Thermometry for semiconductor processing
Thermal processing of semiconductors is a critical, capital intensive step in achieving high yields and profitability in the manufacturing of electronic chips. Many techniques have been developed to control the temperature of the silicon wafer during thermal processing. One way is to monitor the wafer surface temperature using sensors on instrumented test wafers. At NIST, we have supported metrology using instrumented wafers by a) developing thin-film platinum resistance thermometers deposited directly on the wafer, for use in the range 200 °C to 600 °C, and b) developing a non-contact method for the determination of the thermal response time of temperature sensors embedded in semiconductor wafers. The method determines sensor errors when the wafer temperature is changing.
Improvements in temperature measurements of semiconductor wafers will lead to increased efficiency and chip yield for semiconductor manufacturers. We also anticipate that the time-response method will have use in development of new sensor attachment methods, in verifying sensor attachment during production, and in confirming integrity of the attachment after use.
To develop thin-film resistance thermometers on silicon that have an expanded uncertainty of better than 2 °C, and a non-contact method for the determination of the thermal response time of temperature sensors embedded in semiconductor wafers with an uncertainty of less than 20 %.
Fabrication of thin-film resistors
We sputtered Pt thin-film resistors on silicon-wafer and alumina test coupons, using titanium or zirconium as a bond coat to promote adhesion on the oxidized silicon wafers. Calibrated thermocouples determined the film's temperature. The test coupons were cycled between room temperature and up to 700 °C in a tube furnace to determine the thermal coefficient of resistance and the hysteresis.
The tests indicated that the R versus t data could be fitted with a quadratic function similar to the functions described in IEC 60751-1983. The temperature coefficient of resistivity (α) of the films up to 1 μm thick is approximately 15% lower than the industrial specification for bulk platinum wire resistors and decreases with lower thickness. This is likely to be related to a higher film resistivity from impurities, point defects, and surface effects.
Significant hysteresis is present with fast (5 °C/min.) heating or cooling. We hypothesize that much of this hysteresis is related to plastic deformation forced on the Pt film by the Si wafer because of the mismatch in thermal expansion coefficients. Slower rates of heating or cooling (2 °C / min) probably permit recovery and lower stress levels in the Pt thin films. Under these conditions, we obtained an expanded uncertainty (k = 2) of 2 °C, including both thermometer hysteresis and calibration uncertainty.
Sensor response time measurements
In our method, a flash lamp illuminates a spot on the wafer on the opposite side from the sensor under test with a series of periodic flashes. For each flash, the temporal response of the sensor is measured. To determine the response time of the sensor, we combine thermal modeling of the temperature response of the wafer with experimental measurements of the phase lag between the sensor response and the periodic flashes. Experimental data on both PRTs and on thermocouples embedded in wafers show good agreement with the models. To simplify the application of the method, we have produced a table of calculated wafer phases to be used with the measured sensor phase for determining the response time.