As complex compound semiconductor structures find an increasing variety of applications in the markets for optoelectronic, high frequency and high temperature devices, process control for wafer reproducibility and uniformity becomes critical. An example of a key optoelectronic device that will greatly benefit from improvements in epitaxial growth process control is the high-speed telecommunications photodiode. To detect light at the near infrared wavelengths most suitable for transmission of fiber-optic signals (1.31.6 m), the most widely used device is the positive-intrinsic-negative (PIN) diode based on a double heterojunction structure in which an In0.53Ga0.47As absorption region is sandwiched between two layers of InP. Enhanced performance can be obtained from an avalanche photodiode (APD) based on a structure similar to the PIN diode but with additional epilayers designed to create internal avalanche gain. These diodes have become essential elements of all fiber-optic receivers and signal monitors designed for telecommunication applications. They are also instrumental in the test and measurement equipment designed for signal detection in this wavelength range. The push for higher performance in many applications served by compound semiconductors makes it imperative that better epiwafer characterization tech niques are developed. For PINs and APDs, device performance is often critically dependent on wafer parameters such as layer thickness and doping. For instance, device bandwidths can be dictated by critical layer thicknesses, and more consistent device performance requires more consistent epitaxial growth. For the emerging generation of 10 Gb/s receiver products and next-generation 40 Gb/s devices, more challenging photodiode structures will mature in performance, yield, and throughput only if adequate wafer characterization tools are available. Until now, the main characterization methods for measuring thickness and dopant concentrations of film stacks such as PINs or APDs were all destructive: thickness profiles were usually obtained by destructive chemical methods, and dopant concentrations were obtained using destructive CV. Moreover, these tests had to be performed on calibration samples or simultaneously grown witness samples with no option of carrying them out on the shippable wafers themselves. While visible ellipsometry and X-ray reflectometry are used to non-destructively characterize thin layers, the penetration depth at these wavelengths is usually not sufficient to probe thick (microns) layers, and measurements can be very slow. A new infrared metrology for thin film analysis was recently developed to perform fast optical measurements of thickness, free carrier concentration and index of refraction of multilayer compound semiconductor structures such as PINs and APDs. New Infrared Optical Technology The new technology employs state-of-the-art Fourier Transform Infrared (FTIR) reflection spectroscopy combined with advanced model-based analysis, and has been widely used for the characterization of epitaxial silicon [1]. The basis of the FTIR measurement lies in the different index of refraction (or complex dielectric function DF) in the infrared of the various films and the substrate. The index of refraction depends on material composition (e.g. InP, InGaAs) but also on free carrier concentration, as free carriers absorb infrared light. As a probing infrared beam interacts with stacked films with different dielectric functions, the reflected beam displays interference effects from the multiple reflections at the various film interfaces (see Figure 1). The spectrum obtained depends on the thickness and dielectric function of each film as well as of the substrate. Advanced Analysis Algorithms One innovation of the new FTIR metrology is to use powerful multi-layer, multi-parameter analysis algorithms [2] (similar to those typically used in visible ellipsometers) to extract parameters related to the films. This approach differs from "traditional" FTIR measurements, where simpler analysis models of the reflectance or absorbance spectra are used and cannot be applied to complex film stacks. Together with the multi-layer analysis algorithms, an automated solution routine simultaneously solves for the thickness, free carrier concentration and refractive index of the various layers and substrate by varying the model parameters to obtain a match between measured and theoretical reflectance spectra. The whole measurement, including the FTIR spectrum measurement and analysis, typically takes a few seconds per data point. New Optics Another key innovation of the new FTIR metrology is the use of special optics which allow the measurement of front surface reflections only, and eliminate the reflection from the backside of the wafer. This feature greatly simplifies the process of modeling of samples with transparent substrates and not well-controlled back surfaces (such as with glue residues) that would otherwise add artifacts to the measured spectrum. As a consequence, the measurements can be performed without requiring any special surface or backside treatments. The new optics also provide a very small measurement spot (200x800 m on the wafer, with an option for 50x50 m), which is significantly smaller than the typical FTIR s spot size of around 5 mm diameter. The small spot allows measurements very close to the wafer edge, which is important when trying to optimize the wafer "real-estate". Example of PIN analysis A multilayer PIN sample of InP/InGaAs/ InP/InP was measured and analyzed. shows the details of the film stack. The analysis was performed to extract the following parameters: cap InP thickness, InGaAs thickness, buffer InP thickness and free carrier concentration, and InP substrate free carrier concentration. A Drude model was used to represent the free carrier s absorption [3]. Literature spectra for the refractive index n and the extinction coefficient k were used for InP and the InGaAs alloy (note that while the overall shape of the refractive index spectrum does not usually vary significantly with alloy composition, n can however be adjusted linearly if found necessary to account for small differences in composition). Excellent fits were obtained, validating the film stack and dielectric function models used (see ). shows the extracted parameters, which were in excellent agreement with the nominal values. A 3-D thickness map of the InGaAs layer is shown in Figure 3, displaying a fairly radial distribution of thickness across the PIN wafer. Similar thickness maps were obtained for the other layers. Such information is extremely useful to improve the tuning of deposition chambers, ultimately increasing yields across the wafer. These data can also be of great value for optimizing subsequent processes carried out on the wafers, such as the timing of dopant diffusions to create p-n junctions. Low Inter-Parameter Correlation When several fitting parameters are used, correlations between the parameters should be estimated to assess the robustness of the measurements. We investigated the influence of each parameter on the reflectance spectrum for a film stack similar to that of the PIN sample described above. It was found that each parameter has a different influence on the spectrum, which translates into a minimum correlation. Quantitatively, it was determined that film thicknesses had uncertainties from 1 to 5 nm (1s), and free carrier concentrations from 0.11018 to 0.21018 cm-3 (1s). The repeatability for thickness was about 1 nm (1s) and 0.11018 cm-3 (1s) for free carrier concentration. These results are typical for PINs, APDs and similar structures. Correlation with Independent Measurements The accuracy of the FTIR measurements has been tested versus destructive measurements (SIMS, CV, etc.). Thicknesses as measured by SIMS typically agree closely with the FTIR results to better than 5%. An excellent correlation is also obtained between free carrier concentration as measured by the FTIR and CV doping measurements (see ). Detection Limits The detection limits for film thickness depend on the specific film stack to be analyzed. For example, if a thin layer (e.g. 20 nm) is next to a thick layer (microns) of significantly different refractive index, there will be an internal reflection at the interface, which will clearly impact the reflectance spectrum. In such a case, the thicknesses of both layers can be extracted with accuracy. On the other hand, multiple layers which are very thin (