Effective statistical process control for HBTs and PHEMTs

Deciding what to measure

At an early, low-volume production stage in a product s life cycle, engineering input may lead to the definition of key parameters or control points. As the product matures and experience is gained, these can either be enhanced or removed depending on the identified sources of variation. For the epitaxial growth of HBTs and PHEMTs, there is the added complication that not all of the key device parameters can be readily assessed at the final epiwafer level. Calibration growths such as capless PHEMTs are required when the desired characteristic to be controlled is either masked or unreliably measured on the final product wafers. Many measurement techniques are destructive in nature, making MSA a difficult and costly process. Ideally, in situ control can be used to correct and adapt for thickness or alloy drifts. However, at present few optical or electrical properties can be directly assessed in situ.

MOCVD production of PHEMTs makes extensive use of the (004) reflection from X-ray diffraction (XRD) for epitaxial layer assessment. The recent availability of comprehensive fitting software has led to a trend of extracting and reporting all layer thicknesses and alloy mole fractions. However, care must be taken to separate the parameters that can be extracted reliably and thus employed for SPC. The graph in figure 1 shows a representative diffraction spectrum for an MOCVD-grown InGaAs/AlGaAs PHEMT. The lattice distortion in the strained InGaAs layer generates a clear signature that is sensitive to both alloy mole fraction and thickness.

Reproducibility and repeatability

To assess the effectiveness of a measurement system and thus how appropriate its use is for an SPC system, gauge reproducibility and repeatability (R&R) data is required. Reproducibility is a measure of the variability between operators measuring the same parameter. The repeatability is the variability of a single operator measuring a particular parameter a number of times. For SPC to be meaningful, measurement errors must be low, otherwise a value could be solely due to a measurement error, and not a reflection of the process itself.

For a measurement to be effectively employed in SPC, the R&R values should be less than 10% of the allowable tolerance. R&R values greater than 30% make the gauge useless for SPC. Between 10 and 30%, the gauge can be employed for SPC with additional precautions such as averaging.

Table 1 shows the R&R results obtained as part of an MSA for the In composition and layer thickness of an InGaAs PHEMT channel. In line with expectations, both of the parameters are statistically robust, and so the majority of the variability is attributable to the grown material rather than to the measurement technique.

Conversely, extracting the relative thicknesses of the AlGaAs and GaAs layers above the channel is not as robust. The limited strain associated with AlGaAs, and the resolution of a (004) XRD measurement, makes changes in either layer difficult to detect directly. In this case, applying SPC would simply result in noise and numerous false alerts. A far better parameter to track is the total thickness above the channel (AlGaAs plus GaAs), since this is more clearly evident in the fringe patterns of the X-ray spectra. Excursions of the additive growth rate imply a shift of one or more of the growth rates and/ or the alloy mole fraction.

Process capability

Figure 2 shows the PHEMT InGaAs channel thickness (a) and the In composition (b), tracked over a typical growth campaign. The data are taken from 150 mm AlGaAs/InGaAs PHEMT wafers grown in an Emcore E400 reactor using a 5 x 150 mm wafer configuration. One wafer from each growth run is sampled. The data is normalized to a thickness of 150 Å and an In composition of 15%. The plots show the variation in the two parameters between the largest acceptable upper and lower limits. The limit for the In composition is ±1%, and ±15 Å for the InGaAs thickness.

Process capability, or Cpk, measures the relationship between the distribution of measured values and the upper and lower specification limits. A high Cpk value means that the process will yield a large number of good parts. Cpk values are often used as a gate for ramping production. It is generally accepted that a Cpk of more than 1.3 is production capable, corresponding to a predicted reject rate of approximately 6000 ppm. Calculated Cpk values from figure 2 are 1.3 for the thickness and 2.2 for the In mole fraction.

Non-destructive eddy current sheet resistance (Rsh) measurements are a commonly used method for monitoring the electrical properties of epitaxial layers. The speed, simplicity and mapping capability of the eddy current method makes it very attractive for production use. This is born out by gauge R&R data, as in table 2, which shows good gauge capability for both mean Rsh and the cross-wafer uniformity of Rsh in the 40-60 Ω/ sq range of a typical PHEMT.

Figure 3 shows the SPC data of Rsh for a PHEMT growth campaign. The total population of wafers in the campaign was 150 and the target final Rsh was 55 ±3 Ω/sq. The cap (contact layer) of the PHEMT dominates the measurement, but the data shows minimal variation in Rsh, reflecting the combined thickness and doping control inherent in the growth process. The calculated Cpk is greater than 2.5, despite the offset from the target value of 55 Ω/sq. The minimal variation in Rsh enables the mean to be slightly off the target value and still not greatly impact the Cpk value.