Quantitative X-ray analysis provides production control
X-ray diffraction is an established non-destructive technique that enables the determination of layer parameters with high accuracy. Data on epilayer thickness and composition are required for SPC at the epitaxy stage of manufacturing. However, in practice X-ray diffraction analysis becomes qualitative for complex epitaxial structures because it relies on the comparison of fingerprints of "known good samples". While the analysis of these X-ray spectra is often complicated and time consuming, no calculations are done to determine specific layer parameters such as thickness and composition, which are necessary for SPC.
In order to gain quantitative information from InP-based HBT and HEMT epilayers, data from X-ray reflections from the 002 and 004 planes can be used to automatically calculate layer thickness and composition, providing valuable information for R&D and SPC in a production environment.
HBT structure and data model
At IPAG a quantitative X-ray analysis technique is being used to provide SPC data on InP-based HBTs and HEMTs grown by low-pressure MOVPE using a non-gaseous-source configuration (Velling, Velling et al.). The InP/InGaAs:C HBTs have a compositionally graded p-type InGaAs:C base and exhibit ft >120 GHz and fmax >190 GHz. For X-ray diffraction data analysis, the commercial RADS Mercury software tool developed by Bede is used (Bowen, Bede Scientific Instruments). A high-resolution X-ray diffractometer system with a double crystal monochromator setup is used for X-ray measurements in a coupled ω-2Θ mode (see Compound Semiconductor March 2001, p72).
Figure 1 shows the calculated X-ray rocking curve and measured curves for the 004 and 002 reflections. Figure 2 shows the intended InP/InGaAs:C HBT layer stack used by the model to create the calculated X-ray rocking curve. The excellent fit between the measured and calculated X-ray scans in figure 1 is achieved using the same layer model for both reflections. For the initial HBT layer model, a maximum variation in the composition and thickness of each layer of about ±10% around the intended values, is chosen to calculate the best fit to the X-ray spectra.
In order to take interface effects into account, two thin interlayers are included in the layer model between the InP emitter and the emitter cap above and the collector below. These interlayers enable the model to account for effects such as interface roughness and intermixing. It is important to reduce the number of free parameters in the model, as this makes the calculated curve parameters from a complex model more valid. To reduce the number of free parameters for the X-ray spectrum analysis, the HBT layer parameters are coupled and scaled using the growth time.
For example, a ternary layer like InxGa1-xAs is characterized by its thickness and composition. For InxGa1-xAs layers such as the sub-collector, collector and emitter cap, this makes up to six separate layer parameters that need to be determined. It can be assumed that for the HBT structure discussed here, the three layers exhibit the same composition and growth rate. Consequently, the six free parameters corresponding to the three ternary layers can easily be reduced to two free layer parameters by fixing the compositions to the same value and scaling the thickness of each layer to the growth time. For the discussed HBT layer stack, the nine layer parameters for the ternary sub-collector, collector, base and emitter cap layers (two parameters each), and the binary InP emitter layer (one parameter, thickness only), are reduced to five parameters that need to be calculated from the measured X-ray spectra.
From data to epilayer parameters
Nonetheless, the period of the resolved fringes must be associated with an individual HBT layer. The fringe periods correspond to the thickness of the emitter (tE) and the emitter cap (tE-Cap) layers, as shown in figure 1. The calculated values of composition and thickness agree with the measured spectra. The relation between layer thickness (tlayer) and the width of the fringe (Θ2-Θ1) is given by tlayer = λ/2 (sinΘ2-sinΘ1), where λ is the X-ray wavelength, which for Cu Kα X-rays is 0.154 nm.
The fringes in the 002 reflection are dominated by the InP emitter layer. The 002 reflection is quasi-forbidden for GaAs, and signal intensity is also strongly reduced in case of InGaAs. This is shown in figure 1, where the signal intensity corresponding to the emitter-cap, collector and sub-collector In0.532Ga0.468As layers between Θ = 0 and -200 arcseconds in the 004 reflection is strongly reduced in the 002 reflection. Consequently, the broad peak in the range of 0 < Θ < 1500 arcseconds, corresponding to the compositionally graded p-type InGaAs:C base, is suppressed in the 002 reflection.
Taking the 004 and 002 reflections into account in parallel enables a quantitative analysis of complex HBT layer stacks, while the fringe periods and signal intensities can be correlated to individual layers. This HBT layer analysis procedure has also been used at IPAG for InGaP/GaAs:C HBTs, InP-based resonant tunneling diodes (RTDs) and HBT/RTD combinations, resulting in very detailed material characterization. The quantitative X-ray analysis of two intentionally identical HBT layer stacks indicates run-to-run variations of composition and thickness of less than 1% and 0.5%, respectively. These run-to-run variations are equal to the homogeneity variations across a full 2 inch wafer.
As discussed for the HBT structure in figure 2, an X-ray peak at Θ ≅ -200 arcseconds in the 004 reflection is clearly resolved, and corresponds to approximately 0.2% deviation of the indium content in InxGa1-xAs from the composition required for lattice matching (x = 53%). In other words lattice strain corresponding to compositional variations of much less than 1% is clearly resolved.
A good fit between the 004 rocking curves is also achieved for the HEMT structures shown in figure 3. Samples A and B represent two nominally identical HEMT structures, comprising a highly strained In0.6Ga0.4As channel layer with an InP layer acting as a spacer, supply and Schottky barrier. The period of the fringes in the rocking curves from HEMTs A and B deviate from the calculated X-ray spectra by less than 10 arcseconds, and the intensity modulation between A, B and the calculated curve are in agreement. Detailed X-ray analysis of the HEMT structures reveals the same excellent homogeneity characteristics as in the HBT structure.
Fabricated HEMT devices with a gate length of 0.25 µm exhibit ft and fmax values of >100 GHz and a maximum transconductance of 550 mS/mm at Vgs = 0 V. This clearly indicates that the HEMTs are not just test structures for X-ray analysis, but are also well suited for the production of front-end receiver chips operating up to 43 Gbit/s and higher.
Figure 4 shows a very good fit between the measured and modeled common emitter output characteristics of fabricated HBT devices with a 20 µm2 emitter area. A DC gain of up to 35 is reached for SHBT devices, and the extrapolated ft and fmax values are 120 and 190 GHz, respectively, based on the as-measured s-parameters. The collector emitter breakdown voltage (BVceo) exceeds 7 V. These figures demonstrate the high-performance levels of the HBT and HEMT technologies that are the platform for 40 Gbit/s and higher optoelectronic and RF circuits in production and development at IPAG.
Since X-ray analysis revealed the excellent homogeneity of the epitaxial material, device homogeneity has also been investigated. While the thickness and compositional variations across the wafer are less than 1%, the variations of device parameters such as DC gain are less than 5%. This clearly demonstrates that the rapid and quantitative X-ray analysis described here is becoming an integral part of the technological platform for the production of high-speed optoelectronic and RF integrated circuits.
P Velling 2000 Progress of Crystal Growth and Characterization of Materials 41 85-131.
P Velling et al. 2000 Journal of Crystal Growth 221 722-729.
D Bowen & B Tanner 1998 High Resolution X-ray Diffractometry and Topography (Taylor and Francis).
Bede Scientific Instruments Rocking Curve Analysis by Dynamical Simulation (RADS, Handbook, RADS Mercury Software).