is about +7 mft/°C. The drainsource on-resistance primarily senses the temperature in the active drain region of the device.

Source-Gate Voltage, V S G
The source-gate voltage controls the on/off state of the MOSFET. The threshold voltage, V T , is approximately equal to the value of V S G re quired to begin to turn on the MOSFET. When V SG is used as a thermometer in this work, the device is barely "turned on" and V S G a V T .
The drain current. In, is related to Vg G by: Equation (1) is valid when the MOSFET is operat ing in the saturation mode, i.e., when the drainsource voltage, V DS > V SG . In this work, when using V SG as a thermometer, V D g >> VgQ.
From equation (1 ): The apparatus for making a measurement of the device temperature for each of the techniques investigated consists of a temperature-controlled heat sink for the device, an electronic tempera ture control system, and the respective electronic switching and measurement circuitry. Each of these items as well as the measurement procedure will be discussed. The measurement procedure is essentially the same for each of the TSEPs investigated. Only steadystate measurements will be discussed, but most of what follows also applies to transient measure ments. The general procedure used is to repeti tively power the device under test (DUT) with "normal" (heating) operating conditions and to then rapidly switch to a "measurement" (cooling) condition. The duty cycle (heating power time/measurement time) is typically about 99 per cent. The case temperature of the device (only JEDEC TO-3 encased devices were studied) is held constant by the temperature-controlled heat sink and is continuously monitored.

Measurement
During the measurement time, the TSEP is moni tored. An oscilloscope was used in this study to enable the entire TSEP waveform to be measured as a function of time. The ability to monitor the TSEP during the entire measurement time permits one to determine the rate at which the device is cooling. This is important for determining the actual temperature at the instant the device is switched from the heating to the measurement con dition as well as for determining when extraneous, nonthermal effects are interfering with the mea surement. These concepts will be discussed in more detail in a later section.

Temperature-controlled Heat Sink
In order for the measured temperature to be a meaningful parameter, some well-defined reference point associated with the device must be main tained at a constant temperature during measure ment. The reference point chosen in this work is on the bottom outside of the transistor case di rectly below the semiconductor chip. The tempera ture is monitored by either a glass bead thermis tor or a thermocouple using the washer technique (4). The DUT is firmly clasped to a temperaturecontrolled heat sink to maintain a constant refer ence point temperature. The heat sink is made of a copper block which contains water channels for the flow of chilled water as well as several "heating" resistors. The heat-sink temperature control is achieved by electronically controlling the power supplied to the heating resistors. The flow rate of chilled water is manually controlled. The electronic temperature controller uses the glass bead thermistor (thermocouple) to sense the temperature for control purposes. Care is taken to assure that the heat sink surface is flat and "burr free." The DUT attaching screws are screwed down with the same torque, and a thermal grease is used between the device and heat sink for all mea surements • The required calibration of the TSEP is accom plished by monitoring the TSEP as the neat-sink temperature (and thus the DUT temperature) is varied with only measurement conditions applied

Measurement Circuits
Simplified schematics of the measurement circuits used for each of the TSEPs are shown in figure 3.

Not unexpectedly, the most stable parameter is V DB . Only a 2-or 3-percent variation in V DB was usually observed between identical devices, with an even smaller variation observed in dV DB /dT. Most devices exhibited a 2-or 3percent variability in both V SG and R D s(on)
, but a few devices (5 to 10 percent) exhibited a 10-to 20-percent difference from the norm. The variations in dVg G /dT and dRns(°n)/dT were compa rable to but less than the variations in the parameters themselves. These results suggest that a "constant" variability with•temperature for each parameter may be assumed for most devices of a given type, but care should be taken to detect "outliers" and perhaps a separate calibration performed for them if Vg G or Rns(°n) are used as the TSEP.

Measured Temperature Comparisons
The results of temperature measurements on several devices for a number of operating conditions using the three TSEPs as well as the Infrared Microradiometer are given in  TSEPs can be used to measure temperatures between 80 to 100 percent of the peak temperature as dem onstrated by the results in table II.

For four of the devices in table II, a number of different operating conditions are included, but the power dissipation for each (ID x V D S ) is the same. The peak temperature as indicated by
the IRM is the same for each condition. The tem perature of the power MOSFET depends only upon the power level and not upon the combination of I D and VDs used to achieve that power. This is also in contrast to the bipolar power transistor for which the peak temperature depends strongly upon the operating conditions (6).

GENERAL CONSIDERATIONS AND DISCUSSION
There are a number of "universal" difficulties encountered in measuring the temperature of a semiconductor device, whether it is a bipolar device or a MOSFET, a discrete power device or an integrated circuit. Some of these will be dis cussed in this section, and it will be noted how The term "nonthermal switching transient" refers to extraneous components of the measured TSEP waveform that are introduced as a result of switching from the heating to the measurement condition. That is, the measured TSEP has an extraneous, electrical component not present dur ing calibration (no switching occurs during cali bration). The TSEP must usually be measured with a resolution of at least 1 mV. Thus, even though one might consider the device to be fully "switched" for a typical circuit application, this is not the case for temperature measurement appli cations. Nonthermal switching transients are difficult to immediately discern, because the device is naturally cooling during the measurement phase (heating power has been removed) and the TSEP shows a natural thermal transient due to the temperature decay.
Example of waveforms obtained using each of the TSEPs during the measurement phase are shown in figure 5. It is expected that for times greater than t = 0 (the instant the device is switched from heating to measurement), the TSEP should indicate the device to be cooling. (In each dis play, the polarity is adjusted so that a rising voltage indicates cooling.) This is the case for V S G for all t > 0. It is not immediately obvious, though, how much of the decay is due to nonthermal switching transients and how much is due to the cooling. For the case for which V D B is the TSEP, it is obvious that for t < 30 us, nonthermal switching transients dominate the waveform. For these times, V D B indicates the device is heating, which is obviously incor rect. For t > 30 ps, V D B indicates the device is cooling, but how much of the effect is due to nonthermal switching transients is again not obvi ous from the oscilloscope trace.  MOSFET and is primarily limited by the magnitude of the gate capacitance with respect to the source, drain, and channel. On the other hand, the physical area of the drain-body diode is very large, being nearly equal to the total chip area. This means that the voltage dependent capacitance of the diode is intrinsically very large. In addition, the diode must be switched from a large reverse bias (smallest capacitance) to a forward bias (largest capacitance). Also, the carrier lifetime in the diode region is not closely con trolled and is quite long. The end result is that it takes a long time (t > 30 us) to fully switch the diode.
For the case of Rrjg(on), there is also a time during which the device appears to be heating and thus the TSEP is obviously dominated by nonthermal switching transients. Again, it is not obvious when the nonthermal transients totally disappear. The time to fully switch R DS (on) is somewhat longer than V S Q because the drain-body diode voltage is switched for the measurement.
There are techniques for determining the presence of and correcting for nonthermal switching tran sients. It has been demonstrated that for the first 250 Us after switching, the device cools as if the flow of heat were strictly one-dimensional (4). This means that |AT| a t 1/2 where AT is the temperature change from the instant switching occurred (t =0) until time, t. If the measured temperature is plotted versus /t, a straight line relationship occurs if the nonthermal switching transients have subsided. Examples of such a plot for the temperature measured using each TSEP is shown in figure 6. The figure indicates that the linear region of the temperature versus square root of time can be used to extrapolate the re sults to t s o to estimate the temperature at the instant of switching. This procedure was used to obtain the results reported in this work.

Magnetic Leads and the Skin Effect
The nature of the nonthermal switching transient was discussed above, but the physical cause was only briefly touched upon. A portion of the non thermal switching transients is related to stored charge and the time required to charge and dis charge device capacitances (both voltage dependent and voltage independent), to charge transit times, and to measurement circuit switching times. A significant and sometimes dominant portion of the nonthermal switching transients can result from the presence of the skin effect in magnetic leads (7). It is well known that high frequency alter nating currents tend to be forced to the surface of a conductor. The "effective" conductor resis tance due to this skin effect increases as the square root of the magnetic permeability of the conductor. Hermetic transistor packages typically have leads that are made of a material similar to kovar (~15 percent cobalt, 31 percent nickel, 54 percent iron) and are highly magnetic. When try ing to rapidly switch from heating to measurement conditions, the increased lead resistance due to the high frequency components of the switching waveform can contribute significant nonthermal switching transient components to the TSEP wave form.
A technique has been developed for correcting for this effect (7) using a "dummy" package to replace the DUT. The "dummy" consists of a package with the leads used in the measurement internally shorted to one another such that a measurement of the TSEP voltage only measures the voltage across the leads. An example for a bipolar transistor of the "uncorrected" temperature and the temperature measured using the correction technique discussed above is shown in figure 7 single measurement before 50 us is made or if extrapolation to t = 0 using measurements before 50 us is done unless compensation is made for the magnetic material skin effect.

Case Temperature Measurement Probe Location
The temperature at a specific location on the case of the transistor is maintained constant during the measurement of the device temperature. Be cause the temperature of the case is not uniform, it is important to be able to always measure the temperature at the same location on the case.

CONCLUSIONS
Each of the three TSEPs investigated appear to be suitable for measuring the temperature of a power MOSFET device. The simplest circuit is the one using V SG as the TSEP, which requires a single electronic switch, while the other two circuits require two electronic switches. In addition, the V SG circuit is configured identically to the standard circuit used for measuring RQ of a bipolar transistor with the bipolar transistor emitter-base-collector being replaced by the MOSFET source-gate-drain (8). The most stable calibration curve on the other hand is that for V DB which makes V DB perhaps more suited as a TSEP when a very large number of devices are to be measured. The most accurate TSEP, i.e., the one which indicates a temperature most near the peak temperature, is V SG . Also, V SG is typically the most temperature sensitive of the parameters, varying by about -6 mV/°C. Although R DS (on) is a suitable parameter for measuring temperature, there does not appear to be any par ticular circumstances for which it is superior to V SG or V DB-Infrared surface temperature profiles show the temperature of the power MOSFET to be relatively uniform, and the peak temperature as determined by the IRM is not a function of operating conditions; i.e., the peak temperature for a given power dis sipation does not depend upon the drain-currentdrain voltage combinations used to obtain that power. These results are in sharp contrast to what is usually observed for bipolar transistors.
Finally, a number of universal temperature mea surement difficulties have been discussed. Power MOSFET transistors are as susceptible to these difficulties as are power bipolar transistors.